diff --git a/daemons/attrd/attrd_utils.c b/daemons/attrd/attrd_utils.c index 1ea2dccb2b..6f23306b78 100644 --- a/daemons/attrd/attrd_utils.c +++ b/daemons/attrd/attrd_utils.c @@ -1,325 +1,325 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include "pacemaker-attrd.h" cib_t *the_cib = NULL; static bool requesting_shutdown = false; static bool shutting_down = false; static GMainLoop *mloop = NULL; /* A hash table storing information on the protocol version of each peer attrd. * The key is the peer's uname, and the value is the protocol version number. */ GHashTable *peer_protocol_vers = NULL; /*! * \internal * \brief Set requesting_shutdown state */ void attrd_set_requesting_shutdown(void) { requesting_shutdown = true; } /*! * \internal * \brief Clear requesting_shutdown state */ void attrd_clear_requesting_shutdown(void) { requesting_shutdown = false; } /*! * \internal * \brief Check whether local attribute manager is shutting down * * \param[in] if_requested If \c true, also consider presence of * \c PCMK__NODE_ATTR_SHUTDOWN attribute * * \return \c true if local attribute manager has begun shutdown sequence * or (if \p if_requested is \c true) whether local node has a nonzero * \c PCMK__NODE_ATTR_SHUTDOWN attribute set, otherwise \c false * \note Most callers should pass \c false for \p if_requested, because the * attribute manager needs to continue performing while the controller is * shutting down, and even needs to be eligible for election in case all * nodes are shutting down. */ bool attrd_shutting_down(bool if_requested) { return shutting_down || (if_requested && requesting_shutdown); } /*! * \internal * \brief Exit (using mainloop or not, as appropriate) * * \param[in] nsig Ignored */ void attrd_shutdown(int nsig) { // Tell various functions not to do anthing shutting_down = true; // Don't respond to signals while shutting down mainloop_destroy_signal(SIGTERM); mainloop_destroy_signal(SIGCHLD); mainloop_destroy_signal(SIGPIPE); mainloop_destroy_signal(SIGUSR1); mainloop_destroy_signal(SIGUSR2); mainloop_destroy_signal(SIGTRAP); attrd_free_waitlist(); attrd_free_confirmations(); if (peer_protocol_vers != NULL) { g_hash_table_destroy(peer_protocol_vers); peer_protocol_vers = NULL; } if ((mloop == NULL) || !g_main_loop_is_running(mloop)) { /* If there's no main loop active, just exit. This should be possible * only if we get SIGTERM in brief windows at start-up and shutdown. */ crm_exit(CRM_EX_OK); } else { g_main_loop_quit(mloop); g_main_loop_unref(mloop); } } /*! * \internal * \brief Create a main loop for attrd */ void attrd_init_mainloop(void) { mloop = g_main_loop_new(NULL, FALSE); } /*! * \internal * \brief Run attrd main loop */ void attrd_run_mainloop(void) { g_main_loop_run(mloop); } /* strlen("value") */ #define plus_plus_len (5) /*! * \internal * \brief Check whether an attribute value should be expanded * * \param[in] value Attribute value to check * * \return true if value needs expansion, false otherwise */ bool attrd_value_needs_expansion(const char *value) { return ((strlen(value) >= (plus_plus_len + 2)) && (value[plus_plus_len] == '+') && ((value[plus_plus_len + 1] == '+') || (value[plus_plus_len + 1] == '='))); } /*! * \internal * \brief Expand an increment expression into an integer * * \param[in] value Attribute increment expression to expand * \param[in] old_value Previous value of attribute * * \return Expanded value */ int attrd_expand_value(const char *value, const char *old_value) { int offset = 1; int int_value = char2score(old_value); if (value[plus_plus_len + 1] != '+') { const char *offset_s = value + (plus_plus_len + 2); offset = char2score(offset_s); } int_value += offset; - if (int_value > INFINITY) { - int_value = INFINITY; + if (int_value > PCMK_SCORE_INFINITY) { + int_value = PCMK_SCORE_INFINITY; } return int_value; } /*! * \internal * \brief Create regular expression matching failure-related attributes * * \param[out] regex Where to store created regular expression * \param[in] rsc Name of resource to clear (or NULL for all) * \param[in] op Operation to clear if rsc is specified (or NULL for all) * \param[in] interval_ms Interval of operation to clear if op is specified * * \return pcmk_ok on success, -EINVAL if arguments are invalid * * \note The caller is responsible for freeing the result with regfree(). */ int attrd_failure_regex(regex_t *regex, const char *rsc, const char *op, guint interval_ms) { char *pattern = NULL; int rc; /* Create a pattern that matches desired attributes */ if (rsc == NULL) { pattern = strdup(ATTRD_RE_CLEAR_ALL); } else if (op == NULL) { pattern = crm_strdup_printf(ATTRD_RE_CLEAR_ONE, rsc); } else { pattern = crm_strdup_printf(ATTRD_RE_CLEAR_OP, rsc, op, interval_ms); } /* Compile pattern into regular expression */ crm_trace("Clearing attributes matching %s", pattern); rc = regcomp(regex, pattern, REG_EXTENDED|REG_NOSUB); free(pattern); return (rc == 0)? pcmk_ok : -EINVAL; } void attrd_free_attribute_value(gpointer data) { attribute_value_t *v = data; free(v->nodename); free(v->current); free(v->requested); free(v); } void attrd_free_attribute(gpointer data) { attribute_t *a = data; if(a) { free(a->id); free(a->set_id); free(a->set_type); free(a->user); mainloop_timer_del(a->timer); g_hash_table_destroy(a->values); free(a); } } /*! * \internal * \brief When a peer node leaves the cluster, stop tracking its protocol version. * * \param[in] host The peer node's uname to be removed */ void attrd_remove_peer_protocol_ver(const char *host) { if (peer_protocol_vers != NULL) { g_hash_table_remove(peer_protocol_vers, host); } } /*! * \internal * \brief When a peer node broadcasts a message with its protocol version, keep * track of that information. * * We keep track of each peer's protocol version so we know which peers to * expect confirmation messages from when handling cluster-wide sync points. * We additionally keep track of the lowest protocol version supported by all * peers so we know when we can send IPC messages containing more than one * request. * * \param[in] host The peer node's uname to be tracked * \param[in] value The peer node's protocol version */ void attrd_update_minimum_protocol_ver(const char *host, const char *value) { int ver; if (peer_protocol_vers == NULL) { peer_protocol_vers = pcmk__strkey_table(free, NULL); } pcmk__scan_min_int(value, &ver, 0); if (ver > 0) { char *host_name = strdup(host); /* Record the peer attrd's protocol version. */ CRM_ASSERT(host_name != NULL); g_hash_table_insert(peer_protocol_vers, host_name, GINT_TO_POINTER(ver)); /* If the protocol version is a new minimum, record it as such. */ if (minimum_protocol_version == -1 || ver < minimum_protocol_version) { minimum_protocol_version = ver; crm_trace("Set minimum attrd protocol version to %d", minimum_protocol_version); } } } void attrd_copy_xml_attributes(xmlNode *src, xmlNode *dest) { /* Copy attributes from the wrapper parent node into the child node. * We can't just use copy_in_properties because we want to skip any * attributes that are already set on the child. For instance, if * we were told to use a specific node, there will already be a node * attribute on the child. Copying the parent's node attribute over * could result in the wrong value. */ for (xmlAttrPtr a = pcmk__xe_first_attr(src); a != NULL; a = a->next) { const char *p_name = (const char *) a->name; const char *p_value = ((a == NULL) || (a->children == NULL)) ? NULL : (const char *) a->children->content; if (crm_element_value(dest, p_name) == NULL) { crm_xml_add(dest, p_name, p_value); } } } diff --git a/daemons/controld/controld_callbacks.c b/daemons/controld/controld_callbacks.c index fdd44c555b..81f5b13137 100644 --- a/daemons/controld/controld_callbacks.c +++ b/daemons/controld/controld_callbacks.c @@ -1,381 +1,381 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include /* From join_dc... */ extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); void crmd_ha_msg_filter(xmlNode * msg) { if (AM_I_DC) { const char *sys_from = crm_element_value(msg, PCMK__XA_CRM_SYS_FROM); if (pcmk__str_eq(sys_from, CRM_SYSTEM_DC, pcmk__str_casei)) { const char *from = crm_element_value(msg, PCMK__XA_SRC); if (!pcmk__str_eq(from, controld_globals.our_nodename, pcmk__str_casei)) { int level = LOG_INFO; const char *op = crm_element_value(msg, PCMK__XA_CRM_TASK); /* make sure the election happens NOW */ if (controld_globals.fsa_state != S_ELECTION) { ha_msg_input_t new_input; level = LOG_WARNING; new_input.msg = msg; register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, &new_input, __func__); } do_crm_log(level, "Another DC detected: %s (op=%s)", from, op); goto done; } } } else { const char *sys_to = crm_element_value(msg, PCMK__XA_CRM_SYS_TO); if (pcmk__str_eq(sys_to, CRM_SYSTEM_DC, pcmk__str_casei)) { return; } } /* crm_log_xml_trace(msg, "HA[inbound]"); */ route_message(C_HA_MESSAGE, msg); done: controld_trigger_fsa(); } /*! * \internal * \brief Check whether a node is online * * \param[in] node Node to check * * \retval -1 if completely dead * \retval 0 if partially alive * \retval 1 if completely alive */ static int node_alive(const crm_node_t *node) { if (pcmk_is_set(node->flags, crm_remote_node)) { // Pacemaker Remote nodes can't be partially alive return pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei) ? 1: -1; } else if (crm_is_peer_active(node)) { // Completely up cluster node: both cluster member and peer return 1; } else if (!pcmk_is_set(node->processes, crm_get_cluster_proc()) && !pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei)) { // Completely down cluster node: neither cluster member nor peer return -1; } // Partially up cluster node: only cluster member or only peer return 0; } #define state_text(state) ((state)? (const char *)(state) : "in unknown state") void peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data) { uint32_t old = 0; bool appeared = FALSE; bool is_remote = pcmk_is_set(node->flags, crm_remote_node); controld_node_pending_timer(node); /* The controller waits to receive some information from the membership * layer before declaring itself operational. If this is being called for a * cluster node, indicate that we have it. */ if (!is_remote) { controld_set_fsa_input_flags(R_PEER_DATA); } if (type == crm_status_processes && pcmk_is_set(node->processes, crm_get_cluster_proc()) && !AM_I_DC && !is_remote) { /* * This is a hack until we can send to a nodeid and/or we fix node name lookups * These messages are ignored in crmd_ha_msg_filter() */ xmlNode *query = create_request(CRM_OP_HELLO, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); crm_debug("Sending hello to node %u so that it learns our node name", node->id); send_cluster_message(node, crm_msg_crmd, query, FALSE); free_xml(query); } if (node->uname == NULL) { return; } switch (type) { case crm_status_uname: /* If we've never seen the node, then it also won't be in the status section */ crm_info("%s node %s is now %s", (is_remote? "Remote" : "Cluster"), node->uname, state_text(node->state)); return; case crm_status_nstate: /* This callback should not be called unless the state actually * changed, but here's a failsafe just in case. */ CRM_CHECK(!pcmk__str_eq(data, node->state, pcmk__str_casei), return); crm_info("%s node %s is now %s (was %s)", (is_remote? "Remote" : "Cluster"), node->uname, state_text(node->state), state_text(data)); if (pcmk__str_eq(CRM_NODE_MEMBER, node->state, pcmk__str_casei)) { appeared = TRUE; if (!is_remote) { remove_stonith_cleanup(node->uname); } } else { controld_remove_failed_sync_node(node->uname); controld_remove_voter(node->uname); } crmd_alert_node_event(node); break; case crm_status_processes: CRM_CHECK(data != NULL, return); old = *(const uint32_t *)data; appeared = pcmk_is_set(node->processes, crm_get_cluster_proc()); { const char *dc_s = controld_globals.dc_name; if ((dc_s == NULL) && AM_I_DC) { dc_s = PCMK_VALUE_TRUE; } crm_info("Node %s is %s a peer " CRM_XS " DC=%s old=%#07x new=%#07x", node->uname, (appeared? "now" : "no longer"), pcmk__s(dc_s, ""), old, node->processes); } if (!pcmk_is_set((node->processes ^ old), crm_get_cluster_proc())) { /* Peer status did not change. This should not be possible, * since we don't track process flags other than peer status. */ crm_trace("Process flag %#7x did not change from %#7x to %#7x", crm_get_cluster_proc(), old, node->processes); return; } if (!appeared) { node->peer_lost = time(NULL); controld_remove_failed_sync_node(node->uname); controld_remove_voter(node->uname); } if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { crm_trace("Ignoring peer status change because not connected to CIB"); return; } else if (controld_globals.fsa_state == S_STOPPING) { crm_trace("Ignoring peer status change because stopping"); return; } if (!appeared && pcmk__str_eq(node->uname, controld_globals.our_nodename, pcmk__str_casei)) { /* Did we get evicted? */ crm_notice("Our peer connection failed"); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL); } else if (pcmk__str_eq(node->uname, controld_globals.dc_name, pcmk__str_casei) && !crm_is_peer_active(node)) { /* Did the DC leave us? */ crm_notice("Our peer on the DC (%s) is dead", controld_globals.dc_name); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL); /* @COMPAT DC < 1.1.13: If a DC shuts down normally, we don't * want to fence it. Newer DCs will send their shutdown request * to all peers, who will update the DC's expected state to * down, thus avoiding fencing. We can safely erase the DC's * transient attributes when it leaves in that case. However, * the only way to avoid fencing older DCs is to leave the * transient attributes intact until it rejoins. */ if (compare_version(controld_globals.dc_version, "3.0.9") > 0) { controld_delete_node_state(node->uname, controld_section_attrs, cib_scope_local); } } else if (AM_I_DC || pcmk_is_set(controld_globals.flags, controld_dc_left) || (controld_globals.dc_name == NULL)) { /* This only needs to be done once, so normally the DC should do * it. However if there is no DC, every node must do it, since * there is no other way to ensure some one node does it. */ if (appeared) { te_trigger_stonith_history_sync(FALSE); } else { controld_delete_node_state(node->uname, controld_section_attrs, cib_scope_local); } } break; } if (AM_I_DC) { xmlNode *update = NULL; int flags = node_update_peer; int alive = node_alive(node); pcmk__graph_action_t *down = match_down_event(node->uuid); crm_trace("Alive=%d, appeared=%d, down=%d", alive, appeared, (down? down->id : -1)); if (appeared && (alive > 0) && !is_remote) { register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } if (down) { const char *task = crm_element_value(down->xml, PCMK_XA_OPERATION); if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) { /* tengine_stonith_callback() confirms fence actions */ crm_trace("Updating CIB %s fencer reported fencing of %s complete", (pcmk_is_set(down->flags, pcmk__graph_action_confirmed)? "after" : "before"), node->uname); } else if (!appeared && pcmk__str_eq(task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_casei)) { // Shutdown actions are immediately confirmed (i.e. no_wait) if (!is_remote) { flags |= node_update_join | node_update_expected; crmd_peer_down(node, FALSE); check_join_state(controld_globals.fsa_state, __func__); } if (alive >= 0) { crm_info("%s of peer %s is in progress " CRM_XS " action=%d", task, node->uname, down->id); } else { crm_notice("%s of peer %s is complete " CRM_XS " action=%d", task, node->uname, down->id); pcmk__update_graph(controld_globals.transition_graph, down); trigger_graph(); } } else { crm_trace("Node %s is %s, was expected to %s (op %d)", node->uname, ((alive > 0)? "alive" : ((alive < 0)? "dead" : "partially alive")), task, down->id); } } else if (appeared == FALSE) { if ((controld_globals.transition_graph == NULL) || (controld_globals.transition_graph->id == -1)) { crm_info("Stonith/shutdown of node %s is unknown to the " "current DC", node->uname); } else { crm_warn("Stonith/shutdown of node %s was not expected", node->uname); } if (!is_remote) { crm_update_peer_join(__func__, node, crm_join_none); check_join_state(controld_globals.fsa_state, __func__); } - abort_transition(INFINITY, pcmk__graph_restart, "Node failure", - NULL); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Node failure", NULL); fail_incompletable_actions(controld_globals.transition_graph, node->uuid); } else { crm_trace("Node %s came up, was not expected to be down", node->uname); } if (is_remote) { /* A pacemaker_remote node won't have its cluster status updated * in the CIB by membership-layer callbacks, so do it here. */ flags |= node_update_cluster; /* Trigger resource placement on newly integrated nodes */ if (appeared) { - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Pacemaker Remote node integrated", NULL); } } if (!appeared && (type == crm_status_processes) && (node->when_member > 1)) { /* The node left CPG but is still a cluster member. Set its * membership time to 1 to record it in the cluster state as a * boolean, so we don't fence it due to * PCMK_OPT_NODE_PENDING_TIMEOUT. */ node->when_member = 1; flags |= node_update_cluster; controld_node_pending_timer(node); } /* Update the CIB node state */ update = create_node_state_update(node, flags, NULL, __func__); if (update == NULL) { crm_debug("Node state update not yet possible for %s", node->uname); } else { fsa_cib_anon_update(PCMK_XE_STATUS, update); } free_xml(update); } controld_trigger_fsa(); } gboolean crm_fsa_trigger(gpointer user_data) { crm_trace("Invoked (queue len: %d)", g_list_length(controld_globals.fsa_message_queue)); s_crmd_fsa(C_FSA_INTERNAL); crm_trace("Exited (queue len: %d)", g_list_length(controld_globals.fsa_message_queue)); return TRUE; } diff --git a/daemons/controld/controld_fencing.c b/daemons/controld/controld_fencing.c index e27ebdb236..6b743e28f6 100644 --- a/daemons/controld/controld_fencing.c +++ b/daemons/controld/controld_fencing.c @@ -1,1116 +1,1117 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include static void tengine_stonith_history_synced(stonith_t *st, stonith_event_t *st_event); /* * stonith failure counting * * We don't want to get stuck in a permanent fencing loop. Keep track of the * number of fencing failures for each target node, and the most we'll restart a * transition for. */ struct st_fail_rec { int count; }; static bool fence_reaction_panic = false; static unsigned long int stonith_max_attempts = 10; static GHashTable *stonith_failures = NULL; /*! * \internal * \brief Update max fencing attempts before giving up * * \param[in] value New max fencing attempts */ static void update_stonith_max_attempts(const char *value) { stonith_max_attempts = char2score(value); if (stonith_max_attempts < 1UL) { stonith_max_attempts = 10UL; } } /*! * \internal * \brief Configure reaction to notification of local node being fenced * * \param[in] reaction_s Reaction type */ static void set_fence_reaction(const char *reaction_s) { if (pcmk__str_eq(reaction_s, "panic", pcmk__str_casei)) { fence_reaction_panic = true; } else { if (!pcmk__str_eq(reaction_s, PCMK_VALUE_STOP, pcmk__str_casei)) { crm_warn("Invalid value '%s' for %s, using 'stop'", reaction_s, PCMK_OPT_FENCE_REACTION); } fence_reaction_panic = false; } } /*! * \internal * \brief Configure fencing options based on the CIB * * \param[in,out] options Name/value pairs for configured options */ void controld_configure_fencing(GHashTable *options) { const char *value = NULL; value = g_hash_table_lookup(options, PCMK_OPT_FENCE_REACTION); set_fence_reaction(value); value = g_hash_table_lookup(options, PCMK_OPT_STONITH_MAX_ATTEMPTS); update_stonith_max_attempts(value); } static gboolean too_many_st_failures(const char *target) { GHashTableIter iter; const char *key = NULL; struct st_fail_rec *value = NULL; if (stonith_failures == NULL) { return FALSE; } if (target == NULL) { g_hash_table_iter_init(&iter, stonith_failures); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { if (value->count >= stonith_max_attempts) { target = (const char*)key; goto too_many; } } } else { value = g_hash_table_lookup(stonith_failures, target); if ((value != NULL) && (value->count >= stonith_max_attempts)) { goto too_many; } } return FALSE; too_many: crm_warn("Too many failures (%d) to fence %s, giving up", value->count, target); return TRUE; } /*! * \internal * \brief Reset a stonith fail count * * \param[in] target Name of node to reset, or NULL for all */ void st_fail_count_reset(const char *target) { if (stonith_failures == NULL) { return; } if (target) { struct st_fail_rec *rec = NULL; rec = g_hash_table_lookup(stonith_failures, target); if (rec) { rec->count = 0; } } else { GHashTableIter iter; const char *key = NULL; struct st_fail_rec *rec = NULL; g_hash_table_iter_init(&iter, stonith_failures); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &rec)) { rec->count = 0; } } } static void st_fail_count_increment(const char *target) { struct st_fail_rec *rec = NULL; if (stonith_failures == NULL) { stonith_failures = pcmk__strkey_table(free, free); } rec = g_hash_table_lookup(stonith_failures, target); if (rec) { rec->count++; } else { rec = malloc(sizeof(struct st_fail_rec)); if(rec == NULL) { return; } rec->count = 1; g_hash_table_insert(stonith_failures, strdup(target), rec); } } /* end stonith fail count functions */ static void cib_fencing_updated(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { if (rc < pcmk_ok) { crm_err("Fencing update %d for %s: failed - %s (%d)", call_id, (char *)user_data, pcmk_strerror(rc), rc); crm_log_xml_warn(msg, "Failed update"); - abort_transition(INFINITY, pcmk__graph_shutdown, "CIB update failed", - NULL); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_shutdown, + "CIB update failed", NULL); } else { crm_info("Fencing update %d for %s: complete", call_id, (char *)user_data); } } static void send_stonith_update(pcmk__graph_action_t *action, const char *target, const char *uuid) { int rc = pcmk_ok; crm_node_t *peer = NULL; /* We (usually) rely on the membership layer to do node_update_cluster, * and the peer status callback to do node_update_peer, because the node * might have already rejoined before we get the stonith result here. */ int flags = node_update_join | node_update_expected; /* zero out the node-status & remove all LRM status info */ xmlNode *node_state = NULL; CRM_CHECK(target != NULL, return); CRM_CHECK(uuid != NULL, return); /* Make sure the membership and join caches are accurate. * Try getting any existing node cache entry also by node uuid in case it * doesn't have an uname yet. */ peer = pcmk__get_node(0, target, uuid, pcmk__node_search_any); CRM_CHECK(peer != NULL, return); if (peer->state == NULL) { /* Usually, we rely on the membership layer to update the cluster state * in the CIB. However, if the node has never been seen, do it here, so * the node is not considered unclean. */ flags |= node_update_cluster; } if (peer->uuid == NULL) { crm_info("Recording uuid '%s' for node '%s'", uuid, target); peer->uuid = strdup(uuid); } crmd_peer_down(peer, TRUE); /* Generate a node state update for the CIB */ node_state = create_node_state_update(peer, flags, NULL, __func__); /* we have to mark whether or not remote nodes have already been fenced */ if (peer->flags & crm_remote_node) { char *now_s = pcmk__ttoa(time(NULL)); crm_xml_add(node_state, PCMK__XA_NODE_FENCED, now_s); free(now_s); } /* Force our known ID */ crm_xml_add(node_state, PCMK_XA_ID, uuid); rc = controld_globals.cib_conn->cmds->modify(controld_globals.cib_conn, PCMK_XE_STATUS, node_state, cib_scope_local |cib_can_create); /* Delay processing the trigger until the update completes */ crm_debug("Sending fencing update %d for %s", rc, target); fsa_register_cib_callback(rc, strdup(target), cib_fencing_updated); // Make sure it sticks /* controld_globals.cib_conn->cmds->bump_epoch(controld_globals.cib_conn, * cib_scope_local); */ controld_delete_node_state(peer->uname, controld_section_all, cib_scope_local); free_xml(node_state); return; } /*! * \internal * \brief Abort transition due to stonith failure * * \param[in] abort_action Whether to restart or stop transition * \param[in] target Don't restart if this (NULL for any) has too many failures * \param[in] reason Log this stonith action XML as abort reason (or NULL) */ static void abort_for_stonith_failure(enum pcmk__graph_next abort_action, const char *target, const xmlNode *reason) { /* If stonith repeatedly fails, we eventually give up on starting a new * transition for that reason. */ if ((abort_action != pcmk__graph_wait) && too_many_st_failures(target)) { abort_action = pcmk__graph_wait; } - abort_transition(INFINITY, abort_action, "Stonith failed", reason); + abort_transition(PCMK_SCORE_INFINITY, abort_action, "Stonith failed", + reason); } /* * stonith cleanup list * * If the DC is shot, proper notifications might not go out. * The stonith cleanup list allows the cluster to (re-)send * notifications once a new DC is elected. */ static GList *stonith_cleanup_list = NULL; /*! * \internal * \brief Add a node to the stonith cleanup list * * \param[in] target Name of node to add */ void add_stonith_cleanup(const char *target) { stonith_cleanup_list = g_list_append(stonith_cleanup_list, strdup(target)); } /*! * \internal * \brief Remove a node from the stonith cleanup list * * \param[in] Name of node to remove */ void remove_stonith_cleanup(const char *target) { GList *iter = stonith_cleanup_list; while (iter != NULL) { GList *tmp = iter; char *iter_name = tmp->data; iter = iter->next; if (pcmk__str_eq(target, iter_name, pcmk__str_casei)) { crm_trace("Removing %s from the cleanup list", iter_name); stonith_cleanup_list = g_list_delete_link(stonith_cleanup_list, tmp); free(iter_name); } } } /*! * \internal * \brief Purge all entries from the stonith cleanup list */ void purge_stonith_cleanup(void) { if (stonith_cleanup_list) { GList *iter = NULL; for (iter = stonith_cleanup_list; iter != NULL; iter = iter->next) { char *target = iter->data; crm_info("Purging %s from stonith cleanup list", target); free(target); } g_list_free(stonith_cleanup_list); stonith_cleanup_list = NULL; } } /*! * \internal * \brief Send stonith updates for all entries in cleanup list, then purge it */ void execute_stonith_cleanup(void) { GList *iter; for (iter = stonith_cleanup_list; iter != NULL; iter = iter->next) { char *target = iter->data; crm_node_t *target_node = pcmk__get_node(0, target, NULL, pcmk__node_search_cluster); const char *uuid = crm_peer_uuid(target_node); crm_notice("Marking %s, target of a previous stonith action, as clean", target); send_stonith_update(NULL, target, uuid); free(target); } g_list_free(stonith_cleanup_list); stonith_cleanup_list = NULL; } /* end stonith cleanup list functions */ /* stonith API client * * Functions that need to interact directly with the fencer via its API */ static stonith_t *stonith_api = NULL; static mainloop_timer_t *controld_fencer_connect_timer = NULL; static char *te_client_id = NULL; static gboolean fail_incompletable_stonith(pcmk__graph_t *graph) { GList *lpc = NULL; const char *task = NULL; xmlNode *last_action = NULL; if (graph == NULL) { return FALSE; } for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { GList *lpc2 = NULL; pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if (pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) { continue; } for (lpc2 = synapse->actions; lpc2 != NULL; lpc2 = lpc2->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc2->data; if ((action->type != pcmk__cluster_graph_action) || pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { continue; } task = crm_element_value(action->xml, PCMK_XA_OPERATION); if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) { pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); last_action = action->xml; pcmk__update_graph(graph, action); crm_notice("Failing action %d (%s): fencer terminated", action->id, pcmk__xe_id(action->xml)); } } } if (last_action != NULL) { crm_warn("Fencer failure resulted in unrunnable actions"); abort_for_stonith_failure(pcmk__graph_restart, NULL, last_action); return TRUE; } return FALSE; } static void tengine_stonith_connection_destroy(stonith_t *st, stonith_event_t *e) { te_cleanup_stonith_history_sync(st, FALSE); if (pcmk_is_set(controld_globals.fsa_input_register, R_ST_REQUIRED)) { crm_err("Lost fencer connection (will attempt to reconnect)"); if (!mainloop_timer_running(controld_fencer_connect_timer)) { mainloop_timer_start(controld_fencer_connect_timer); } } else { crm_info("Disconnected from fencer"); } if (stonith_api) { /* the client API won't properly reconnect notifications * if they are still in the table - so remove them */ if (stonith_api->state != stonith_disconnected) { stonith_api->cmds->disconnect(st); } stonith_api->cmds->remove_notification(stonith_api, NULL); } if (AM_I_DC) { fail_incompletable_stonith(controld_globals.transition_graph); trigger_graph(); } } /*! * \internal * \brief Handle an event notification from the fencing API * * \param[in] st Fencing API connection (ignored) * \param[in] event Fencing API event notification */ static void handle_fence_notification(stonith_t *st, stonith_event_t *event) { bool succeeded = true; const char *executioner = "the cluster"; const char *client = "a client"; const char *reason = NULL; int exec_status; if (te_client_id == NULL) { te_client_id = crm_strdup_printf("%s.%lu", crm_system_name, (unsigned long) getpid()); } if (event == NULL) { crm_err("Notify data not found"); return; } if (event->executioner != NULL) { executioner = event->executioner; } if (event->client_origin != NULL) { client = event->client_origin; } exec_status = stonith__event_execution_status(event); if ((stonith__event_exit_status(event) != CRM_EX_OK) || (exec_status != PCMK_EXEC_DONE)) { succeeded = false; if (exec_status == PCMK_EXEC_DONE) { exec_status = PCMK_EXEC_ERROR; } } reason = stonith__event_exit_reason(event); crmd_alert_fencing_op(event); if (pcmk__str_eq(PCMK_ACTION_ON, event->action, pcmk__str_none)) { // Unfencing doesn't need special handling, just a log message if (succeeded) { crm_notice("%s was unfenced by %s at the request of %s@%s", event->target, executioner, client, event->origin); } else { crm_err("Unfencing of %s by %s failed (%s%s%s) with exit status %d", event->target, executioner, pcmk_exec_status_str(exec_status), ((reason == NULL)? "" : ": "), ((reason == NULL)? "" : reason), stonith__event_exit_status(event)); } return; } if (succeeded && pcmk__str_eq(event->target, controld_globals.our_nodename, pcmk__str_casei)) { /* We were notified of our own fencing. Most likely, either fencing was * misconfigured, or fabric fencing that doesn't cut cluster * communication is in use. * * Either way, shutting down the local host is a good idea, to require * administrator intervention. Also, other nodes would otherwise likely * set our status to lost because of the fencing callback and discard * our subsequent election votes as "not part of our cluster". */ crm_crit("We were allegedly just fenced by %s for %s!", executioner, event->origin); // Dumps blackbox if enabled if (fence_reaction_panic) { pcmk__panic(__func__); } else { crm_exit(CRM_EX_FATAL); } return; // Should never get here } /* Update the count of fencing failures for this target, in case we become * DC later. The current DC has already updated its fail count in * tengine_stonith_callback(). */ if (!AM_I_DC) { if (succeeded) { st_fail_count_reset(event->target); } else { st_fail_count_increment(event->target); } } crm_notice("Peer %s was%s terminated (%s) by %s on behalf of %s@%s: " "%s%s%s%s " CRM_XS " event=%s", event->target, (succeeded? "" : " not"), event->action, executioner, client, event->origin, (succeeded? "OK" : pcmk_exec_status_str(exec_status)), ((reason == NULL)? "" : " ("), ((reason == NULL)? "" : reason), ((reason == NULL)? "" : ")"), event->id); if (succeeded) { crm_node_t *peer = pcmk__search_node_caches(0, event->target, pcmk__node_search_any |pcmk__node_search_known); const char *uuid = NULL; if (peer == NULL) { return; } uuid = crm_peer_uuid(peer); if (AM_I_DC) { /* The DC always sends updates */ send_stonith_update(NULL, event->target, uuid); /* @TODO Ideally, at this point, we'd check whether the fenced node * hosted any guest nodes, and call remote_node_down() for them. * Unfortunately, the controller doesn't have a simple, reliable way * to map hosts to guests. It might be possible to track this in the * peer cache via crm_remote_peer_cache_refresh(). For now, we rely * on the scheduler creating fence pseudo-events for the guests. */ if (!pcmk__str_eq(client, te_client_id, pcmk__str_casei)) { /* Abort the current transition if it wasn't the cluster that * initiated fencing. */ crm_info("External fencing operation from %s fenced %s", client, event->target); - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "External Fencing Operation", NULL); } } else if (pcmk__str_eq(controld_globals.dc_name, event->target, pcmk__str_null_matches|pcmk__str_casei) && !pcmk_is_set(peer->flags, crm_remote_node)) { // Assume the target was our DC if we don't currently have one if (controld_globals.dc_name != NULL) { crm_notice("Fencing target %s was our DC", event->target); } else { crm_notice("Fencing target %s may have been our DC", event->target); } /* Given the CIB resyncing that occurs around elections, * have one node update the CIB now and, if the new DC is different, * have them do so too after the election */ if (pcmk__str_eq(event->executioner, controld_globals.our_nodename, pcmk__str_casei)) { send_stonith_update(NULL, event->target, uuid); } add_stonith_cleanup(event->target); } /* If the target is a remote node, and we host its connection, * immediately fail all monitors so it can be recovered quickly. * The connection won't necessarily drop when a remote node is fenced, * so the failure might not otherwise be detected until the next poke. */ if (pcmk_is_set(peer->flags, crm_remote_node)) { remote_ra_fail(event->target); } crmd_peer_down(peer, TRUE); } } /*! * \brief Connect to fencer * * \param[in] user_data If NULL, retry failures now, otherwise retry in mainloop timer * * \return G_SOURCE_REMOVE on success, G_SOURCE_CONTINUE to retry * \note If user_data is NULL, this will wait 2s between attempts, for up to * 30 attempts, meaning the controller could be blocked as long as 58s. */ gboolean controld_timer_fencer_connect(gpointer user_data) { int rc = pcmk_ok; if (stonith_api == NULL) { stonith_api = stonith_api_new(); if (stonith_api == NULL) { crm_err("Could not connect to fencer: API memory allocation failed"); return G_SOURCE_REMOVE; } } if (stonith_api->state != stonith_disconnected) { crm_trace("Already connected to fencer, no need to retry"); return G_SOURCE_REMOVE; } if (user_data == NULL) { // Blocking (retry failures now until successful) rc = stonith_api_connect_retry(stonith_api, crm_system_name, 30); if (rc != pcmk_ok) { crm_err("Could not connect to fencer in 30 attempts: %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); } } else { // Non-blocking (retry failures later in main loop) rc = stonith_api->cmds->connect(stonith_api, crm_system_name, NULL); if (controld_fencer_connect_timer == NULL) { controld_fencer_connect_timer = mainloop_timer_add("controld_fencer_connect", 1000, TRUE, controld_timer_fencer_connect, GINT_TO_POINTER(TRUE)); } if (rc != pcmk_ok) { if (pcmk_is_set(controld_globals.fsa_input_register, R_ST_REQUIRED)) { crm_notice("Fencer connection failed (will retry): %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); if (!mainloop_timer_running(controld_fencer_connect_timer)) { mainloop_timer_start(controld_fencer_connect_timer); } return G_SOURCE_CONTINUE; } else { crm_info("Fencer connection failed (ignoring because no longer required): %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); } return G_SOURCE_REMOVE; } } if (rc == pcmk_ok) { stonith_api->cmds->register_notification(stonith_api, T_STONITH_NOTIFY_DISCONNECT, tengine_stonith_connection_destroy); stonith_api->cmds->register_notification(stonith_api, T_STONITH_NOTIFY_FENCE, handle_fence_notification); stonith_api->cmds->register_notification(stonith_api, T_STONITH_NOTIFY_HISTORY_SYNCED, tengine_stonith_history_synced); te_trigger_stonith_history_sync(TRUE); crm_notice("Fencer successfully connected"); } return G_SOURCE_REMOVE; } void controld_disconnect_fencer(bool destroy) { if (stonith_api) { // Prevent fencer connection from coming up again controld_clear_fsa_input_flags(R_ST_REQUIRED); if (stonith_api->state != stonith_disconnected) { stonith_api->cmds->disconnect(stonith_api); } stonith_api->cmds->remove_notification(stonith_api, NULL); } if (destroy) { if (stonith_api) { stonith_api->cmds->free(stonith_api); stonith_api = NULL; } if (controld_fencer_connect_timer) { mainloop_timer_del(controld_fencer_connect_timer); controld_fencer_connect_timer = NULL; } if (te_client_id) { free(te_client_id); te_client_id = NULL; } } } static gboolean do_stonith_history_sync(gpointer user_data) { if (stonith_api && (stonith_api->state != stonith_disconnected)) { stonith_history_t *history = NULL; te_cleanup_stonith_history_sync(stonith_api, FALSE); stonith_api->cmds->history(stonith_api, st_opt_sync_call | st_opt_broadcast, NULL, &history, 5); stonith_history_free(history); return TRUE; } else { crm_info("Skip triggering stonith history-sync as stonith is disconnected"); return FALSE; } } static void tengine_stonith_callback(stonith_t *stonith, stonith_callback_data_t *data) { char *uuid = NULL; int stonith_id = -1; int transition_id = -1; pcmk__graph_action_t *action = NULL; const char *target = NULL; if ((data == NULL) || (data->userdata == NULL)) { crm_err("Ignoring fence operation %d result: " "No transition key given (bug?)", ((data == NULL)? -1 : data->call_id)); return; } if (!AM_I_DC) { const char *reason = stonith__exit_reason(data); if (reason == NULL) { reason = pcmk_exec_status_str(stonith__execution_status(data)); } crm_notice("Result of fence operation %d: %d (%s) " CRM_XS " key=%s", data->call_id, stonith__exit_status(data), reason, (const char *) data->userdata); return; } CRM_CHECK(decode_transition_key(data->userdata, &uuid, &transition_id, &stonith_id, NULL), goto bail); if (controld_globals.transition_graph->complete || (stonith_id < 0) || !pcmk__str_eq(uuid, controld_globals.te_uuid, pcmk__str_none) || (controld_globals.transition_graph->id != transition_id)) { crm_info("Ignoring fence operation %d result: " "Not from current transition " CRM_XS " complete=%s action=%d uuid=%s (vs %s) transition=%d (vs %d)", data->call_id, pcmk__btoa(controld_globals.transition_graph->complete), stonith_id, uuid, controld_globals.te_uuid, transition_id, controld_globals.transition_graph->id); goto bail; } action = controld_get_action(stonith_id); if (action == NULL) { crm_err("Ignoring fence operation %d result: " "Action %d not found in transition graph (bug?) " CRM_XS " uuid=%s transition=%d", data->call_id, stonith_id, uuid, transition_id); goto bail; } target = crm_element_value(action->xml, PCMK__META_ON_NODE); if (target == NULL) { crm_err("Ignoring fence operation %d result: No target given (bug?)", data->call_id); goto bail; } stop_te_timer(action); if (stonith__exit_status(data) == CRM_EX_OK) { const char *uuid = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID); const char *op = crm_meta_value(action->params, PCMK__META_STONITH_ACTION); crm_info("Fence operation %d for %s succeeded", data->call_id, target); if (!(pcmk_is_set(action->flags, pcmk__graph_action_confirmed))) { te_action_confirmed(action, NULL); if (pcmk__str_eq(PCMK_ACTION_ON, op, pcmk__str_casei)) { const char *value = NULL; char *now = pcmk__ttoa(time(NULL)); gboolean is_remote_node = FALSE; /* This check is not 100% reliable, since this node is not * guaranteed to have the remote node cached. However, it * doesn't have to be reliable, since the attribute manager can * learn a node's "remoteness" by other means sooner or later. * This allows it to learn more quickly if this node does have * the information. */ if (g_hash_table_lookup(crm_remote_peer_cache, uuid) != NULL) { is_remote_node = TRUE; } update_attrd(target, CRM_ATTR_UNFENCED, now, NULL, is_remote_node); free(now); value = crm_meta_value(action->params, PCMK__META_DIGESTS_ALL); update_attrd(target, CRM_ATTR_DIGESTS_ALL, value, NULL, is_remote_node); value = crm_meta_value(action->params, PCMK__META_DIGESTS_SECURE); update_attrd(target, CRM_ATTR_DIGESTS_SECURE, value, NULL, is_remote_node); } else if (!(pcmk_is_set(action->flags, pcmk__graph_action_sent_update))) { send_stonith_update(action, target, uuid); pcmk__set_graph_action_flags(action, pcmk__graph_action_sent_update); } } st_fail_count_reset(target); } else { enum pcmk__graph_next abort_action = pcmk__graph_restart; int status = stonith__execution_status(data); const char *reason = stonith__exit_reason(data); if (reason == NULL) { if (status == PCMK_EXEC_DONE) { reason = "Agent returned error"; } else { reason = pcmk_exec_status_str(status); } } pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); /* If no fence devices were available, there's no use in immediately * checking again, so don't start a new transition in that case. */ if (status == PCMK_EXEC_NO_FENCE_DEVICE) { crm_warn("Fence operation %d for %s failed: %s " "(aborting transition and giving up for now)", data->call_id, target, reason); abort_action = pcmk__graph_wait; } else { crm_notice("Fence operation %d for %s failed: %s " "(aborting transition)", data->call_id, target, reason); } /* Increment the fail count now, so abort_for_stonith_failure() can * check it. Non-DC nodes will increment it in * handle_fence_notification(). */ st_fail_count_increment(target); abort_for_stonith_failure(abort_action, target, NULL); } pcmk__update_graph(controld_globals.transition_graph, action); trigger_graph(); bail: free(data->userdata); free(uuid); return; } static int fence_with_delay(const char *target, const char *type, int delay) { uint32_t options = st_opt_none; // Group of enum stonith_call_options int timeout_sec = (int) (controld_globals.transition_graph->stonith_timeout / 1000); if (crmd_join_phase_count(crm_join_confirmed) == 1) { stonith__set_call_options(options, target, st_opt_allow_suicide); } return stonith_api->cmds->fence_with_delay(stonith_api, options, target, type, timeout_sec, 0, delay); } /*! * \internal * \brief Execute a fencing action from a transition graph * * \param[in] graph Transition graph being executed (ignored) * \param[in] action Fencing action to execute * * \return Standard Pacemaker return code */ int controld_execute_fence_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { int rc = 0; const char *id = pcmk__xe_id(action->xml); const char *uuid = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID); const char *target = crm_element_value(action->xml, PCMK__META_ON_NODE); const char *type = crm_meta_value(action->params, PCMK__META_STONITH_ACTION); char *transition_key = NULL; const char *priority_delay = NULL; int delay_i = 0; gboolean invalid_action = FALSE; int stonith_timeout = (int) (controld_globals.transition_graph->stonith_timeout / 1000); CRM_CHECK(id != NULL, invalid_action = TRUE); CRM_CHECK(uuid != NULL, invalid_action = TRUE); CRM_CHECK(type != NULL, invalid_action = TRUE); CRM_CHECK(target != NULL, invalid_action = TRUE); if (invalid_action) { crm_log_xml_warn(action->xml, "BadAction"); return EPROTO; } priority_delay = crm_meta_value(action->params, PCMK_OPT_PRIORITY_FENCING_DELAY); crm_notice("Requesting fencing (%s) targeting node %s " CRM_XS " action=%s timeout=%i%s%s", type, target, id, stonith_timeout, priority_delay ? " priority_delay=" : "", priority_delay ? priority_delay : ""); /* Passing NULL means block until we can connect... */ controld_timer_fencer_connect(NULL); pcmk__scan_min_int(priority_delay, &delay_i, 0); rc = fence_with_delay(target, type, delay_i); transition_key = pcmk__transition_key(controld_globals.transition_graph->id, action->id, 0, controld_globals.te_uuid), stonith_api->cmds->register_callback(stonith_api, rc, (stonith_timeout + (delay_i > 0 ? delay_i : 0)), st_opt_timeout_updates, transition_key, "tengine_stonith_callback", tengine_stonith_callback); return pcmk_rc_ok; } bool controld_verify_stonith_watchdog_timeout(const char *value) { long long st_timeout = (value != NULL)? crm_get_msec(value) : 0; const char *our_nodename = controld_globals.our_nodename; if (st_timeout == 0 || (stonith_api && (stonith_api->state != stonith_disconnected) && stonith__watchdog_fencing_enabled_for_node_api(stonith_api, our_nodename))) { return pcmk__valid_stonith_watchdog_timeout(value); } return true; } /* end stonith API client functions */ /* * stonith history synchronization * * Each node's fencer keeps track of a cluster-wide fencing history. When a node * joins or leaves, we need to synchronize the history across all nodes. */ static crm_trigger_t *stonith_history_sync_trigger = NULL; static mainloop_timer_t *stonith_history_sync_timer_short = NULL; static mainloop_timer_t *stonith_history_sync_timer_long = NULL; void te_cleanup_stonith_history_sync(stonith_t *st, bool free_timers) { if (free_timers) { mainloop_timer_del(stonith_history_sync_timer_short); stonith_history_sync_timer_short = NULL; mainloop_timer_del(stonith_history_sync_timer_long); stonith_history_sync_timer_long = NULL; } else { mainloop_timer_stop(stonith_history_sync_timer_short); mainloop_timer_stop(stonith_history_sync_timer_long); } if (st) { st->cmds->remove_notification(st, T_STONITH_NOTIFY_HISTORY_SYNCED); } } static void tengine_stonith_history_synced(stonith_t *st, stonith_event_t *st_event) { te_cleanup_stonith_history_sync(st, FALSE); crm_debug("Fence-history synced - cancel all timers"); } static gboolean stonith_history_sync_set_trigger(gpointer user_data) { mainloop_set_trigger(stonith_history_sync_trigger); return FALSE; } void te_trigger_stonith_history_sync(bool long_timeout) { /* trigger a sync in 5s to give more nodes the * chance to show up so that we don't create * unnecessary stonith-history-sync traffic * * the long timeout of 30s is there as a fallback * so that after a successful connection to fenced * we will wait for 30s for the DC to trigger a * history-sync * if this doesn't happen we trigger a sync locally * (e.g. fenced segfaults and is restarted by pacemakerd) */ /* as we are finally checking the stonith-connection * in do_stonith_history_sync we should be fine * leaving stonith_history_sync_time & stonith_history_sync_trigger * around */ if (stonith_history_sync_trigger == NULL) { stonith_history_sync_trigger = mainloop_add_trigger(G_PRIORITY_LOW, do_stonith_history_sync, NULL); } if (long_timeout) { if(stonith_history_sync_timer_long == NULL) { stonith_history_sync_timer_long = mainloop_timer_add("history_sync_long", 30000, FALSE, stonith_history_sync_set_trigger, NULL); } crm_info("Fence history will be synchronized cluster-wide within 30 seconds"); mainloop_timer_start(stonith_history_sync_timer_long); } else { if(stonith_history_sync_timer_short == NULL) { stonith_history_sync_timer_short = mainloop_timer_add("history_sync_short", 5000, FALSE, stonith_history_sync_set_trigger, NULL); } crm_info("Fence history will be synchronized cluster-wide within 5 seconds"); mainloop_timer_start(stonith_history_sync_timer_short); } } /* end stonith history synchronization functions */ diff --git a/daemons/controld/controld_join_dc.c b/daemons/controld/controld_join_dc.c index 596edbb53b..1fab3883b0 100644 --- a/daemons/controld/controld_join_dc.c +++ b/daemons/controld/controld_join_dc.c @@ -1,1043 +1,1044 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include static char *max_generation_from = NULL; static xmlNodePtr max_generation_xml = NULL; /*! * \internal * \brief Nodes from which a CIB sync has failed since the peer joined * * This table is of the form (node_name -> join_id). \p node_name is * the name of a client node from which a CIB \p sync_from() call has failed in * \p do_dc_join_finalize() since the client joined the cluster as a peer. * \p join_id is the ID of the join round in which the \p sync_from() failed, * and is intended for use in nack log messages. */ static GHashTable *failed_sync_nodes = NULL; void finalize_join_for(gpointer key, gpointer value, gpointer user_data); void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data); gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); /* Numeric counter used to identify join rounds (an unsigned int would be * appropriate, except we get and set it in XML as int) */ static int current_join_id = 0; /*! * \internal * \brief Destroy the hash table containing failed sync nodes */ void controld_destroy_failed_sync_table(void) { if (failed_sync_nodes != NULL) { g_hash_table_destroy(failed_sync_nodes); failed_sync_nodes = NULL; } } /*! * \internal * \brief Remove a node from the failed sync nodes table if present * * \param[in] node_name Node name to remove */ void controld_remove_failed_sync_node(const char *node_name) { if (failed_sync_nodes != NULL) { g_hash_table_remove(failed_sync_nodes, (gchar *) node_name); } } /*! * \internal * \brief Add to a hash table a node whose CIB failed to sync * * \param[in] node_name Name of node whose CIB failed to sync * \param[in] join_id Join round when the failure occurred */ static void record_failed_sync_node(const char *node_name, gint join_id) { if (failed_sync_nodes == NULL) { failed_sync_nodes = pcmk__strikey_table(g_free, NULL); } /* If the node is already in the table then we failed to nack it during the * filter offer step */ CRM_LOG_ASSERT(g_hash_table_insert(failed_sync_nodes, g_strdup(node_name), GINT_TO_POINTER(join_id))); } /*! * \internal * \brief Look up a node name in the failed sync table * * \param[in] node_name Name of node to look up * \param[out] join_id Where to store the join ID of when the sync failed * * \return Standard Pacemaker return code. Specifically, \p pcmk_rc_ok if the * node name was found, or \p pcmk_rc_node_unknown otherwise. * \note \p *join_id is set to -1 if the node is not found. */ static int lookup_failed_sync_node(const char *node_name, gint *join_id) { *join_id = -1; if (failed_sync_nodes != NULL) { gpointer result = g_hash_table_lookup(failed_sync_nodes, (gchar *) node_name); if (result != NULL) { *join_id = GPOINTER_TO_INT(result); return pcmk_rc_ok; } } return pcmk_rc_node_unknown; } void crm_update_peer_join(const char *source, crm_node_t * node, enum crm_join_phase phase) { enum crm_join_phase last = 0; CRM_CHECK(node != NULL, return); /* Remote nodes do not participate in joins */ if (pcmk_is_set(node->flags, crm_remote_node)) { return; } last = node->join; if(phase == last) { crm_trace("Node %s join-%d phase is still %s " CRM_XS " nodeid=%u source=%s", node->uname, current_join_id, crm_join_phase_str(last), node->id, source); } else if ((phase <= crm_join_none) || (phase == (last + 1))) { node->join = phase; crm_trace("Node %s join-%d phase is now %s (was %s) " CRM_XS " nodeid=%u source=%s", node->uname, current_join_id, crm_join_phase_str(phase), crm_join_phase_str(last), node->id, source); } else { crm_warn("Rejecting join-%d phase update for node %s because " "can't go from %s to %s " CRM_XS " nodeid=%u source=%s", current_join_id, node->uname, crm_join_phase_str(last), crm_join_phase_str(phase), node->id, source); } } static void start_join_round(void) { GHashTableIter iter; crm_node_t *peer = NULL; crm_debug("Starting new join round join-%d", current_join_id); g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { crm_update_peer_join(__func__, peer, crm_join_none); } if (max_generation_from != NULL) { free(max_generation_from); max_generation_from = NULL; } if (max_generation_xml != NULL) { free_xml(max_generation_xml); max_generation_xml = NULL; } controld_clear_fsa_input_flags(R_HAVE_CIB); } /*! * \internal * \brief Create a join message from the DC * * \param[in] join_op Join operation name * \param[in] host_to Recipient of message */ static xmlNode * create_dc_message(const char *join_op, const char *host_to) { xmlNode *msg = create_request(join_op, NULL, host_to, CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL); /* Identify which election this is a part of */ crm_xml_add_int(msg, PCMK__XA_JOIN_ID, current_join_id); /* Add a field specifying whether the DC is shutting down. This keeps the * joining node from fencing the old DC if it becomes the new DC. */ pcmk__xe_set_bool_attr(msg, PCMK__XA_DC_LEAVING, pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)); return msg; } static void join_make_offer(gpointer key, gpointer value, gpointer user_data) { xmlNode *offer = NULL; crm_node_t *member = (crm_node_t *)value; CRM_ASSERT(member != NULL); if (crm_is_peer_active(member) == FALSE) { crm_info("Not making join-%d offer to inactive node %s", current_join_id, (member->uname? member->uname : "with unknown name")); if(member->expected == NULL && pcmk__str_eq(member->state, CRM_NODE_LOST, pcmk__str_casei)) { /* You would think this unsafe, but in fact this plus an * active resource is what causes it to be fenced. * * Yes, this does mean that any node that dies at the same * time as the old DC and is not running resource (still) * won't be fenced. * * I'm not happy about this either. */ pcmk__update_peer_expected(__func__, member, CRMD_JOINSTATE_DOWN); } return; } if (member->uname == NULL) { crm_info("Not making join-%d offer to node uuid %s with unknown name", current_join_id, member->uuid); return; } if (controld_globals.membership_id != crm_peer_seq) { controld_globals.membership_id = crm_peer_seq; crm_info("Making join-%d offers based on membership event %llu", current_join_id, crm_peer_seq); } if(user_data && member->join > crm_join_none) { crm_info("Not making join-%d offer to already known node %s (%s)", current_join_id, member->uname, crm_join_phase_str(member->join)); return; } crm_update_peer_join(__func__, (crm_node_t*)member, crm_join_none); offer = create_dc_message(CRM_OP_JOIN_OFFER, member->uname); // Advertise our feature set so the joining node can bail if not compatible crm_xml_add(offer, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); crm_info("Sending join-%d offer to %s", current_join_id, member->uname); send_cluster_message(member, crm_msg_crmd, offer, TRUE); free_xml(offer); crm_update_peer_join(__func__, member, crm_join_welcomed); } /* A_DC_JOIN_OFFER_ALL */ void do_dc_join_offer_all(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { int count; /* Reset everyone's status back to down or in_ccm in the CIB. * Any nodes that are active in the CIB but not in the cluster membership * will be seen as offline by the scheduler anyway. */ current_join_id++; start_join_round(); update_dc(NULL); if (cause == C_HA_MESSAGE && current_input == I_NODE_JOIN) { crm_info("A new node joined the cluster"); } g_hash_table_foreach(crm_peer_cache, join_make_offer, NULL); count = crmd_join_phase_count(crm_join_welcomed); crm_info("Waiting on join-%d requests from %d outstanding node%s", current_join_id, count, pcmk__plural_s(count)); // Don't waste time by invoking the scheduler yet } /* A_DC_JOIN_OFFER_ONE */ void do_dc_join_offer_one(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_node_t *member; ha_msg_input_t *welcome = NULL; int count; const char *join_to = NULL; if (msg_data->data == NULL) { crm_info("Making join-%d offers to any unconfirmed nodes " "because an unknown node joined", current_join_id); g_hash_table_foreach(crm_peer_cache, join_make_offer, &member); check_join_state(cur_state, __func__); return; } welcome = fsa_typed_data(fsa_dt_ha_msg); if (welcome == NULL) { // fsa_typed_data() already logged an error return; } join_to = crm_element_value(welcome->msg, PCMK__XA_SRC); if (join_to == NULL) { crm_err("Can't make join-%d offer to unknown node", current_join_id); return; } member = pcmk__get_node(0, join_to, NULL, pcmk__node_search_cluster); /* It is possible that a node will have been sick or starting up when the * original offer was made. However, it will either re-announce itself in * due course, or we can re-store the original offer on the client. */ crm_update_peer_join(__func__, member, crm_join_none); join_make_offer(NULL, member, NULL); /* If the offer isn't to the local node, make an offer to the local node as * well, to ensure the correct value for max_generation_from. */ if (strcasecmp(join_to, controld_globals.our_nodename) != 0) { member = pcmk__get_node(0, controld_globals.our_nodename, NULL, pcmk__node_search_cluster); join_make_offer(NULL, member, NULL); } /* This was a genuine join request; cancel any existing transition and * invoke the scheduler. */ - abort_transition(INFINITY, pcmk__graph_restart, "Node join", NULL); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Node join", + NULL); count = crmd_join_phase_count(crm_join_welcomed); crm_info("Waiting on join-%d requests from %d outstanding node%s", current_join_id, count, pcmk__plural_s(count)); // Don't waste time by invoking the scheduler yet } static int compare_int_fields(xmlNode * left, xmlNode * right, const char *field) { const char *elem_l = crm_element_value(left, field); const char *elem_r = crm_element_value(right, field); long long int_elem_l; long long int_elem_r; pcmk__scan_ll(elem_l, &int_elem_l, -1LL); pcmk__scan_ll(elem_r, &int_elem_r, -1LL); if (int_elem_l < int_elem_r) { return -1; } else if (int_elem_l > int_elem_r) { return 1; } return 0; } /* A_DC_JOIN_PROCESS_REQ */ void do_dc_join_filter_offer(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { xmlNode *generation = NULL; int cmp = 0; int join_id = -1; int count = 0; gint value = 0; gboolean ack_nack_bool = TRUE; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *join_from = crm_element_value(join_ack->msg, PCMK__XA_SRC); const char *ref = crm_element_value(join_ack->msg, PCMK_XA_REFERENCE); const char *join_version = crm_element_value(join_ack->msg, PCMK_XA_CRM_FEATURE_SET); crm_node_t *join_node = NULL; if (join_from == NULL) { crm_err("Ignoring invalid join request without node name"); return; } join_node = pcmk__get_node(0, join_from, NULL, pcmk__node_search_cluster); crm_element_value_int(join_ack->msg, PCMK__XA_JOIN_ID, &join_id); if (join_id != current_join_id) { crm_debug("Ignoring join-%d request from %s because we are on join-%d", join_id, join_from, current_join_id); check_join_state(cur_state, __func__); return; } generation = join_ack->xml; if (max_generation_xml != NULL && generation != NULL) { int lpc = 0; const char *attributes[] = { PCMK_XA_ADMIN_EPOCH, PCMK_XA_EPOCH, PCMK_XA_NUM_UPDATES, }; /* It's not obvious that join_ack->xml is the PCMK__XE_GENERATION_TUPLE * element from the join client. The "if" guard is for clarity. */ if (pcmk__xe_is(generation, PCMK__XE_GENERATION_TUPLE)) { for (lpc = 0; cmp == 0 && lpc < PCMK__NELEM(attributes); lpc++) { cmp = compare_int_fields(max_generation_xml, generation, attributes[lpc]); } } else { // Should always be PCMK__XE_GENERATION_TUPLE CRM_LOG_ASSERT(false); } } if (ref == NULL) { ref = "none"; // for logging only } if (lookup_failed_sync_node(join_from, &value) == pcmk_rc_ok) { crm_err("Rejecting join-%d request from node %s because we failed to " "sync its CIB in join-%d " CRM_XS " ref=%s", join_id, join_from, value, ref); ack_nack_bool = FALSE; } else if (!crm_is_peer_active(join_node)) { if (match_down_event(join_from) != NULL) { /* The join request was received after the node was fenced or * otherwise shutdown in a way that we're aware of. No need to log * an error in this rare occurrence; we know the client was recently * shut down, and receiving a lingering in-flight request is not * cause for alarm. */ crm_debug("Rejecting join-%d request from inactive node %s " CRM_XS " ref=%s", join_id, join_from, ref); } else { crm_err("Rejecting join-%d request from inactive node %s " CRM_XS " ref=%s", join_id, join_from, ref); } ack_nack_bool = FALSE; } else if (generation == NULL) { crm_err("Rejecting invalid join-%d request from node %s " "missing CIB generation " CRM_XS " ref=%s", join_id, join_from, ref); ack_nack_bool = FALSE; } else if ((join_version == NULL) || !feature_set_compatible(CRM_FEATURE_SET, join_version)) { crm_err("Rejecting join-%d request from node %s because feature set %s" " is incompatible with ours (%s) " CRM_XS " ref=%s", join_id, join_from, (join_version? join_version : "pre-3.1.0"), CRM_FEATURE_SET, ref); ack_nack_bool = FALSE; } else if (max_generation_xml == NULL) { const char *validation = crm_element_value(generation, PCMK_XA_VALIDATE_WITH); if (get_schema_version(validation) < 0) { crm_err("Rejecting join-%d request from %s (with first CIB " "generation) due to unknown schema version %s " CRM_XS " ref=%s", join_id, join_from, validation, ref); ack_nack_bool = FALSE; } else { crm_debug("Accepting join-%d request from %s (with first CIB " "generation) " CRM_XS " ref=%s", join_id, join_from, ref); max_generation_xml = copy_xml(generation); pcmk__str_update(&max_generation_from, join_from); } } else if ((cmp < 0) || ((cmp == 0) && pcmk__str_eq(join_from, controld_globals.our_nodename, pcmk__str_casei))) { const char *validation = crm_element_value(generation, PCMK_XA_VALIDATE_WITH); if (get_schema_version(validation) < 0) { crm_err("Rejecting join-%d request from %s (with better CIB " "generation than current best from %s) due to unknown " "schema version %s " CRM_XS " ref=%s", join_id, join_from, max_generation_from, validation, ref); ack_nack_bool = FALSE; } else { crm_debug("Accepting join-%d request from %s (with better CIB " "generation than current best from %s) " CRM_XS " ref=%s", join_id, join_from, max_generation_from, ref); crm_log_xml_debug(max_generation_xml, "Old max generation"); crm_log_xml_debug(generation, "New max generation"); free_xml(max_generation_xml); max_generation_xml = copy_xml(join_ack->xml); pcmk__str_update(&max_generation_from, join_from); } } else { crm_debug("Accepting join-%d request from %s " CRM_XS " ref=%s", join_id, join_from, ref); } if (!ack_nack_bool) { if (compare_version(join_version, "3.17.0") < 0) { /* Clients with CRM_FEATURE_SET < 3.17.0 may respawn infinitely * after a nack message, don't send one */ crm_update_peer_join(__func__, join_node, crm_join_nack_quiet); } else { crm_update_peer_join(__func__, join_node, crm_join_nack); } pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_NACK); } else { crm_update_peer_join(__func__, join_node, crm_join_integrated); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER); } count = crmd_join_phase_count(crm_join_integrated); crm_debug("%d node%s currently integrated in join-%d", count, pcmk__plural_s(count), join_id); if (check_join_state(cur_state, __func__) == FALSE) { // Don't waste time by invoking the scheduler yet count = crmd_join_phase_count(crm_join_welcomed); crm_debug("Waiting on join-%d requests from %d outstanding node%s", join_id, count, pcmk__plural_s(count)); } } /* A_DC_JOIN_FINALIZE */ void do_dc_join_finalize(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { char *sync_from = NULL; int rc = pcmk_ok; int count_welcomed = crmd_join_phase_count(crm_join_welcomed); int count_finalizable = crmd_join_phase_count(crm_join_integrated) + crmd_join_phase_count(crm_join_nack) + crmd_join_phase_count(crm_join_nack_quiet); /* This we can do straight away and avoid clients timing us out * while we compute the latest CIB */ if (count_welcomed != 0) { crm_debug("Waiting on join-%d requests from %d outstanding node%s " "before finalizing join", current_join_id, count_welcomed, pcmk__plural_s(count_welcomed)); crmd_join_phase_log(LOG_DEBUG); /* crmd_fsa_stall(FALSE); Needed? */ return; } else if (count_finalizable == 0) { crm_debug("Finalization not needed for join-%d at the current time", current_join_id); crmd_join_phase_log(LOG_DEBUG); check_join_state(controld_globals.fsa_state, __func__); return; } controld_clear_fsa_input_flags(R_HAVE_CIB); if (pcmk__str_eq(max_generation_from, controld_globals.our_nodename, pcmk__str_null_matches|pcmk__str_casei)) { controld_set_fsa_input_flags(R_HAVE_CIB); } if (!controld_globals.transition_graph->complete) { crm_warn("Delaying join-%d finalization while transition in progress", current_join_id); crmd_join_phase_log(LOG_DEBUG); crmd_fsa_stall(FALSE); return; } if (pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) { // Send our CIB out to everyone pcmk__str_update(&sync_from, controld_globals.our_nodename); crm_debug("Finalizing join-%d for %d node%s (sync'ing from local CIB)", current_join_id, count_finalizable, pcmk__plural_s(count_finalizable)); crm_log_xml_debug(max_generation_xml, "Requested CIB version"); } else { // Ask for the agreed best CIB pcmk__str_update(&sync_from, max_generation_from); crm_notice("Finalizing join-%d for %d node%s (sync'ing CIB from %s)", current_join_id, count_finalizable, pcmk__plural_s(count_finalizable), sync_from); crm_log_xml_notice(max_generation_xml, "Requested CIB version"); } crmd_join_phase_log(LOG_DEBUG); rc = controld_globals.cib_conn->cmds->sync_from(controld_globals.cib_conn, sync_from, NULL, cib_none); fsa_register_cib_callback(rc, sync_from, finalize_sync_callback); } void free_max_generation(void) { free(max_generation_from); max_generation_from = NULL; free_xml(max_generation_xml); max_generation_xml = NULL; } void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { CRM_LOG_ASSERT(-EPERM != rc); if (rc != pcmk_ok) { const char *sync_from = (const char *) user_data; do_crm_log(((rc == -pcmk_err_old_data)? LOG_WARNING : LOG_ERR), "Could not sync CIB from %s in join-%d: %s", sync_from, current_join_id, pcmk_strerror(rc)); if (rc != -pcmk_err_old_data) { record_failed_sync_node(sync_from, current_join_id); } /* restart the whole join process */ register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION_DC, NULL, NULL, __func__); } else if (!AM_I_DC) { crm_debug("Sync'ed CIB for join-%d but no longer DC", current_join_id); } else if (controld_globals.fsa_state != S_FINALIZE_JOIN) { crm_debug("Sync'ed CIB for join-%d but no longer in S_FINALIZE_JOIN " "(%s)", current_join_id, fsa_state2string(controld_globals.fsa_state)); } else { controld_set_fsa_input_flags(R_HAVE_CIB); /* make sure dc_uuid is re-set to us */ if (!check_join_state(controld_globals.fsa_state, __func__)) { int count_finalizable = 0; count_finalizable = crmd_join_phase_count(crm_join_integrated) + crmd_join_phase_count(crm_join_nack) + crmd_join_phase_count(crm_join_nack_quiet); crm_debug("Notifying %d node%s of join-%d results", count_finalizable, pcmk__plural_s(count_finalizable), current_join_id); g_hash_table_foreach(crm_peer_cache, finalize_join_for, NULL); } } } static void join_node_state_commit_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { const char *node = user_data; if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; // for register_fsa_error() macro crm_crit("join-%d node history update (via CIB call %d) for node %s " "failed: %s", current_join_id, call_id, node, pcmk_strerror(rc)); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } crm_debug("join-%d node history update (via CIB call %d) for node %s " "complete", current_join_id, call_id, node); check_join_state(controld_globals.fsa_state, __func__); } /* A_DC_JOIN_PROCESS_ACK */ void do_dc_join_ack(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { int join_id = -1; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *op = crm_element_value(join_ack->msg, PCMK__XA_CRM_TASK); char *join_from = crm_element_value_copy(join_ack->msg, PCMK__XA_SRC); crm_node_t *peer = NULL; enum controld_section_e section = controld_section_lrm; char *xpath = NULL; xmlNode *state = join_ack->xml; xmlNode *execd_state = NULL; cib_t *cib = controld_globals.cib_conn; int rc = pcmk_ok; // Sanity checks if (join_from == NULL) { crm_warn("Ignoring message received without node identification"); goto done; } if (op == NULL) { crm_warn("Ignoring message received from %s without task", join_from); goto done; } if (strcmp(op, CRM_OP_JOIN_CONFIRM)) { crm_debug("Ignoring '%s' message from %s while waiting for '%s'", op, join_from, CRM_OP_JOIN_CONFIRM); goto done; } if (crm_element_value_int(join_ack->msg, PCMK__XA_JOIN_ID, &join_id) != 0) { crm_warn("Ignoring join confirmation from %s without valid join ID", join_from); goto done; } peer = pcmk__get_node(0, join_from, NULL, pcmk__node_search_cluster); if (peer->join != crm_join_finalized) { crm_info("Ignoring out-of-sequence join-%d confirmation from %s " "(currently %s not %s)", join_id, join_from, crm_join_phase_str(peer->join), crm_join_phase_str(crm_join_finalized)); goto done; } if (join_id != current_join_id) { crm_err("Rejecting join-%d confirmation from %s " "because currently on join-%d", join_id, join_from, current_join_id); crm_update_peer_join(__func__, peer, crm_join_nack); goto done; } crm_update_peer_join(__func__, peer, crm_join_confirmed); /* Update CIB with node's current executor state. A new transition will be * triggered later, when the CIB manager notifies us of the change. * * The delete and modify requests are part of an atomic transaction. */ rc = cib->cmds->init_transaction(cib); if (rc != pcmk_ok) { goto done; } // Delete relevant parts of node's current executor state from CIB if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { section = controld_section_lrm_unlocked; } controld_node_state_deletion_strings(join_from, section, &xpath, NULL); rc = cib->cmds->remove(cib, xpath, NULL, cib_scope_local |cib_xpath |cib_multiple |cib_transaction); if (rc != pcmk_ok) { goto done; } // Update CIB with node's latest known executor state if (pcmk__str_eq(join_from, controld_globals.our_nodename, pcmk__str_casei)) { // Use the latest possible state if processing our own join ack execd_state = controld_query_executor_state(); if (execd_state != NULL) { crm_debug("Updating local node history for join-%d from query " "result", current_join_id); state = execd_state; } else { crm_warn("Updating local node history from join-%d confirmation " "because query failed", current_join_id); } } else { crm_debug("Updating node history for %s from join-%d confirmation", join_from, current_join_id); } rc = cib->cmds->modify(cib, PCMK_XE_STATUS, state, cib_scope_local|cib_can_create|cib_transaction); free_xml(execd_state); if (rc != pcmk_ok) { goto done; } // Commit the transaction rc = cib->cmds->end_transaction(cib, true, cib_scope_local); fsa_register_cib_callback(rc, join_from, join_node_state_commit_callback); if (rc > 0) { // join_from will be freed after callback join_from = NULL; rc = pcmk_ok; } done: if (rc != pcmk_ok) { crm_crit("join-%d node history update for node %s failed: %s", current_join_id, join_from, pcmk_strerror(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } free(join_from); free(xpath); } void finalize_join_for(gpointer key, gpointer value, gpointer user_data) { xmlNode *acknak = NULL; xmlNode *tmp1 = NULL; crm_node_t *join_node = value; const char *join_to = join_node->uname; bool integrated = false; switch (join_node->join) { case crm_join_integrated: integrated = true; break; case crm_join_nack: case crm_join_nack_quiet: break; default: crm_trace("Not updating non-integrated and non-nacked node %s (%s) " "for join-%d", join_to, crm_join_phase_str(join_node->join), current_join_id); return; } /* Update the element with the node's name and UUID, in case they * weren't known before */ crm_trace("Updating node name and UUID in CIB for %s", join_to); tmp1 = create_xml_node(NULL, PCMK_XE_NODE); crm_xml_add(tmp1, PCMK_XA_ID, crm_peer_uuid(join_node)); crm_xml_add(tmp1, PCMK_XA_UNAME, join_to); fsa_cib_anon_update(PCMK_XE_NODES, tmp1); free_xml(tmp1); if (join_node->join == crm_join_nack_quiet) { crm_trace("Not sending nack message to node %s with feature set older " "than 3.17.0", join_to); return; } join_node = pcmk__get_node(0, join_to, NULL, pcmk__node_search_cluster); if (!crm_is_peer_active(join_node)) { /* * NACK'ing nodes that the membership layer doesn't know about yet * simply creates more churn * * Better to leave them waiting and let the join restart when * the new membership event comes in * * All other NACKs (due to versions etc) should still be processed */ pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_PENDING); return; } // Acknowledge or nack node's join request crm_debug("%sing join-%d request from %s", integrated? "Acknowledg" : "Nack", current_join_id, join_to); acknak = create_dc_message(CRM_OP_JOIN_ACKNAK, join_to); pcmk__xe_set_bool_attr(acknak, CRM_OP_JOIN_ACKNAK, integrated); if (integrated) { // No change needed for a nacked node crm_update_peer_join(__func__, join_node, crm_join_finalized); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER); /* Iterate through the remote peer cache and add information on which * node hosts each to the ACK message. This keeps new controllers in * sync with what has already happened. */ if (crm_remote_peer_cache_size() != 0) { GHashTableIter iter; crm_node_t *node = NULL; xmlNode *remotes = create_xml_node(acknak, PCMK_XE_NODES); g_hash_table_iter_init(&iter, crm_remote_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *remote = NULL; if (!node->conn_host) { continue; } remote = create_xml_node(remotes, PCMK_XE_NODE); pcmk__xe_set_props(remote, PCMK_XA_ID, node->uname, PCMK__XA_NODE_STATE, node->state, PCMK__XA_CONNECTION_HOST, node->conn_host, NULL); } } } send_cluster_message(join_node, crm_msg_crmd, acknak, TRUE); free_xml(acknak); return; } gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source) { static unsigned long long highest_seq = 0; if (controld_globals.membership_id != crm_peer_seq) { crm_debug("join-%d: Membership changed from %llu to %llu " CRM_XS " highest=%llu state=%s for=%s", current_join_id, controld_globals.membership_id, crm_peer_seq, highest_seq, fsa_state2string(cur_state), source); if(highest_seq < crm_peer_seq) { /* Don't spam the FSA with duplicates */ highest_seq = crm_peer_seq; register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } } else if (cur_state == S_INTEGRATION) { if (crmd_join_phase_count(crm_join_welcomed) == 0) { int count = crmd_join_phase_count(crm_join_integrated); crm_debug("join-%d: Integration of %d peer%s complete " CRM_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); register_fsa_input_before(C_FSA_INTERNAL, I_INTEGRATED, NULL); return TRUE; } } else if (cur_state == S_FINALIZE_JOIN) { if (!pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) { crm_debug("join-%d: Delaying finalization until we have CIB " CRM_XS " state=%s for=%s", current_join_id, fsa_state2string(cur_state), source); return TRUE; } else if (crmd_join_phase_count(crm_join_welcomed) != 0) { int count = crmd_join_phase_count(crm_join_welcomed); crm_debug("join-%d: Still waiting on %d welcomed node%s " CRM_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else if (crmd_join_phase_count(crm_join_integrated) != 0) { int count = crmd_join_phase_count(crm_join_integrated); crm_debug("join-%d: Still waiting on %d integrated node%s " CRM_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else if (crmd_join_phase_count(crm_join_finalized) != 0) { int count = crmd_join_phase_count(crm_join_finalized); crm_debug("join-%d: Still waiting on %d finalized node%s " CRM_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else { crm_debug("join-%d: Complete " CRM_XS " state=%s for=%s", current_join_id, fsa_state2string(cur_state), source); register_fsa_input_later(C_FSA_INTERNAL, I_FINALIZED, NULL); return TRUE; } } return FALSE; } void do_dc_join_final(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_debug("Ensuring DC, quorum and node attributes are up-to-date"); crm_update_quorum(crm_have_quorum, TRUE); } int crmd_join_phase_count(enum crm_join_phase phase) { int count = 0; crm_node_t *peer; GHashTableIter iter; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { if(peer->join == phase) { count++; } } return count; } void crmd_join_phase_log(int level) { crm_node_t *peer; GHashTableIter iter; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { do_crm_log(level, "join-%d: %s=%s", current_join_id, peer->uname, crm_join_phase_str(peer->join)); } } diff --git a/daemons/controld/controld_membership.c b/daemons/controld/controld_membership.c index 8e5e7f62a7..02b61327cb 100644 --- a/daemons/controld/controld_membership.c +++ b/daemons/controld/controld_membership.c @@ -1,467 +1,467 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ /* put these first so that uuid_t is defined without conflicts */ #include #include #include #include #include #include #include void post_cache_update(int instance); extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); static void reap_dead_nodes(gpointer key, gpointer value, gpointer user_data) { crm_node_t *node = value; if (crm_is_peer_active(node) == FALSE) { crm_update_peer_join(__func__, node, crm_join_none); if(node && node->uname) { if (pcmk__str_eq(controld_globals.our_nodename, node->uname, pcmk__str_casei)) { crm_err("We're not part of the cluster anymore"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); } else if (!AM_I_DC && pcmk__str_eq(node->uname, controld_globals.dc_name, pcmk__str_casei)) { crm_warn("Our DC node (%s) left the cluster", node->uname); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } } if ((controld_globals.fsa_state == S_INTEGRATION) || (controld_globals.fsa_state == S_FINALIZE_JOIN)) { check_join_state(controld_globals.fsa_state, __func__); } if ((node != NULL) && (node->uuid != NULL)) { fail_incompletable_actions(controld_globals.transition_graph, node->uuid); } } } void post_cache_update(int instance) { xmlNode *no_op = NULL; crm_peer_seq = instance; crm_debug("Updated cache after membership event %d.", instance); g_hash_table_foreach(crm_peer_cache, reap_dead_nodes, NULL); controld_set_fsa_input_flags(R_MEMBERSHIP); if (AM_I_DC) { populate_cib_nodes(node_update_quick | node_update_cluster | node_update_peer | node_update_expected, __func__); } /* * If we lost nodes, we should re-check the election status * Safe to call outside of an election */ controld_set_fsa_action_flags(A_ELECTION_CHECK); controld_trigger_fsa(); /* Membership changed, remind everyone we're here. * This will aid detection of duplicate DCs */ no_op = create_request(CRM_OP_NOOP, NULL, NULL, CRM_SYSTEM_CRMD, AM_I_DC ? CRM_SYSTEM_DC : CRM_SYSTEM_CRMD, NULL); send_cluster_message(NULL, crm_msg_crmd, no_op, FALSE); free_xml(no_op); } static void crmd_node_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if (rc == pcmk_ok) { crm_trace("Node update %d complete", call_id); } else if(call_id < pcmk_ok) { crm_err("Node update failed: %s (%d)", pcmk_strerror(call_id), call_id); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else { crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } /*! * \internal * \brief Create an XML node state tag with updates * * \param[in,out] node Node whose state will be used for update * \param[in] flags Bitmask of node_update_flags indicating what to update * \param[in,out] parent XML node to contain update (or NULL) * \param[in] source Who requested the update (only used for logging) * * \return Pointer to created node state tag */ xmlNode * create_node_state_update(crm_node_t *node, int flags, xmlNode *parent, const char *source) { const char *value = NULL; xmlNode *node_state; if (!node->state) { crm_info("Node update for %s cancelled: no state, not seen yet", node->uname); return NULL; } node_state = create_xml_node(parent, PCMK__XE_NODE_STATE); if (pcmk_is_set(node->flags, crm_remote_node)) { pcmk__xe_set_bool_attr(node_state, PCMK_XA_REMOTE_NODE, true); } if (crm_xml_add(node_state, PCMK_XA_ID, crm_peer_uuid(node)) == NULL) { crm_info("Node update for %s cancelled: no ID", node->uname); free_xml(node_state); return NULL; } crm_xml_add(node_state, PCMK_XA_UNAME, node->uname); if ((flags & node_update_cluster) && node->state) { if (compare_version(controld_globals.dc_version, "3.18.0") >= 0) { // A value 0 means the node is not a cluster member. crm_xml_add_ll(node_state, PCMK__XA_IN_CCM, node->when_member); } else { pcmk__xe_set_bool_attr(node_state, PCMK__XA_IN_CCM, pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei)); } } if (!pcmk_is_set(node->flags, crm_remote_node)) { if (flags & node_update_peer) { if (compare_version(controld_globals.dc_version, "3.18.0") >= 0) { // A value 0 means the peer is offline in CPG. crm_xml_add_ll(node_state, PCMK_XA_CRMD, node->when_online); } else { // @COMPAT DCs < 2.1.7 use online/offline rather than timestamp value = PCMK_VALUE_OFFLINE; if (pcmk_is_set(node->processes, crm_get_cluster_proc())) { value = PCMK_VALUE_ONLINE; } crm_xml_add(node_state, PCMK_XA_CRMD, value); } } if (flags & node_update_join) { if (node->join <= crm_join_none) { value = CRMD_JOINSTATE_DOWN; } else { value = CRMD_JOINSTATE_MEMBER; } crm_xml_add(node_state, PCMK__XA_JOIN, value); } if (flags & node_update_expected) { crm_xml_add(node_state, PCMK_XA_EXPECTED, node->expected); } } crm_xml_add(node_state, PCMK_XA_CRM_DEBUG_ORIGIN, source); return node_state; } static void remove_conflicting_node_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { char *node_uuid = user_data; do_crm_log_unlikely(rc == 0 ? LOG_DEBUG : LOG_NOTICE, "Deletion of the unknown conflicting node \"%s\": %s (rc=%d)", node_uuid, pcmk_strerror(rc), rc); } static void search_conflicting_node_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { char *new_node_uuid = user_data; xmlNode *node_xml = NULL; if (rc != pcmk_ok) { if (rc != -ENXIO) { crm_notice("Searching conflicting nodes for %s failed: %s (%d)", new_node_uuid, pcmk_strerror(rc), rc); } return; } else if (output == NULL) { return; } if (pcmk__xe_is(output, PCMK_XE_NODE)) { node_xml = output; } else { node_xml = pcmk__xml_first_child(output); } for (; node_xml != NULL; node_xml = pcmk__xml_next(node_xml)) { const char *node_uuid = NULL; const char *node_uname = NULL; GHashTableIter iter; crm_node_t *node = NULL; gboolean known = FALSE; if (!pcmk__xe_is(node_xml, PCMK_XE_NODE)) { continue; } node_uuid = crm_element_value(node_xml, PCMK_XA_ID); node_uname = crm_element_value(node_xml, PCMK_XA_UNAME); if (node_uuid == NULL || node_uname == NULL) { continue; } g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (node->uuid && pcmk__str_eq(node->uuid, node_uuid, pcmk__str_casei) && node->uname && pcmk__str_eq(node->uname, node_uname, pcmk__str_casei)) { known = TRUE; break; } } if (known == FALSE) { cib_t *cib_conn = controld_globals.cib_conn; int delete_call_id = 0; xmlNode *node_state_xml = NULL; crm_notice("Deleting unknown node %s/%s which has conflicting uname with %s", node_uuid, node_uname, new_node_uuid); delete_call_id = cib_conn->cmds->remove(cib_conn, PCMK_XE_NODES, node_xml, cib_scope_local); fsa_register_cib_callback(delete_call_id, strdup(node_uuid), remove_conflicting_node_callback); node_state_xml = create_xml_node(NULL, PCMK__XE_NODE_STATE); crm_xml_add(node_state_xml, PCMK_XA_ID, node_uuid); crm_xml_add(node_state_xml, PCMK_XA_UNAME, node_uname); delete_call_id = cib_conn->cmds->remove(cib_conn, PCMK_XE_STATUS, node_state_xml, cib_scope_local); fsa_register_cib_callback(delete_call_id, strdup(node_uuid), remove_conflicting_node_callback); free_xml(node_state_xml); } } } static void node_list_update_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if(call_id < pcmk_ok) { crm_err("Node list update failed: %s (%d)", pcmk_strerror(call_id), call_id); crm_log_xml_debug(msg, "update:failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else if(rc < pcmk_ok) { crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "update:failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } void populate_cib_nodes(enum node_update_flags flags, const char *source) { cib_t *cib_conn = controld_globals.cib_conn; int call_id = 0; gboolean from_hashtable = TRUE; xmlNode *node_list = create_xml_node(NULL, PCMK_XE_NODES); #if SUPPORT_COROSYNC if (!pcmk_is_set(flags, node_update_quick) && is_corosync_cluster()) { from_hashtable = pcmk__corosync_add_nodes(node_list); } #endif if (from_hashtable) { GHashTableIter iter; crm_node_t *node = NULL; GString *xpath = NULL; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *new_node = NULL; if ((node->uuid != NULL) && (node->uname != NULL)) { crm_trace("Creating node entry for %s/%s", node->uname, node->uuid); if (xpath == NULL) { xpath = g_string_sized_new(512); } else { g_string_truncate(xpath, 0); } /* We need both to be valid */ new_node = create_xml_node(node_list, PCMK_XE_NODE); crm_xml_add(new_node, PCMK_XA_ID, node->uuid); crm_xml_add(new_node, PCMK_XA_UNAME, node->uname); /* Search and remove unknown nodes with the conflicting uname from CIB */ pcmk__g_strcat(xpath, "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_NODES "/" PCMK_XE_NODE "[@" PCMK_XA_UNAME "='", node->uname, "']" "[@" PCMK_XA_ID "!='", node->uuid, "']", NULL); call_id = cib_conn->cmds->query(cib_conn, (const char *) xpath->str, NULL, cib_scope_local|cib_xpath); fsa_register_cib_callback(call_id, strdup(node->uuid), search_conflicting_node_callback); } } if (xpath != NULL) { g_string_free(xpath, TRUE); } } crm_trace("Populating section from %s", from_hashtable ? "hashtable" : "cluster"); if ((controld_update_cib(PCMK_XE_NODES, node_list, cib_scope_local, node_list_update_callback) == pcmk_rc_ok) && (crm_peer_cache != NULL) && AM_I_DC) { /* * There is no need to update the local CIB with our values if * we've not seen valid membership data */ GHashTableIter iter; crm_node_t *node = NULL; free_xml(node_list); node_list = create_xml_node(NULL, PCMK_XE_STATUS); g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { create_node_state_update(node, flags, node_list, source); } if (crm_remote_peer_cache) { g_hash_table_iter_init(&iter, crm_remote_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { create_node_state_update(node, flags, node_list, source); } } controld_update_cib(PCMK_XE_STATUS, node_list, cib_scope_local, crmd_node_update_complete); } free_xml(node_list); } static void cib_quorum_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if (rc == pcmk_ok) { crm_trace("Quorum update %d complete", call_id); } else { crm_err("Quorum update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } void crm_update_quorum(gboolean quorum, gboolean force_update) { bool has_quorum = pcmk_is_set(controld_globals.flags, controld_has_quorum); if (quorum) { controld_set_global_flags(controld_ever_had_quorum); } else if (pcmk_all_flags_set(controld_globals.flags, controld_ever_had_quorum |controld_no_quorum_suicide)) { pcmk__panic(__func__); } if (AM_I_DC && ((has_quorum && !quorum) || (!has_quorum && quorum) || force_update)) { xmlNode *update = NULL; update = create_xml_node(NULL, PCMK_XE_CIB); crm_xml_add_int(update, PCMK_XA_HAVE_QUORUM, quorum); crm_xml_add(update, PCMK_XA_DC_UUID, controld_globals.our_uuid); crm_debug("Updating quorum status to %s", pcmk__btoa(quorum)); controld_update_cib(PCMK_XE_CIB, update, cib_scope_local, cib_quorum_update_complete); free_xml(update); /* Quorum changes usually cause a new transition via other activity: * quorum gained via a node joining will abort via the node join, * and quorum lost via a node leaving will usually abort via resource * activity and/or fencing. * * However, it is possible that nothing else causes a transition (e.g. * someone forces quorum via corosync-cmaptcl, or quorum is lost due to * a node in standby shutting down cleanly), so here ensure a new * transition is triggered. */ if (quorum) { /* If quorum was gained, abort after a short delay, in case multiple * nodes are joining around the same time, so the one that brings us * to quorum doesn't cause all the remaining ones to be fenced. */ - abort_after_delay(INFINITY, pcmk__graph_restart, "Quorum gained", - 5000); + abort_after_delay(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Quorum gained", 5000); } else { - abort_transition(INFINITY, pcmk__graph_restart, "Quorum lost", - NULL); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Quorum lost", NULL); } } if (quorum) { controld_set_global_flags(controld_has_quorum); } else { controld_clear_global_flags(controld_has_quorum); } } diff --git a/daemons/controld/controld_te_callbacks.c b/daemons/controld/controld_te_callbacks.c index 44626b57ae..c52482268c 100644 --- a/daemons/controld/controld_te_callbacks.c +++ b/daemons/controld/controld_te_callbacks.c @@ -1,700 +1,704 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include void te_update_confirm(const char *event, xmlNode * msg); #define RSC_OP_PREFIX "//" PCMK__XE_DIFF_ADDED "//" PCMK_XE_CIB \ "//" PCMK__XE_LRM_RSC_OP "[@" PCMK_XA_ID "='" // An explicit PCMK_OPT_SHUTDOWN_LOCK of 0 means the lock has been cleared static bool shutdown_lock_cleared(xmlNode *lrm_resource) { time_t shutdown_lock = 0; return (crm_element_value_epoch(lrm_resource, PCMK_OPT_SHUTDOWN_LOCK, &shutdown_lock) == pcmk_ok) && (shutdown_lock == 0); } static void te_update_diff_v1(const char *event, xmlNode *diff) { int lpc, max; xmlXPathObject *xpathObj = NULL; GString *rsc_op_xpath = NULL; CRM_CHECK(diff != NULL, return); pcmk__output_set_log_level(controld_globals.logger_out, LOG_TRACE); controld_globals.logger_out->message(controld_globals.logger_out, "xml-patchset", diff); if (cib__config_changed_v1(NULL, NULL, &diff)) { - abort_transition(INFINITY, pcmk__graph_restart, "Non-status change", - diff); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Non-status change", diff); goto bail; /* configuration changed */ } /* Tickets Attributes - Added/Updated */ xpathObj = xpath_search(diff, "//" PCMK__XA_CIB_UPDATE_RESULT "//" PCMK__XE_DIFF_ADDED "//" PCMK_XE_TICKETS); if (numXpathResults(xpathObj) > 0) { xmlNode *aborted = getXpathResult(xpathObj, 0); - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Ticket attribute: update", aborted); goto bail; } freeXpathObject(xpathObj); /* Tickets Attributes - Removed */ xpathObj = xpath_search(diff, "//" PCMK__XA_CIB_UPDATE_RESULT "//" PCMK__XE_DIFF_REMOVED "//" PCMK_XE_TICKETS); if (numXpathResults(xpathObj) > 0) { xmlNode *aborted = getXpathResult(xpathObj, 0); - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Ticket attribute: removal", aborted); goto bail; } freeXpathObject(xpathObj); /* Transient Attributes - Removed */ xpathObj = xpath_search(diff, "//" PCMK__XA_CIB_UPDATE_RESULT "//" PCMK__XE_DIFF_REMOVED "//" PCMK__XE_TRANSIENT_ATTRIBUTES); if (numXpathResults(xpathObj) > 0) { xmlNode *aborted = getXpathResult(xpathObj, 0); - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Transient attribute: removal", aborted); goto bail; } freeXpathObject(xpathObj); // Check for PCMK__XE_LRM_RESOURCE entries xpathObj = xpath_search(diff, "//" PCMK__XA_CIB_UPDATE_RESULT "//" PCMK__XE_DIFF_ADDED "//" PCMK__XE_LRM_RESOURCE); max = numXpathResults(xpathObj); /* * Updates by, or in response to, graph actions will never affect more than * one resource at a time, so such updates indicate an LRM refresh. In that * case, start a new transition rather than check each result individually, * which can result in _huge_ speedups in large clusters. * * Unfortunately, we can only do so when there are no pending actions. * Otherwise, we could mistakenly throw away those results here, and * the cluster will stall waiting for them and time out the operation. */ if ((controld_globals.transition_graph->pending == 0) && (max > 1)) { crm_debug("Ignoring resource operation updates due to history refresh of %d resources", max); crm_log_xml_trace(diff, "lrm-refresh"); - abort_transition(INFINITY, pcmk__graph_restart, "History refresh", - NULL); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "History refresh", NULL); goto bail; } if (max == 1) { xmlNode *lrm_resource = getXpathResult(xpathObj, 0); if (shutdown_lock_cleared(lrm_resource)) { // @TODO would be more efficient to abort once after transition done - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Shutdown lock cleared", lrm_resource); // Still process results, so we stop timers and update failcounts } } freeXpathObject(xpathObj); /* Process operation updates */ xpathObj = xpath_search(diff, "//" PCMK__XA_CIB_UPDATE_RESULT "//" PCMK__XE_DIFF_ADDED "//" PCMK__XE_LRM_RSC_OP); max = numXpathResults(xpathObj); if (max > 0) { int lpc = 0; for (lpc = 0; lpc < max; lpc++) { xmlNode *rsc_op = getXpathResult(xpathObj, lpc); const char *node = get_node_id(rsc_op); process_graph_event(rsc_op, node); } } freeXpathObject(xpathObj); /* Detect deleted (as opposed to replaced or added) actions - eg. crm_resource -C */ xpathObj = xpath_search(diff, "//" PCMK__XE_DIFF_REMOVED "//" PCMK__XE_LRM_RSC_OP); max = numXpathResults(xpathObj); for (lpc = 0; lpc < max; lpc++) { const char *op_id = NULL; xmlXPathObject *op_match = NULL; xmlNode *match = getXpathResult(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if(match == NULL) { continue; }; op_id = pcmk__xe_id(match); if (rsc_op_xpath == NULL) { rsc_op_xpath = g_string_new(RSC_OP_PREFIX); } else { g_string_truncate(rsc_op_xpath, sizeof(RSC_OP_PREFIX) - 1); } pcmk__g_strcat(rsc_op_xpath, op_id, "']", NULL); op_match = xpath_search(diff, (const char *) rsc_op_xpath->str); if (numXpathResults(op_match) == 0) { /* Prevent false positives by matching cancelations too */ const char *node = get_node_id(match); pcmk__graph_action_t *cancelled = get_cancel_action(op_id, node); if (cancelled == NULL) { crm_debug("No match for deleted action %s (%s on %s)", (const char *) rsc_op_xpath->str, op_id, node); - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Resource op removal", match); freeXpathObject(op_match); goto bail; } else { crm_debug("Deleted " PCMK__XE_LRM_RSC_OP " %s on %s was for " "graph event %d", op_id, node, cancelled->id); } } freeXpathObject(op_match); } bail: freeXpathObject(xpathObj); if (rsc_op_xpath != NULL) { g_string_free(rsc_op_xpath, TRUE); } } static void process_lrm_resource_diff(xmlNode *lrm_resource, const char *node) { for (xmlNode *rsc_op = pcmk__xml_first_child(lrm_resource); rsc_op != NULL; rsc_op = pcmk__xml_next(rsc_op)) { process_graph_event(rsc_op, node); } if (shutdown_lock_cleared(lrm_resource)) { // @TODO would be more efficient to abort once after transition done - abort_transition(INFINITY, pcmk__graph_restart, "Shutdown lock cleared", - lrm_resource); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Shutdown lock cleared", lrm_resource); } } static void process_resource_updates(const char *node, xmlNode *xml, xmlNode *change, const char *op, const char *xpath) { xmlNode *rsc = NULL; if (xml == NULL) { return; } if (pcmk__xe_is(xml, PCMK__XE_LRM)) { xml = first_named_child(xml, PCMK__XE_LRM_RESOURCES); CRM_CHECK(xml != NULL, return); } CRM_CHECK(pcmk__xe_is(xml, PCMK__XE_LRM_RESOURCES), return); /* * Updates by, or in response to, TE actions will never contain updates * for more than one resource at a time, so such updates indicate an * LRM refresh. * * In that case, start a new transition rather than check each result * individually, which can result in _huge_ speedups in large clusters. * * Unfortunately, we can only do so when there are no pending actions. * Otherwise, we could mistakenly throw away those results here, and * the cluster will stall waiting for them and time out the operation. */ if ((controld_globals.transition_graph->pending == 0) && (xml->children != NULL) && (xml->children->next != NULL)) { crm_log_xml_trace(change, "lrm-refresh"); - abort_transition(INFINITY, pcmk__graph_restart, "History refresh", - NULL); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "History refresh", NULL); return; } for (rsc = pcmk__xml_first_child(xml); rsc != NULL; rsc = pcmk__xml_next(rsc)) { crm_trace("Processing %s", pcmk__xe_id(rsc)); process_lrm_resource_diff(rsc, node); } } static char *extract_node_uuid(const char *xpath) { char *mutable_path = strdup(xpath); char *node_uuid = NULL; char *search = NULL; char *match = NULL; match = strstr(mutable_path, PCMK__XE_NODE_STATE "[@" PCMK_XA_ID "=\'"); if (match == NULL) { free(mutable_path); return NULL; } match += strlen(PCMK__XE_NODE_STATE "[@" PCMK_XA_ID "=\'"); search = strchr(match, '\''); if (search == NULL) { free(mutable_path); return NULL; } search[0] = 0; node_uuid = strdup(match); free(mutable_path); return node_uuid; } static void abort_unless_down(const char *xpath, const char *op, xmlNode *change, const char *reason) { char *node_uuid = NULL; pcmk__graph_action_t *down = NULL; if (!pcmk__str_eq(op, PCMK_VALUE_DELETE, pcmk__str_none)) { - abort_transition(INFINITY, pcmk__graph_restart, reason, change); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, reason, + change); return; } node_uuid = extract_node_uuid(xpath); if(node_uuid == NULL) { crm_err("Could not extract node ID from %s", xpath); - abort_transition(INFINITY, pcmk__graph_restart, reason, change); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, reason, + change); return; } down = match_down_event(node_uuid); if (down == NULL) { crm_trace("Not expecting %s to be down (%s)", node_uuid, xpath); - abort_transition(INFINITY, pcmk__graph_restart, reason, change); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, reason, + change); } else { crm_trace("Expecting changes to %s (%s)", node_uuid, xpath); } free(node_uuid); } static void process_op_deletion(const char *xpath, xmlNode *change) { char *mutable_key = strdup(xpath); char *key; char *node_uuid; // Extract the part of xpath between last pair of single quotes key = strrchr(mutable_key, '\''); if (key != NULL) { *key = '\0'; key = strrchr(mutable_key, '\''); } if (key == NULL) { crm_warn("Ignoring malformed CIB update (resource deletion of %s)", xpath); free(mutable_key); return; } ++key; node_uuid = extract_node_uuid(xpath); if (confirm_cancel_action(key, node_uuid) == FALSE) { - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Resource operation removal", change); } free(mutable_key); free(node_uuid); } static void process_delete_diff(const char *xpath, const char *op, xmlNode *change) { if (strstr(xpath, "/" PCMK__XE_LRM_RSC_OP "[")) { process_op_deletion(xpath, change); } else if (strstr(xpath, "/" PCMK__XE_LRM "[")) { abort_unless_down(xpath, op, change, "Resource state removal"); } else if (strstr(xpath, "/" PCMK__XE_NODE_STATE "[")) { abort_unless_down(xpath, op, change, "Node state removal"); } else { crm_trace("Ignoring delete of %s", xpath); } } static void process_node_state_diff(xmlNode *state, xmlNode *change, const char *op, const char *xpath) { xmlNode *lrm = first_named_child(state, PCMK__XE_LRM); process_resource_updates(pcmk__xe_id(state), lrm, change, op, xpath); } static void process_status_diff(xmlNode *status, xmlNode *change, const char *op, const char *xpath) { for (xmlNode *state = pcmk__xml_first_child(status); state != NULL; state = pcmk__xml_next(state)) { process_node_state_diff(state, change, op, xpath); } } static void process_cib_diff(xmlNode *cib, xmlNode *change, const char *op, const char *xpath) { xmlNode *status = first_named_child(cib, PCMK_XE_STATUS); xmlNode *config = first_named_child(cib, PCMK_XE_CONFIGURATION); if (status) { process_status_diff(status, change, op, xpath); } if (config) { - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Non-status-only change", change); } } static void te_update_diff_v2(xmlNode *diff) { crm_log_xml_trace(diff, "Patch:Raw"); for (xmlNode *change = pcmk__xml_first_child(diff); change != NULL; change = pcmk__xml_next(change)) { xmlNode *match = NULL; const char *name = NULL; const char *xpath = crm_element_value(change, PCMK_XA_PATH); // Possible ops: create, modify, delete, move const char *op = crm_element_value(change, PCMK_XA_OPERATION); // Ignore uninteresting updates if (op == NULL) { continue; } else if (xpath == NULL) { crm_trace("Ignoring %s change for version field", op); continue; } else if ((strcmp(op, PCMK_VALUE_MOVE) == 0) && (strstr(xpath, "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES) == NULL)) { /* We still need to consider moves within the resources section, * since they affect placement order. */ crm_trace("Ignoring move change at %s", xpath); continue; } // Find the result of create/modify ops if (strcmp(op, PCMK_VALUE_CREATE) == 0) { match = change->children; } else if (strcmp(op, PCMK_VALUE_MODIFY) == 0) { match = first_named_child(change, PCMK_XE_CHANGE_RESULT); if(match) { match = match->children; } } else if (!pcmk__str_any_of(op, PCMK_VALUE_DELETE, PCMK_VALUE_MOVE, NULL)) { crm_warn("Ignoring malformed CIB update (%s operation on %s is unrecognized)", op, xpath); continue; } if (match) { if (match->type == XML_COMMENT_NODE) { crm_trace("Ignoring %s operation for comment at %s", op, xpath); continue; } name = (const char *)match->name; } crm_trace("Handling %s operation for %s%s%s", op, (xpath? xpath : "CIB"), (name? " matched by " : ""), (name? name : "")); if (strstr(xpath, "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION)) { - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Configuration change", change); break; // Won't be packaged with operation results we may be waiting for } else if (strstr(xpath, "/" PCMK_XE_TICKETS) || pcmk__str_eq(name, PCMK_XE_TICKETS, pcmk__str_none)) { - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Ticket attribute change", change); break; // Won't be packaged with operation results we may be waiting for } else if (strstr(xpath, "/" PCMK__XE_TRANSIENT_ATTRIBUTES "[") || pcmk__str_eq(name, PCMK__XE_TRANSIENT_ATTRIBUTES, pcmk__str_none)) { abort_unless_down(xpath, op, change, "Transient attribute change"); break; // Won't be packaged with operation results we may be waiting for } else if (strcmp(op, PCMK_VALUE_DELETE) == 0) { process_delete_diff(xpath, op, change); } else if (name == NULL) { crm_warn("Ignoring malformed CIB update (%s at %s has no result)", op, xpath); } else if (strcmp(name, PCMK_XE_CIB) == 0) { process_cib_diff(match, change, op, xpath); } else if (strcmp(name, PCMK_XE_STATUS) == 0) { process_status_diff(match, change, op, xpath); } else if (strcmp(name, PCMK__XE_NODE_STATE) == 0) { process_node_state_diff(match, change, op, xpath); } else if (strcmp(name, PCMK__XE_LRM) == 0) { process_resource_updates(pcmk__xe_id(match), match, change, op, xpath); } else if (strcmp(name, PCMK__XE_LRM_RESOURCES) == 0) { char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM); process_resource_updates(local_node, match, change, op, xpath); free(local_node); } else if (strcmp(name, PCMK__XE_LRM_RESOURCE) == 0) { char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM); process_lrm_resource_diff(match, local_node); free(local_node); } else if (strcmp(name, PCMK__XE_LRM_RSC_OP) == 0) { char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM); process_graph_event(match, local_node); free(local_node); } else { crm_warn("Ignoring malformed CIB update (%s at %s has unrecognized result %s)", op, xpath, name); } } } void te_update_diff(const char *event, xmlNode * msg) { xmlNode *diff = NULL; const char *op = NULL; int rc = -EINVAL; int format = 1; int p_add[] = { 0, 0, 0 }; int p_del[] = { 0, 0, 0 }; CRM_CHECK(msg != NULL, return); crm_element_value_int(msg, PCMK__XA_CIB_RC, &rc); if (controld_globals.transition_graph == NULL) { crm_trace("No graph"); return; } else if (rc < pcmk_ok) { crm_trace("Filter rc=%d (%s)", rc, pcmk_strerror(rc)); return; } else if (controld_globals.transition_graph->complete && (controld_globals.fsa_state != S_IDLE) && (controld_globals.fsa_state != S_TRANSITION_ENGINE) && (controld_globals.fsa_state != S_POLICY_ENGINE)) { crm_trace("Filter state=%s (complete)", fsa_state2string(controld_globals.fsa_state)); return; } op = crm_element_value(msg, PCMK__XA_CIB_OP); diff = get_message_xml(msg, PCMK__XA_CIB_UPDATE_RESULT); xml_patch_versions(diff, p_add, p_del); crm_debug("Processing (%s) diff: %d.%d.%d -> %d.%d.%d (%s)", op, p_del[0], p_del[1], p_del[2], p_add[0], p_add[1], p_add[2], fsa_state2string(controld_globals.fsa_state)); crm_element_value_int(diff, PCMK_XA_FORMAT, &format); switch (format) { case 1: te_update_diff_v1(event, diff); break; case 2: te_update_diff_v2(diff); break; default: crm_warn("Ignoring malformed CIB update (unknown patch format %d)", format); } controld_remove_all_outside_events(); } void process_te_message(xmlNode * msg, xmlNode * xml_data) { const char *value = NULL; xmlXPathObject *xpathObj = NULL; int nmatches = 0; CRM_CHECK(msg != NULL, return); // Transition requests must specify transition engine as subsystem value = crm_element_value(msg, PCMK__XA_CRM_SYS_TO); if (pcmk__str_empty(value) || !pcmk__str_eq(value, CRM_SYSTEM_TENGINE, pcmk__str_none)) { crm_info("Received invalid transition request: subsystem '%s' not '" CRM_SYSTEM_TENGINE "'", pcmk__s(value, "")); return; } // Only the lrm_invoke command is supported as a transition request value = crm_element_value(msg, PCMK__XA_CRM_TASK); if (!pcmk__str_eq(value, CRM_OP_INVOKE_LRM, pcmk__str_none)) { crm_info("Received invalid transition request: command '%s' not '" CRM_OP_INVOKE_LRM "'", pcmk__s(value, "")); return; } // Transition requests must be marked as coming from the executor value = crm_element_value(msg, PCMK__XA_CRM_SYS_FROM); if (!pcmk__str_eq(value, CRM_SYSTEM_LRMD, pcmk__str_none)) { crm_info("Received invalid transition request: from '%s' not '" CRM_SYSTEM_LRMD "'", pcmk__s(value, "")); return; } crm_debug("Processing transition request with ref='%s' origin='%s'", pcmk__s(crm_element_value(msg, PCMK_XA_REFERENCE), ""), pcmk__s(crm_element_value(msg, PCMK__XA_SRC), "")); xpathObj = xpath_search(xml_data, "//" PCMK__XE_LRM_RSC_OP); nmatches = numXpathResults(xpathObj); if (nmatches == 0) { crm_err("Received transition request with no results (bug?)"); } else { for (int lpc = 0; lpc < nmatches; lpc++) { xmlNode *rsc_op = getXpathResult(xpathObj, lpc); const char *node = get_node_id(rsc_op); process_graph_event(rsc_op, node); } } freeXpathObject(xpathObj); } void cib_action_updated(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { if (rc < pcmk_ok) { crm_err("Update %d FAILED: %s", call_id, pcmk_strerror(rc)); } } /*! * \brief Handle a timeout in node-to-node communication * * \param[in,out] data Pointer to graph action * * \return FALSE (indicating that source should be not be re-added) */ gboolean action_timer_callback(gpointer data) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) data; const char *task = NULL; const char *on_node = NULL; const char *via_node = NULL; CRM_CHECK(data != NULL, return FALSE); stop_te_timer(action); task = crm_element_value(action->xml, PCMK_XA_OPERATION); on_node = crm_element_value(action->xml, PCMK__META_ON_NODE); via_node = crm_element_value(action->xml, PCMK__XA_ROUTER_NODE); if (controld_globals.transition_graph->complete) { crm_notice("Node %s did not send %s result (via %s) within %dms " "(ignoring because transition not in progress)", (on_node? on_node : ""), (task? task : "unknown action"), (via_node? via_node : "controller"), action->timeout); } else { /* fail the action */ crm_err("Node %s did not send %s result (via %s) within %dms " "(action timeout plus " PCMK_OPT_CLUSTER_DELAY ")", (on_node? on_node : ""), (task? task : "unknown action"), (via_node? via_node : "controller"), (action->timeout + controld_globals.transition_graph->network_delay)); pcmk__log_graph_action(LOG_ERR, action); pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); te_action_confirmed(action, controld_globals.transition_graph); - abort_transition(INFINITY, pcmk__graph_restart, "Action lost", NULL); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Action lost", NULL); // Record timeout in the CIB if appropriate if ((action->type == pcmk__rsc_graph_action) && controld_action_is_recordable(task)) { controld_record_action_timeout(action); } } return FALSE; } diff --git a/daemons/controld/controld_te_events.c b/daemons/controld/controld_te_events.c index 6f71b80441..93bd95cb2d 100644 --- a/daemons/controld/controld_te_events.c +++ b/daemons/controld/controld_te_events.c @@ -1,611 +1,614 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include /*! * \internal * \brief Action numbers of outside events processed in current update diff * * This table is to be used as a set. It should be empty when the transitioner * begins processing a CIB update diff. It ensures that if there are multiple * events (for example, "_last_0" and "_last_failure_0") for the same action, * only one of them updates the failcount. Events that originate outside the * cluster can't be confirmed, since they're not in the transition graph. */ static GHashTable *outside_events = NULL; /*! * \internal * \brief Empty the hash table containing action numbers of outside events */ void controld_remove_all_outside_events(void) { if (outside_events != NULL) { g_hash_table_remove_all(outside_events); } } /*! * \internal * \brief Destroy the hash table containing action numbers of outside events */ void controld_destroy_outside_events_table(void) { if (outside_events != NULL) { g_hash_table_destroy(outside_events); outside_events = NULL; } } /*! * \internal * \brief Add an outside event's action number to a set * * \return Standard Pacemaker return code. Specifically, \p pcmk_rc_ok if the * event was not already in the set, or \p pcmk_rc_already otherwise. */ static int record_outside_event(gint action_num) { if (outside_events == NULL) { outside_events = g_hash_table_new(NULL, NULL); } if (g_hash_table_add(outside_events, GINT_TO_POINTER(action_num))) { return pcmk_rc_ok; } return pcmk_rc_already; } gboolean fail_incompletable_actions(pcmk__graph_t *graph, const char *down_node) { const char *target_uuid = NULL; const char *router = NULL; const char *router_uuid = NULL; xmlNode *last_action = NULL; GList *gIter = NULL; GList *gIter2 = NULL; if (graph == NULL || graph->complete) { return FALSE; } gIter = graph->synapses; for (; gIter != NULL; gIter = gIter->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) gIter->data; if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed|pcmk__synapse_failed)) { /* We've already been here */ continue; } gIter2 = synapse->actions; for (; gIter2 != NULL; gIter2 = gIter2->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) gIter2->data; if ((action->type == pcmk__pseudo_graph_action) || pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { continue; } else if (action->type == pcmk__cluster_graph_action) { const char *task = crm_element_value(action->xml, PCMK_XA_OPERATION); if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) { continue; } } target_uuid = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID); router = crm_element_value(action->xml, PCMK__XA_ROUTER_NODE); if (router) { crm_node_t *node = pcmk__get_node(0, router, NULL, pcmk__node_search_cluster); if (node) { router_uuid = node->uuid; } } if (pcmk__str_eq(target_uuid, down_node, pcmk__str_casei) || pcmk__str_eq(router_uuid, down_node, pcmk__str_casei)) { pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); pcmk__set_synapse_flags(synapse, pcmk__synapse_failed); last_action = action->xml; stop_te_timer(action); pcmk__update_graph(graph, action); if (pcmk_is_set(synapse->flags, pcmk__synapse_executed)) { crm_notice("Action %d (%s) was pending on %s (offline)", action->id, crm_element_value(action->xml, PCMK__XA_OPERATION_KEY), down_node); } else { crm_info("Action %d (%s) is scheduled for %s (offline)", action->id, crm_element_value(action->xml, PCMK__XA_OPERATION_KEY), down_node); } } } } if (last_action != NULL) { crm_info("Node %s shutdown resulted in un-runnable actions", down_node); - abort_transition(INFINITY, pcmk__graph_restart, "Node failure", - last_action); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Node failure", last_action); return TRUE; } return FALSE; } /*! * \internal * \brief Update failure-related node attributes if warranted * * \param[in] event XML describing operation that (maybe) failed * \param[in] event_node_uuid Node that event occurred on * \param[in] rc Actual operation return code * \param[in] target_rc Expected operation return code * \param[in] do_update If TRUE, do update regardless of operation type * \param[in] ignore_failures If TRUE, update last failure but not fail count * * \return TRUE if this was not a direct nack, success or lrm status refresh */ static gboolean update_failcount(const xmlNode *event, const char *event_node_uuid, int rc, int target_rc, gboolean do_update, gboolean ignore_failures) { guint interval_ms = 0; char *task = NULL; char *rsc_id = NULL; const char *value = NULL; const char *id = crm_element_value(event, PCMK__XA_OPERATION_KEY); const char *on_uname = crm_peer_uname(event_node_uuid); const char *origin = crm_element_value(event, PCMK_XA_CRM_DEBUG_ORIGIN); // Nothing needs to be done for success or status refresh if (rc == target_rc) { return FALSE; } else if (pcmk__str_eq(origin, "build_active_RAs", pcmk__str_casei)) { crm_debug("No update for %s (rc=%d) on %s: Old failure from lrm status refresh", id, rc, on_uname); return FALSE; } /* Sanity check */ CRM_CHECK(on_uname != NULL, return TRUE); CRM_CHECK(parse_op_key(id, &rsc_id, &task, &interval_ms), crm_err("Couldn't parse: %s", pcmk__xe_id(event)); goto bail); /* Decide whether update is necessary and what value to use */ if ((interval_ms > 0) || pcmk__str_eq(task, PCMK_ACTION_PROMOTE, pcmk__str_none) || pcmk__str_eq(task, PCMK_ACTION_DEMOTE, pcmk__str_none)) { do_update = TRUE; } else if (pcmk__str_eq(task, PCMK_ACTION_START, pcmk__str_none)) { do_update = TRUE; value = pcmk__s(controld_globals.transition_graph->failed_start_offset, PCMK_VALUE_INFINITY); } else if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_none)) { do_update = TRUE; value = pcmk__s(controld_globals.transition_graph->failed_stop_offset, PCMK_VALUE_INFINITY); } if (do_update) { pcmk__attrd_query_pair_t *fail_pair = NULL; pcmk__attrd_query_pair_t *last_pair = NULL; char *fail_name = NULL; char *last_name = NULL; GList *attrs = NULL; uint32_t opts = pcmk__node_attr_none; char *now = pcmk__ttoa(time(NULL)); // Fail count will be either incremented or set to infinity if (!pcmk_str_is_infinity(value)) { value = PCMK_XA_VALUE "++"; } if (g_hash_table_lookup(crm_remote_peer_cache, event_node_uuid)) { opts |= pcmk__node_attr_remote; } crm_info("Updating %s for %s on %s after failed %s: rc=%d (update=%s, time=%s)", (ignore_failures? "last failure" : "failcount"), rsc_id, on_uname, task, rc, value, now); /* Update the fail count, if we're not ignoring failures */ if (!ignore_failures) { fail_pair = calloc(1, sizeof(pcmk__attrd_query_pair_t)); CRM_ASSERT(fail_pair != NULL); fail_name = pcmk__failcount_name(rsc_id, task, interval_ms); fail_pair->name = fail_name; fail_pair->value = value; fail_pair->node = on_uname; attrs = g_list_prepend(attrs, fail_pair); } /* Update the last failure time (even if we're ignoring failures, * so that failure can still be detected and shown, e.g. by crm_mon) */ last_pair = calloc(1, sizeof(pcmk__attrd_query_pair_t)); CRM_ASSERT(last_pair != NULL); last_name = pcmk__lastfailure_name(rsc_id, task, interval_ms); last_pair->name = last_name; last_pair->value = now; last_pair->node = on_uname; attrs = g_list_prepend(attrs, last_pair); update_attrd_list(attrs, opts); free(fail_name); free(fail_pair); free(last_name); free(last_pair); g_list_free(attrs); free(now); } bail: free(rsc_id); free(task); return TRUE; } pcmk__graph_action_t * controld_get_action(int id) { for (GList *item = controld_globals.transition_graph->synapses; item != NULL; item = item->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) item->data; for (GList *item2 = synapse->actions; item2; item2 = item2->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) item2->data; if (action->id == id) { return action; } } } return NULL; } pcmk__graph_action_t * get_cancel_action(const char *id, const char *node) { GList *gIter = NULL; GList *gIter2 = NULL; gIter = controld_globals.transition_graph->synapses; for (; gIter != NULL; gIter = gIter->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) gIter->data; gIter2 = synapse->actions; for (; gIter2 != NULL; gIter2 = gIter2->next) { const char *task = NULL; const char *target = NULL; pcmk__graph_action_t *action = (pcmk__graph_action_t *) gIter2->data; task = crm_element_value(action->xml, PCMK_XA_OPERATION); if (!pcmk__str_eq(PCMK_ACTION_CANCEL, task, pcmk__str_casei)) { continue; } task = crm_element_value(action->xml, PCMK__XA_OPERATION_KEY); if (!pcmk__str_eq(task, id, pcmk__str_casei)) { crm_trace("Wrong key %s for %s on %s", task, id, node); continue; } target = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID); if (node && !pcmk__str_eq(target, node, pcmk__str_casei)) { crm_trace("Wrong node %s for %s on %s", target, id, node); continue; } crm_trace("Found %s on %s", id, node); return action; } } return NULL; } bool confirm_cancel_action(const char *id, const char *node_id) { const char *op_key = NULL; const char *node_name = NULL; pcmk__graph_action_t *cancel = get_cancel_action(id, node_id); if (cancel == NULL) { return FALSE; } op_key = crm_element_value(cancel->xml, PCMK__XA_OPERATION_KEY); node_name = crm_element_value(cancel->xml, PCMK__META_ON_NODE); stop_te_timer(cancel); te_action_confirmed(cancel, controld_globals.transition_graph); crm_info("Cancellation of %s on %s confirmed (action %d)", op_key, node_name, cancel->id); return TRUE; } /* downed nodes are listed like: ... */ #define XPATH_DOWNED "//" PCMK__XE_DOWNED \ "/" PCMK_XE_NODE "[@" PCMK_XA_ID "='%s']" /*! * \brief Find a transition event that would have made a specified node down * * \param[in] target UUID of node to match * * \return Matching event if found, NULL otherwise */ pcmk__graph_action_t * match_down_event(const char *target) { pcmk__graph_action_t *match = NULL; xmlXPathObjectPtr xpath_ret = NULL; GList *gIter, *gIter2; char *xpath = crm_strdup_printf(XPATH_DOWNED, target); for (gIter = controld_globals.transition_graph->synapses; gIter != NULL && match == NULL; gIter = gIter->next) { for (gIter2 = ((pcmk__graph_synapse_t * ) gIter->data)->actions; gIter2 != NULL && match == NULL; gIter2 = gIter2->next) { match = (pcmk__graph_action_t *) gIter2->data; if (pcmk_is_set(match->flags, pcmk__graph_action_executed)) { xpath_ret = xpath_search(match->xml, xpath); if (numXpathResults(xpath_ret) < 1) { match = NULL; } freeXpathObject(xpath_ret); } else { // Only actions that were actually started can match match = NULL; } } } free(xpath); if (match != NULL) { crm_debug("Shutdown action %d (%s) found for node %s", match->id, crm_element_value(match->xml, PCMK__XA_OPERATION_KEY), target); } else { crm_debug("No reason to expect node %s to be down", target); } return match; } void process_graph_event(xmlNode *event, const char *event_node) { int rc = -1; // Actual result int target_rc = -1; // Expected result int status = -1; // Executor status int callid = -1; // Executor call ID int transition_num = -1; // Transition number int action_num = -1; // Action number within transition char *update_te_uuid = NULL; bool ignore_failures = FALSE; const char *id = NULL; const char *desc = NULL; const char *magic = NULL; const char *uname = NULL; CRM_ASSERT(event != NULL); /* */ magic = crm_element_value(event, PCMK__XA_TRANSITION_KEY); if (magic == NULL) { /* non-change */ return; } crm_element_value_int(event, PCMK__XA_OP_STATUS, &status); if (status == PCMK_EXEC_PENDING) { return; } id = crm_element_value(event, PCMK__XA_OPERATION_KEY); crm_element_value_int(event, PCMK__XA_RC_CODE, &rc); crm_element_value_int(event, PCMK__XA_CALL_ID, &callid); rc = pcmk__effective_rc(rc); if (decode_transition_key(magic, &update_te_uuid, &transition_num, &action_num, &target_rc) == FALSE) { // decode_transition_key() already logged the bad key crm_err("Can't process action %s result: Incompatible versions? " CRM_XS " call-id=%d", id, callid); - abort_transition(INFINITY, pcmk__graph_restart, "Bad event", event); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Bad event", event); return; } if (transition_num == -1) { // E.g. crm_resource --fail if (record_outside_event(action_num) != pcmk_rc_ok) { crm_debug("Outside event with transition key '%s' has already been " "processed", magic); goto bail; } desc = "initiated outside of the cluster"; - abort_transition(INFINITY, pcmk__graph_restart, "Unexpected event", - event); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Unexpected event", event); } else if ((action_num < 0) || !pcmk__str_eq(update_te_uuid, controld_globals.te_uuid, pcmk__str_none)) { desc = "initiated by a different DC"; - abort_transition(INFINITY, pcmk__graph_restart, "Foreign event", event); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Foreign event", event); } else if ((controld_globals.transition_graph->id != transition_num) || controld_globals.transition_graph->complete) { // Action is not from currently active transition guint interval_ms = 0; if (parse_op_key(id, NULL, NULL, &interval_ms) && (interval_ms != 0)) { /* Recurring actions have the transition number they were first * scheduled in. */ if (status == PCMK_EXEC_CANCELLED) { confirm_cancel_action(id, get_node_id(event)); goto bail; } desc = "arrived after initial scheduling"; - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Change in recurring result", event); } else if (controld_globals.transition_graph->id != transition_num) { desc = "arrived really late"; - abort_transition(INFINITY, pcmk__graph_restart, "Old event", event); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Old event", event); } else { desc = "arrived late"; - abort_transition(INFINITY, pcmk__graph_restart, "Inactive graph", - event); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Inactive graph", event); } } else { // Event is result of an action from currently active transition pcmk__graph_action_t *action = controld_get_action(action_num); if (action == NULL) { // Should never happen desc = "unknown"; - abort_transition(INFINITY, pcmk__graph_restart, "Unknown event", - event); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Unknown event", event); } else if (pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { /* Nothing further needs to be done if the action has already been * confirmed. This can happen e.g. when processing both an * "xxx_last_0" or "xxx_last_failure_0" record as well as the main * history record, which would otherwise result in incorrectly * bumping the fail count twice. */ crm_log_xml_debug(event, "Event already confirmed:"); goto bail; } else { /* An action result needs to be confirmed. * (This is the only case where desc == NULL.) */ if (pcmk__str_eq(crm_meta_value(action->params, PCMK_META_ON_FAIL), PCMK_VALUE_IGNORE, pcmk__str_casei)) { ignore_failures = TRUE; } else if (rc != target_rc) { pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); } stop_te_timer(action); te_action_confirmed(action, controld_globals.transition_graph); if (pcmk_is_set(action->flags, pcmk__graph_action_failed)) { abort_transition(action->synapse->priority + 1, pcmk__graph_restart, "Event failed", event); } } } if (id == NULL) { id = "unknown action"; } uname = crm_element_value(event, PCMK__META_ON_NODE); if (uname == NULL) { uname = "unknown node"; } if (status == PCMK_EXEC_INVALID) { // We couldn't attempt the action crm_info("Transition %d action %d (%s on %s): %s", transition_num, action_num, id, uname, pcmk_exec_status_str(status)); } else if (desc && update_failcount(event, event_node, rc, target_rc, (transition_num == -1), FALSE)) { crm_notice("Transition %d action %d (%s on %s): expected '%s' but got '%s' " CRM_XS " target-rc=%d rc=%d call-id=%d event='%s'", transition_num, action_num, id, uname, services_ocf_exitcode_str(target_rc), services_ocf_exitcode_str(rc), target_rc, rc, callid, desc); } else if (desc) { crm_info("Transition %d action %d (%s on %s): %s " CRM_XS " rc=%d target-rc=%d call-id=%d", transition_num, action_num, id, uname, desc, rc, target_rc, callid); } else if (rc == target_rc) { crm_info("Transition %d action %d (%s on %s) confirmed: %s " CRM_XS " rc=%d call-id=%d", transition_num, action_num, id, uname, services_ocf_exitcode_str(rc), rc, callid); } else { update_failcount(event, event_node, rc, target_rc, (transition_num == -1), ignore_failures); crm_notice("Transition %d action %d (%s on %s): expected '%s' but got '%s' " CRM_XS " target-rc=%d rc=%d call-id=%d", transition_num, action_num, id, uname, services_ocf_exitcode_str(target_rc), services_ocf_exitcode_str(rc), target_rc, rc, callid); } bail: free(update_te_uuid); } diff --git a/daemons/controld/controld_te_utils.c b/daemons/controld/controld_te_utils.c index 39b03789d5..c0c7bea1ba 100644 --- a/daemons/controld/controld_te_utils.c +++ b/daemons/controld/controld_te_utils.c @@ -1,509 +1,509 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include //! Triggers transition graph processing static crm_trigger_t *transition_trigger = NULL; static GHashTable *node_pending_timers = NULL; gboolean stop_te_timer(pcmk__graph_action_t *action) { if (action == NULL) { return FALSE; } if (action->timer != 0) { crm_trace("Stopping action timer"); g_source_remove(action->timer); action->timer = 0; } else { crm_trace("Action timer was already stopped"); return FALSE; } return TRUE; } static gboolean te_graph_trigger(gpointer user_data) { if (controld_globals.transition_graph == NULL) { crm_debug("Nothing to do"); return TRUE; } crm_trace("Invoking graph %d in state %s", controld_globals.transition_graph->id, fsa_state2string(controld_globals.fsa_state)); switch (controld_globals.fsa_state) { case S_STARTING: case S_PENDING: case S_NOT_DC: case S_HALT: case S_ILLEGAL: case S_STOPPING: case S_TERMINATE: return TRUE; default: break; } if (!controld_globals.transition_graph->complete) { enum pcmk__graph_status graph_rc; int orig_limit = controld_globals.transition_graph->batch_limit; int throttled_limit = throttle_get_total_job_limit(orig_limit); controld_globals.transition_graph->batch_limit = throttled_limit; graph_rc = pcmk__execute_graph(controld_globals.transition_graph); controld_globals.transition_graph->batch_limit = orig_limit; if (graph_rc == pcmk__graph_active) { crm_trace("Transition not yet complete"); return TRUE; } else if (graph_rc == pcmk__graph_pending) { crm_trace("Transition not yet complete - no actions fired"); return TRUE; } if (graph_rc != pcmk__graph_complete) { crm_warn("Transition failed: %s", pcmk__graph_status2text(graph_rc)); pcmk__log_graph(LOG_NOTICE, controld_globals.transition_graph); } } crm_debug("Transition %d is now complete", controld_globals.transition_graph->id); controld_globals.transition_graph->complete = true; notify_crmd(controld_globals.transition_graph); return TRUE; } /*! * \internal * \brief Initialize transition trigger */ void controld_init_transition_trigger(void) { transition_trigger = mainloop_add_trigger(G_PRIORITY_LOW, te_graph_trigger, NULL); } /*! * \internal * \brief Destroy transition trigger */ void controld_destroy_transition_trigger(void) { mainloop_destroy_trigger(transition_trigger); transition_trigger = NULL; } void controld_trigger_graph_as(const char *fn, int line) { crm_trace("%s:%d - Triggered graph processing", fn, line); mainloop_set_trigger(transition_trigger); } static struct abort_timer_s { bool aborted; guint id; int priority; enum pcmk__graph_next action; const char *text; } abort_timer = { 0, }; static gboolean abort_timer_popped(gpointer data) { struct abort_timer_s *abort_timer = (struct abort_timer_s *) data; if (AM_I_DC && (abort_timer->aborted == FALSE)) { abort_transition(abort_timer->priority, abort_timer->action, abort_timer->text, NULL); } abort_timer->id = 0; return FALSE; // do not immediately reschedule timer } /*! * \internal * \brief Abort transition after delay, if not already aborted in that time * * \param[in] abort_text Must be literal string */ void abort_after_delay(int abort_priority, enum pcmk__graph_next abort_action, const char *abort_text, guint delay_ms) { if (abort_timer.id) { // Timer already in progress, stop and reschedule g_source_remove(abort_timer.id); } abort_timer.aborted = FALSE; abort_timer.priority = abort_priority; abort_timer.action = abort_action; abort_timer.text = abort_text; abort_timer.id = g_timeout_add(delay_ms, abort_timer_popped, &abort_timer); } static void free_node_pending_timer(gpointer data) { struct abort_timer_s *node_pending_timer = (struct abort_timer_s *) data; if (node_pending_timer->id != 0) { g_source_remove(node_pending_timer->id); node_pending_timer->id = 0; } free(node_pending_timer); } static gboolean node_pending_timer_popped(gpointer key) { struct abort_timer_s *node_pending_timer = NULL; if (node_pending_timers == NULL) { return FALSE; } node_pending_timer = g_hash_table_lookup(node_pending_timers, key); if (node_pending_timer == NULL) { return FALSE; } crm_warn("Node with id '%s' pending timed out (%us) on joining the process " "group", (const char *) key, controld_globals.node_pending_timeout); if (controld_globals.node_pending_timeout > 0) { abort_timer_popped(node_pending_timer); } g_hash_table_remove(node_pending_timers, key); return FALSE; // do not reschedule timer } static void init_node_pending_timer(const crm_node_t *node, guint timeout) { struct abort_timer_s *node_pending_timer = NULL; char *key = NULL; if (node->uuid == NULL) { return; } if (node_pending_timers == NULL) { node_pending_timers = pcmk__strikey_table(free, free_node_pending_timer); // The timer is somehow already existing } else if (g_hash_table_lookup(node_pending_timers, node->uuid) != NULL) { return; } crm_notice("Waiting for pending %s with id '%s' to join the process " "group (timeout=%us)", node->uname ? node->uname : "node", node->uuid, controld_globals.node_pending_timeout); node_pending_timer = calloc(1, sizeof(struct abort_timer_s)); CRM_ASSERT(node_pending_timer != NULL); node_pending_timer->aborted = FALSE; - node_pending_timer->priority = INFINITY; + node_pending_timer->priority = PCMK_SCORE_INFINITY; node_pending_timer->action = pcmk__graph_restart; node_pending_timer->text = "Node pending timed out"; key = strdup(node->uuid); CRM_ASSERT(key != NULL); g_hash_table_replace(node_pending_timers, key, node_pending_timer); node_pending_timer->id = g_timeout_add_seconds(timeout, node_pending_timer_popped, key); CRM_ASSERT(node_pending_timer->id != 0); } static void remove_node_pending_timer(const char *node_uuid) { if (node_pending_timers == NULL) { return; } g_hash_table_remove(node_pending_timers, node_uuid); } void controld_node_pending_timer(const crm_node_t *node) { long long remaining_timeout = 0; /* If the node is not an active cluster node, is leaving the cluster, or is * already part of CPG, or PCMK_OPT_NODE_PENDING_TIMEOUT is disabled, free * any node pending timer for it. */ if (pcmk_is_set(node->flags, crm_remote_node) || (node->when_member <= 1) || (node->when_online > 0) || (controld_globals.node_pending_timeout == 0)) { remove_node_pending_timer(node->uuid); return; } // Node is a cluster member but offline in CPG remaining_timeout = node->when_member - time(NULL) + controld_globals.node_pending_timeout; /* It already passed node pending timeout somehow. * Free any node pending timer of it. */ if (remaining_timeout <= 0) { remove_node_pending_timer(node->uuid); return; } init_node_pending_timer(node, remaining_timeout); } void controld_free_node_pending_timers(void) { if (node_pending_timers == NULL) { return; } g_hash_table_destroy(node_pending_timers); node_pending_timers = NULL; } static const char * abort2text(enum pcmk__graph_next abort_action) { switch (abort_action) { case pcmk__graph_done: return "done"; case pcmk__graph_wait: return "stop"; case pcmk__graph_restart: return "restart"; case pcmk__graph_shutdown: return "shutdown"; } return "unknown"; } static bool update_abort_priority(pcmk__graph_t *graph, int priority, enum pcmk__graph_next action, const char *abort_reason) { bool change = FALSE; if (graph == NULL) { return change; } if (graph->abort_priority < priority) { crm_debug("Abort priority upgraded from %d to %d", graph->abort_priority, priority); graph->abort_priority = priority; if (graph->abort_reason != NULL) { crm_debug("'%s' abort superseded by %s", graph->abort_reason, abort_reason); } graph->abort_reason = abort_reason; change = TRUE; } if (graph->completion_action < action) { crm_debug("Abort action %s superseded by %s: %s", abort2text(graph->completion_action), abort2text(action), abort_reason); graph->completion_action = action; change = TRUE; } return change; } void abort_transition_graph(int abort_priority, enum pcmk__graph_next abort_action, const char *abort_text, const xmlNode *reason, const char *fn, int line) { int add[] = { 0, 0, 0 }; int del[] = { 0, 0, 0 }; int level = LOG_INFO; const xmlNode *diff = NULL; const xmlNode *change = NULL; CRM_CHECK(controld_globals.transition_graph != NULL, return); switch (controld_globals.fsa_state) { case S_STARTING: case S_PENDING: case S_NOT_DC: case S_HALT: case S_ILLEGAL: case S_STOPPING: case S_TERMINATE: crm_info("Abort %s suppressed: state=%s (%scomplete)", abort_text, fsa_state2string(controld_globals.fsa_state), (controld_globals.transition_graph->complete? "" : "in")); return; default: break; } abort_timer.aborted = TRUE; controld_expect_sched_reply(NULL); if (!controld_globals.transition_graph->complete && update_abort_priority(controld_globals.transition_graph, abort_priority, abort_action, abort_text)) { level = LOG_NOTICE; } if (reason != NULL) { const xmlNode *search = NULL; for(search = reason; search; search = search->parent) { if (pcmk__xe_is(search, PCMK_XE_DIFF)) { diff = search; break; } } if(diff) { xml_patch_versions(diff, add, del); for(search = reason; search; search = search->parent) { if (pcmk__xe_is(search, PCMK_XE_CHANGE)) { change = search; break; } } } } if (reason == NULL) { do_crm_log(level, "Transition %d aborted: %s " CRM_XS " source=%s:%d " "complete=%s", controld_globals.transition_graph->id, abort_text, fn, line, pcmk__btoa(controld_globals.transition_graph->complete)); } else if(change == NULL) { GString *local_path = pcmk__element_xpath(reason); CRM_ASSERT(local_path != NULL); do_crm_log(level, "Transition %d aborted by %s.%s: %s " CRM_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s", controld_globals.transition_graph->id, reason->name, pcmk__xe_id(reason), abort_text, add[0], add[1], add[2], fn, line, (const char *) local_path->str, pcmk__btoa(controld_globals.transition_graph->complete)); g_string_free(local_path, TRUE); } else { const char *op = crm_element_value(change, PCMK_XA_OPERATION); const char *path = crm_element_value(change, PCMK_XA_PATH); if(change == reason) { if (strcmp(op, PCMK_VALUE_CREATE) == 0) { reason = reason->children; } else if (strcmp(op, PCMK_VALUE_MODIFY) == 0) { reason = first_named_child(reason, PCMK_XE_CHANGE_RESULT); if(reason) { reason = reason->children; } } CRM_CHECK(reason != NULL, goto done); } if (strcmp(op, PCMK_VALUE_DELETE) == 0) { const char *shortpath = strrchr(path, '/'); do_crm_log(level, "Transition %d aborted by deletion of %s: %s " CRM_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s", controld_globals.transition_graph->id, (shortpath? (shortpath + 1) : path), abort_text, add[0], add[1], add[2], fn, line, path, pcmk__btoa(controld_globals.transition_graph->complete)); } else if (pcmk__xe_is(reason, PCMK_XE_NVPAIR)) { do_crm_log(level, "Transition %d aborted by %s doing %s %s=%s: %s " CRM_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s", controld_globals.transition_graph->id, crm_element_value(reason, PCMK_XA_ID), op, crm_element_value(reason, PCMK_XA_NAME), crm_element_value(reason, PCMK_XA_VALUE), abort_text, add[0], add[1], add[2], fn, line, path, pcmk__btoa(controld_globals.transition_graph->complete)); } else if (pcmk__xe_is(reason, PCMK__XE_LRM_RSC_OP)) { const char *magic = crm_element_value(reason, PCMK__XA_TRANSITION_MAGIC); do_crm_log(level, "Transition %d aborted by operation %s '%s' on %s: %s " CRM_XS " magic=%s cib=%d.%d.%d source=%s:%d complete=%s", controld_globals.transition_graph->id, crm_element_value(reason, PCMK__XA_OPERATION_KEY), op, crm_element_value(reason, PCMK__META_ON_NODE), abort_text, magic, add[0], add[1], add[2], fn, line, pcmk__btoa(controld_globals.transition_graph->complete)); } else if (pcmk__str_any_of((const char *) reason->name, PCMK__XE_NODE_STATE, PCMK_XE_NODE, NULL)) { const char *uname = crm_peer_uname(pcmk__xe_id(reason)); do_crm_log(level, "Transition %d aborted by %s '%s' on %s: %s " CRM_XS " cib=%d.%d.%d source=%s:%d complete=%s", controld_globals.transition_graph->id, reason->name, op, pcmk__s(uname, pcmk__xe_id(reason)), abort_text, add[0], add[1], add[2], fn, line, pcmk__btoa(controld_globals.transition_graph->complete)); } else { const char *id = pcmk__xe_id(reason); do_crm_log(level, "Transition %d aborted by %s.%s '%s': %s " CRM_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s", controld_globals.transition_graph->id, reason->name, pcmk__s(id, ""), pcmk__s(op, "change"), abort_text, add[0], add[1], add[2], fn, line, path, pcmk__btoa(controld_globals.transition_graph->complete)); } } done: if (controld_globals.transition_graph->complete) { if (controld_get_period_transition_timer() > 0) { controld_stop_transition_timer(); controld_start_transition_timer(); } else { register_fsa_input(C_FSA_INTERNAL, I_PE_CALC, NULL); } return; } trigger_graph(); } diff --git a/daemons/controld/controld_transition.c b/daemons/controld/controld_transition.c index 241a6c81eb..6a17f6e45b 100644 --- a/daemons/controld/controld_transition.c +++ b/daemons/controld/controld_transition.c @@ -1,181 +1,183 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include static pcmk__graph_t * create_blank_graph(void) { pcmk__graph_t *a_graph = pcmk__unpack_graph(NULL, NULL); a_graph->complete = true; a_graph->abort_reason = "DC Takeover"; a_graph->completion_action = pcmk__graph_restart; return a_graph; } /* A_TE_START, A_TE_STOP, O_TE_RESTART */ void do_te_control(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { cib_t *cib_conn = controld_globals.cib_conn; gboolean init_ok = TRUE; if (pcmk_is_set(action, A_TE_STOP)) { pcmk__free_graph(controld_globals.transition_graph); controld_globals.transition_graph = NULL; if (cib_conn != NULL) { cib_conn->cmds->del_notify_callback(cib_conn, T_CIB_DIFF_NOTIFY, te_update_diff); } controld_clear_fsa_input_flags(R_TE_CONNECTED); crm_info("Transitioner is now inactive"); } if ((action & A_TE_START) == 0) { return; } else if (pcmk_is_set(controld_globals.fsa_input_register, R_TE_CONNECTED)) { crm_debug("The transitioner is already active"); return; } else if ((action & A_TE_START) && cur_state == S_STOPPING) { crm_info("Ignoring request to start the transitioner while shutting down"); return; } if (controld_globals.te_uuid == NULL) { controld_globals.te_uuid = crm_generate_uuid(); crm_info("Registering TE UUID: %s", controld_globals.te_uuid); } if (cib_conn == NULL) { crm_err("Could not set CIB callbacks"); init_ok = FALSE; } else { if (cib_conn->cmds->add_notify_callback(cib_conn, T_CIB_DIFF_NOTIFY, te_update_diff) != pcmk_ok) { crm_err("Could not set CIB notification callback"); init_ok = FALSE; } } if (init_ok) { controld_register_graph_functions(); pcmk__free_graph(controld_globals.transition_graph); /* create a blank one */ crm_debug("Transitioner is now active"); controld_globals.transition_graph = create_blank_graph(); controld_set_fsa_input_flags(R_TE_CONNECTED); } } /* A_TE_INVOKE, A_TE_CANCEL */ void do_te_invoke(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { if (!AM_I_DC || ((controld_globals.fsa_state != S_TRANSITION_ENGINE) && pcmk_is_set(action, A_TE_INVOKE))) { crm_notice("No need to invoke the TE (%s) in state %s", fsa_action2string(action), fsa_state2string(controld_globals.fsa_state)); return; } if (action & A_TE_CANCEL) { crm_debug("Cancelling the transition: %sactive", controld_globals.transition_graph->complete? "in" : ""); - abort_transition(INFINITY, pcmk__graph_restart, "Peer Cancelled", NULL); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Peer Cancelled", NULL); if (!controld_globals.transition_graph->complete) { crmd_fsa_stall(FALSE); } } else if (action & A_TE_HALT) { - abort_transition(INFINITY, pcmk__graph_wait, "Peer Halt", NULL); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_wait, "Peer Halt", + NULL); if (!controld_globals.transition_graph->complete) { crmd_fsa_stall(FALSE); } } else if (action & A_TE_INVOKE) { ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); xmlNode *graph_data = input->xml; const char *ref = crm_element_value(input->msg, PCMK_XA_REFERENCE); const char *graph_input = crm_element_value(input->msg, PCMK__XA_CRM_TGRAPH_IN); if (graph_data == NULL) { crm_log_xml_err(input->msg, "Bad command"); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); return; } if (!controld_globals.transition_graph->complete) { crm_info("Another transition is already active"); - abort_transition(INFINITY, pcmk__graph_restart, "Transition Active", - NULL); + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, + "Transition Active", NULL); return; } if ((controld_globals.fsa_pe_ref == NULL) || !pcmk__str_eq(controld_globals.fsa_pe_ref, ref, pcmk__str_none)) { crm_info("Transition is redundant: %s expected but %s received", pcmk__s(controld_globals.fsa_pe_ref, "no reference"), pcmk__s(ref, "no reference")); - abort_transition(INFINITY, pcmk__graph_restart, + abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Transition Redundant", NULL); } if (controld_is_started_transition_timer()) { crm_debug("The transitioner wait for a transition timer"); return; } CRM_CHECK(graph_data != NULL, crm_err("Input raised by %s is invalid", msg_data->origin); crm_log_xml_err(input->msg, "Bad command"); return); pcmk__free_graph(controld_globals.transition_graph); controld_globals.transition_graph = pcmk__unpack_graph(graph_data, graph_input); CRM_CHECK(controld_globals.transition_graph != NULL, controld_globals.transition_graph = create_blank_graph(); return); crm_info("Processing graph %d (ref=%s) derived from %s", controld_globals.transition_graph->id, ref, graph_input); te_reset_job_counts(); trigger_graph(); pcmk__log_graph(LOG_TRACE, controld_globals.transition_graph); if (graph_data != input->xml) { free_xml(graph_data); } } } diff --git a/include/crm/common/scores.h b/include/crm/common/scores.h index 6fd333fa8d..4b73f6610d 100644 --- a/include/crm/common/scores.h +++ b/include/crm/common/scores.h @@ -1,30 +1,33 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_SCORES__H #define PCMK__CRM_COMMON_SCORES__H #ifdef __cplusplus extern "C" { #endif /** * \file * \brief Pacemaker APIs related to scores * \ingroup core */ +//! Integer score to use to represent "infinity" +#define PCMK_SCORE_INFINITY 1000000 + const char *pcmk_readable_score(int score); int char2score(const char *score); #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_SCORES__H diff --git a/lib/common/scores.c b/lib/common/scores.c index 0f3047b911..33da7ee4b6 100644 --- a/lib/common/scores.c +++ b/lib/common/scores.c @@ -1,166 +1,166 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #ifndef _GNU_SOURCE # define _GNU_SOURCE #endif #include // snprintf(), NULL #include // strcpy(), strdup() #include // size_t int pcmk__score_red = 0; int pcmk__score_green = 0; int pcmk__score_yellow = 0; /*! * \brief Get the integer value of a score string * * Given a string representation of a score, return the integer equivalent. * This accepts infinity strings as well as red, yellow, and green, and * bounds the result to +/-INFINITY. * * \param[in] score Score as string * * \return Integer value corresponding to \p score */ int char2score(const char *score) { if (score == NULL) { return 0; } else if (pcmk_str_is_minus_infinity(score)) { - return -CRM_SCORE_INFINITY; + return -PCMK_SCORE_INFINITY; } else if (pcmk_str_is_infinity(score)) { - return CRM_SCORE_INFINITY; + return PCMK_SCORE_INFINITY; } else if (pcmk__str_eq(score, PCMK_VALUE_RED, pcmk__str_casei)) { return pcmk__score_red; } else if (pcmk__str_eq(score, PCMK_VALUE_YELLOW, pcmk__str_casei)) { return pcmk__score_yellow; } else if (pcmk__str_eq(score, PCMK_VALUE_GREEN, pcmk__str_casei)) { return pcmk__score_green; } else { long long score_ll; pcmk__scan_ll(score, &score_ll, 0LL); - if (score_ll > CRM_SCORE_INFINITY) { - return CRM_SCORE_INFINITY; + if (score_ll > PCMK_SCORE_INFINITY) { + return PCMK_SCORE_INFINITY; - } else if (score_ll < -CRM_SCORE_INFINITY) { - return -CRM_SCORE_INFINITY; + } else if (score_ll < -PCMK_SCORE_INFINITY) { + return -PCMK_SCORE_INFINITY; } else { return (int) score_ll; } } } /*! * \brief Return a displayable static string for a score value * * Given a score value, return a pointer to a static string representation of * the score suitable for log messages, output, etc. * * \param[in] score Score to display * * \return Pointer to static memory containing string representation of \p score * \note Subsequent calls to this function will overwrite the returned value, so * it should be used only in a local context such as a printf()-style * statement. */ const char * pcmk_readable_score(int score) { // The longest possible result is "-INFINITY" static char score_s[sizeof(PCMK_VALUE_MINUS_INFINITY)]; - if (score >= CRM_SCORE_INFINITY) { + if (score >= PCMK_SCORE_INFINITY) { strcpy(score_s, PCMK_VALUE_INFINITY); - } else if (score <= -CRM_SCORE_INFINITY) { + } else if (score <= -PCMK_SCORE_INFINITY) { strcpy(score_s, PCMK_VALUE_MINUS_INFINITY); } else { // Range is limited to +/-1000000, so no chance of overflow snprintf(score_s, sizeof(score_s), "%d", score); } return score_s; } /*! * \internal * \brief Add two scores, bounding to +/-INFINITY * * \param[in] score1 First score to add * \param[in] score2 Second score to add * * \note This function does not have context about what the scores mean, so it * does not log any messages. */ int pcmk__add_scores(int score1, int score2) { - /* As long as CRM_SCORE_INFINITY is less than half of the maximum integer, + /* As long as PCMK_SCORE_INFINITY is less than half of the maximum integer, * we can ignore the possibility of integer overflow. */ int result = score1 + score2; // First handle the cases where one or both is infinite - if ((score1 <= -CRM_SCORE_INFINITY) || (score2 <= -CRM_SCORE_INFINITY)) { - return -CRM_SCORE_INFINITY; + if ((score1 <= -PCMK_SCORE_INFINITY) || (score2 <= -PCMK_SCORE_INFINITY)) { + return -PCMK_SCORE_INFINITY; } - if ((score1 >= CRM_SCORE_INFINITY) || (score2 >= CRM_SCORE_INFINITY)) { - return CRM_SCORE_INFINITY; + if ((score1 >= PCMK_SCORE_INFINITY) || (score2 >= PCMK_SCORE_INFINITY)) { + return PCMK_SCORE_INFINITY; } // Bound result to infinity. - if (result >= CRM_SCORE_INFINITY) { - return CRM_SCORE_INFINITY; + if (result >= PCMK_SCORE_INFINITY) { + return PCMK_SCORE_INFINITY; } - if (result <= -CRM_SCORE_INFINITY) { - return -CRM_SCORE_INFINITY; + if (result <= -PCMK_SCORE_INFINITY) { + return -PCMK_SCORE_INFINITY; } return result; } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include char * score2char(int score) { char *result = strdup(pcmk_readable_score(score)); CRM_ASSERT(result != NULL); return result; } char * score2char_stack(int score, char *buf, size_t len) { CRM_CHECK((buf != NULL) && (len >= sizeof(PCMK_VALUE_MINUS_INFINITY)), return NULL); strcpy(buf, pcmk_readable_score(score)); return buf; } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/common/tests/scores/char2score_test.c b/lib/common/tests/scores/char2score_test.c index fbba12a067..5d7252f4a2 100644 --- a/lib/common/tests/scores/char2score_test.c +++ b/lib/common/tests/scores/char2score_test.c @@ -1,75 +1,77 @@ /* - * Copyright 2022 the Pacemaker project contributors + * Copyright 2022-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include extern int pcmk__score_red; extern int pcmk__score_green; extern int pcmk__score_yellow; static void empty_input(void **state) { assert_int_equal(char2score(NULL), 0); } static void bad_input(void **state) { assert_int_equal(char2score("PQRST"), 0); assert_int_equal(char2score("3.141592"), 3); assert_int_equal(char2score("0xf00d"), 0); } static void special_values(void **state) { - assert_int_equal(char2score("-INFINITY"), -CRM_SCORE_INFINITY); - assert_int_equal(char2score("INFINITY"), CRM_SCORE_INFINITY); - assert_int_equal(char2score("+INFINITY"), CRM_SCORE_INFINITY); + assert_int_equal(char2score("-INFINITY"), -PCMK_SCORE_INFINITY); + assert_int_equal(char2score("INFINITY"), PCMK_SCORE_INFINITY); + assert_int_equal(char2score("+INFINITY"), PCMK_SCORE_INFINITY); pcmk__score_red = 10; pcmk__score_green = 20; pcmk__score_yellow = 30; assert_int_equal(char2score("red"), pcmk__score_red); assert_int_equal(char2score("green"), pcmk__score_green); assert_int_equal(char2score("yellow"), pcmk__score_yellow); assert_int_equal(char2score("ReD"), pcmk__score_red); assert_int_equal(char2score("GrEeN"), pcmk__score_green); assert_int_equal(char2score("yElLoW"), pcmk__score_yellow); } /* These ridiculous macros turn an integer constant into a string constant. */ #define A(x) #x #define B(x) A(x) static void outside_limits(void **state) { - assert_int_equal(char2score(B(CRM_SCORE_INFINITY) "00"), CRM_SCORE_INFINITY); - assert_int_equal(char2score("-" B(CRM_SCORE_INFINITY) "00"), -CRM_SCORE_INFINITY); + assert_int_equal(char2score(B(PCMK_SCORE_INFINITY) "00"), + PCMK_SCORE_INFINITY); + assert_int_equal(char2score("-" B(PCMK_SCORE_INFINITY) "00"), + -PCMK_SCORE_INFINITY); } static void inside_limits(void **state) { assert_int_equal(char2score("1234"), 1234); assert_int_equal(char2score("-1234"), -1234); } PCMK__UNIT_TEST(NULL, NULL, cmocka_unit_test(empty_input), cmocka_unit_test(bad_input), cmocka_unit_test(special_values), cmocka_unit_test(outside_limits), cmocka_unit_test(inside_limits)) diff --git a/lib/common/tests/scores/pcmk__add_scores_test.c b/lib/common/tests/scores/pcmk__add_scores_test.c index 1309659fb3..952cf978ad 100644 --- a/lib/common/tests/scores/pcmk__add_scores_test.c +++ b/lib/common/tests/scores/pcmk__add_scores_test.c @@ -1,76 +1,99 @@ /* - * Copyright 2022-2023 the Pacemaker project contributors + * Copyright 2022-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include static void score1_minus_inf(void **state) { - assert_int_equal(pcmk__add_scores(-CRM_SCORE_INFINITY, -CRM_SCORE_INFINITY), -CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(-CRM_SCORE_INFINITY, -1), -CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(-CRM_SCORE_INFINITY, 0), -CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(-CRM_SCORE_INFINITY, 1), -CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(-CRM_SCORE_INFINITY, CRM_SCORE_INFINITY), -CRM_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(-PCMK_SCORE_INFINITY, + -PCMK_SCORE_INFINITY), + -PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(-PCMK_SCORE_INFINITY, -1), + -PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(-PCMK_SCORE_INFINITY, 0), + -PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(-PCMK_SCORE_INFINITY, 1), + -PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(-PCMK_SCORE_INFINITY, + PCMK_SCORE_INFINITY), + -PCMK_SCORE_INFINITY); } static void score2_minus_inf(void **state) { - assert_int_equal(pcmk__add_scores(-1, -CRM_SCORE_INFINITY), -CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(0, -CRM_SCORE_INFINITY), -CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(1, -CRM_SCORE_INFINITY), -CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(CRM_SCORE_INFINITY, -CRM_SCORE_INFINITY), -CRM_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(-1, -PCMK_SCORE_INFINITY), + -PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(0, -PCMK_SCORE_INFINITY), + -PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(1, -PCMK_SCORE_INFINITY), + -PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(PCMK_SCORE_INFINITY, + -PCMK_SCORE_INFINITY), + -PCMK_SCORE_INFINITY); } static void score1_pos_inf(void **state) { - assert_int_equal(pcmk__add_scores(CRM_SCORE_INFINITY, CRM_SCORE_INFINITY), CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(CRM_SCORE_INFINITY, -1), CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(CRM_SCORE_INFINITY, 0), CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(CRM_SCORE_INFINITY, 1), CRM_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(PCMK_SCORE_INFINITY, PCMK_SCORE_INFINITY), + PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(PCMK_SCORE_INFINITY, -1), + PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(PCMK_SCORE_INFINITY, 0), + PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(PCMK_SCORE_INFINITY, 1), + PCMK_SCORE_INFINITY); } static void score2_pos_inf(void **state) { - assert_int_equal(pcmk__add_scores(-1, CRM_SCORE_INFINITY), CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(0, CRM_SCORE_INFINITY), CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(1, CRM_SCORE_INFINITY), CRM_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(-1, PCMK_SCORE_INFINITY), + PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(0, PCMK_SCORE_INFINITY), + PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(1, PCMK_SCORE_INFINITY), + PCMK_SCORE_INFINITY); } static void result_infinite(void **state) { - assert_int_equal(pcmk__add_scores(INT_MAX, INT_MAX), CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(INT_MIN, INT_MIN), -CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(2000000, 50), CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(CRM_SCORE_INFINITY/2, CRM_SCORE_INFINITY/2), CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(-CRM_SCORE_INFINITY/2, -CRM_SCORE_INFINITY/2), -CRM_SCORE_INFINITY); - assert_int_equal(pcmk__add_scores(-4000000, 50), -CRM_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(INT_MAX, INT_MAX), PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(INT_MIN, INT_MIN), -PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(2000000, 50), PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(PCMK_SCORE_INFINITY/2, + PCMK_SCORE_INFINITY/2), + PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(-PCMK_SCORE_INFINITY/2, + -PCMK_SCORE_INFINITY/2), + -PCMK_SCORE_INFINITY); + assert_int_equal(pcmk__add_scores(-4000000, 50), -PCMK_SCORE_INFINITY); } static void result_finite(void **state) { assert_int_equal(pcmk__add_scores(0, 0), 0); assert_int_equal(pcmk__add_scores(0, 100), 100); assert_int_equal(pcmk__add_scores(200, 0), 200); assert_int_equal(pcmk__add_scores(200, -50), 150); } PCMK__UNIT_TEST(NULL, NULL, cmocka_unit_test(score1_minus_inf), cmocka_unit_test(score2_minus_inf), cmocka_unit_test(score1_pos_inf), cmocka_unit_test(score2_pos_inf), cmocka_unit_test(result_infinite), cmocka_unit_test(result_finite)) diff --git a/lib/common/tests/scores/pcmk_readable_score_test.c b/lib/common/tests/scores/pcmk_readable_score_test.c index 444b3878a0..c3d66f6ab1 100644 --- a/lib/common/tests/scores/pcmk_readable_score_test.c +++ b/lib/common/tests/scores/pcmk_readable_score_test.c @@ -1,33 +1,33 @@ /* * Copyright 2022-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include static void outside_limits(void **state) { - assert_string_equal(pcmk_readable_score(CRM_SCORE_INFINITY * 2), + assert_string_equal(pcmk_readable_score(PCMK_SCORE_INFINITY * 2), PCMK_VALUE_INFINITY); - assert_string_equal(pcmk_readable_score(-CRM_SCORE_INFINITY * 2), + assert_string_equal(pcmk_readable_score(-PCMK_SCORE_INFINITY * 2), PCMK_VALUE_MINUS_INFINITY); } static void inside_limits(void **state) { assert_string_equal(pcmk_readable_score(0), "0"); assert_string_equal(pcmk_readable_score(1024), "1024"); assert_string_equal(pcmk_readable_score(-1024), "-1024"); } PCMK__UNIT_TEST(NULL, NULL, cmocka_unit_test(outside_limits), cmocka_unit_test(inside_limits)) diff --git a/lib/common/xml.c b/lib/common/xml.c index f23e96e58b..1948c08e27 100644 --- a/lib/common/xml.c +++ b/lib/common/xml.c @@ -1,3001 +1,3001 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include /* xmlAllocOutputBuffer */ #include #include #include // PCMK__XML_LOG_BASE, etc. #include "crmcommon_private.h" // Define this as 1 in development to get insanely verbose trace messages #ifndef XML_PARSER_DEBUG #define XML_PARSER_DEBUG 0 #endif /* @TODO XML_PARSE_RECOVER allows some XML errors to be silently worked around * by libxml2, which is potentially ambiguous and dangerous. We should drop it * when we can break backward compatibility with configurations that might be * relying on it (i.e. pacemaker 3.0.0). * * It might be a good idea to have a transitional period where we first try * parsing without XML_PARSE_RECOVER, and if that fails, try parsing again with * it, logging a warning if it succeeds. */ #define PCMK__XML_PARSE_OPTS_WITHOUT_RECOVER (XML_PARSE_NOBLANKS) #define PCMK__XML_PARSE_OPTS_WITH_RECOVER (XML_PARSE_NOBLANKS | XML_PARSE_RECOVER) bool pcmk__tracking_xml_changes(xmlNode *xml, bool lazy) { if(xml == NULL || xml->doc == NULL || xml->doc->_private == NULL) { return FALSE; } else if (!pcmk_is_set(((xml_doc_private_t *)xml->doc->_private)->flags, pcmk__xf_tracking)) { return FALSE; } else if (lazy && !pcmk_is_set(((xml_doc_private_t *)xml->doc->_private)->flags, pcmk__xf_lazy)) { return FALSE; } return TRUE; } static inline void set_parent_flag(xmlNode *xml, long flag) { for(; xml; xml = xml->parent) { xml_node_private_t *nodepriv = xml->_private; if (nodepriv == NULL) { /* During calls to xmlDocCopyNode(), _private will be unset for parent nodes */ } else { pcmk__set_xml_flags(nodepriv, flag); } } } void pcmk__set_xml_doc_flag(xmlNode *xml, enum xml_private_flags flag) { if(xml && xml->doc && xml->doc->_private){ /* During calls to xmlDocCopyNode(), xml->doc may be unset */ xml_doc_private_t *docpriv = xml->doc->_private; pcmk__set_xml_flags(docpriv, flag); } } // Mark document, element, and all element's parents as changed void pcmk__mark_xml_node_dirty(xmlNode *xml) { pcmk__set_xml_doc_flag(xml, pcmk__xf_dirty); set_parent_flag(xml, pcmk__xf_dirty); } // Clear flags on XML node and its children static void reset_xml_node_flags(xmlNode *xml) { xmlNode *cIter = NULL; xml_node_private_t *nodepriv = xml->_private; if (nodepriv) { nodepriv->flags = 0; } for (cIter = pcmk__xml_first_child(xml); cIter != NULL; cIter = pcmk__xml_next(cIter)) { reset_xml_node_flags(cIter); } } // Set xpf_created flag on XML node and any children void pcmk__mark_xml_created(xmlNode *xml) { xmlNode *cIter = NULL; xml_node_private_t *nodepriv = NULL; CRM_ASSERT(xml != NULL); nodepriv = xml->_private; if (nodepriv && pcmk__tracking_xml_changes(xml, FALSE)) { if (!pcmk_is_set(nodepriv->flags, pcmk__xf_created)) { pcmk__set_xml_flags(nodepriv, pcmk__xf_created); pcmk__mark_xml_node_dirty(xml); } for (cIter = pcmk__xml_first_child(xml); cIter != NULL; cIter = pcmk__xml_next(cIter)) { pcmk__mark_xml_created(cIter); } } } #define XML_DOC_PRIVATE_MAGIC 0x81726354UL #define XML_NODE_PRIVATE_MAGIC 0x54637281UL // Free an XML object previously marked as deleted static void free_deleted_object(void *data) { if(data) { pcmk__deleted_xml_t *deleted_obj = data; free(deleted_obj->path); free(deleted_obj); } } // Free and NULL user, ACLs, and deleted objects in an XML node's private data static void reset_xml_private_data(xml_doc_private_t *docpriv) { if (docpriv != NULL) { CRM_ASSERT(docpriv->check == XML_DOC_PRIVATE_MAGIC); free(docpriv->user); docpriv->user = NULL; if (docpriv->acls != NULL) { pcmk__free_acls(docpriv->acls); docpriv->acls = NULL; } if(docpriv->deleted_objs) { g_list_free_full(docpriv->deleted_objs, free_deleted_object); docpriv->deleted_objs = NULL; } } } // Free all private data associated with an XML node static void free_private_data(xmlNode *node) { /* Note: This function frees private data assosciated with an XML node, unless the function is being called as a result of internal XSLT cleanup. That could happen through, for example, the following chain of function calls: xsltApplyStylesheetInternal -> xsltFreeTransformContext -> xsltFreeRVTs -> xmlFreeDoc And in that case, the node would fulfill three conditions: 1. It would be a standalone document (i.e. it wouldn't be part of a document) 2. It would have a space-prefixed name (for reference, please see xsltInternals.h: XSLT_MARK_RES_TREE_FRAG) 3. It would carry its own payload in the _private field. We do not free data in this circumstance to avoid a failed assertion on the XML_*_PRIVATE_MAGIC later. */ if (node->name == NULL || node->name[0] != ' ') { if (node->_private) { if (node->type == XML_DOCUMENT_NODE) { reset_xml_private_data(node->_private); } else { CRM_ASSERT(((xml_node_private_t *) node->_private)->check == XML_NODE_PRIVATE_MAGIC); /* nothing dynamically allocated nested */ } free(node->_private); node->_private = NULL; } } } // Allocate and initialize private data for an XML node static void new_private_data(xmlNode *node) { switch (node->type) { case XML_DOCUMENT_NODE: { xml_doc_private_t *docpriv = NULL; docpriv = calloc(1, sizeof(xml_doc_private_t)); CRM_ASSERT(docpriv != NULL); docpriv->check = XML_DOC_PRIVATE_MAGIC; /* Flags will be reset if necessary when tracking is enabled */ pcmk__set_xml_flags(docpriv, pcmk__xf_dirty|pcmk__xf_created); node->_private = docpriv; break; } case XML_ELEMENT_NODE: case XML_ATTRIBUTE_NODE: case XML_COMMENT_NODE: { xml_node_private_t *nodepriv = NULL; nodepriv = calloc(1, sizeof(xml_node_private_t)); CRM_ASSERT(nodepriv != NULL); nodepriv->check = XML_NODE_PRIVATE_MAGIC; /* Flags will be reset if necessary when tracking is enabled */ pcmk__set_xml_flags(nodepriv, pcmk__xf_dirty|pcmk__xf_created); node->_private = nodepriv; if (pcmk__tracking_xml_changes(node, FALSE)) { /* XML_ELEMENT_NODE doesn't get picked up here, node->doc is * not hooked up at the point we are called */ pcmk__mark_xml_node_dirty(node); } break; } case XML_TEXT_NODE: case XML_DTD_NODE: case XML_CDATA_SECTION_NODE: break; default: /* Ignore */ crm_trace("Ignoring %p %d", node, node->type); CRM_LOG_ASSERT(node->type == XML_ELEMENT_NODE); break; } } void xml_track_changes(xmlNode * xml, const char *user, xmlNode *acl_source, bool enforce_acls) { xml_accept_changes(xml); crm_trace("Tracking changes%s to %p", enforce_acls?" with ACLs":"", xml); pcmk__set_xml_doc_flag(xml, pcmk__xf_tracking); if(enforce_acls) { if(acl_source == NULL) { acl_source = xml; } pcmk__set_xml_doc_flag(xml, pcmk__xf_acl_enabled); pcmk__unpack_acl(acl_source, xml, user); pcmk__apply_acl(xml); } } bool xml_tracking_changes(xmlNode * xml) { return (xml != NULL) && (xml->doc != NULL) && (xml->doc->_private != NULL) && pcmk_is_set(((xml_doc_private_t *)(xml->doc->_private))->flags, pcmk__xf_tracking); } bool xml_document_dirty(xmlNode *xml) { return (xml != NULL) && (xml->doc != NULL) && (xml->doc->_private != NULL) && pcmk_is_set(((xml_doc_private_t *)(xml->doc->_private))->flags, pcmk__xf_dirty); } /*! * \internal * \brief Return ordinal position of an XML node among its siblings * * \param[in] xml XML node to check * \param[in] ignore_if_set Don't count siblings with this flag set * * \return Ordinal position of \p xml (starting with 0) */ int pcmk__xml_position(const xmlNode *xml, enum xml_private_flags ignore_if_set) { int position = 0; for (const xmlNode *cIter = xml; cIter->prev; cIter = cIter->prev) { xml_node_private_t *nodepriv = ((xmlNode*)cIter->prev)->_private; if (!pcmk_is_set(nodepriv->flags, ignore_if_set)) { position++; } } return position; } // Remove all attributes marked as deleted from an XML node static void accept_attr_deletions(xmlNode *xml) { // Clear XML node's flags ((xml_node_private_t *) xml->_private)->flags = pcmk__xf_none; // Remove this XML node's attributes that were marked as deleted pcmk__xe_remove_matching_attrs(xml, pcmk__marked_as_deleted, NULL); // Recursively do the same for this XML node's children for (xmlNodePtr cIter = pcmk__xml_first_child(xml); cIter != NULL; cIter = pcmk__xml_next(cIter)) { accept_attr_deletions(cIter); } } /*! * \internal * \brief Find first child XML node matching another given XML node * * \param[in] haystack XML whose children should be checked * \param[in] needle XML to match (comment content or element name and ID) * \param[in] exact If true and needle is a comment, position must match */ xmlNode * pcmk__xml_match(const xmlNode *haystack, const xmlNode *needle, bool exact) { CRM_CHECK(needle != NULL, return NULL); if (needle->type == XML_COMMENT_NODE) { return pcmk__xc_match(haystack, needle, exact); } else { const char *id = pcmk__xe_id(needle); const char *attr = (id == NULL)? NULL : PCMK_XA_ID; return pcmk__xe_match(haystack, (const char *) needle->name, attr, id); } } void xml_accept_changes(xmlNode * xml) { xmlNode *top = NULL; xml_doc_private_t *docpriv = NULL; if(xml == NULL) { return; } crm_trace("Accepting changes to %p", xml); docpriv = xml->doc->_private; top = xmlDocGetRootElement(xml->doc); reset_xml_private_data(xml->doc->_private); if (!pcmk_is_set(docpriv->flags, pcmk__xf_dirty)) { docpriv->flags = pcmk__xf_none; return; } docpriv->flags = pcmk__xf_none; accept_attr_deletions(top); } xmlNode * find_xml_node(const xmlNode *root, const char *search_path, gboolean must_find) { xmlNode *a_child = NULL; const char *name = (root == NULL)? "" : (const char *) root->name; if (search_path == NULL) { crm_warn("Will never find "); return NULL; } for (a_child = pcmk__xml_first_child(root); a_child != NULL; a_child = pcmk__xml_next(a_child)) { if (strcmp((const char *)a_child->name, search_path) == 0) { return a_child; } } if (must_find) { crm_warn("Could not find %s in %s.", search_path, name); } else if (root != NULL) { crm_trace("Could not find %s in %s.", search_path, name); } else { crm_trace("Could not find %s in .", search_path); } return NULL; } #define attr_matches(c, n, v) pcmk__str_eq(crm_element_value((c), (n)), \ (v), pcmk__str_none) /*! * \internal * \brief Find first XML child element matching given criteria * * \param[in] parent XML element to search * \param[in] node_name If not NULL, only match children of this type * \param[in] attr_n If not NULL, only match children with an attribute * of this name. * \param[in] attr_v If \p attr_n and this are not NULL, only match children * with an attribute named \p attr_n and this value * * \return Matching XML child element, or NULL if none found */ xmlNode * pcmk__xe_match(const xmlNode *parent, const char *node_name, const char *attr_n, const char *attr_v) { CRM_CHECK(parent != NULL, return NULL); CRM_CHECK(attr_v == NULL || attr_n != NULL, return NULL); for (xmlNode *child = pcmk__xml_first_child(parent); child != NULL; child = pcmk__xml_next(child)) { if (((node_name == NULL) || pcmk__xe_is(child, node_name)) && ((attr_n == NULL) || (attr_v == NULL && xmlHasProp(child, (pcmkXmlStr) attr_n)) || (attr_v != NULL && attr_matches(child, attr_n, attr_v)))) { return child; } } crm_trace("XML child node <%s%s%s%s%s> not found in %s", (node_name? node_name : "(any)"), (attr_n? " " : ""), (attr_n? attr_n : ""), (attr_n? "=" : ""), (attr_n? attr_v : ""), (const char *) parent->name); return NULL; } void copy_in_properties(xmlNode *target, const xmlNode *src) { if (src == NULL) { crm_warn("No node to copy properties from"); } else if (target == NULL) { crm_err("No node to copy properties into"); } else { for (xmlAttrPtr a = pcmk__xe_first_attr(src); a != NULL; a = a->next) { const char *p_name = (const char *) a->name; const char *p_value = pcmk__xml_attr_value(a); expand_plus_plus(target, p_name, p_value); if (xml_acl_denied(target)) { crm_trace("Cannot copy %s=%s to %s", p_name, p_value, target->name); return; } } } return; } /*! * \brief Parse integer assignment statements on this node and all its child * nodes * * \param[in,out] target Root XML node to be processed * * \note This function is recursive */ void fix_plus_plus_recursive(xmlNode *target) { /* TODO: Remove recursion and use xpath searches for value++ */ xmlNode *child = NULL; for (xmlAttrPtr a = pcmk__xe_first_attr(target); a != NULL; a = a->next) { const char *p_name = (const char *) a->name; const char *p_value = pcmk__xml_attr_value(a); expand_plus_plus(target, p_name, p_value); } for (child = pcmk__xml_first_child(target); child != NULL; child = pcmk__xml_next(child)) { fix_plus_plus_recursive(child); } } /*! * \brief Update current XML attribute value per parsed integer assignment statement * * \param[in,out] target an XML node, containing a XML attribute that is * initialized to some numeric value, to be processed * \param[in] name name of the XML attribute, e.g. X, whose value * should be updated * \param[in] value assignment statement, e.g. "X++" or * "X+=5", to be applied to the initialized value. * * \note The original XML attribute value is treated as 0 if non-numeric and * truncated to be an integer if decimal-point-containing. * \note The final XML attribute value is truncated to not exceed 1000000. * \note Undefined behavior if unexpected input. */ void expand_plus_plus(xmlNode * target, const char *name, const char *value) { int offset = 1; int name_len = 0; int int_value = 0; int value_len = 0; const char *old_value = NULL; if (target == NULL || value == NULL || name == NULL) { return; } old_value = crm_element_value(target, name); if (old_value == NULL) { /* if no previous value, set unexpanded */ goto set_unexpanded; } else if (strstr(value, name) != value) { goto set_unexpanded; } name_len = strlen(name); value_len = strlen(value); if (value_len < (name_len + 2) || value[name_len] != '+' || (value[name_len + 1] != '+' && value[name_len + 1] != '=')) { goto set_unexpanded; } /* if we are expanding ourselves, * then no previous value was set and leave int_value as 0 */ if (old_value != value) { int_value = char2score(old_value); } if (value[name_len + 1] != '+') { const char *offset_s = value + (name_len + 2); offset = char2score(offset_s); } int_value += offset; - if (int_value > INFINITY) { - int_value = (int)INFINITY; + if (int_value > PCMK_SCORE_INFINITY) { + int_value = PCMK_SCORE_INFINITY; } crm_xml_add_int(target, name, int_value); return; set_unexpanded: if (old_value == value) { /* the old value is already set, nothing to do */ return; } crm_xml_add(target, name, value); return; } /*! * \internal * \brief Remove an XML element's attributes that match some criteria * * \param[in,out] element XML element to modify * \param[in] match If not NULL, only remove attributes for which * this function returns true * \param[in,out] user_data Data to pass to \p match */ void pcmk__xe_remove_matching_attrs(xmlNode *element, bool (*match)(xmlAttrPtr, void *), void *user_data) { xmlAttrPtr next = NULL; for (xmlAttrPtr a = pcmk__xe_first_attr(element); a != NULL; a = next) { next = a->next; // Grab now because attribute might get removed if ((match == NULL) || match(a, user_data)) { if (!pcmk__check_acl(element, NULL, pcmk__xf_acl_write)) { crm_trace("ACLs prevent removal of attributes (%s and " "possibly others) from %s element", (const char *) a->name, (const char *) element->name); return; // ACLs apply to element, not particular attributes } if (pcmk__tracking_xml_changes(element, false)) { // Leave (marked for removal) until after diff is calculated set_parent_flag(element, pcmk__xf_dirty); pcmk__set_xml_flags((xml_node_private_t *) a->_private, pcmk__xf_deleted); } else { xmlRemoveProp(a); } } } } xmlNode * add_node_copy(xmlNode * parent, xmlNode * src_node) { xmlNode *child = NULL; CRM_CHECK((parent != NULL) && (src_node != NULL), return NULL); child = xmlDocCopyNode(src_node, parent->doc, 1); if (child == NULL) { return NULL; } xmlAddChild(parent, child); pcmk__mark_xml_created(child); return child; } xmlNode * create_xml_node(xmlNode * parent, const char *name) { xmlDoc *doc = NULL; xmlNode *node = NULL; if (pcmk__str_empty(name)) { CRM_CHECK(name != NULL && name[0] == 0, return NULL); return NULL; } if (parent == NULL) { doc = xmlNewDoc((pcmkXmlStr) "1.0"); if (doc == NULL) { return NULL; } node = xmlNewDocRawNode(doc, NULL, (pcmkXmlStr) name, NULL); if (node == NULL) { xmlFreeDoc(doc); return NULL; } xmlDocSetRootElement(doc, node); } else { node = xmlNewChild(parent, NULL, (pcmkXmlStr) name, NULL); if (node == NULL) { return NULL; } } pcmk__mark_xml_created(node); return node; } /*! * \internal * \brief Set a given string as an XML node's content * * \param[in,out] node Node whose content to set * \param[in] content String to set as the content * * \note \c xmlNodeSetContent() does not escape special characters. */ void pcmk__xe_set_content(xmlNode *node, const char *content) { if (node != NULL) { char *escaped = pcmk__xml_escape(content, false); xmlNodeSetContent(node, (pcmkXmlStr) escaped); free(escaped); } } xmlNode * pcmk_create_xml_text_node(xmlNode * parent, const char *name, const char *content) { xmlNode *node = create_xml_node(parent, name); pcmk__xe_set_content(node, content); return node; } xmlNode * pcmk_create_html_node(xmlNode * parent, const char *element_name, const char *id, const char *class_name, const char *text) { xmlNode *node = pcmk_create_xml_text_node(parent, element_name, text); if (class_name != NULL) { crm_xml_add(node, PCMK_XA_CLASS, class_name); } if (id != NULL) { crm_xml_add(node, PCMK_XA_ID, id); } return node; } /*! * Free an XML element and all of its children, removing it from its parent * * \param[in,out] xml XML element to free */ void pcmk_free_xml_subtree(xmlNode *xml) { xmlUnlinkNode(xml); // Detaches from parent and siblings xmlFreeNode(xml); // Frees } static void free_xml_with_position(xmlNode * child, int position) { if (child != NULL) { xmlNode *top = NULL; xmlDoc *doc = child->doc; xml_node_private_t *nodepriv = child->_private; xml_doc_private_t *docpriv = NULL; if (doc != NULL) { top = xmlDocGetRootElement(doc); } if (doc != NULL && top == child) { /* Free everything */ xmlFreeDoc(doc); } else if (pcmk__check_acl(child, NULL, pcmk__xf_acl_write) == FALSE) { GString *xpath = NULL; pcmk__if_tracing({}, return); xpath = pcmk__element_xpath(child); qb_log_from_external_source(__func__, __FILE__, "Cannot remove %s %x", LOG_TRACE, __LINE__, 0, (const char *) xpath->str, nodepriv->flags); g_string_free(xpath, TRUE); return; } else { if (doc && pcmk__tracking_xml_changes(child, FALSE) && !pcmk_is_set(nodepriv->flags, pcmk__xf_created)) { GString *xpath = pcmk__element_xpath(child); if (xpath != NULL) { pcmk__deleted_xml_t *deleted_obj = NULL; crm_trace("Deleting %s %p from %p", (const char *) xpath->str, child, doc); deleted_obj = calloc(1, sizeof(pcmk__deleted_xml_t)); deleted_obj->path = strdup((const char *) xpath->str); CRM_ASSERT(deleted_obj->path != NULL); g_string_free(xpath, TRUE); deleted_obj->position = -1; /* Record the "position" only for XML comments for now */ if (child->type == XML_COMMENT_NODE) { if (position >= 0) { deleted_obj->position = position; } else { deleted_obj->position = pcmk__xml_position(child, pcmk__xf_skip); } } docpriv = doc->_private; docpriv->deleted_objs = g_list_append(docpriv->deleted_objs, deleted_obj); pcmk__set_xml_doc_flag(child, pcmk__xf_dirty); } } pcmk_free_xml_subtree(child); } } } void free_xml(xmlNode * child) { free_xml_with_position(child, -1); } xmlNode * copy_xml(xmlNode * src) { xmlDoc *doc = xmlNewDoc((pcmkXmlStr) "1.0"); xmlNode *copy = xmlDocCopyNode(src, doc, 1); CRM_ASSERT(copy != NULL); xmlDocSetRootElement(doc, copy); return copy; } xmlNode * string2xml(const char *input) { xmlNode *xml = NULL; xmlDocPtr output = NULL; xmlParserCtxtPtr ctxt = NULL; const xmlError *last_error = NULL; if (input == NULL) { crm_err("Can't parse NULL input"); return NULL; } /* create a parser context */ ctxt = xmlNewParserCtxt(); CRM_CHECK(ctxt != NULL, return NULL); xmlCtxtResetLastError(ctxt); xmlSetGenericErrorFunc(ctxt, pcmk__log_xmllib_err); output = xmlCtxtReadDoc(ctxt, (pcmkXmlStr) input, NULL, NULL, PCMK__XML_PARSE_OPTS_WITHOUT_RECOVER); if (output == NULL) { output = xmlCtxtReadDoc(ctxt, (pcmkXmlStr) input, NULL, NULL, PCMK__XML_PARSE_OPTS_WITH_RECOVER); if (output) { crm_warn("Successfully recovered from XML errors " "(note: a future release will treat this as a fatal failure)"); } } if (output) { xml = xmlDocGetRootElement(output); } last_error = xmlCtxtGetLastError(ctxt); if (last_error && last_error->code != XML_ERR_OK) { /* crm_abort(__FILE__,__func__,__LINE__, "last_error->code != XML_ERR_OK", TRUE, TRUE); */ /* * http://xmlsoft.org/html/libxml-xmlerror.html#xmlErrorLevel * http://xmlsoft.org/html/libxml-xmlerror.html#xmlParserErrors */ crm_warn("Parsing failed (domain=%d, level=%d, code=%d): %s", last_error->domain, last_error->level, last_error->code, last_error->message); if (last_error->code == XML_ERR_DOCUMENT_EMPTY) { CRM_LOG_ASSERT("Cannot parse an empty string"); } else if (last_error->code != XML_ERR_DOCUMENT_END) { crm_err("Couldn't%s parse %d chars: %s", xml ? " fully" : "", (int)strlen(input), input); if (xml != NULL) { crm_log_xml_err(xml, "Partial"); } } else { int len = strlen(input); int lpc = 0; while(lpc < len) { crm_warn("Parse error[+%.3d]: %.80s", lpc, input+lpc); lpc += 80; } CRM_LOG_ASSERT("String parsing error"); } } xmlFreeParserCtxt(ctxt); return xml; } xmlNode * stdin2xml(void) { size_t data_length = 0; size_t read_chars = 0; char *xml_buffer = NULL; xmlNode *xml_obj = NULL; do { xml_buffer = pcmk__realloc(xml_buffer, data_length + PCMK__BUFFER_SIZE); read_chars = fread(xml_buffer + data_length, 1, PCMK__BUFFER_SIZE, stdin); data_length += read_chars; } while (read_chars == PCMK__BUFFER_SIZE); if (data_length == 0) { crm_warn("No XML supplied on stdin"); free(xml_buffer); return NULL; } xml_buffer[data_length] = '\0'; xml_obj = string2xml(xml_buffer); free(xml_buffer); crm_log_xml_trace(xml_obj, "Created fragment"); return xml_obj; } static char * decompress_file(const char *filename) { char *buffer = NULL; int rc = 0; size_t length = 0, read_len = 0; BZFILE *bz_file = NULL; FILE *input = fopen(filename, "r"); if (input == NULL) { crm_perror(LOG_ERR, "Could not open %s for reading", filename); return NULL; } bz_file = BZ2_bzReadOpen(&rc, input, 0, 0, NULL, 0); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_err("Could not prepare to read compressed %s: %s " CRM_XS " rc=%d", filename, pcmk_rc_str(rc), rc); BZ2_bzReadClose(&rc, bz_file); fclose(input); return NULL; } rc = BZ_OK; // cppcheck seems not to understand the abort-logic in pcmk__realloc // cppcheck-suppress memleak while (rc == BZ_OK) { buffer = pcmk__realloc(buffer, PCMK__BUFFER_SIZE + length + 1); read_len = BZ2_bzRead(&rc, bz_file, buffer + length, PCMK__BUFFER_SIZE); crm_trace("Read %ld bytes from file: %d", (long)read_len, rc); if (rc == BZ_OK || rc == BZ_STREAM_END) { length += read_len; } } buffer[length] = '\0'; rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_err("Could not read compressed %s: %s " CRM_XS " rc=%d", filename, pcmk_rc_str(rc), rc); free(buffer); buffer = NULL; } BZ2_bzReadClose(&rc, bz_file); fclose(input); return buffer; } /*! * \internal * \brief Remove XML text nodes from specified XML and all its children * * \param[in,out] xml XML to strip text from */ void pcmk__strip_xml_text(xmlNode *xml) { xmlNode *iter = xml->children; while (iter) { xmlNode *next = iter->next; switch (iter->type) { case XML_TEXT_NODE: /* Remove it */ pcmk_free_xml_subtree(iter); break; case XML_ELEMENT_NODE: /* Search it */ pcmk__strip_xml_text(iter); break; default: /* Leave it */ break; } iter = next; } } xmlNode * filename2xml(const char *filename) { xmlNode *xml = NULL; xmlDocPtr output = NULL; bool uncompressed = true; xmlParserCtxtPtr ctxt = NULL; const xmlError *last_error = NULL; /* create a parser context */ ctxt = xmlNewParserCtxt(); CRM_CHECK(ctxt != NULL, return NULL); xmlCtxtResetLastError(ctxt); xmlSetGenericErrorFunc(ctxt, pcmk__log_xmllib_err); if (filename) { uncompressed = !pcmk__ends_with_ext(filename, ".bz2"); } if (pcmk__str_eq(filename, "-", pcmk__str_null_matches)) { /* STDIN_FILENO == fileno(stdin) */ output = xmlCtxtReadFd(ctxt, STDIN_FILENO, "unknown.xml", NULL, PCMK__XML_PARSE_OPTS_WITHOUT_RECOVER); if (output == NULL) { output = xmlCtxtReadFd(ctxt, STDIN_FILENO, "unknown.xml", NULL, PCMK__XML_PARSE_OPTS_WITH_RECOVER); if (output) { crm_warn("Successfully recovered from XML errors " "(note: a future release will treat this as a fatal failure)"); } } } else if (uncompressed) { output = xmlCtxtReadFile(ctxt, filename, NULL, PCMK__XML_PARSE_OPTS_WITHOUT_RECOVER); if (output == NULL) { output = xmlCtxtReadFile(ctxt, filename, NULL, PCMK__XML_PARSE_OPTS_WITH_RECOVER); if (output) { crm_warn("Successfully recovered from XML errors " "(note: a future release will treat this as a fatal failure)"); } } } else { char *input = decompress_file(filename); output = xmlCtxtReadDoc(ctxt, (pcmkXmlStr) input, NULL, NULL, PCMK__XML_PARSE_OPTS_WITHOUT_RECOVER); if (output == NULL) { output = xmlCtxtReadDoc(ctxt, (pcmkXmlStr) input, NULL, NULL, PCMK__XML_PARSE_OPTS_WITH_RECOVER); if (output) { crm_warn("Successfully recovered from XML errors " "(note: a future release will treat this as a fatal failure)"); } } free(input); } if (output && (xml = xmlDocGetRootElement(output))) { pcmk__strip_xml_text(xml); } last_error = xmlCtxtGetLastError(ctxt); if (last_error && last_error->code != XML_ERR_OK) { /* crm_abort(__FILE__,__func__,__LINE__, "last_error->code != XML_ERR_OK", TRUE, TRUE); */ /* * http://xmlsoft.org/html/libxml-xmlerror.html#xmlErrorLevel * http://xmlsoft.org/html/libxml-xmlerror.html#xmlParserErrors */ crm_err("Parsing failed (domain=%d, level=%d, code=%d): %s", last_error->domain, last_error->level, last_error->code, last_error->message); if (last_error && last_error->code != XML_ERR_OK) { crm_err("Couldn't%s parse %s", xml ? " fully" : "", filename); if (xml != NULL) { crm_log_xml_err(xml, "Partial"); } } } xmlFreeParserCtxt(ctxt); return xml; } /*! * \internal * \brief Add a "last written" attribute to an XML element, set to current time * * \param[in,out] xe XML element to add attribute to * * \return Value that was set, or NULL on error */ const char * pcmk__xe_add_last_written(xmlNode *xe) { char *now_s = pcmk__epoch2str(NULL, 0); const char *result = NULL; result = crm_xml_add(xe, PCMK_XA_CIB_LAST_WRITTEN, pcmk__s(now_s, "Could not determine current time")); free(now_s); return result; } /*! * \brief Sanitize a string so it is usable as an XML ID * * \param[in,out] id String to sanitize */ void crm_xml_sanitize_id(char *id) { char *c; for (c = id; *c; ++c) { /* @TODO Sanitize more comprehensively */ switch (*c) { case ':': case '#': *c = '.'; } } } /*! * \brief Set the ID of an XML element using a format * * \param[in,out] xml XML element * \param[in] fmt printf-style format * \param[in] ... any arguments required by format */ void crm_xml_set_id(xmlNode *xml, const char *format, ...) { va_list ap; int len = 0; char *id = NULL; /* equivalent to crm_strdup_printf() */ va_start(ap, format); len = vasprintf(&id, format, ap); va_end(ap); CRM_ASSERT(len > 0); crm_xml_sanitize_id(id); crm_xml_add(xml, PCMK_XA_ID, id); free(id); } /*! * \internal * \brief Write XML to a file stream * * \param[in] xml XML to write * \param[in] filename Name of file being written (for logging only) * \param[in,out] stream Open file stream corresponding to filename * \param[in] compress Whether to compress XML before writing * \param[out] nbytes Number of bytes written * * \return Standard Pacemaker return code */ static int write_xml_stream(const xmlNode *xml, const char *filename, FILE *stream, bool compress, unsigned int *nbytes) { int rc = pcmk_rc_ok; char *buffer = NULL; *nbytes = 0; crm_log_xml_trace(xml, "writing"); buffer = dump_xml_formatted(xml); CRM_CHECK(buffer && strlen(buffer), crm_log_xml_warn(xml, "formatting failed"); rc = pcmk_rc_error; goto bail); if (compress) { unsigned int in = 0; BZFILE *bz_file = NULL; rc = BZ_OK; bz_file = BZ2_bzWriteOpen(&rc, stream, 5, 0, 30); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_warn("Not compressing %s: could not prepare file stream: %s " CRM_XS " rc=%d", filename, pcmk_rc_str(rc), rc); } else { BZ2_bzWrite(&rc, bz_file, buffer, strlen(buffer)); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_warn("Not compressing %s: could not compress data: %s " CRM_XS " rc=%d errno=%d", filename, pcmk_rc_str(rc), rc, errno); } } if (rc == pcmk_rc_ok) { BZ2_bzWriteClose(&rc, bz_file, 0, &in, nbytes); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_warn("Not compressing %s: could not write compressed data: %s " CRM_XS " rc=%d errno=%d", filename, pcmk_rc_str(rc), rc, errno); *nbytes = 0; // retry without compression } else { crm_trace("Compressed XML for %s from %u bytes to %u", filename, in, *nbytes); } } rc = pcmk_rc_ok; // Either true, or we'll retry without compression } if (*nbytes == 0) { rc = fprintf(stream, "%s", buffer); if (rc < 0) { rc = errno; crm_perror(LOG_ERR, "writing %s", filename); } else { *nbytes = (unsigned int) rc; rc = pcmk_rc_ok; } } bail: if (fflush(stream) != 0) { rc = errno; crm_perror(LOG_ERR, "flushing %s", filename); } /* Don't report error if the file does not support synchronization */ if (fsync(fileno(stream)) < 0 && errno != EROFS && errno != EINVAL) { rc = errno; crm_perror(LOG_ERR, "synchronizing %s", filename); } fclose(stream); crm_trace("Saved %d bytes to %s as XML", *nbytes, filename); free(buffer); return rc; } /*! * \brief Write XML to a file descriptor * * \param[in] xml XML to write * \param[in] filename Name of file being written (for logging only) * \param[in] fd Open file descriptor corresponding to filename * \param[in] compress Whether to compress XML before writing * * \return Number of bytes written on success, -errno otherwise */ int write_xml_fd(const xmlNode *xml, const char *filename, int fd, gboolean compress) { FILE *stream = NULL; unsigned int nbytes = 0; int rc = pcmk_rc_ok; CRM_CHECK((xml != NULL) && (fd > 0), return -EINVAL); stream = fdopen(fd, "w"); if (stream == NULL) { return -errno; } rc = write_xml_stream(xml, filename, stream, compress, &nbytes); if (rc != pcmk_rc_ok) { return pcmk_rc2legacy(rc); } return (int) nbytes; } /*! * \brief Write XML to a file * * \param[in] xml XML to write * \param[in] filename Name of file to write * \param[in] compress Whether to compress XML before writing * * \return Number of bytes written on success, -errno otherwise */ int write_xml_file(const xmlNode *xml, const char *filename, gboolean compress) { FILE *stream = NULL; unsigned int nbytes = 0; int rc = pcmk_rc_ok; CRM_CHECK((xml != NULL) && (filename != NULL), return -EINVAL); stream = fopen(filename, "w"); if (stream == NULL) { return -errno; } rc = write_xml_stream(xml, filename, stream, compress, &nbytes); if (rc != pcmk_rc_ok) { return pcmk_rc2legacy(rc); } return (int) nbytes; } /*! * \internal * \brief Get consecutive bytes encoding non-ASCII UTF-8 characters * * \param[in] text String to check * * \return Number of non-ASCII UTF-8 bytes at the beginning of \p text */ static size_t utf8_bytes(const char *text) { // Total number of consecutive bytes containing UTF-8 characters size_t c_bytes = 0; if (text == NULL) { return 0; } /* UTF-8 uses one to four 8-bit bytes per character. The first byte * indicates the width of the character. A byte beginning with a '0' bit is * a one-byte ASCII character. * * A C byte is 8 bits on most systems, but this is not guaranteed. * * Count until we find an ASCII character or an invalid byte. Check bytes * aligned with the C byte boundary. */ for (const uint8_t *utf8_byte = (const uint8_t *) text; (*utf8_byte & 0x80) != 0; utf8_byte = (const uint8_t *) (text + c_bytes)) { size_t utf8_bits = 0; if ((*utf8_byte & 0xf0) == 0xf0) { // Four-byte character (first byte: 11110xxx) utf8_bits = 32; } else if ((*utf8_byte & 0xe0) == 0xe0) { // Three-byte character (first byte: 1110xxxx) utf8_bits = 24; } else if ((*utf8_byte & 0xc0) == 0xc0) { // Two-byte character (first byte: 110xxxxx) utf8_bits = 16; } else { crm_warn("Found invalid UTF-8 character %.2x", (unsigned char) *utf8_byte); return c_bytes; } c_bytes += utf8_bits / CHAR_BIT; #if (CHAR_BIT != 8) // Coverity complains about dead code without this CPP guard if ((utf8_bits % CHAR_BIT) > 0) { c_bytes++; } #endif // CHAR_BIT != 8 } return c_bytes; } /*! * \internal * \brief Replace a character in a dynamically allocated string, reallocating * memory * * \param[in,out] text String to replace a character in * \param[in,out] index Index of character to replace with new string; on * return, reset to index of end of replacement string * \param[in,out] length Length of \p text * \param[in] replace String to replace character at \p index with (must * not be empty) * * \return \p text, with the character at \p index replaced by \p replace */ static char * replace_text(char *text, size_t *index, size_t *length, const char *replace) { /* @TODO Replace with GString? Or at least copy char-by-char, escaping * characters as needed, instead of shifting characters on every replacement */ // We have space for 1 char already size_t offset = strlen(replace) - 1; if (offset > 0) { *length += offset; text = pcmk__realloc(text, *length + 1); // Shift characters to the right to make room for the replacement string for (size_t i = *length; i > (*index + offset); i--) { text[i] = text[i - offset]; } } // Replace the character at index by the replacement string memcpy(text + *index, replace, offset + 1); // Reset index to the end of replacement string *index += offset; return text; } /*! * \internal * \brief Check whether a string has XML special characters that must be escaped * * See \c pcmk__xml_escape() for more details. * * \param[in] text String to check * \param[in] escape_quote If \c true, double quotes must be escaped * * \return \c true if \p text has special characters that need to be escaped, or * \c false otherwise */ bool pcmk__xml_needs_escape(const char *text, bool escape_quote) { size_t length = 0; if (text == NULL) { return false; } length = strlen(text); for (size_t index = 0; index < length; index++) { // Don't escape any non-ASCII characters index += utf8_bytes(&(text[index])); switch (text[index]) { case '\0': // Reached end of string by skipping UTF-8 bytes return false; case '<': return true; case '>': // Not necessary, but for symmetry with '<' return true; case '&': return true; case '"': if (escape_quote) { return true; } break; case '\n': case '\t': // Don't escape newline or tab break; default: if ((text[index] < 0x20) || (text[index] >= 0x7f)) { // Escape non-printing characters return true; } break; } } return false; } /*! * \internal * \brief Replace special characters with their XML escape sequences * * XML allows the escaping of special characters by replacing them with entity * references (for example, """) or character references (for * example, " "). * * The special characters '<' and '&' are not allowed in their * literal forms in XML character data. Character data is non-markup text (for * example, the content of a text node). * * Additionally, if an attribute value is delimited by single quotes, then * single quotes must be escaped within the value. Similarly, if an attribute * value is delimited by double quotes, then double quotes must be escaped * within the value. * * For more details, see the "Character Data and Markup" section of the XML * spec, currently section 2.4: * https://www.w3.org/TR/xml/#dt-markup * * Pacemaker always delimits attribute values with double quotes, so this * function doesn't escape single quotes. * * \param[in] text Text to escape * \param[in] escape_quote If \c true, escape double quotes (should be enabled * for attribute values) * * \return Newly allocated string equivalent to \p text but with special * characters replaced with XML escape sequences (or \c NULL if \p text * is \c NULL). If \p text is not \c NULL, the return value is * guaranteed not to be \c NULL. * * \note There are libxml functions that purport to do this: * \c xmlEncodeEntitiesReentrant() and \c xmlEncodeSpecialChars(). * However, their escaping is incomplete. See: * https://discourse.gnome.org/t/intended-use-of-xmlencodeentitiesreentrant-vs-xmlencodespecialchars/19252 */ char * pcmk__xml_escape(const char *text, bool escape_quote) { size_t length = 0; char *copy = NULL; char buf[32] = { '\0', }; if (text == NULL) { return NULL; } length = strlen(text); pcmk__str_update(©, text); for (size_t index = 0; index < length; index++) { // Don't escape any non-ASCII characters index += utf8_bytes(&(copy[index])); switch (copy[index]) { case '\0': // Reached end of string by skipping UTF-8 bytes break; case '<': copy = replace_text(copy, &index, &length, "<"); break; case '>': // Not necessary, but for symmetry with '<' copy = replace_text(copy, &index, &length, ">"); break; case '&': copy = replace_text(copy, &index, &length, "&"); break; case '"': if (escape_quote) { copy = replace_text(copy, &index, &length, """); } break; case '\n': case '\t': // Don't escape newlines and tabs break; default: if ((copy[index] < 0x20) || (copy[index] >= 0x7f)) { // Escape non-printing characters snprintf(buf, sizeof(buf), "&#%.2x;", copy[index]); copy = replace_text(copy, &index, &length, buf); } break; } } return copy; } /*! * \internal * \brief Append a string representation of an XML element to a buffer * * \param[in] data XML whose representation to append * \param[in] options Group of \p pcmk__xml_fmt_options flags * \param[in,out] buffer Where to append the content (must not be \p NULL) * \param[in] depth Current indentation level */ static void dump_xml_element(const xmlNode *data, uint32_t options, GString *buffer, int depth) { bool pretty = pcmk_is_set(options, pcmk__xml_fmt_pretty); bool filtered = pcmk_is_set(options, pcmk__xml_fmt_filtered); int spaces = pretty? (2 * depth) : 0; for (int lpc = 0; lpc < spaces; lpc++) { g_string_append_c(buffer, ' '); } pcmk__g_strcat(buffer, "<", data->name, NULL); for (const xmlAttr *attr = pcmk__xe_first_attr(data); attr != NULL; attr = attr->next) { if (!filtered || !pcmk__xa_filterable((const char *) (attr->name))) { pcmk__dump_xml_attr(attr, buffer); } } if (data->children == NULL) { g_string_append(buffer, "/>"); } else { g_string_append_c(buffer, '>'); } if (pretty) { g_string_append_c(buffer, '\n'); } if (data->children) { for (const xmlNode *child = data->children; child != NULL; child = child->next) { pcmk__xml2text(child, options, buffer, depth + 1); } for (int lpc = 0; lpc < spaces; lpc++) { g_string_append_c(buffer, ' '); } pcmk__g_strcat(buffer, "name, ">", NULL); if (pretty) { g_string_append_c(buffer, '\n'); } } } /*! * \internal * \brief Append XML text content to a buffer * * \param[in] data XML whose content to append * \param[in] options Group of \p xml_log_options flags * \param[in,out] buffer Where to append the content (must not be \p NULL) * \param[in] depth Current indentation level */ static void dump_xml_text(const xmlNode *data, uint32_t options, GString *buffer, int depth) { bool pretty = pcmk_is_set(options, pcmk__xml_fmt_pretty); int spaces = pretty? (2 * depth) : 0; const char *content = (const char *) data->content; char *content_esc = NULL; if (pcmk__xml_needs_escape(content, false)) { content_esc = pcmk__xml_escape(content, false); content = content_esc; } for (int lpc = 0; lpc < spaces; lpc++) { g_string_append_c(buffer, ' '); } g_string_append(buffer, content); if (pretty) { g_string_append_c(buffer, '\n'); } free(content_esc); } /*! * \internal * \brief Append XML CDATA content to a buffer * * \param[in] data XML whose content to append * \param[in] options Group of \p pcmk__xml_fmt_options flags * \param[in,out] buffer Where to append the content (must not be \p NULL) * \param[in] depth Current indentation level */ static void dump_xml_cdata(const xmlNode *data, uint32_t options, GString *buffer, int depth) { bool pretty = pcmk_is_set(options, pcmk__xml_fmt_pretty); int spaces = pretty? (2 * depth) : 0; for (int lpc = 0; lpc < spaces; lpc++) { g_string_append_c(buffer, ' '); } pcmk__g_strcat(buffer, "content, "]]>", NULL); if (pretty) { g_string_append_c(buffer, '\n'); } } /*! * \internal * \brief Append an XML comment to a buffer * * \param[in] data XML whose content to append * \param[in] options Group of \p pcmk__xml_fmt_options flags * \param[in,out] buffer Where to append the content (must not be \p NULL) * \param[in] depth Current indentation level */ static void dump_xml_comment(const xmlNode *data, uint32_t options, GString *buffer, int depth) { bool pretty = pcmk_is_set(options, pcmk__xml_fmt_pretty); int spaces = pretty? (2 * depth) : 0; for (int lpc = 0; lpc < spaces; lpc++) { g_string_append_c(buffer, ' '); } pcmk__g_strcat(buffer, "", NULL); if (pretty) { g_string_append_c(buffer, '\n'); } } /*! * \internal * \brief Get a string representation of an XML element type * * \param[in] type XML element type * * \return String representation of \p type */ static const char * xml_element_type2str(xmlElementType type) { static const char *const element_type_names[] = { [XML_ELEMENT_NODE] = "element", [XML_ATTRIBUTE_NODE] = "attribute", [XML_TEXT_NODE] = "text", [XML_CDATA_SECTION_NODE] = "CDATA section", [XML_ENTITY_REF_NODE] = "entity reference", [XML_ENTITY_NODE] = "entity", [XML_PI_NODE] = "PI", [XML_COMMENT_NODE] = "comment", [XML_DOCUMENT_NODE] = "document", [XML_DOCUMENT_TYPE_NODE] = "document type", [XML_DOCUMENT_FRAG_NODE] = "document fragment", [XML_NOTATION_NODE] = "notation", [XML_HTML_DOCUMENT_NODE] = "HTML document", [XML_DTD_NODE] = "DTD", [XML_ELEMENT_DECL] = "element declaration", [XML_ATTRIBUTE_DECL] = "attribute declaration", [XML_ENTITY_DECL] = "entity declaration", [XML_NAMESPACE_DECL] = "namespace declaration", [XML_XINCLUDE_START] = "XInclude start", [XML_XINCLUDE_END] = "XInclude end", }; if ((type < 0) || (type >= PCMK__NELEM(element_type_names))) { return "unrecognized type"; } return element_type_names[type]; } /*! * \internal * \brief Create a text representation of an XML object * * \param[in] data XML to convert * \param[in] options Group of \p pcmk__xml_fmt_options flags * \param[in,out] buffer Where to store the text (must not be \p NULL) * \param[in] depth Current indentation level */ void pcmk__xml2text(const xmlNode *data, uint32_t options, GString *buffer, int depth) { if (data == NULL) { crm_trace("Nothing to dump"); return; } CRM_ASSERT(buffer != NULL); CRM_CHECK(depth >= 0, depth = 0); switch(data->type) { case XML_ELEMENT_NODE: /* Handle below */ dump_xml_element(data, options, buffer, depth); break; case XML_TEXT_NODE: if (pcmk_is_set(options, pcmk__xml_fmt_text)) { dump_xml_text(data, options, buffer, depth); } break; case XML_COMMENT_NODE: dump_xml_comment(data, options, buffer, depth); break; case XML_CDATA_SECTION_NODE: dump_xml_cdata(data, options, buffer, depth); break; default: crm_warn("Cannot convert XML %s node to text " CRM_XS " type=%d", xml_element_type2str(data->type), data->type); break; } } char * dump_xml_formatted_with_text(const xmlNode *xml) { /* libxml's xmlNodeDumpOutput() would work here since we're not specifically * filtering out any nodes. However, use pcmk__xml2text() for consistency, * to escape attribute values, and to allow a const argument. */ char *buffer = NULL; GString *g_buffer = g_string_sized_new(1024); pcmk__xml2text(xml, pcmk__xml_fmt_pretty|pcmk__xml_fmt_text, g_buffer, 0); pcmk__str_update(&buffer, g_buffer->str); g_string_free(g_buffer, TRUE); return buffer; } char * dump_xml_formatted(const xmlNode *xml) { char *buffer = NULL; GString *g_buffer = g_string_sized_new(1024); pcmk__xml2text(xml, pcmk__xml_fmt_pretty, g_buffer, 0); pcmk__str_update(&buffer, g_buffer->str); g_string_free(g_buffer, TRUE); return buffer; } char * dump_xml_unformatted(const xmlNode *xml) { char *buffer = NULL; GString *g_buffer = g_string_sized_new(1024); pcmk__xml2text(xml, 0, g_buffer, 0); pcmk__str_update(&buffer, g_buffer->str); g_string_free(g_buffer, TRUE); return buffer; } int pcmk__xml2fd(int fd, xmlNode *cur) { bool success; xmlOutputBuffer *fd_out = xmlOutputBufferCreateFd(fd, NULL); CRM_ASSERT(fd_out != NULL); xmlNodeDumpOutput(fd_out, cur->doc, cur, 0, pcmk__xml_fmt_pretty, NULL); success = xmlOutputBufferWrite(fd_out, sizeof("\n") - 1, "\n") != -1; success = xmlOutputBufferClose(fd_out) != -1 && success; if (!success) { return EIO; } fsync(fd); return pcmk_rc_ok; } void xml_remove_prop(xmlNode * obj, const char *name) { if (crm_element_value(obj, name) == NULL) { return; } if (pcmk__check_acl(obj, NULL, pcmk__xf_acl_write) == FALSE) { crm_trace("Cannot remove %s from %s", name, obj->name); } else if (pcmk__tracking_xml_changes(obj, FALSE)) { /* Leave in place (marked for removal) until after the diff is calculated */ xmlAttr *attr = xmlHasProp(obj, (pcmkXmlStr) name); xml_node_private_t *nodepriv = attr->_private; set_parent_flag(obj, pcmk__xf_dirty); pcmk__set_xml_flags(nodepriv, pcmk__xf_deleted); } else { xmlUnsetProp(obj, (pcmkXmlStr) name); } } void save_xml_to_file(const xmlNode *xml, const char *desc, const char *filename) { char *f = NULL; if (filename == NULL) { char *uuid = crm_generate_uuid(); f = crm_strdup_printf("%s/%s", pcmk__get_tmpdir(), uuid); filename = f; free(uuid); } crm_info("Saving %s to %s", desc, filename); write_xml_file(xml, filename, FALSE); free(f); } /*! * \internal * \brief Set a flag on all attributes of an XML element * * \param[in,out] xml XML node to set flags on * \param[in] flag XML private flag to set */ static void set_attrs_flag(xmlNode *xml, enum xml_private_flags flag) { for (xmlAttr *attr = pcmk__xe_first_attr(xml); attr; attr = attr->next) { pcmk__set_xml_flags((xml_node_private_t *) (attr->_private), flag); } } /*! * \internal * \brief Add an XML attribute to a node, marked as deleted * * When calculating XML changes, we need to know when an attribute has been * deleted. Add the attribute back to the new XML, so that we can check the * removal against ACLs, and mark it as deleted for later removal after * differences have been calculated. * * \param[in,out] new_xml XML to modify * \param[in] element Name of XML element that changed (for logging) * \param[in] attr_name Name of attribute that was deleted * \param[in] old_value Value of attribute that was deleted */ static void mark_attr_deleted(xmlNode *new_xml, const char *element, const char *attr_name, const char *old_value) { xml_doc_private_t *docpriv = new_xml->doc->_private; xmlAttr *attr = NULL; xml_node_private_t *nodepriv; // Prevent the dirty flag being set recursively upwards pcmk__clear_xml_flags(docpriv, pcmk__xf_tracking); // Restore the old value (and the tracking flag) attr = xmlSetProp(new_xml, (pcmkXmlStr) attr_name, (pcmkXmlStr) old_value); pcmk__set_xml_flags(docpriv, pcmk__xf_tracking); // Reset flags (so the attribute doesn't appear as newly created) nodepriv = attr->_private; nodepriv->flags = 0; // Check ACLs and mark restored value for later removal xml_remove_prop(new_xml, attr_name); crm_trace("XML attribute %s=%s was removed from %s", attr_name, old_value, element); } /* * \internal * \brief Check ACLs for a changed XML attribute */ static void mark_attr_changed(xmlNode *new_xml, const char *element, const char *attr_name, const char *old_value) { char *vcopy = crm_element_value_copy(new_xml, attr_name); crm_trace("XML attribute %s was changed from '%s' to '%s' in %s", attr_name, old_value, vcopy, element); // Restore the original value xmlSetProp(new_xml, (pcmkXmlStr) attr_name, (pcmkXmlStr) old_value); // Change it back to the new value, to check ACLs crm_xml_add(new_xml, attr_name, vcopy); free(vcopy); } /*! * \internal * \brief Mark an XML attribute as having changed position * * \param[in,out] new_xml XML to modify * \param[in] element Name of XML element that changed (for logging) * \param[in,out] old_attr Attribute that moved, in original XML * \param[in,out] new_attr Attribute that moved, in \p new_xml * \param[in] p_old Ordinal position of \p old_attr in original XML * \param[in] p_new Ordinal position of \p new_attr in \p new_xml */ static void mark_attr_moved(xmlNode *new_xml, const char *element, xmlAttr *old_attr, xmlAttr *new_attr, int p_old, int p_new) { xml_node_private_t *nodepriv = new_attr->_private; crm_trace("XML attribute %s moved from position %d to %d in %s", old_attr->name, p_old, p_new, element); // Mark document, element, and all element's parents as changed pcmk__mark_xml_node_dirty(new_xml); // Mark attribute as changed pcmk__set_xml_flags(nodepriv, pcmk__xf_dirty|pcmk__xf_moved); nodepriv = (p_old > p_new)? old_attr->_private : new_attr->_private; pcmk__set_xml_flags(nodepriv, pcmk__xf_skip); } /*! * \internal * \brief Calculate differences in all previously existing XML attributes * * \param[in,out] old_xml Original XML to compare * \param[in,out] new_xml New XML to compare */ static void xml_diff_old_attrs(xmlNode *old_xml, xmlNode *new_xml) { xmlAttr *attr_iter = pcmk__xe_first_attr(old_xml); while (attr_iter != NULL) { const char *name = (const char *) attr_iter->name; xmlAttr *old_attr = attr_iter; xmlAttr *new_attr = xmlHasProp(new_xml, attr_iter->name); const char *old_value = pcmk__xml_attr_value(attr_iter); attr_iter = attr_iter->next; if (new_attr == NULL) { mark_attr_deleted(new_xml, (const char *) old_xml->name, name, old_value); } else { xml_node_private_t *nodepriv = new_attr->_private; int new_pos = pcmk__xml_position((xmlNode*) new_attr, pcmk__xf_skip); int old_pos = pcmk__xml_position((xmlNode*) old_attr, pcmk__xf_skip); const char *new_value = crm_element_value(new_xml, name); // This attribute isn't new pcmk__clear_xml_flags(nodepriv, pcmk__xf_created); if (strcmp(new_value, old_value) != 0) { mark_attr_changed(new_xml, (const char *) old_xml->name, name, old_value); } else if ((old_pos != new_pos) && !pcmk__tracking_xml_changes(new_xml, TRUE)) { mark_attr_moved(new_xml, (const char *) old_xml->name, old_attr, new_attr, old_pos, new_pos); } } } } /*! * \internal * \brief Check all attributes in new XML for creation * * For each of a given XML element's attributes marked as newly created, accept * (and mark as dirty) or reject the creation according to ACLs. * * \param[in,out] new_xml XML to check */ static void mark_created_attrs(xmlNode *new_xml) { xmlAttr *attr_iter = pcmk__xe_first_attr(new_xml); while (attr_iter != NULL) { xmlAttr *new_attr = attr_iter; xml_node_private_t *nodepriv = attr_iter->_private; attr_iter = attr_iter->next; if (pcmk_is_set(nodepriv->flags, pcmk__xf_created)) { const char *attr_name = (const char *) new_attr->name; crm_trace("Created new attribute %s=%s in %s", attr_name, pcmk__xml_attr_value(new_attr), new_xml->name); /* Check ACLs (we can't use the remove-then-create trick because it * would modify the attribute position). */ if (pcmk__check_acl(new_xml, attr_name, pcmk__xf_acl_write)) { pcmk__mark_xml_attr_dirty(new_attr); } else { // Creation was not allowed, so remove the attribute xmlUnsetProp(new_xml, new_attr->name); } } } } /*! * \internal * \brief Calculate differences in attributes between two XML nodes * * \param[in,out] old_xml Original XML to compare * \param[in,out] new_xml New XML to compare */ static void xml_diff_attrs(xmlNode *old_xml, xmlNode *new_xml) { set_attrs_flag(new_xml, pcmk__xf_created); // cleared later if not really new xml_diff_old_attrs(old_xml, new_xml); mark_created_attrs(new_xml); } /*! * \internal * \brief Add an XML child element to a node, marked as deleted * * When calculating XML changes, we need to know when a child element has been * deleted. Add the child back to the new XML, so that we can check the removal * against ACLs, and mark it as deleted for later removal after differences have * been calculated. * * \param[in,out] old_child Child element from original XML * \param[in,out] new_parent New XML to add marked copy to */ static void mark_child_deleted(xmlNode *old_child, xmlNode *new_parent) { // Re-create the child element so we can check ACLs xmlNode *candidate = add_node_copy(new_parent, old_child); // Clear flags on new child and its children reset_xml_node_flags(candidate); // Check whether ACLs allow the deletion pcmk__apply_acl(xmlDocGetRootElement(candidate->doc)); // Remove the child again (which will track it in document's deleted_objs) free_xml_with_position(candidate, pcmk__xml_position(old_child, pcmk__xf_skip)); if (pcmk__xml_match(new_parent, old_child, true) == NULL) { pcmk__set_xml_flags((xml_node_private_t *) (old_child->_private), pcmk__xf_skip); } } static void mark_child_moved(xmlNode *old_child, xmlNode *new_parent, xmlNode *new_child, int p_old, int p_new) { xml_node_private_t *nodepriv = new_child->_private; crm_trace("Child element %s with id='%s' moved from position %d to %d under %s", new_child->name, pcmk__s(pcmk__xe_id(new_child), ""), p_old, p_new, new_parent->name); pcmk__mark_xml_node_dirty(new_parent); pcmk__set_xml_flags(nodepriv, pcmk__xf_moved); if (p_old > p_new) { nodepriv = old_child->_private; } else { nodepriv = new_child->_private; } pcmk__set_xml_flags(nodepriv, pcmk__xf_skip); } // Given original and new XML, mark new XML portions that have changed static void mark_xml_changes(xmlNode *old_xml, xmlNode *new_xml, bool check_top) { xmlNode *cIter = NULL; xml_node_private_t *nodepriv = NULL; CRM_CHECK(new_xml != NULL, return); if (old_xml == NULL) { pcmk__mark_xml_created(new_xml); pcmk__apply_creation_acl(new_xml, check_top); return; } nodepriv = new_xml->_private; CRM_CHECK(nodepriv != NULL, return); if(nodepriv->flags & pcmk__xf_processed) { /* Avoid re-comparing nodes */ return; } pcmk__set_xml_flags(nodepriv, pcmk__xf_processed); xml_diff_attrs(old_xml, new_xml); // Check for differences in the original children for (cIter = pcmk__xml_first_child(old_xml); cIter != NULL; ) { xmlNode *old_child = cIter; xmlNode *new_child = pcmk__xml_match(new_xml, cIter, true); cIter = pcmk__xml_next(cIter); if(new_child) { mark_xml_changes(old_child, new_child, TRUE); } else { mark_child_deleted(old_child, new_xml); } } // Check for moved or created children for (cIter = pcmk__xml_first_child(new_xml); cIter != NULL; ) { xmlNode *new_child = cIter; xmlNode *old_child = pcmk__xml_match(old_xml, cIter, true); cIter = pcmk__xml_next(cIter); if(old_child == NULL) { // This is a newly created child nodepriv = new_child->_private; pcmk__set_xml_flags(nodepriv, pcmk__xf_skip); mark_xml_changes(old_child, new_child, TRUE); } else { /* Check for movement, we already checked for differences */ int p_new = pcmk__xml_position(new_child, pcmk__xf_skip); int p_old = pcmk__xml_position(old_child, pcmk__xf_skip); if(p_old != p_new) { mark_child_moved(old_child, new_xml, new_child, p_old, p_new); } } } } void xml_calculate_significant_changes(xmlNode *old_xml, xmlNode *new_xml) { pcmk__set_xml_doc_flag(new_xml, pcmk__xf_lazy); xml_calculate_changes(old_xml, new_xml); } // Called functions may set the \p pcmk__xf_skip flag on parts of \p old_xml void xml_calculate_changes(xmlNode *old_xml, xmlNode *new_xml) { CRM_CHECK((old_xml != NULL) && (new_xml != NULL) && pcmk__xe_is(old_xml, (const char *) new_xml->name) && pcmk__str_eq(pcmk__xe_id(old_xml), pcmk__xe_id(new_xml), pcmk__str_none), return); if(xml_tracking_changes(new_xml) == FALSE) { xml_track_changes(new_xml, NULL, NULL, FALSE); } mark_xml_changes(old_xml, new_xml, FALSE); } gboolean can_prune_leaf(xmlNode * xml_node) { xmlNode *cIter = NULL; gboolean can_prune = TRUE; CRM_CHECK(xml_node != NULL, return FALSE); /* @COMPAT PCMK__XE_ROLE_REF was deprecated in Pacemaker 1.1.12 (needed for * rolling upgrades) */ if (pcmk__strcase_any_of((const char *) xml_node->name, PCMK_XE_RESOURCE_REF, PCMK_XE_OBJ_REF, PCMK_XE_ROLE, PCMK__XE_ROLE_REF, NULL)) { return FALSE; } for (xmlAttrPtr a = pcmk__xe_first_attr(xml_node); a != NULL; a = a->next) { const char *p_name = (const char *) a->name; if (strcmp(p_name, PCMK_XA_ID) == 0) { continue; } can_prune = FALSE; } cIter = pcmk__xml_first_child(xml_node); while (cIter) { xmlNode *child = cIter; cIter = pcmk__xml_next(cIter); if (can_prune_leaf(child)) { free_xml(child); } else { can_prune = FALSE; } } return can_prune; } /*! * \internal * \brief Find a comment with matching content in specified XML * * \param[in] root XML to search * \param[in] search_comment Comment whose content should be searched for * \param[in] exact If true, comment must also be at same position */ xmlNode * pcmk__xc_match(const xmlNode *root, const xmlNode *search_comment, bool exact) { xmlNode *a_child = NULL; int search_offset = pcmk__xml_position(search_comment, pcmk__xf_skip); CRM_CHECK(search_comment->type == XML_COMMENT_NODE, return NULL); for (a_child = pcmk__xml_first_child(root); a_child != NULL; a_child = pcmk__xml_next(a_child)) { if (exact) { int offset = pcmk__xml_position(a_child, pcmk__xf_skip); xml_node_private_t *nodepriv = a_child->_private; if (offset < search_offset) { continue; } else if (offset > search_offset) { return NULL; } if (pcmk_is_set(nodepriv->flags, pcmk__xf_skip)) { continue; } } if (a_child->type == XML_COMMENT_NODE && pcmk__str_eq((const char *)a_child->content, (const char *)search_comment->content, pcmk__str_casei)) { return a_child; } else if (exact) { return NULL; } } return NULL; } /*! * \internal * \brief Make one XML comment match another (in content) * * \param[in,out] parent If \p target is NULL and this is not, add or update * comment child of this XML node that matches \p update * \param[in,out] target If not NULL, update this XML comment node * \param[in] update Make comment content match this (must not be NULL) * * \note At least one of \parent and \target must be non-NULL */ void pcmk__xc_update(xmlNode *parent, xmlNode *target, xmlNode *update) { CRM_CHECK(update != NULL, return); CRM_CHECK(update->type == XML_COMMENT_NODE, return); if (target == NULL) { target = pcmk__xc_match(parent, update, false); } if (target == NULL) { add_node_copy(parent, update); } else if (!pcmk__str_eq((const char *)target->content, (const char *)update->content, pcmk__str_casei)) { xmlFree(target->content); target->content = xmlStrdup(update->content); } } /*! * \internal * \brief Make one XML tree match another (in children and attributes) * * \param[in,out] parent If \p target is NULL and this is not, add or update * child of this XML node that matches \p update * \param[in,out] target If not NULL, update this XML * \param[in] update Make the desired XML match this (must not be NULL) * \param[in] as_diff If false, expand "++" when making attributes match * * \note At least one of \p parent and \p target must be non-NULL */ void pcmk__xml_update(xmlNode *parent, xmlNode *target, xmlNode *update, bool as_diff) { xmlNode *a_child = NULL; const char *object_name = NULL, *object_href = NULL, *object_href_val = NULL; #if XML_PARSER_DEBUG crm_log_xml_trace(update, "update:"); crm_log_xml_trace(target, "target:"); #endif CRM_CHECK(update != NULL, return); if (update->type == XML_COMMENT_NODE) { pcmk__xc_update(parent, target, update); return; } object_name = (const char *) update->name; object_href_val = pcmk__xe_id(update); if (object_href_val != NULL) { object_href = PCMK_XA_ID; } else { object_href_val = crm_element_value(update, PCMK_XA_ID_REF); object_href = (object_href_val == NULL)? NULL : PCMK_XA_ID_REF; } CRM_CHECK(object_name != NULL, return); CRM_CHECK(target != NULL || parent != NULL, return); if (target == NULL) { target = pcmk__xe_match(parent, object_name, object_href, object_href_val); } if (target == NULL) { target = create_xml_node(parent, object_name); CRM_CHECK(target != NULL, return); #if XML_PARSER_DEBUG crm_trace("Added <%s%s%s%s%s/>", pcmk__s(object_name, ""), object_href ? " " : "", object_href ? object_href : "", object_href ? "=" : "", object_href ? object_href_val : ""); } else { crm_trace("Found node <%s%s%s%s%s/> to update", pcmk__s(object_name, ""), object_href ? " " : "", object_href ? object_href : "", object_href ? "=" : "", object_href ? object_href_val : ""); #endif } CRM_CHECK(pcmk__xe_is(target, (const char *) update->name), return); if (as_diff == FALSE) { /* So that expand_plus_plus() gets called */ copy_in_properties(target, update); } else { /* No need for expand_plus_plus(), just raw speed */ for (xmlAttrPtr a = pcmk__xe_first_attr(update); a != NULL; a = a->next) { const char *p_value = pcmk__xml_attr_value(a); /* Remove it first so the ordering of the update is preserved */ xmlUnsetProp(target, a->name); xmlSetProp(target, a->name, (pcmkXmlStr) p_value); } } for (a_child = pcmk__xml_first_child(update); a_child != NULL; a_child = pcmk__xml_next(a_child)) { #if XML_PARSER_DEBUG crm_trace("Updating child <%s%s%s%s%s/>", pcmk__s(object_name, ""), object_href ? " " : "", object_href ? object_href : "", object_href ? "=" : "", object_href ? object_href_val : ""); #endif pcmk__xml_update(target, NULL, a_child, as_diff); } #if XML_PARSER_DEBUG crm_trace("Finished with <%s%s%s%s%s/>", pcmk__s(object_name, ""), object_href ? " " : "", object_href ? object_href : "", object_href ? "=" : "", object_href ? object_href_val : ""); #endif } gboolean update_xml_child(xmlNode * child, xmlNode * to_update) { gboolean can_update = TRUE; xmlNode *child_of_child = NULL; CRM_CHECK(child != NULL, return FALSE); CRM_CHECK(to_update != NULL, return FALSE); if (!pcmk__xe_is(to_update, (const char *) child->name)) { can_update = FALSE; } else if (!pcmk__str_eq(pcmk__xe_id(to_update), pcmk__xe_id(child), pcmk__str_none)) { can_update = FALSE; } else if (can_update) { #if XML_PARSER_DEBUG crm_log_xml_trace(child, "Update match found..."); #endif pcmk__xml_update(NULL, child, to_update, false); } for (child_of_child = pcmk__xml_first_child(child); child_of_child != NULL; child_of_child = pcmk__xml_next(child_of_child)) { /* only update the first one */ if (can_update) { break; } can_update = update_xml_child(child_of_child, to_update); } return can_update; } int find_xml_children(xmlNode ** children, xmlNode * root, const char *tag, const char *field, const char *value, gboolean search_matches) { int match_found = 0; CRM_CHECK(root != NULL, return FALSE); CRM_CHECK(children != NULL, return FALSE); if ((tag != NULL) && !pcmk__xe_is(root, tag)) { } else if (value != NULL && !pcmk__str_eq(value, crm_element_value(root, field), pcmk__str_casei)) { } else { if (*children == NULL) { *children = create_xml_node(NULL, __func__); } add_node_copy(*children, root); match_found = 1; } if (search_matches || match_found == 0) { xmlNode *child = NULL; for (child = pcmk__xml_first_child(root); child != NULL; child = pcmk__xml_next(child)) { match_found += find_xml_children(children, child, tag, field, value, search_matches); } } return match_found; } gboolean replace_xml_child(xmlNode * parent, xmlNode * child, xmlNode * update, gboolean delete_only) { gboolean can_delete = FALSE; xmlNode *child_of_child = NULL; const char *up_id = NULL; const char *child_id = NULL; const char *right_val = NULL; CRM_CHECK(child != NULL, return FALSE); CRM_CHECK(update != NULL, return FALSE); up_id = pcmk__xe_id(update); child_id = pcmk__xe_id(child); if (up_id == NULL || (child_id && strcmp(child_id, up_id) == 0)) { can_delete = TRUE; } if (!pcmk__xe_is(update, (const char *) child->name)) { can_delete = FALSE; } if (can_delete && delete_only) { for (xmlAttrPtr a = pcmk__xe_first_attr(update); a != NULL; a = a->next) { const char *p_name = (const char *) a->name; const char *p_value = pcmk__xml_attr_value(a); right_val = crm_element_value(child, p_name); if (!pcmk__str_eq(p_value, right_val, pcmk__str_casei)) { can_delete = FALSE; } } } if (can_delete && parent != NULL) { crm_log_xml_trace(child, "Delete match found..."); if (delete_only || update == NULL) { free_xml(child); } else { xmlNode *old = child; xmlNode *new = xmlCopyNode(update, 1); CRM_ASSERT(new != NULL); // May be unnecessary but avoids slight changes to some test outputs reset_xml_node_flags(new); old = xmlReplaceNode(old, new); if (xml_tracking_changes(new)) { // Replaced sections may have included relevant ACLs pcmk__apply_acl(new); } xml_calculate_changes(old, new); xmlFreeNode(old); } return TRUE; } else if (can_delete) { crm_log_xml_debug(child, "Cannot delete the search root"); can_delete = FALSE; } child_of_child = pcmk__xml_first_child(child); while (child_of_child) { xmlNode *next = pcmk__xml_next(child_of_child); can_delete = replace_xml_child(child, child_of_child, update, delete_only); /* only delete the first one */ if (can_delete) { child_of_child = NULL; } else { child_of_child = next; } } return can_delete; } xmlNode * sorted_xml(xmlNode *input, xmlNode *parent, gboolean recursive) { xmlNode *child = NULL; GSList *nvpairs = NULL; xmlNode *result = NULL; CRM_CHECK(input != NULL, return NULL); result = create_xml_node(parent, (const char *) input->name); nvpairs = pcmk_xml_attrs2nvpairs(input); nvpairs = pcmk_sort_nvpairs(nvpairs); pcmk_nvpairs2xml_attrs(nvpairs, result); pcmk_free_nvpairs(nvpairs); for (child = pcmk__xml_first_child(input); child != NULL; child = pcmk__xml_next(child)) { if (recursive) { sorted_xml(child, result, recursive); } else { add_node_copy(result, child); } } return result; } xmlNode * first_named_child(const xmlNode *parent, const char *name) { xmlNode *match = NULL; for (match = pcmk__xe_first_child(parent); match != NULL; match = pcmk__xe_next(match)) { /* * name == NULL gives first child regardless of name; this is * semantically incorrect in this function, but may be necessary * due to prior use of xml_child_iter_filter */ if ((name == NULL) || pcmk__xe_is(match, name)) { return match; } } return NULL; } /*! * \brief Get next instance of same XML tag * * \param[in] sibling XML tag to start from * * \return Next sibling XML tag with same name */ xmlNode * crm_next_same_xml(const xmlNode *sibling) { xmlNode *match = pcmk__xe_next(sibling); while (match != NULL) { if (pcmk__xe_is(match, (const char *) sibling->name)) { return match; } match = pcmk__xe_next(match); } return NULL; } void crm_xml_init(void) { static bool init = true; if(init) { init = false; /* The default allocator XML_BUFFER_ALLOC_EXACT does far too many * pcmk__realloc()s and it can take upwards of 18 seconds (yes, seconds) * to dump a 28kb tree which XML_BUFFER_ALLOC_DOUBLEIT can do in * less than 1 second. */ xmlSetBufferAllocationScheme(XML_BUFFER_ALLOC_DOUBLEIT); /* Populate and free the _private field when nodes are created and destroyed */ xmlDeregisterNodeDefault(free_private_data); xmlRegisterNodeDefault(new_private_data); crm_schema_init(); } } void crm_xml_cleanup(void) { crm_schema_cleanup(); xmlCleanupParser(); } #define XPATH_MAX 512 xmlNode * expand_idref(xmlNode * input, xmlNode * top) { char *xpath = NULL; const char *ref = NULL; xmlNode *result = NULL; if (input == NULL) { return NULL; } ref = crm_element_value(input, PCMK_XA_ID_REF); if (ref == NULL) { return input; } if (top == NULL) { top = input; } xpath = crm_strdup_printf("//%s[@" PCMK_XA_ID "='%s']", input->name, ref); result = get_xpath_object(xpath, top, LOG_DEBUG); if (result == NULL) { // Not possible with schema validation enabled pcmk__config_err("Ignoring invalid %s configuration: " PCMK_XA_ID_REF " '%s' does not reference " "a valid object " CRM_XS " xpath=%s", input->name, ref, xpath); } free(xpath); return result; } char * pcmk__xml_artefact_root(enum pcmk__xml_artefact_ns ns) { static const char *base = NULL; char *ret = NULL; if (base == NULL) { base = pcmk__env_option(PCMK__ENV_SCHEMA_DIRECTORY); } if (pcmk__str_empty(base)) { base = CRM_SCHEMA_DIRECTORY; } switch (ns) { case pcmk__xml_artefact_ns_legacy_rng: case pcmk__xml_artefact_ns_legacy_xslt: ret = strdup(base); break; case pcmk__xml_artefact_ns_base_rng: case pcmk__xml_artefact_ns_base_xslt: ret = crm_strdup_printf("%s/base", base); break; default: crm_err("XML artefact family specified as %u not recognized", ns); } return ret; } static char * find_artefact(enum pcmk__xml_artefact_ns ns, const char *path, const char *filespec) { char *ret = NULL; switch (ns) { case pcmk__xml_artefact_ns_legacy_rng: case pcmk__xml_artefact_ns_base_rng: if (pcmk__ends_with(filespec, ".rng")) { ret = crm_strdup_printf("%s/%s", path, filespec); } else { ret = crm_strdup_printf("%s/%s.rng", path, filespec); } break; case pcmk__xml_artefact_ns_legacy_xslt: case pcmk__xml_artefact_ns_base_xslt: if (pcmk__ends_with(filespec, ".xsl")) { ret = crm_strdup_printf("%s/%s", path, filespec); } else { ret = crm_strdup_printf("%s/%s.xsl", path, filespec); } break; default: crm_err("XML artefact family specified as %u not recognized", ns); } return ret; } char * pcmk__xml_artefact_path(enum pcmk__xml_artefact_ns ns, const char *filespec) { struct stat sb; char *base = pcmk__xml_artefact_root(ns); char *ret = NULL; ret = find_artefact(ns, base, filespec); free(base); if (stat(ret, &sb) != 0 || !S_ISREG(sb.st_mode)) { const char *remote_schema_dir = pcmk__remote_schema_dir(); ret = find_artefact(ns, remote_schema_dir, filespec); } return ret; } void pcmk__xe_set_propv(xmlNodePtr node, va_list pairs) { while (true) { const char *name, *value; name = va_arg(pairs, const char *); if (name == NULL) { return; } value = va_arg(pairs, const char *); if (value != NULL) { crm_xml_add(node, name, value); } } } void pcmk__xe_set_props(xmlNodePtr node, ...) { va_list pairs; va_start(pairs, node); pcmk__xe_set_propv(node, pairs); va_end(pairs); } int pcmk__xe_foreach_child(xmlNode *xml, const char *child_element_name, int (*handler)(xmlNode *xml, void *userdata), void *userdata) { xmlNode *children = (xml? xml->children : NULL); CRM_ASSERT(handler != NULL); for (xmlNode *node = children; node != NULL; node = node->next) { if ((node->type == XML_ELEMENT_NODE) && ((child_element_name == NULL) || pcmk__xe_is(node, child_element_name))) { int rc = handler(node, userdata); if (rc != pcmk_rc_ok) { return rc; } } } return pcmk_rc_ok; } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include xmlNode * find_entity(xmlNode *parent, const char *node_name, const char *id) { return pcmk__xe_match(parent, node_name, ((id == NULL)? id : PCMK_XA_ID), id); } void crm_destroy_xml(gpointer data) { free_xml(data); } xmlDoc * getDocPtr(xmlNode *node) { xmlDoc *doc = NULL; CRM_CHECK(node != NULL, return NULL); doc = node->doc; if (doc == NULL) { doc = xmlNewDoc((pcmkXmlStr) "1.0"); xmlDocSetRootElement(doc, node); } return doc; } int add_node_nocopy(xmlNode *parent, const char *name, xmlNode *child) { add_node_copy(parent, child); free_xml(child); return 1; } gboolean xml_has_children(const xmlNode * xml_root) { if (xml_root != NULL && xml_root->children != NULL) { return TRUE; } return FALSE; } char * crm_xml_escape(const char *text) { size_t length = 0; char *copy = NULL; if (text == NULL) { return NULL; } length = strlen(text); copy = strdup(text); CRM_ASSERT(copy != NULL); for (size_t index = 0; index <= length; index++) { if(copy[index] & 0x80 && copy[index+1] & 0x80){ index++; continue; } switch (copy[index]) { case 0: // Sanity only; loop should stop at the last non-null byte break; case '<': copy = replace_text(copy, &index, &length, "<"); break; case '>': copy = replace_text(copy, &index, &length, ">"); break; case '"': copy = replace_text(copy, &index, &length, """); break; case '\'': copy = replace_text(copy, &index, &length, "'"); break; case '&': copy = replace_text(copy, &index, &length, "&"); break; case '\t': /* Might as well just expand to a few spaces... */ copy = replace_text(copy, &index, &length, " "); break; case '\n': copy = replace_text(copy, &index, &length, "\\n"); break; case '\r': copy = replace_text(copy, &index, &length, "\\r"); break; default: /* Check for and replace non-printing characters with their octal equivalent */ if(copy[index] < ' ' || copy[index] > '~') { char *replace = crm_strdup_printf("\\%.3o", copy[index]); copy = replace_text(copy, &index, &length, replace); free(replace); } } } return copy; } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/pacemaker/pcmk_graph_consumer.c b/lib/pacemaker/pcmk_graph_consumer.c index 1b16814ef1..9ceb24ce64 100644 --- a/lib/pacemaker/pcmk_graph_consumer.c +++ b/lib/pacemaker/pcmk_graph_consumer.c @@ -1,882 +1,882 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include /* * Functions for updating graph */ /*! * \internal * \brief Update synapse after completed prerequisite * * A synapse is ready to be executed once all its prerequisite actions (inputs) * complete. Given a completed action, check whether it is an input for a given * synapse, and if so, mark the input as confirmed, and mark the synapse as * ready if appropriate. * * \param[in,out] synapse Transition graph synapse to update * \param[in] action_id ID of an action that completed * * \note The only substantial effect here is confirming synapse inputs. * should_fire_synapse() will recalculate pcmk__synapse_ready, so the only * thing that uses the pcmk__synapse_ready from here is * synapse_state_str(). */ static void update_synapse_ready(pcmk__graph_synapse_t *synapse, int action_id) { if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) { return; // All inputs have already been confirmed } // Presume ready until proven otherwise pcmk__set_synapse_flags(synapse, pcmk__synapse_ready); for (GList *lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *prereq = (pcmk__graph_action_t *) lpc->data; if (prereq->id == action_id) { crm_trace("Confirming input %d of synapse %d", action_id, synapse->id); pcmk__set_graph_action_flags(prereq, pcmk__graph_action_confirmed); } else if (!pcmk_is_set(prereq->flags, pcmk__graph_action_confirmed)) { pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready); crm_trace("Synapse %d still not ready after action %d", synapse->id, action_id); } } if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) { crm_trace("Synapse %d is now ready to execute", synapse->id); } } /*! * \internal * \brief Update action and synapse confirmation after action completion * * \param[in,out] synapse Transition graph synapse that action belongs to * \param[in] action_id ID of action that completed */ static void update_synapse_confirmed(pcmk__graph_synapse_t *synapse, int action_id) { bool all_confirmed = true; for (GList *lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc->data; if (action->id == action_id) { crm_trace("Confirmed action %d of synapse %d", action_id, synapse->id); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); } else if (all_confirmed && !pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { all_confirmed = false; crm_trace("Synapse %d still not confirmed after action %d", synapse->id, action_id); } } if (all_confirmed && !pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) { crm_trace("Confirmed synapse %d", synapse->id); pcmk__set_synapse_flags(synapse, pcmk__synapse_confirmed); } } /*! * \internal * \brief Update the transition graph with a completed action result * * \param[in,out] graph Transition graph to update * \param[in] action Action that completed */ void pcmk__update_graph(pcmk__graph_t *graph, const pcmk__graph_action_t *action) { for (GList *lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed|pcmk__synapse_failed)) { continue; // This synapse already completed } else if (pcmk_is_set(synapse->flags, pcmk__synapse_executed)) { update_synapse_confirmed(synapse, action->id); } else if (!pcmk_is_set(action->flags, pcmk__graph_action_failed) - || (synapse->priority == INFINITY)) { + || (synapse->priority == PCMK_SCORE_INFINITY)) { update_synapse_ready(synapse, action->id); } } } /* * Functions for executing graph */ /* A transition graph consists of various types of actions. The library caller * registers execution functions for each action type, which will be stored * here. */ static pcmk__graph_functions_t *graph_fns = NULL; /*! * \internal * \brief Set transition graph execution functions * * \param[in] Execution functions to use */ void pcmk__set_graph_functions(pcmk__graph_functions_t *fns) { crm_debug("Setting custom functions for executing transition graphs"); graph_fns = fns; CRM_ASSERT(graph_fns != NULL); CRM_ASSERT(graph_fns->rsc != NULL); CRM_ASSERT(graph_fns->cluster != NULL); CRM_ASSERT(graph_fns->pseudo != NULL); CRM_ASSERT(graph_fns->fence != NULL); } /*! * \internal * \brief Check whether a graph synapse is ready to be executed * * \param[in,out] graph Transition graph that synapse is part of * \param[in,out] synapse Synapse to check * * \return true if synapse is ready, false otherwise */ static bool should_fire_synapse(pcmk__graph_t *graph, pcmk__graph_synapse_t *synapse) { GList *lpc = NULL; pcmk__set_synapse_flags(synapse, pcmk__synapse_ready); for (lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *prereq = (pcmk__graph_action_t *) lpc->data; if (!(pcmk_is_set(prereq->flags, pcmk__graph_action_confirmed))) { crm_trace("Input %d for synapse %d not yet confirmed", prereq->id, synapse->id); pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready); break; } else if (pcmk_is_set(prereq->flags, pcmk__graph_action_failed) && !pcmk_is_set(prereq->flags, pcmk__graph_action_can_fail)) { crm_trace("Input %d for synapse %d confirmed but failed", prereq->id, synapse->id); pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready); break; } } if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) { crm_trace("Synapse %d is ready to execute", synapse->id); } else { return false; } for (lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *a = (pcmk__graph_action_t *) lpc->data; if (a->type == pcmk__pseudo_graph_action) { /* None of the below applies to pseudo ops */ } else if (synapse->priority < graph->abort_priority) { crm_trace("Skipping synapse %d: priority %d is less than " "abort priority %d", synapse->id, synapse->priority, graph->abort_priority); graph->skipped++; return false; } else if (graph_fns->allowed && !(graph_fns->allowed(graph, a))) { crm_trace("Deferring synapse %d: not allowed", synapse->id); return false; } } return true; } /*! * \internal * \brief Initiate an action from a transition graph * * \param[in,out] graph Transition graph containing action * \param[in,out] action Action to execute * * \return Standard Pacemaker return code */ static int initiate_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *id = pcmk__xe_id(action->xml); CRM_CHECK(id != NULL, return EINVAL); CRM_CHECK(!pcmk_is_set(action->flags, pcmk__graph_action_executed), return pcmk_rc_already); pcmk__set_graph_action_flags(action, pcmk__graph_action_executed); switch (action->type) { case pcmk__pseudo_graph_action: crm_trace("Executing pseudo-action %d (%s)", action->id, id); return graph_fns->pseudo(graph, action); case pcmk__rsc_graph_action: crm_trace("Executing resource action %d (%s)", action->id, id); return graph_fns->rsc(graph, action); case pcmk__cluster_graph_action: if (pcmk__str_eq(crm_element_value(action->xml, PCMK_XA_OPERATION), PCMK_ACTION_STONITH, pcmk__str_none)) { crm_trace("Executing fencing action %d (%s)", action->id, id); return graph_fns->fence(graph, action); } crm_trace("Executing cluster action %d (%s)", action->id, id); return graph_fns->cluster(graph, action); default: crm_err("Unsupported graph action type <%s " PCMK_XA_ID "='%s'> " "(bug?)", action->xml->name, id); return EINVAL; } } /*! * \internal * \brief Execute a graph synapse * * \param[in,out] graph Transition graph with synapse to execute * \param[in,out] synapse Synapse to execute * * \return Standard Pacemaker return value */ static int fire_synapse(pcmk__graph_t *graph, pcmk__graph_synapse_t *synapse) { pcmk__set_synapse_flags(synapse, pcmk__synapse_executed); for (GList *lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc->data; int rc = initiate_action(graph, action); if (rc != pcmk_rc_ok) { crm_err("Failed initiating <%s " PCMK_XA_ID "=%d> in synapse %d: " "%s", action->xml->name, action->id, synapse->id, pcmk_rc_str(rc)); pcmk__set_synapse_flags(synapse, pcmk__synapse_confirmed); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed |pcmk__graph_action_failed); return pcmk_rc_error; } } return pcmk_rc_ok; } /*! * \internal * \brief Dummy graph method that can be used with simulations * * \param[in,out] graph Transition graph containing action * \param[in,out] action Graph action to be initiated * * \return Standard Pacemaker return code * \note If the PE_fail environment variable is set to the action ID, * then the graph action will be marked as failed. */ static int pseudo_action_dummy(pcmk__graph_t *graph, pcmk__graph_action_t *action) { static int fail = -1; if (fail < 0) { long long fail_ll; if ((pcmk__scan_ll(getenv("PE_fail"), &fail_ll, 0LL) == pcmk_rc_ok) && (fail_ll > 0LL) && (fail_ll <= INT_MAX)) { fail = (int) fail_ll; } else { fail = 0; } } if (action->id == fail) { crm_err("Dummy event handler: pretending action %d failed", action->id); pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); - graph->abort_priority = INFINITY; + graph->abort_priority = PCMK_SCORE_INFINITY; } else { crm_trace("Dummy event handler: action %d initiated", action->id); } pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); pcmk__update_graph(graph, action); return pcmk_rc_ok; } static pcmk__graph_functions_t default_fns = { pseudo_action_dummy, pseudo_action_dummy, pseudo_action_dummy, pseudo_action_dummy }; /*! * \internal * \brief Execute all actions in a transition graph * * \param[in,out] graph Transition graph to execute * * \return Status of transition after execution */ enum pcmk__graph_status pcmk__execute_graph(pcmk__graph_t *graph) { GList *lpc = NULL; int log_level = LOG_DEBUG; enum pcmk__graph_status pass_result = pcmk__graph_active; const char *status = "In progress"; if (graph_fns == NULL) { graph_fns = &default_fns; } if (graph == NULL) { return pcmk__graph_complete; } graph->fired = 0; graph->pending = 0; graph->skipped = 0; graph->completed = 0; graph->incomplete = 0; // Count completed and in-flight synapses for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if (pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) { graph->completed++; } else if (!pcmk_is_set(synapse->flags, pcmk__synapse_failed) && pcmk_is_set(synapse->flags, pcmk__synapse_executed)) { graph->pending++; } } crm_trace("Executing graph %d (%d synapses already completed, %d pending)", graph->id, graph->completed, graph->pending); // Execute any synapses that are ready for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if ((graph->batch_limit > 0) && (graph->pending >= graph->batch_limit)) { crm_debug("Throttling graph execution: batch limit (%d) reached", graph->batch_limit); break; } else if (pcmk_is_set(synapse->flags, pcmk__synapse_failed)) { graph->skipped++; continue; } else if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed |pcmk__synapse_executed)) { continue; // Already handled } else if (should_fire_synapse(graph, synapse)) { graph->fired++; if (fire_synapse(graph, synapse) != pcmk_rc_ok) { crm_err("Synapse %d failed to fire", synapse->id); log_level = LOG_ERR; - graph->abort_priority = INFINITY; + graph->abort_priority = PCMK_SCORE_INFINITY; graph->incomplete++; graph->fired--; } if (!(pcmk_is_set(synapse->flags, pcmk__synapse_confirmed))) { graph->pending++; } } else { crm_trace("Synapse %d cannot fire", synapse->id); graph->incomplete++; } } if ((graph->pending == 0) && (graph->fired == 0)) { graph->complete = true; if ((graph->incomplete != 0) && (graph->abort_priority <= 0)) { log_level = LOG_WARNING; pass_result = pcmk__graph_terminated; status = "Terminated"; } else if (graph->skipped != 0) { log_level = LOG_NOTICE; pass_result = pcmk__graph_complete; status = "Stopped"; } else { log_level = LOG_NOTICE; pass_result = pcmk__graph_complete; status = "Complete"; } } else if (graph->fired == 0) { pass_result = pcmk__graph_pending; } do_crm_log(log_level, "Transition %d (Complete=%d, Pending=%d," " Fired=%d, Skipped=%d, Incomplete=%d, Source=%s): %s", graph->id, graph->completed, graph->pending, graph->fired, graph->skipped, graph->incomplete, graph->source, status); return pass_result; } /* * Functions for unpacking transition graph XML into structs */ /*! * \internal * \brief Unpack a transition graph action from XML * * \param[in] parent Synapse that action is part of * \param[in] xml_action Action XML to unparse * * \return Newly allocated action on success, or NULL otherwise */ static pcmk__graph_action_t * unpack_action(pcmk__graph_synapse_t *parent, xmlNode *xml_action) { enum pcmk__graph_action_type action_type; pcmk__graph_action_t *action = NULL; const char *value = pcmk__xe_id(xml_action); if (value == NULL) { crm_err("Ignoring transition graph action without id (bug?)"); crm_log_xml_trace(xml_action, "invalid"); return NULL; } if (pcmk__xe_is(xml_action, PCMK__XE_RSC_OP)) { action_type = pcmk__rsc_graph_action; } else if (pcmk__xe_is(xml_action, PCMK__XE_PSEUDO_EVENT)) { action_type = pcmk__pseudo_graph_action; } else if (pcmk__xe_is(xml_action, PCMK__XE_CRM_EVENT)) { action_type = pcmk__cluster_graph_action; } else { crm_err("Ignoring transition graph action of unknown type '%s' (bug?)", xml_action->name); crm_log_xml_trace(xml_action, "invalid"); return NULL; } action = calloc(1, sizeof(pcmk__graph_action_t)); if (action == NULL) { crm_perror(LOG_CRIT, "Cannot unpack transition graph action"); crm_log_xml_trace(xml_action, "lost"); return NULL; } pcmk__scan_min_int(value, &(action->id), -1); action->type = pcmk__rsc_graph_action; action->xml = copy_xml(xml_action); action->synapse = parent; action->type = action_type; action->params = xml2list(action->xml); value = crm_meta_value(action->params, PCMK_META_TIMEOUT); pcmk__scan_min_int(value, &(action->timeout), 0); /* Take PCMK_META_START_DELAY into account for the timeout of the action * timer */ value = crm_meta_value(action->params, PCMK_META_START_DELAY); { int start_delay; pcmk__scan_min_int(value, &start_delay, 0); action->timeout += start_delay; } if (pcmk__guint_from_hash(action->params, CRM_META "_" PCMK_META_INTERVAL, 0, &(action->interval_ms)) != pcmk_rc_ok) { action->interval_ms = 0; } value = crm_meta_value(action->params, PCMK__META_CAN_FAIL); if (value != NULL) { int can_fail = 0; if ((crm_str_to_boolean(value, &can_fail) > 0) && (can_fail > 0)) { pcmk__set_graph_action_flags(action, pcmk__graph_action_can_fail); } else { pcmk__clear_graph_action_flags(action, pcmk__graph_action_can_fail); } #ifndef PCMK__COMPAT_2_0 if (pcmk_is_set(action->flags, pcmk__graph_action_can_fail)) { crm_warn("Support for the " PCMK__META_CAN_FAIL " meta-attribute " "is deprecated and will be removed in a future release"); } #endif } crm_trace("Action %d has timer set to %dms", action->id, action->timeout); return action; } /*! * \internal * \brief Unpack transition graph synapse from XML * * \param[in,out] new_graph Transition graph that synapse is part of * \param[in] xml_synapse Synapse XML * * \return Newly allocated synapse on success, or NULL otherwise */ static pcmk__graph_synapse_t * unpack_synapse(pcmk__graph_t *new_graph, const xmlNode *xml_synapse) { const char *value = NULL; xmlNode *action_set = NULL; pcmk__graph_synapse_t *new_synapse = NULL; crm_trace("Unpacking synapse %s", pcmk__xe_id(xml_synapse)); new_synapse = calloc(1, sizeof(pcmk__graph_synapse_t)); if (new_synapse == NULL) { return NULL; } pcmk__scan_min_int(pcmk__xe_id(xml_synapse), &(new_synapse->id), 0); value = crm_element_value(xml_synapse, PCMK__XA_PRIORITY); pcmk__scan_min_int(value, &(new_synapse->priority), 0); CRM_CHECK(new_synapse->id >= 0, free(new_synapse); return NULL); new_graph->num_synapses++; crm_trace("Unpacking synapse %s action sets", crm_element_value(xml_synapse, PCMK_XA_ID)); for (action_set = first_named_child(xml_synapse, "action_set"); action_set != NULL; action_set = crm_next_same_xml(action_set)) { for (xmlNode *action = pcmk__xml_first_child(action_set); action != NULL; action = pcmk__xml_next(action)) { pcmk__graph_action_t *new_action = unpack_action(new_synapse, action); if (new_action == NULL) { continue; } crm_trace("Adding action %d to synapse %d", new_action->id, new_synapse->id); new_graph->num_actions++; new_synapse->actions = g_list_append(new_synapse->actions, new_action); } } crm_trace("Unpacking synapse %s inputs", pcmk__xe_id(xml_synapse)); for (xmlNode *inputs = first_named_child(xml_synapse, "inputs"); inputs != NULL; inputs = crm_next_same_xml(inputs)) { for (xmlNode *trigger = first_named_child(inputs, "trigger"); trigger != NULL; trigger = crm_next_same_xml(trigger)) { for (xmlNode *input = pcmk__xml_first_child(trigger); input != NULL; input = pcmk__xml_next(input)) { pcmk__graph_action_t *new_input = unpack_action(new_synapse, input); if (new_input == NULL) { continue; } crm_trace("Adding input %d to synapse %d", new_input->id, new_synapse->id); new_synapse->inputs = g_list_append(new_synapse->inputs, new_input); } } } return new_synapse; } /*! * \internal * \brief Unpack transition graph XML * * \param[in] xml_graph Transition graph XML to unpack * \param[in] reference Where the XML came from (for logging) * * \return Newly allocated transition graph on success, NULL otherwise * \note The caller is responsible for freeing the return value using * pcmk__free_graph(). * \note The XML is expected to be structured like: ... ... */ pcmk__graph_t * pcmk__unpack_graph(const xmlNode *xml_graph, const char *reference) { pcmk__graph_t *new_graph = NULL; new_graph = calloc(1, sizeof(pcmk__graph_t)); if (new_graph == NULL) { return NULL; } new_graph->source = strdup(pcmk__s(reference, "unknown")); if (new_graph->source == NULL) { free(new_graph); return NULL; } new_graph->id = -1; new_graph->abort_priority = 0; new_graph->network_delay = 0; new_graph->stonith_timeout = 0; new_graph->completion_action = pcmk__graph_done; // Parse top-level attributes from PCMK__XE_TRANSITION_GRAPH if (xml_graph != NULL) { const char *buf = crm_element_value(xml_graph, "transition_id"); CRM_CHECK(buf != NULL, free(new_graph); return NULL); pcmk__scan_min_int(buf, &(new_graph->id), -1); buf = crm_element_value(xml_graph, PCMK_OPT_CLUSTER_DELAY); CRM_CHECK(buf != NULL, free(new_graph); return NULL); pcmk_parse_interval_spec(buf, &(new_graph->network_delay)); buf = crm_element_value(xml_graph, PCMK_OPT_STONITH_TIMEOUT); if (buf == NULL) { new_graph->stonith_timeout = new_graph->network_delay; } else { pcmk_parse_interval_spec(buf, &(new_graph->stonith_timeout)); } // Use 0 (dynamic limit) as default/invalid, -1 (no limit) as minimum buf = crm_element_value(xml_graph, PCMK_OPT_BATCH_LIMIT); if ((buf == NULL) || (pcmk__scan_min_int(buf, &(new_graph->batch_limit), -1) != pcmk_rc_ok)) { new_graph->batch_limit = 0; } buf = crm_element_value(xml_graph, PCMK_OPT_MIGRATION_LIMIT); pcmk__scan_min_int(buf, &(new_graph->migration_limit), -1); pcmk__str_update(&(new_graph->failed_stop_offset), crm_element_value(xml_graph, "failed-stop-offset")); pcmk__str_update(&(new_graph->failed_start_offset), crm_element_value(xml_graph, "failed-start-offset")); if (crm_element_value_epoch(xml_graph, "recheck-by", &(new_graph->recheck_by)) != pcmk_ok) { new_graph->recheck_by = 0; } } // Unpack each child element for (const xmlNode *synapse_xml = first_named_child(xml_graph, "synapse"); synapse_xml != NULL; synapse_xml = crm_next_same_xml(synapse_xml)) { pcmk__graph_synapse_t *new_synapse = unpack_synapse(new_graph, synapse_xml); if (new_synapse != NULL) { new_graph->synapses = g_list_append(new_graph->synapses, new_synapse); } } crm_debug("Unpacked transition %d from %s: %d actions in %d synapses", new_graph->id, new_graph->source, new_graph->num_actions, new_graph->num_synapses); return new_graph; } /* * Functions for freeing transition graph objects */ /*! * \internal * \brief Free a transition graph action object * * \param[in,out] user_data Action to free */ static void free_graph_action(gpointer user_data) { pcmk__graph_action_t *action = user_data; if (action->timer != 0) { crm_warn("Cancelling timer for graph action %d", action->id); g_source_remove(action->timer); } if (action->params != NULL) { g_hash_table_destroy(action->params); } free_xml(action->xml); free(action); } /*! * \internal * \brief Free a transition graph synapse object * * \param[in,out] user_data Synapse to free */ static void free_graph_synapse(gpointer user_data) { pcmk__graph_synapse_t *synapse = user_data; g_list_free_full(synapse->actions, free_graph_action); g_list_free_full(synapse->inputs, free_graph_action); free(synapse); } /*! * \internal * \brief Free a transition graph object * * \param[in,out] graph Transition graph to free */ void pcmk__free_graph(pcmk__graph_t *graph) { if (graph != NULL) { g_list_free_full(graph->synapses, free_graph_synapse); free(graph->source); free(graph->failed_stop_offset); free(graph->failed_start_offset); free(graph); } } /* * Other transition graph utilities */ /*! * \internal * \brief Synthesize an executor event from a graph action * * \param[in] resource If not NULL, use greater call ID than in this XML * \param[in] action Graph action * \param[in] status What to use as event execution status * \param[in] rc What to use as event exit status * \param[in] exit_reason What to use as event exit reason * * \return Newly allocated executor event on success, or NULL otherwise */ lrmd_event_data_t * pcmk__event_from_graph_action(const xmlNode *resource, const pcmk__graph_action_t *action, int status, int rc, const char *exit_reason) { lrmd_event_data_t *op = NULL; GHashTableIter iter; const char *name = NULL; const char *value = NULL; xmlNode *action_resource = NULL; CRM_CHECK(action != NULL, return NULL); CRM_CHECK(action->type == pcmk__rsc_graph_action, return NULL); action_resource = first_named_child(action->xml, PCMK_XE_PRIMITIVE); CRM_CHECK(action_resource != NULL, crm_log_xml_warn(action->xml, "invalid"); return NULL); op = lrmd_new_event(pcmk__xe_id(action_resource), crm_element_value(action->xml, PCMK_XA_OPERATION), action->interval_ms); lrmd__set_result(op, rc, status, exit_reason); op->t_run = time(NULL); op->t_rcchange = op->t_run; op->params = pcmk__strkey_table(free, free); g_hash_table_iter_init(&iter, action->params); while (g_hash_table_iter_next(&iter, (void **)&name, (void **)&value)) { pcmk__insert_dup(op->params, name, value); } for (xmlNode *xop = pcmk__xml_first_child(resource); xop != NULL; xop = pcmk__xml_next(xop)) { int tmp = 0; crm_element_value_int(xop, PCMK__XA_CALL_ID, &tmp); crm_debug("Got call_id=%d for %s", tmp, pcmk__xe_id(resource)); if (tmp > op->call_id) { op->call_id = tmp; } } op->call_id++; return op; } diff --git a/lib/pacemaker/pcmk_sched_bundle.c b/lib/pacemaker/pcmk_sched_bundle.c index 091bc012c1..1d1466be9f 100644 --- a/lib/pacemaker/pcmk_sched_bundle.c +++ b/lib/pacemaker/pcmk_sched_bundle.c @@ -1,1058 +1,1059 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "libpacemaker_private.h" struct assign_data { const pcmk_node_t *prefer; bool stop_if_fail; }; /*! * \internal * \brief Assign a single bundle replica's resources (other than container) * * \param[in,out] replica Replica to assign * \param[in] user_data Preferred node, if any * * \return true (to indicate that any further replicas should be processed) */ static bool assign_replica(pcmk__bundle_replica_t *replica, void *user_data) { pcmk_node_t *container_host = NULL; struct assign_data *assign_data = user_data; const pcmk_node_t *prefer = assign_data->prefer; bool stop_if_fail = assign_data->stop_if_fail; const pcmk_resource_t *bundle = pe__const_top_resource(replica->container, true); if (replica->ip != NULL) { pcmk__rsc_trace(bundle, "Assigning bundle %s IP %s", bundle->id, replica->ip->id); replica->ip->cmds->assign(replica->ip, prefer, stop_if_fail); } container_host = replica->container->allocated_to; if (replica->remote != NULL) { if (pcmk__is_pacemaker_remote_node(container_host)) { /* REMOTE_CONTAINER_HACK: "Nested" connection resources must be on * the same host because Pacemaker Remote only supports a single * active connection. */ pcmk__new_colocation("#replica-remote-with-host-remote", NULL, - INFINITY, replica->remote, + PCMK_SCORE_INFINITY, replica->remote, container_host->details->remote_rsc, NULL, NULL, pcmk__coloc_influence); } pcmk__rsc_trace(bundle, "Assigning bundle %s connection %s", bundle->id, replica->remote->id); replica->remote->cmds->assign(replica->remote, prefer, stop_if_fail); } if (replica->child != NULL) { pcmk_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, replica->child->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (!pcmk__same_node(node, replica->node)) { - node->weight = -INFINITY; + node->weight = -PCMK_SCORE_INFINITY; } else if (!pcmk__threshold_reached(replica->child, node, NULL)) { - node->weight = INFINITY; + node->weight = PCMK_SCORE_INFINITY; } } pcmk__set_rsc_flags(replica->child->parent, pcmk_rsc_assigning); pcmk__rsc_trace(bundle, "Assigning bundle %s replica child %s", bundle->id, replica->child->id); replica->child->cmds->assign(replica->child, replica->node, stop_if_fail); pcmk__clear_rsc_flags(replica->child->parent, pcmk_rsc_assigning); } return true; } /*! * \internal * \brief Assign a bundle resource to a node * * \param[in,out] rsc Resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * \param[in] stop_if_fail If \c true and a primitive descendant of \p rsc * can't be assigned to a node, set the * descendant's next role to stopped and update * existing actions * * \return Node that \p rsc is assigned to, if assigned entirely to one node * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ pcmk_node_t * pcmk__bundle_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { GList *containers = NULL; pcmk_resource_t *bundled_resource = NULL; struct assign_data assign_data = { prefer, stop_if_fail }; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pcmk__rsc_trace(rsc, "Assigning bundle %s", rsc->id); pcmk__set_rsc_flags(rsc, pcmk_rsc_assigning); pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); // Assign all containers first, so we know what nodes the bundle will be on containers = g_list_sort(pe__bundle_containers(rsc), pcmk__cmp_instance); pcmk__assign_instances(rsc, containers, pe__bundle_max(rsc), rsc->fns->max_per_node(rsc)); g_list_free(containers); // Then assign remaining replica resources pe__foreach_bundle_replica(rsc, assign_replica, (void *) &assign_data); // Finally, assign the bundled resources to each bundle node bundled_resource = pe__bundled_resource(rsc); if (bundled_resource != NULL) { pcmk_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, bundled_resource->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { if (pe__node_is_bundle_instance(rsc, node)) { node->weight = 0; } else { - node->weight = -INFINITY; + node->weight = -PCMK_SCORE_INFINITY; } } bundled_resource->cmds->assign(bundled_resource, prefer, stop_if_fail); } pcmk__clear_rsc_flags(rsc, pcmk_rsc_assigning|pcmk_rsc_unassigned); return NULL; } /*! * \internal * \brief Create actions for a bundle replica's resources (other than child) * * \param[in,out] replica Replica to create actions for * \param[in] user_data Unused * * \return true (to indicate that any further replicas should be processed) */ static bool create_replica_actions(pcmk__bundle_replica_t *replica, void *user_data) { if (replica->ip != NULL) { replica->ip->cmds->create_actions(replica->ip); } if (replica->container != NULL) { replica->container->cmds->create_actions(replica->container); } if (replica->remote != NULL) { replica->remote->cmds->create_actions(replica->remote); } return true; } /*! * \internal * \brief Create all actions needed for a given bundle resource * * \param[in,out] rsc Bundle resource to create actions for */ void pcmk__bundle_create_actions(pcmk_resource_t *rsc) { pcmk_action_t *action = NULL; GList *containers = NULL; pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, create_replica_actions, NULL); containers = pe__bundle_containers(rsc); pcmk__create_instance_actions(rsc, containers); g_list_free(containers); bundled_resource = pe__bundled_resource(rsc); if (bundled_resource != NULL) { bundled_resource->cmds->create_actions(bundled_resource); if (pcmk_is_set(bundled_resource->flags, pcmk_rsc_promotable)) { pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_PROMOTE, true, true); action = pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_PROMOTED, true, true); - action->priority = INFINITY; + action->priority = PCMK_SCORE_INFINITY; pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_DEMOTE, true, true); action = pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_DEMOTED, true, true); - action->priority = INFINITY; + action->priority = PCMK_SCORE_INFINITY; } } } /*! * \internal * \brief Create internal constraints for a bundle replica's resources * * \param[in,out] replica Replica to create internal constraints for * \param[in,out] user_data Replica's parent bundle * * \return true (to indicate that any further replicas should be processed) */ static bool replica_internal_constraints(pcmk__bundle_replica_t *replica, void *user_data) { pcmk_resource_t *bundle = user_data; replica->container->cmds->internal_constraints(replica->container); // Start bundle -> start replica container pcmk__order_starts(bundle, replica->container, pcmk__ar_unrunnable_first_blocks |pcmk__ar_then_implies_first_graphed); // Stop bundle -> stop replica child and container if (replica->child != NULL) { pcmk__order_stops(bundle, replica->child, pcmk__ar_then_implies_first_graphed); } pcmk__order_stops(bundle, replica->container, pcmk__ar_then_implies_first_graphed); // Start replica container -> bundle is started pcmk__order_resource_actions(replica->container, PCMK_ACTION_START, bundle, PCMK_ACTION_RUNNING, pcmk__ar_first_implies_then_graphed); // Stop replica container -> bundle is stopped pcmk__order_resource_actions(replica->container, PCMK_ACTION_STOP, bundle, PCMK_ACTION_STOPPED, pcmk__ar_first_implies_then_graphed); if (replica->ip != NULL) { replica->ip->cmds->internal_constraints(replica->ip); // Replica IP address -> replica container (symmetric) pcmk__order_starts(replica->ip, replica->container, pcmk__ar_unrunnable_first_blocks |pcmk__ar_guest_allowed); pcmk__order_stops(replica->container, replica->ip, pcmk__ar_then_implies_first|pcmk__ar_guest_allowed); - pcmk__new_colocation("#ip-with-container", NULL, INFINITY, replica->ip, - replica->container, NULL, NULL, + pcmk__new_colocation("#ip-with-container", NULL, PCMK_SCORE_INFINITY, + replica->ip, replica->container, NULL, NULL, pcmk__coloc_influence); } if (replica->remote != NULL) { /* This handles ordering and colocating remote relative to container * (via "#resource-with-container"). Since IP is also ordered and * colocated relative to the container, we don't need to do anything * explicit here with IP. */ replica->remote->cmds->internal_constraints(replica->remote); } if (replica->child != NULL) { CRM_ASSERT(replica->remote != NULL); // "Start remote then child" is implicit in scheduler's remote logic } return true; } /*! * \internal * \brief Create implicit constraints needed for a bundle resource * * \param[in,out] rsc Bundle resource to create implicit constraints for */ void pcmk__bundle_internal_constraints(pcmk_resource_t *rsc) { pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, replica_internal_constraints, rsc); bundled_resource = pe__bundled_resource(rsc); if (bundled_resource == NULL) { return; } // Start bundle -> start bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_START, bundled_resource, PCMK_ACTION_START, pcmk__ar_then_implies_first_graphed); // Bundled clone is started -> bundle is started pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_RUNNING, rsc, PCMK_ACTION_RUNNING, pcmk__ar_first_implies_then_graphed); // Stop bundle -> stop bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, bundled_resource, PCMK_ACTION_STOP, pcmk__ar_then_implies_first_graphed); // Bundled clone is stopped -> bundle is stopped pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_STOPPED, rsc, PCMK_ACTION_STOPPED, pcmk__ar_first_implies_then_graphed); bundled_resource->cmds->internal_constraints(bundled_resource); if (!pcmk_is_set(bundled_resource->flags, pcmk_rsc_promotable)) { return; } pcmk__promotable_restart_ordering(rsc); // Demote bundle -> demote bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_DEMOTE, bundled_resource, PCMK_ACTION_DEMOTE, pcmk__ar_then_implies_first_graphed); // Bundled clone is demoted -> bundle is demoted pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_DEMOTED, rsc, PCMK_ACTION_DEMOTED, pcmk__ar_first_implies_then_graphed); // Promote bundle -> promote bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_PROMOTE, bundled_resource, PCMK_ACTION_PROMOTE, pcmk__ar_then_implies_first_graphed); // Bundled clone is promoted -> bundle is promoted pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_PROMOTED, rsc, PCMK_ACTION_PROMOTED, pcmk__ar_first_implies_then_graphed); } struct match_data { const pcmk_node_t *node; // Node to compare against replica pcmk_resource_t *container; // Replica container corresponding to node }; /*! * \internal * \brief Check whether a replica container is assigned to a given node * * \param[in] replica Replica to check * \param[in,out] user_data struct match_data with node to compare against * * \return true if the replica does not match (to indicate further replicas * should be processed), otherwise false */ static bool match_replica_container(const pcmk__bundle_replica_t *replica, void *user_data) { struct match_data *match_data = user_data; if (pcmk__instance_matches(replica->container, match_data->node, pcmk_role_unknown, false)) { match_data->container = replica->container; return false; // Match found, don't bother searching further replicas } return true; // No match, keep searching } /*! * \internal * \brief Get the host to which a bundle node is assigned * * \param[in] node Possible bundle node to check * * \return Node to which the container for \p node is assigned if \p node is a * bundle node, otherwise \p node itself */ static const pcmk_node_t * get_bundle_node_host(const pcmk_node_t *node) { if (pcmk__is_bundle_node(node)) { const pcmk_resource_t *container = node->details->remote_rsc->container; return container->fns->location(container, NULL, 0); } return node; } /*! * \internal * \brief Find a bundle container compatible with a dependent resource * * \param[in] dependent Dependent resource in colocation with bundle * \param[in] bundle Bundle that \p dependent is colocated with * * \return A container from \p bundle assigned to the same node as \p dependent * if assigned, otherwise assigned to any of dependent's allowed nodes, * otherwise NULL. */ static pcmk_resource_t * compatible_container(const pcmk_resource_t *dependent, const pcmk_resource_t *bundle) { GList *scratch = NULL; struct match_data match_data = { NULL, NULL }; // If dependent is assigned, only check there match_data.node = dependent->fns->location(dependent, NULL, 0); match_data.node = get_bundle_node_host(match_data.node); if (match_data.node != NULL) { pe__foreach_const_bundle_replica(bundle, match_replica_container, &match_data); return match_data.container; } // Otherwise, check for any of the dependent's allowed nodes scratch = g_hash_table_get_values(dependent->allowed_nodes); scratch = pcmk__sort_nodes(scratch, NULL); for (const GList *iter = scratch; iter != NULL; iter = iter->next) { match_data.node = iter->data; match_data.node = get_bundle_node_host(match_data.node); if (match_data.node == NULL) { continue; } pe__foreach_const_bundle_replica(bundle, match_replica_container, &match_data); if (match_data.container != NULL) { break; } } g_list_free(scratch); return match_data.container; } struct coloc_data { const pcmk__colocation_t *colocation; pcmk_resource_t *dependent; GList *container_hosts; }; /*! * \internal * \brief Apply a colocation score to replica node scores or resource priority * * \param[in] replica Replica of primary bundle resource in colocation * \param[in,out] user_data struct coloc_data for colocation being applied * * \return true (to indicate that any further replicas should be processed) */ static bool replica_apply_coloc_score(const pcmk__bundle_replica_t *replica, void *user_data) { struct coloc_data *coloc_data = user_data; pcmk_node_t *chosen = NULL; - if (coloc_data->colocation->score < INFINITY) { + if (coloc_data->colocation->score < PCMK_SCORE_INFINITY) { replica->container->cmds->apply_coloc_score(coloc_data->dependent, replica->container, coloc_data->colocation, false); return true; } chosen = replica->container->fns->location(replica->container, NULL, 0); if ((chosen == NULL) || is_set_recursive(replica->container, pcmk_rsc_blocked, true)) { return true; } if ((coloc_data->colocation->primary_role >= pcmk_role_promoted) && ((replica->child == NULL) || (replica->child->next_role < pcmk_role_promoted))) { return true; } pcmk__rsc_trace(pe__const_top_resource(replica->container, true), "Allowing mandatory colocation %s using %s @%d", coloc_data->colocation->id, pcmk__node_name(chosen), chosen->weight); coloc_data->container_hosts = g_list_prepend(coloc_data->container_hosts, chosen); return true; } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__bundle_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { struct coloc_data coloc_data = { colocation, dependent, NULL }; /* This should never be called for the bundle itself as a dependent. * Instead, we add its colocation constraints to its containers and bundled * primitive and call the apply_coloc_score() method for them as dependents. */ CRM_ASSERT((primary != NULL) && (primary->variant == pcmk_rsc_variant_bundle) && (dependent != NULL) && (dependent->variant == pcmk_rsc_variant_primitive) && (colocation != NULL) && !for_dependent); if (pcmk_is_set(primary->flags, pcmk_rsc_unassigned)) { pcmk__rsc_trace(primary, "Skipping applying colocation %s " "because %s is still provisional", colocation->id, primary->id); return; } pcmk__rsc_trace(primary, "Applying colocation %s (%s with %s at %s)", colocation->id, dependent->id, primary->id, pcmk_readable_score(colocation->score)); /* If the constraint dependent is a clone or bundle, "dependent" here is one * of its instances. Look for a compatible instance of this bundle. */ if (colocation->dependent->variant > pcmk_rsc_variant_group) { const pcmk_resource_t *primary_container = NULL; primary_container = compatible_container(dependent, primary); if (primary_container != NULL) { // Success, we found one pcmk__rsc_debug(primary, "Pairing %s with %s", dependent->id, primary_container->id); dependent->cmds->apply_coloc_score(dependent, primary_container, colocation, true); - } else if (colocation->score >= INFINITY) { // Failure, and it's fatal + } else if (colocation->score >= PCMK_SCORE_INFINITY) { + // Failure, and it's fatal crm_notice("%s cannot run because there is no compatible " "instance of %s to colocate with", dependent->id, primary->id); pcmk__assign_resource(dependent, NULL, true, true); } else { // Failure, but we can ignore it pcmk__rsc_debug(primary, "%s cannot be colocated with any instance of %s", dependent->id, primary->id); } return; } pe__foreach_const_bundle_replica(primary, replica_apply_coloc_score, &coloc_data); - if (colocation->score >= INFINITY) { + if (colocation->score >= PCMK_SCORE_INFINITY) { pcmk__colocation_intersect_nodes(dependent, primary, colocation, coloc_data.container_hosts, false); } g_list_free(coloc_data.container_hosts); } // Bundle implementation of pcmk_assignment_methods_t:with_this_colocations() void pcmk__with_bundle_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *bundled_rsc = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle) && (orig_rsc != NULL) && (list != NULL)); // The bundle itself and its containers always get its colocations if ((orig_rsc == rsc) || pcmk_is_set(orig_rsc->flags, pcmk_rsc_replica_container)) { pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); return; } /* The bundled resource gets the colocations if it's promotable and we've * begun choosing roles */ bundled_rsc = pe__bundled_resource(rsc); if ((bundled_rsc == NULL) || !pcmk_is_set(bundled_rsc->flags, pcmk_rsc_promotable) || (pe__const_top_resource(orig_rsc, false) != bundled_rsc)) { return; } if (orig_rsc == bundled_rsc) { if (pe__clone_flag_is_set(orig_rsc, pcmk__clone_promotion_constrained)) { /* orig_rsc is the clone and we're setting roles (or have already * done so) */ pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); } } else if (!pcmk_is_set(orig_rsc->flags, pcmk_rsc_unassigned)) { /* orig_rsc is an instance and is already assigned. If something * requests colocations for orig_rsc now, it's for setting roles. */ pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); } } // Bundle implementation of pcmk_assignment_methods_t:this_with_colocations() void pcmk__bundle_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *bundled_rsc = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle) && (orig_rsc != NULL) && (list != NULL)); // The bundle itself and its containers always get its colocations if ((orig_rsc == rsc) || pcmk_is_set(orig_rsc->flags, pcmk_rsc_replica_container)) { pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); return; } /* The bundled resource gets the colocations if it's promotable and we've * begun choosing roles */ bundled_rsc = pe__bundled_resource(rsc); if ((bundled_rsc == NULL) || !pcmk_is_set(bundled_rsc->flags, pcmk_rsc_promotable) || (pe__const_top_resource(orig_rsc, false) != bundled_rsc)) { return; } if (orig_rsc == bundled_rsc) { if (pe__clone_flag_is_set(orig_rsc, pcmk__clone_promotion_constrained)) { /* orig_rsc is the clone and we're setting roles (or have already * done so) */ pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); } } else if (!pcmk_is_set(orig_rsc->flags, pcmk_rsc_unassigned)) { /* orig_rsc is an instance and is already assigned. If something * requests colocations for orig_rsc now, it's for setting roles. */ pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); } } /*! * \internal * \brief Return action flags for a given bundle resource action * * \param[in,out] action Bundle resource action to get flags for * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__bundle_action_flags(pcmk_action_t *action, const pcmk_node_t *node) { GList *containers = NULL; uint32_t flags = 0; pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((action != NULL) && (action->rsc != NULL) && (action->rsc->variant == pcmk_rsc_variant_bundle)); bundled_resource = pe__bundled_resource(action->rsc); if (bundled_resource != NULL) { // Clone actions are done on the bundled clone resource, not container switch (get_complex_task(bundled_resource, action->task)) { case pcmk_action_unspecified: case pcmk_action_notify: case pcmk_action_notified: case pcmk_action_promote: case pcmk_action_promoted: case pcmk_action_demote: case pcmk_action_demoted: return pcmk__collective_action_flags(action, bundled_resource->children, node); default: break; } } containers = pe__bundle_containers(action->rsc); flags = pcmk__collective_action_flags(action, containers, node); g_list_free(containers); return flags; } /*! * \internal * \brief Apply a location constraint to a bundle replica * * \param[in,out] replica Replica to apply constraint to * \param[in,out] user_data Location constraint to apply * * \return true (to indicate that any further replicas should be processed) */ static bool apply_location_to_replica(pcmk__bundle_replica_t *replica, void *user_data) { pcmk__location_t *location = user_data; if (replica->container != NULL) { replica->container->cmds->apply_location(replica->container, location); } if (replica->ip != NULL) { replica->ip->cmds->apply_location(replica->ip, location); } return true; } /*! * \internal * \brief Apply a location constraint to a bundle resource's allowed node scores * * \param[in,out] rsc Bundle resource to apply constraint to * \param[in,out] location Location constraint to apply */ void pcmk__bundle_apply_location(pcmk_resource_t *rsc, pcmk__location_t *location) { pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle) && (location != NULL)); pcmk__apply_location(rsc, location); pe__foreach_bundle_replica(rsc, apply_location_to_replica, location); bundled_resource = pe__bundled_resource(rsc); if ((bundled_resource != NULL) && ((location->role_filter == pcmk_role_unpromoted) || (location->role_filter == pcmk_role_promoted))) { bundled_resource->cmds->apply_location(bundled_resource, location); bundled_resource->rsc_location = g_list_prepend( bundled_resource->rsc_location, location); } } #define XPATH_REMOTE "//nvpair[@name='" PCMK_REMOTE_RA_ADDR "']" /*! * \internal * \brief Add a bundle replica's actions to transition graph * * \param[in,out] replica Replica to add to graph * \param[in] user_data Bundle that replica belongs to (for logging only) * * \return true (to indicate that any further replicas should be processed) */ static bool add_replica_actions_to_graph(pcmk__bundle_replica_t *replica, void *user_data) { if ((replica->remote != NULL) && (replica->container != NULL) && pe__bundle_needs_remote_name(replica->remote)) { /* REMOTE_CONTAINER_HACK: Allow remote nodes to run containers that * run pacemaker-remoted inside, without needing a separate IP for * the container. This is done by configuring the inner remote's * connection host as the magic string "#uname", then * replacing it with the underlying host when needed. */ xmlNode *nvpair = get_xpath_object(XPATH_REMOTE, replica->remote->xml, LOG_ERR); const char *calculated_addr = NULL; // Replace the value in replica->remote->xml (if appropriate) calculated_addr = pe__add_bundle_remote_name(replica->remote, replica->remote->cluster, nvpair, PCMK_XA_VALUE); if (calculated_addr != NULL) { /* Since this is for the bundle as a resource, and not any * particular action, replace the value in the default * parameters (not evaluated for node). create_graph_action() * will grab it from there to replace it in node-evaluated * parameters. */ GHashTable *params = pe_rsc_params(replica->remote, NULL, replica->remote->cluster); pcmk__insert_dup(params, PCMK_REMOTE_RA_ADDR, calculated_addr); } else { pcmk_resource_t *bundle = user_data; /* The only way to get here is if the remote connection is * neither currently running nor scheduled to run. That means we * won't be doing any operations that require addr (only start * requires it; we additionally use it to compare digests when * unpacking status, promote, and migrate_from history, but * that's already happened by this point). */ pcmk__rsc_info(bundle, "Unable to determine address for bundle %s " "remote connection", bundle->id); } } if (replica->ip != NULL) { replica->ip->cmds->add_actions_to_graph(replica->ip); } if (replica->container != NULL) { replica->container->cmds->add_actions_to_graph(replica->container); } if (replica->remote != NULL) { replica->remote->cmds->add_actions_to_graph(replica->remote); } return true; } /*! * \internal * \brief Add a bundle resource's actions to the transition graph * * \param[in,out] rsc Bundle resource whose actions should be added */ void pcmk__bundle_add_actions_to_graph(pcmk_resource_t *rsc) { pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); bundled_resource = pe__bundled_resource(rsc); if (bundled_resource != NULL) { bundled_resource->cmds->add_actions_to_graph(bundled_resource); } pe__foreach_bundle_replica(rsc, add_replica_actions_to_graph, rsc); } struct probe_data { pcmk_resource_t *bundle; // Bundle being probed pcmk_node_t *node; // Node to create probes on bool any_created; // Whether any probes have been created }; /*! * \internal * \brief Order a bundle replica's start after another replica's probe * * \param[in,out] replica Replica to order start for * \param[in,out] user_data Replica with probe to order after * * \return true (to indicate that any further replicas should be processed) */ static bool order_replica_start_after(pcmk__bundle_replica_t *replica, void *user_data) { pcmk__bundle_replica_t *probed_replica = user_data; if ((replica == probed_replica) || (replica->container == NULL)) { return true; } pcmk__new_ordering(probed_replica->container, pcmk__op_key(probed_replica->container->id, PCMK_ACTION_MONITOR, 0), NULL, replica->container, pcmk__op_key(replica->container->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered|pcmk__ar_if_on_same_node, replica->container->cluster); return true; } /*! * \internal * \brief Create probes for a bundle replica's resources * * \param[in,out] replica Replica to create probes for * \param[in,out] user_data struct probe_data * * \return true (to indicate that any further replicas should be processed) */ static bool create_replica_probes(pcmk__bundle_replica_t *replica, void *user_data) { struct probe_data *probe_data = user_data; if ((replica->ip != NULL) && replica->ip->cmds->create_probe(replica->ip, probe_data->node)) { probe_data->any_created = true; } if ((replica->child != NULL) && pcmk__same_node(probe_data->node, replica->node) && replica->child->cmds->create_probe(replica->child, probe_data->node)) { probe_data->any_created = true; } if ((replica->container != NULL) && replica->container->cmds->create_probe(replica->container, probe_data->node)) { probe_data->any_created = true; /* If we're limited to one replica per host (due to * the lack of an IP range probably), then we don't * want any of our peer containers starting until * we've established that no other copies are already * running. * * Partly this is to ensure that the maximum replicas per host is * observed, but also to ensure that the containers * don't fail to start because the necessary port * mappings (which won't include an IP for uniqueness) * are already taken */ if (probe_data->bundle->fns->max_per_node(probe_data->bundle) == 1) { pe__foreach_bundle_replica(probe_data->bundle, order_replica_start_after, replica); } } if ((replica->container != NULL) && (replica->remote != NULL) && replica->remote->cmds->create_probe(replica->remote, probe_data->node)) { /* Do not probe the remote resource until we know where the container is * running. This is required for REMOTE_CONTAINER_HACK to correctly * probe remote resources. */ char *probe_uuid = pcmk__op_key(replica->remote->id, PCMK_ACTION_MONITOR, 0); pcmk_action_t *probe = find_first_action(replica->remote->actions, probe_uuid, NULL, probe_data->node); free(probe_uuid); if (probe != NULL) { probe_data->any_created = true; pcmk__rsc_trace(probe_data->bundle, "Ordering %s probe on %s", replica->remote->id, pcmk__node_name(probe_data->node)); pcmk__new_ordering(replica->container, pcmk__op_key(replica->container->id, PCMK_ACTION_START, 0), NULL, replica->remote, NULL, probe, pcmk__ar_nested_remote_probe, probe_data->bundle->cluster); } } return true; } /*! * \internal * * \brief Schedule any probes needed for a bundle resource on a node * * \param[in,out] rsc Bundle resource to create probes for * \param[in,out] node Node to create probe on * * \return true if any probe was created, otherwise false */ bool pcmk__bundle_create_probe(pcmk_resource_t *rsc, pcmk_node_t *node) { struct probe_data probe_data = { rsc, node, false }; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, create_replica_probes, &probe_data); return probe_data.any_created; } /*! * \internal * \brief Output actions for one bundle replica * * \param[in,out] replica Replica to output actions for * \param[in] user_data Unused * * \return true (to indicate that any further replicas should be processed) */ static bool output_replica_actions(pcmk__bundle_replica_t *replica, void *user_data) { if (replica->ip != NULL) { replica->ip->cmds->output_actions(replica->ip); } if (replica->container != NULL) { replica->container->cmds->output_actions(replica->container); } if (replica->remote != NULL) { replica->remote->cmds->output_actions(replica->remote); } if (replica->child != NULL) { replica->child->cmds->output_actions(replica->child); } return true; } /*! * \internal * \brief Output a summary of scheduled actions for a bundle resource * * \param[in,out] rsc Bundle resource to output actions for */ void pcmk__output_bundle_actions(pcmk_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, output_replica_actions, NULL); } // Bundle implementation of pcmk_assignment_methods_t:add_utilization() void pcmk__bundle_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { pcmk_resource_t *container = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { return; } /* All bundle replicas are identical, so using the utilization of the first * is sufficient for any. Only the implicit container resource can have * utilization values. */ container = pe__first_container(rsc); if (container != NULL) { container->cmds->add_utilization(container, orig_rsc, all_rscs, utilization); } } // Bundle implementation of pcmk_assignment_methods_t:shutdown_lock() void pcmk__bundle_shutdown_lock(pcmk_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); // Bundles currently don't support shutdown locks } diff --git a/lib/pacemaker/pcmk_sched_clone.c b/lib/pacemaker/pcmk_sched_clone.c index 2c6238bc54..379f666576 100644 --- a/lib/pacemaker/pcmk_sched_clone.c +++ b/lib/pacemaker/pcmk_sched_clone.c @@ -1,709 +1,709 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Assign a clone resource's instances to nodes * * \param[in,out] rsc Clone resource to assign * \param[in] prefer Node to prefer, if all else is equal * \param[in] stop_if_fail If \c true and a primitive descendant of \p rsc * can't be assigned to a node, set the * descendant's next role to stopped and update * existing actions * * \return NULL (clones are not assigned to a single node) * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ pcmk_node_t * pcmk__clone_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { GList *colocations = NULL; CRM_ASSERT(pcmk__is_clone(rsc)); if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { return NULL; // Assignment has already been done } // Detect assignment loops if (pcmk_is_set(rsc->flags, pcmk_rsc_assigning)) { pcmk__rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id); return NULL; } pcmk__set_rsc_flags(rsc, pcmk_rsc_assigning); // If this clone is promotable, consider nodes' promotion scores if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { pcmk__add_promotion_scores(rsc); } // If this clone is colocated with any other resources, assign those first colocations = pcmk__this_with_colocations(rsc); for (GList *iter = colocations; iter != NULL; iter = iter->next) { pcmk__colocation_t *constraint = (pcmk__colocation_t *) iter->data; pcmk__rsc_trace(rsc, "%s: Assigning colocation %s primary %s first", rsc->id, constraint->id, constraint->primary->id); constraint->primary->cmds->assign(constraint->primary, prefer, stop_if_fail); } g_list_free(colocations); // If any resources are colocated with this one, consider their preferences colocations = pcmk__with_this_colocations(rsc); g_list_foreach(colocations, pcmk__add_dependent_scores, rsc); g_list_free(colocations); pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance); pcmk__assign_instances(rsc, rsc->children, pe__clone_max(rsc), pe__clone_node_max(rsc)); if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { pcmk__set_instance_roles(rsc); } pcmk__clear_rsc_flags(rsc, pcmk_rsc_unassigned|pcmk_rsc_assigning); pcmk__rsc_trace(rsc, "Assigned clone %s", rsc->id); return NULL; } /*! * \internal * \brief Create all actions needed for a given clone resource * * \param[in,out] rsc Clone resource to create actions for */ void pcmk__clone_create_actions(pcmk_resource_t *rsc) { CRM_ASSERT(pcmk__is_clone(rsc)); pcmk__rsc_trace(rsc, "Creating actions for clone %s", rsc->id); pcmk__create_instance_actions(rsc, rsc->children); if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { pcmk__create_promotable_actions(rsc); } } /*! * \internal * \brief Create implicit constraints needed for a clone resource * * \param[in,out] rsc Clone resource to create implicit constraints for */ void pcmk__clone_internal_constraints(pcmk_resource_t *rsc) { bool ordered = false; CRM_ASSERT(pcmk__is_clone(rsc)); pcmk__rsc_trace(rsc, "Creating internal constraints for clone %s", rsc->id); // Restart ordering: Stop -> stopped -> start -> started pcmk__order_resource_actions(rsc, PCMK_ACTION_STOPPED, rsc, PCMK_ACTION_START, pcmk__ar_ordered); pcmk__order_resource_actions(rsc, PCMK_ACTION_START, rsc, PCMK_ACTION_RUNNING, pcmk__ar_unrunnable_first_blocks); pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, rsc, PCMK_ACTION_STOPPED, pcmk__ar_unrunnable_first_blocks); // Demoted -> stop and started -> promote if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { pcmk__order_resource_actions(rsc, PCMK_ACTION_DEMOTED, rsc, PCMK_ACTION_STOP, pcmk__ar_ordered); pcmk__order_resource_actions(rsc, PCMK_ACTION_RUNNING, rsc, PCMK_ACTION_PROMOTE, pcmk__ar_unrunnable_first_blocks); } ordered = pe__clone_is_ordered(rsc); if (ordered) { /* Ordered clone instances must start and stop by instance number. The * instances might have been previously shuffled for assignment or * promotion purposes, so re-sort them. */ rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance_number); } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; instance->cmds->internal_constraints(instance); // Start clone -> start instance -> clone started pcmk__order_starts(rsc, instance, pcmk__ar_unrunnable_first_blocks |pcmk__ar_then_implies_first_graphed); pcmk__order_resource_actions(instance, PCMK_ACTION_START, rsc, PCMK_ACTION_RUNNING, pcmk__ar_first_implies_then_graphed); // Stop clone -> stop instance -> clone stopped pcmk__order_stops(rsc, instance, pcmk__ar_then_implies_first_graphed); pcmk__order_resource_actions(instance, PCMK_ACTION_STOP, rsc, PCMK_ACTION_STOPPED, pcmk__ar_first_implies_then_graphed); /* Instances of ordered clones must be started and stopped by instance * number. Since only some instances may be starting or stopping, order * each instance relative to every later instance. */ if (ordered) { for (GList *later = iter->next; later != NULL; later = later->next) { pcmk__order_starts(instance, (pcmk_resource_t *) later->data, pcmk__ar_ordered); pcmk__order_stops((pcmk_resource_t *) later->data, instance, pcmk__ar_ordered); } } } if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { pcmk__order_promotable_instances(rsc); } } /*! * \internal * \brief Check whether colocated resources can be interleaved * * \param[in] colocation Colocation constraint with clone as primary * * \return true if colocated resources can be interleaved, otherwise false */ static bool can_interleave(const pcmk__colocation_t *colocation) { const pcmk_resource_t *dependent = colocation->dependent; // Only colocations between clone or bundle resources use interleaving if (dependent->variant <= pcmk_rsc_variant_group) { return false; } // Only the dependent needs to be marked for interleaving if (!crm_is_true(g_hash_table_lookup(dependent->meta, PCMK_META_INTERLEAVE))) { return false; } /* @TODO Do we actually care about multiple primary instances sharing a * dependent instance? */ if (dependent->fns->max_per_node(dependent) != colocation->primary->fns->max_per_node(colocation->primary)) { pcmk__config_err("Cannot interleave %s and %s because they do not " "support the same number of instances per node", dependent->id, colocation->primary->id); return false; } return true; } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__clone_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { const GList *iter = NULL; /* This should never be called for the clone itself as a dependent. Instead, * we add its colocation constraints to its instances and call the * apply_coloc_score() method for the instances as dependents. */ CRM_ASSERT(!for_dependent); CRM_ASSERT((colocation != NULL) && pcmk__is_clone(primary) && (dependent != NULL) && (dependent->variant == pcmk_rsc_variant_primitive)); if (pcmk_is_set(primary->flags, pcmk_rsc_unassigned)) { pcmk__rsc_trace(primary, "Delaying processing colocation %s " "because cloned primary %s is still provisional", colocation->id, primary->id); return; } pcmk__rsc_trace(primary, "Processing colocation %s (%s with clone %s @%s)", colocation->id, dependent->id, primary->id, pcmk_readable_score(colocation->score)); // Apply role-specific colocations if (pcmk_is_set(primary->flags, pcmk_rsc_promotable) && (colocation->primary_role != pcmk_role_unknown)) { if (pcmk_is_set(dependent->flags, pcmk_rsc_unassigned)) { // We're assigning the dependent to a node pcmk__update_dependent_with_promotable(primary, dependent, colocation); return; } if (colocation->dependent_role == pcmk_role_promoted) { // We're choosing a role for the dependent pcmk__update_promotable_dependent_priority(primary, dependent, colocation); return; } } // Apply interleaved colocations if (can_interleave(colocation)) { const pcmk_resource_t *primary_instance = NULL; primary_instance = pcmk__find_compatible_instance(dependent, primary, pcmk_role_unknown, false); if (primary_instance != NULL) { pcmk__rsc_debug(primary, "Interleaving %s with %s", dependent->id, primary_instance->id); dependent->cmds->apply_coloc_score(dependent, primary_instance, colocation, true); - } else if (colocation->score >= INFINITY) { + } else if (colocation->score >= PCMK_SCORE_INFINITY) { crm_notice("%s cannot run because it cannot interleave with " "any instance of %s", dependent->id, primary->id); pcmk__assign_resource(dependent, NULL, true, true); } else { pcmk__rsc_debug(primary, "%s will not colocate with %s " "because no instance can interleave with it", dependent->id, primary->id); } return; } // Apply mandatory colocations - if (colocation->score >= INFINITY) { + if (colocation->score >= PCMK_SCORE_INFINITY) { GList *primary_nodes = NULL; // Dependent can run only where primary will have unblocked instances for (iter = primary->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *instance = iter->data; pcmk_node_t *chosen = instance->fns->location(instance, NULL, 0); if ((chosen != NULL) && !is_set_recursive(instance, pcmk_rsc_blocked, TRUE)) { pcmk__rsc_trace(primary, "Allowing %s: %s %d", colocation->id, pcmk__node_name(chosen), chosen->weight); primary_nodes = g_list_prepend(primary_nodes, chosen); } } pcmk__colocation_intersect_nodes(dependent, primary, colocation, primary_nodes, false); g_list_free(primary_nodes); return; } // Apply optional colocations for (iter = primary->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *instance = iter->data; instance->cmds->apply_coloc_score(dependent, instance, colocation, false); } } // Clone implementation of pcmk_assignment_methods_t:with_this_colocations() void pcmk__with_clone_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { CRM_CHECK((rsc != NULL) && (orig_rsc != NULL) && (list != NULL), return); pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); if (rsc->parent != NULL) { rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list); } } // Clone implementation of pcmk_assignment_methods_t:this_with_colocations() void pcmk__clone_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { CRM_CHECK((rsc != NULL) && (orig_rsc != NULL) && (list != NULL), return); pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); if (rsc->parent != NULL) { rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list); } } /*! * \internal * \brief Return action flags for a given clone resource action * * \param[in,out] action Action to get flags for * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__clone_action_flags(pcmk_action_t *action, const pcmk_node_t *node) { CRM_ASSERT((action != NULL) && pcmk__is_clone(action->rsc)); return pcmk__collective_action_flags(action, action->rsc->children, node); } /*! * \internal * \brief Apply a location constraint to a clone resource's allowed node scores * * \param[in,out] rsc Clone resource to apply constraint to * \param[in,out] location Location constraint to apply */ void pcmk__clone_apply_location(pcmk_resource_t *rsc, pcmk__location_t *location) { CRM_CHECK((location != NULL) && pcmk__is_clone(rsc), return); pcmk__apply_location(rsc, location); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; instance->cmds->apply_location(instance, location); } } // GFunc wrapper for calling the action_flags() resource method static void call_action_flags(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = user_data; rsc->cmds->action_flags((pcmk_action_t *) data, NULL); } /*! * \internal * \brief Add a clone resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void pcmk__clone_add_actions_to_graph(pcmk_resource_t *rsc) { CRM_ASSERT(pcmk__is_clone(rsc)); g_list_foreach(rsc->actions, call_action_flags, rsc); pe__create_clone_notifications(rsc); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) iter->data; child_rsc->cmds->add_actions_to_graph(child_rsc); } pcmk__add_rsc_actions_to_graph(rsc); pe__free_clone_notification_data(rsc); } /*! * \internal * \brief Check whether a resource or any children have been probed on a node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p node is in the known_on table of \p rsc or any of its * children, otherwise false */ static bool rsc_probed_on(const pcmk_resource_t *rsc, const pcmk_node_t *node) { if (rsc->children != NULL) { for (GList *child_iter = rsc->children; child_iter != NULL; child_iter = child_iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) child_iter->data; if (rsc_probed_on(child, node)) { return true; } } return false; } if (rsc->known_on != NULL) { GHashTableIter iter; pcmk_node_t *known_node = NULL; g_hash_table_iter_init(&iter, rsc->known_on); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &known_node)) { if (pcmk__same_node(node, known_node)) { return true; } } } return false; } /*! * \internal * \brief Find clone instance that has been probed on given node * * \param[in] clone Clone resource to check * \param[in] node Node to check * * \return Instance of \p clone that has been probed on \p node if any, * otherwise NULL */ static pcmk_resource_t * find_probed_instance_on(const pcmk_resource_t *clone, const pcmk_node_t *node) { for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; if (rsc_probed_on(instance, node)) { return instance; } } return NULL; } /*! * \internal * \brief Probe an anonymous clone on a node * * \param[in,out] clone Anonymous clone to probe * \param[in,out] node Node to probe \p clone on */ static bool probe_anonymous_clone(pcmk_resource_t *clone, pcmk_node_t *node) { // Check whether we already probed an instance on this node pcmk_resource_t *child = find_probed_instance_on(clone, node); // Otherwise, check if we plan to start an instance on this node for (GList *iter = clone->children; (iter != NULL) && (child == NULL); iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; const pcmk_node_t *instance_node = NULL; instance_node = instance->fns->location(instance, NULL, 0); if (pcmk__same_node(instance_node, node)) { child = instance; } } // Otherwise, use the first clone instance if (child == NULL) { child = clone->children->data; } // Anonymous clones only need to probe a single instance return child->cmds->create_probe(child, node); } /*! * \internal * \brief Schedule any probes needed for a resource on a node * * \param[in,out] rsc Resource to create probe for * \param[in,out] node Node to create probe on * * \return true if any probe was created, otherwise false */ bool pcmk__clone_create_probe(pcmk_resource_t *rsc, pcmk_node_t *node) { CRM_ASSERT((node != NULL) && pcmk__is_clone(rsc)); if (rsc->exclusive_discover) { /* The clone is configured to be probed only where a location constraint * exists with PCMK_XA_RESOURCE_DISCOVERY set to exclusive. * * This check is not strictly necessary here since the instance's * create_probe() method would also check, but doing it here is more * efficient (especially for unique clones with a large number of * instances), and affects the CRM_meta_notify_available_uname variable * passed with notify actions. */ pcmk_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if ((allowed == NULL) || (allowed->rsc_discover_mode != pcmk_probe_exclusive)) { /* This node is not marked for resource discovery. Remove it from * allowed_nodes so that notifications contain only nodes that the * clone can possibly run on. */ pcmk__rsc_trace(rsc, "Skipping probe for %s on %s because resource has " "exclusive discovery but is not allowed on node", rsc->id, pcmk__node_name(node)); g_hash_table_remove(rsc->allowed_nodes, node->details->id); return false; } } rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance_number); if (pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { return pcmk__probe_resource_list(rsc->children, node); } else { return probe_anonymous_clone(rsc, node); } } /*! * \internal * \brief Add meta-attributes relevant to transition graph actions to XML * * Add clone-specific meta-attributes needed for transition graph actions. * * \param[in] rsc Clone resource whose meta-attributes should be added * \param[in,out] xml Transition graph action attributes XML to add to */ void pcmk__clone_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml) { char *name = NULL; CRM_ASSERT(pcmk__is_clone(rsc) && (xml != NULL)); name = crm_meta_name(PCMK_META_GLOBALLY_UNIQUE); crm_xml_add(xml, name, pcmk__flag_text(rsc->flags, pcmk_rsc_unique)); free(name); name = crm_meta_name(PCMK_META_NOTIFY); crm_xml_add(xml, name, pcmk__flag_text(rsc->flags, pcmk_rsc_notify)); free(name); name = crm_meta_name(PCMK_META_CLONE_MAX); crm_xml_add_int(xml, name, pe__clone_max(rsc)); free(name); name = crm_meta_name(PCMK_META_CLONE_NODE_MAX); crm_xml_add_int(xml, name, pe__clone_node_max(rsc)); free(name); if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { int promoted_max = pe__clone_promoted_max(rsc); int promoted_node_max = pe__clone_promoted_node_max(rsc); name = crm_meta_name(PCMK_META_PROMOTED_MAX); crm_xml_add_int(xml, name, promoted_max); free(name); name = crm_meta_name(PCMK_META_PROMOTED_NODE_MAX); crm_xml_add_int(xml, name, promoted_node_max); free(name); /* @COMPAT Maintain backward compatibility with resource agents that * expect the old names (deprecated since 2.0.0). */ name = crm_meta_name(PCMK__META_PROMOTED_MAX_LEGACY); crm_xml_add_int(xml, name, promoted_max); free(name); name = crm_meta_name(PCMK__META_PROMOTED_NODE_MAX_LEGACY); crm_xml_add_int(xml, name, promoted_node_max); free(name); } } // Clone implementation of pcmk_assignment_methods_t:add_utilization() void pcmk__clone_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { bool existing = false; pcmk_resource_t *child = NULL; CRM_ASSERT(pcmk__is_clone(rsc) && (orig_rsc != NULL) && (utilization != NULL)); if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { return; } // Look for any child already existing in the list for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { child = (pcmk_resource_t *) iter->data; if (g_list_find(all_rscs, child)) { existing = true; // Keep checking remaining children } else { // If this is a clone of a group, look for group's members for (GList *member_iter = child->children; member_iter != NULL; member_iter = member_iter->next) { pcmk_resource_t *member = (pcmk_resource_t *) member_iter->data; if (g_list_find(all_rscs, member) != NULL) { // Add *child's* utilization, not group member's child->cmds->add_utilization(child, orig_rsc, all_rscs, utilization); existing = true; break; } } } } if (!existing && (rsc->children != NULL)) { // If nothing was found, still add first child's utilization child = (pcmk_resource_t *) rsc->children->data; child->cmds->add_utilization(child, orig_rsc, all_rscs, utilization); } } // Clone implementation of pcmk_assignment_methods_t:shutdown_lock() void pcmk__clone_shutdown_lock(pcmk_resource_t *rsc) { CRM_ASSERT(pcmk__is_clone(rsc)); return; // Clones currently don't support shutdown locks } diff --git a/lib/pacemaker/pcmk_sched_colocation.c b/lib/pacemaker/pcmk_sched_colocation.c index 9108fffc7b..0aee0c5c54 100644 --- a/lib/pacemaker/pcmk_sched_colocation.c +++ b/lib/pacemaker/pcmk_sched_colocation.c @@ -1,1927 +1,1929 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include "crm/common/util.h" #include "crm/common/xml_internal.h" #include "crm/common/xml.h" #include "libpacemaker_private.h" // Used to temporarily mark a node as unusable -#define INFINITY_HACK (INFINITY * -100) +#define INFINITY_HACK (PCMK_SCORE_INFINITY * -100) /*! * \internal * \brief Compare two colocations according to priority * * Compare two colocations according to the order in which they should be * considered, based on either their dependent resources or their primary * resources -- preferring (in order): * * Colocation that is not \c NULL * * Colocation whose resource has higher priority * * Colocation whose resource is of a higher-level variant * (bundle > clone > group > primitive) * * Colocation whose resource is promotable, if both are clones * * Colocation whose resource has lower ID in lexicographic order * * \param[in] colocation1 First colocation to compare * \param[in] colocation2 Second colocation to compare * \param[in] dependent If \c true, compare colocations by dependent * priority; otherwise compare them by primary priority * * \return A negative number if \p colocation1 should be considered first, * a positive number if \p colocation2 should be considered first, * or 0 if order doesn't matter */ static gint cmp_colocation_priority(const pcmk__colocation_t *colocation1, const pcmk__colocation_t *colocation2, bool dependent) { const pcmk_resource_t *rsc1 = NULL; const pcmk_resource_t *rsc2 = NULL; if (colocation1 == NULL) { return 1; } if (colocation2 == NULL) { return -1; } if (dependent) { rsc1 = colocation1->dependent; rsc2 = colocation2->dependent; CRM_ASSERT(colocation1->primary != NULL); } else { rsc1 = colocation1->primary; rsc2 = colocation2->primary; CRM_ASSERT(colocation1->dependent != NULL); } CRM_ASSERT((rsc1 != NULL) && (rsc2 != NULL)); if (rsc1->priority > rsc2->priority) { return -1; } if (rsc1->priority < rsc2->priority) { return 1; } // Process clones before primitives and groups if (rsc1->variant > rsc2->variant) { return -1; } if (rsc1->variant < rsc2->variant) { return 1; } /* @COMPAT scheduler <2.0.0: Process promotable clones before nonpromotable * clones (probably unnecessary, but avoids having to update regression * tests) */ if (rsc1->variant == pcmk_rsc_variant_clone) { if (pcmk_is_set(rsc1->flags, pcmk_rsc_promotable) && !pcmk_is_set(rsc2->flags, pcmk_rsc_promotable)) { return -1; } if (!pcmk_is_set(rsc1->flags, pcmk_rsc_promotable) && pcmk_is_set(rsc2->flags, pcmk_rsc_promotable)) { return 1; } } return strcmp(rsc1->id, rsc2->id); } /*! * \internal * \brief Compare two colocations according to priority based on dependents * * Compare two colocations according to the order in which they should be * considered, based on their dependent resources -- preferring (in order): * * Colocation that is not \c NULL * * Colocation whose resource has higher priority * * Colocation whose resource is of a higher-level variant * (bundle > clone > group > primitive) * * Colocation whose resource is promotable, if both are clones * * Colocation whose resource has lower ID in lexicographic order * * \param[in] a First colocation to compare * \param[in] b Second colocation to compare * * \return A negative number if \p a should be considered first, * a positive number if \p b should be considered first, * or 0 if order doesn't matter */ static gint cmp_dependent_priority(gconstpointer a, gconstpointer b) { return cmp_colocation_priority(a, b, true); } /*! * \internal * \brief Compare two colocations according to priority based on primaries * * Compare two colocations according to the order in which they should be * considered, based on their primary resources -- preferring (in order): * * Colocation that is not \c NULL * * Colocation whose primary has higher priority * * Colocation whose primary is of a higher-level variant * (bundle > clone > group > primitive) * * Colocation whose primary is promotable, if both are clones * * Colocation whose primary has lower ID in lexicographic order * * \param[in] a First colocation to compare * \param[in] b Second colocation to compare * * \return A negative number if \p a should be considered first, * a positive number if \p b should be considered first, * or 0 if order doesn't matter */ static gint cmp_primary_priority(gconstpointer a, gconstpointer b) { return cmp_colocation_priority(a, b, false); } /*! * \internal * \brief Add a "this with" colocation constraint to a sorted list * * \param[in,out] list List of constraints to add \p colocation to * \param[in] colocation Colocation constraint to add to \p list * \param[in] rsc Resource whose colocations we're getting (for * logging only) * * \note The list will be sorted using cmp_primary_priority(). */ void pcmk__add_this_with(GList **list, const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc) { CRM_ASSERT((list != NULL) && (colocation != NULL) && (rsc != NULL)); pcmk__rsc_trace(rsc, "Adding colocation %s (%s with %s using %s @%s) to " "'this with' list for %s", colocation->id, colocation->dependent->id, colocation->primary->id, colocation->node_attribute, pcmk_readable_score(colocation->score), rsc->id); *list = g_list_insert_sorted(*list, (gpointer) colocation, cmp_primary_priority); } /*! * \internal * \brief Add a list of "this with" colocation constraints to a list * * \param[in,out] list List of constraints to add \p addition to * \param[in] addition List of colocation constraints to add to \p list * \param[in] rsc Resource whose colocations we're getting (for * logging only) * * \note The lists must be pre-sorted by cmp_primary_priority(). */ void pcmk__add_this_with_list(GList **list, GList *addition, const pcmk_resource_t *rsc) { CRM_ASSERT((list != NULL) && (rsc != NULL)); pcmk__if_tracing( {}, // Always add each colocation individually if tracing { if (*list == NULL) { // Trivial case for efficiency if not tracing *list = g_list_copy(addition); return; } } ); for (const GList *iter = addition; iter != NULL; iter = iter->next) { pcmk__add_this_with(list, addition->data, rsc); } } /*! * \internal * \brief Add a "with this" colocation constraint to a sorted list * * \param[in,out] list List of constraints to add \p colocation to * \param[in] colocation Colocation constraint to add to \p list * \param[in] rsc Resource whose colocations we're getting (for * logging only) * * \note The list will be sorted using cmp_dependent_priority(). */ void pcmk__add_with_this(GList **list, const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc) { CRM_ASSERT((list != NULL) && (colocation != NULL) && (rsc != NULL)); pcmk__rsc_trace(rsc, "Adding colocation %s (%s with %s using %s @%s) to " "'with this' list for %s", colocation->id, colocation->dependent->id, colocation->primary->id, colocation->node_attribute, pcmk_readable_score(colocation->score), rsc->id); *list = g_list_insert_sorted(*list, (gpointer) colocation, cmp_dependent_priority); } /*! * \internal * \brief Add a list of "with this" colocation constraints to a list * * \param[in,out] list List of constraints to add \p addition to * \param[in] addition List of colocation constraints to add to \p list * \param[in] rsc Resource whose colocations we're getting (for * logging only) * * \note The lists must be pre-sorted by cmp_dependent_priority(). */ void pcmk__add_with_this_list(GList **list, GList *addition, const pcmk_resource_t *rsc) { CRM_ASSERT((list != NULL) && (rsc != NULL)); pcmk__if_tracing( {}, // Always add each colocation individually if tracing { if (*list == NULL) { // Trivial case for efficiency if not tracing *list = g_list_copy(addition); return; } } ); for (const GList *iter = addition; iter != NULL; iter = iter->next) { pcmk__add_with_this(list, addition->data, rsc); } } /*! * \internal * \brief Add orderings necessary for an anti-colocation constraint * * \param[in,out] first_rsc One resource in an anti-colocation * \param[in] first_role Anti-colocation role of \p first_rsc * \param[in] then_rsc Other resource in the anti-colocation * \param[in] then_role Anti-colocation role of \p then_rsc */ static void anti_colocation_order(pcmk_resource_t *first_rsc, int first_role, pcmk_resource_t *then_rsc, int then_role) { const char *first_tasks[] = { NULL, NULL }; const char *then_tasks[] = { NULL, NULL }; /* Actions to make first_rsc lose first_role */ if (first_role == pcmk_role_promoted) { first_tasks[0] = PCMK_ACTION_DEMOTE; } else { first_tasks[0] = PCMK_ACTION_STOP; if (first_role == pcmk_role_unpromoted) { first_tasks[1] = PCMK_ACTION_PROMOTE; } } /* Actions to make then_rsc gain then_role */ if (then_role == pcmk_role_promoted) { then_tasks[0] = PCMK_ACTION_PROMOTE; } else { then_tasks[0] = PCMK_ACTION_START; if (then_role == pcmk_role_unpromoted) { then_tasks[1] = PCMK_ACTION_DEMOTE; } } for (int first_lpc = 0; (first_lpc <= 1) && (first_tasks[first_lpc] != NULL); first_lpc++) { for (int then_lpc = 0; (then_lpc <= 1) && (then_tasks[then_lpc] != NULL); then_lpc++) { pcmk__order_resource_actions(first_rsc, first_tasks[first_lpc], then_rsc, then_tasks[then_lpc], pcmk__ar_if_required_on_same_node); } } } /*! * \internal * \brief Add a new colocation constraint to scheduler data * * \param[in] id XML ID for this constraint * \param[in] node_attr Colocate by this attribute (NULL for #uname) * \param[in] score Constraint score * \param[in,out] dependent Resource to be colocated * \param[in,out] primary Resource to colocate \p dependent with * \param[in] dependent_role Current role of \p dependent * \param[in] primary_role Current role of \p primary * \param[in] flags Group of enum pcmk__coloc_flags */ void pcmk__new_colocation(const char *id, const char *node_attr, int score, pcmk_resource_t *dependent, pcmk_resource_t *primary, const char *dependent_role, const char *primary_role, uint32_t flags) { pcmk__colocation_t *new_con = NULL; CRM_CHECK(id != NULL, return); if ((dependent == NULL) || (primary == NULL)) { pcmk__config_err("Ignoring colocation '%s' because resource " "does not exist", id); return; } if (score == 0) { pcmk__rsc_trace(dependent, "Ignoring colocation '%s' (%s with %s) because score is 0", id, dependent->id, primary->id); return; } new_con = calloc(1, sizeof(pcmk__colocation_t)); CRM_ASSERT(new_con != NULL); if (pcmk__str_eq(dependent_role, PCMK__ROLE_STARTED, pcmk__str_null_matches|pcmk__str_casei)) { dependent_role = PCMK__ROLE_UNKNOWN; } if (pcmk__str_eq(primary_role, PCMK__ROLE_STARTED, pcmk__str_null_matches|pcmk__str_casei)) { primary_role = PCMK__ROLE_UNKNOWN; } new_con->id = id; new_con->dependent = dependent; new_con->primary = primary; new_con->score = score; new_con->dependent_role = pcmk_parse_role(dependent_role); new_con->primary_role = pcmk_parse_role(primary_role); new_con->node_attribute = pcmk__s(node_attr, CRM_ATTR_UNAME); new_con->flags = flags; pcmk__add_this_with(&(dependent->rsc_cons), new_con, dependent); pcmk__add_with_this(&(primary->rsc_cons_lhs), new_con, primary); dependent->cluster->colocation_constraints = g_list_prepend( dependent->cluster->colocation_constraints, new_con); - if (score <= -INFINITY) { + if (score <= -PCMK_SCORE_INFINITY) { anti_colocation_order(dependent, new_con->dependent_role, primary, new_con->primary_role); anti_colocation_order(primary, new_con->primary_role, dependent, new_con->dependent_role); } } /*! * \internal * \brief Return the boolean influence corresponding to configuration * * \param[in] coloc_id Colocation XML ID (for error logging) * \param[in] rsc Resource involved in constraint (for default) * \param[in] influence_s String value of \c PCMK_XA_INFLUENCE option * * \return \c pcmk__coloc_influence if string evaluates true, or string is * \c NULL or invalid and resource's \c PCMK_META_CRITICAL option * evaluates true, otherwise \c pcmk__coloc_none */ static uint32_t unpack_influence(const char *coloc_id, const pcmk_resource_t *rsc, const char *influence_s) { if (influence_s != NULL) { int influence_i = 0; if (crm_str_to_boolean(influence_s, &influence_i) < 0) { pcmk__config_err("Constraint '%s' has invalid value for " PCMK_XA_INFLUENCE " (using default)", coloc_id); } else { return (influence_i == 0)? pcmk__coloc_none : pcmk__coloc_influence; } } if (pcmk_is_set(rsc->flags, pcmk_rsc_critical)) { return pcmk__coloc_influence; } return pcmk__coloc_none; } static void unpack_colocation_set(xmlNode *set, int score, const char *coloc_id, const char *influence_s, pcmk_scheduler_t *scheduler) { xmlNode *xml_rsc = NULL; pcmk_resource_t *other = NULL; pcmk_resource_t *resource = NULL; const char *set_id = pcmk__xe_id(set); const char *role = crm_element_value(set, PCMK_XA_ROLE); bool with_previous = false; int local_score = score; bool sequential = false; uint32_t flags = pcmk__coloc_none; const char *xml_rsc_id = NULL; const char *score_s = crm_element_value(set, PCMK_XA_SCORE); if (score_s) { local_score = char2score(score_s); } if (local_score == 0) { crm_trace("Ignoring colocation '%s' for set '%s' because score is 0", coloc_id, set_id); return; } /* @COMPAT The deprecated PCMK__XA_ORDERING attribute specifies whether * resources in a positive-score set are colocated with the previous or next * resource. */ if (pcmk__str_eq(crm_element_value(set, PCMK__XA_ORDERING), PCMK__VALUE_GROUP, pcmk__str_null_matches|pcmk__str_casei)) { with_previous = true; } else { pcmk__warn_once(pcmk__wo_set_ordering, "Support for '" PCMK__XA_ORDERING "' other than" " '" PCMK__VALUE_GROUP "' in " PCMK_XE_RESOURCE_SET " (such as %s) is deprecated and will be removed in a" " future release", set_id); } if ((pcmk__xe_get_bool_attr(set, PCMK_XA_SEQUENTIAL, &sequential) == pcmk_rc_ok) && !sequential) { return; } if (local_score > 0) { for (xml_rsc = first_named_child(set, PCMK_XE_RESOURCE_REF); xml_rsc != NULL; xml_rsc = crm_next_same_xml(xml_rsc)) { xml_rsc_id = pcmk__xe_id(xml_rsc); resource = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (resource == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring %s and later resources in set %s: " "No such resource", xml_rsc_id, set_id); return; } if (other != NULL) { flags = pcmk__coloc_explicit | unpack_influence(coloc_id, resource, influence_s); if (with_previous) { pcmk__rsc_trace(resource, "Colocating %s with %s in set %s", resource->id, other->id, set_id); pcmk__new_colocation(set_id, NULL, local_score, resource, other, role, role, flags); } else { pcmk__rsc_trace(resource, "Colocating %s with %s in set %s", other->id, resource->id, set_id); pcmk__new_colocation(set_id, NULL, local_score, other, resource, role, role, flags); } } other = resource; } } else { /* Anti-colocating with every prior resource is * the only way to ensure the intuitive result * (i.e. that no one in the set can run with anyone else in the set) */ for (xml_rsc = first_named_child(set, PCMK_XE_RESOURCE_REF); xml_rsc != NULL; xml_rsc = crm_next_same_xml(xml_rsc)) { xmlNode *xml_rsc_with = NULL; xml_rsc_id = pcmk__xe_id(xml_rsc); resource = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (resource == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring %s and later resources in set %s: " "No such resource", xml_rsc_id, set_id); return; } flags = pcmk__coloc_explicit | unpack_influence(coloc_id, resource, influence_s); for (xml_rsc_with = first_named_child(set, PCMK_XE_RESOURCE_REF); xml_rsc_with != NULL; xml_rsc_with = crm_next_same_xml(xml_rsc_with)) { xml_rsc_id = pcmk__xe_id(xml_rsc_with); if (pcmk__str_eq(resource->id, xml_rsc_id, pcmk__str_none)) { break; } other = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); CRM_ASSERT(other != NULL); // We already processed it pcmk__new_colocation(set_id, NULL, local_score, resource, other, role, role, flags); } } } } /*! * \internal * \brief Colocate two resource sets relative to each other * * \param[in] id Colocation XML ID * \param[in] set1 Dependent set * \param[in] set2 Primary set * \param[in] score Colocation score * \param[in] influence_s Value of colocation's \c PCMK_XA_INFLUENCE * attribute * \param[in,out] scheduler Scheduler data */ static void colocate_rsc_sets(const char *id, const xmlNode *set1, const xmlNode *set2, int score, const char *influence_s, pcmk_scheduler_t *scheduler) { xmlNode *xml_rsc = NULL; pcmk_resource_t *rsc_1 = NULL; pcmk_resource_t *rsc_2 = NULL; const char *xml_rsc_id = NULL; const char *role_1 = crm_element_value(set1, PCMK_XA_ROLE); const char *role_2 = crm_element_value(set2, PCMK_XA_ROLE); int rc = pcmk_rc_ok; bool sequential = false; uint32_t flags = pcmk__coloc_none; if (score == 0) { crm_trace("Ignoring colocation '%s' between sets %s and %s " "because score is 0", id, pcmk__xe_id(set1), pcmk__xe_id(set2)); return; } rc = pcmk__xe_get_bool_attr(set1, PCMK_XA_SEQUENTIAL, &sequential); if ((rc != pcmk_rc_ok) || sequential) { // Get the first one xml_rsc = first_named_child(set1, PCMK_XE_RESOURCE_REF); if (xml_rsc != NULL) { xml_rsc_id = pcmk__xe_id(xml_rsc); rsc_1 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_1 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring colocation of set %s with set %s " "because first resource %s not found", pcmk__xe_id(set1), pcmk__xe_id(set2), xml_rsc_id); return; } } } rc = pcmk__xe_get_bool_attr(set2, PCMK_XA_SEQUENTIAL, &sequential); if ((rc != pcmk_rc_ok) || sequential) { // Get the last one for (xml_rsc = first_named_child(set2, PCMK_XE_RESOURCE_REF); xml_rsc != NULL; xml_rsc = crm_next_same_xml(xml_rsc)) { xml_rsc_id = pcmk__xe_id(xml_rsc); } rsc_2 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_2 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring colocation of set %s with set %s " "because last resource %s not found", pcmk__xe_id(set1), pcmk__xe_id(set2), xml_rsc_id); return; } } if ((rsc_1 != NULL) && (rsc_2 != NULL)) { // Both sets are sequential flags = pcmk__coloc_explicit | unpack_influence(id, rsc_1, influence_s); pcmk__new_colocation(id, NULL, score, rsc_1, rsc_2, role_1, role_2, flags); } else if (rsc_1 != NULL) { // Only set1 is sequential flags = pcmk__coloc_explicit | unpack_influence(id, rsc_1, influence_s); for (xml_rsc = first_named_child(set2, PCMK_XE_RESOURCE_REF); xml_rsc != NULL; xml_rsc = crm_next_same_xml(xml_rsc)) { xml_rsc_id = pcmk__xe_id(xml_rsc); rsc_2 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_2 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring set %s colocation with resource %s " "in set %s: No such resource", pcmk__xe_id(set1), xml_rsc_id, pcmk__xe_id(set2)); continue; } pcmk__new_colocation(id, NULL, score, rsc_1, rsc_2, role_1, role_2, flags); } } else if (rsc_2 != NULL) { // Only set2 is sequential for (xml_rsc = first_named_child(set1, PCMK_XE_RESOURCE_REF); xml_rsc != NULL; xml_rsc = crm_next_same_xml(xml_rsc)) { xml_rsc_id = pcmk__xe_id(xml_rsc); rsc_1 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_1 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring colocation of set %s resource %s " "with set %s: No such resource", pcmk__xe_id(set1), xml_rsc_id, pcmk__xe_id(set2)); continue; } flags = pcmk__coloc_explicit | unpack_influence(id, rsc_1, influence_s); pcmk__new_colocation(id, NULL, score, rsc_1, rsc_2, role_1, role_2, flags); } } else { // Neither set is sequential for (xml_rsc = first_named_child(set1, PCMK_XE_RESOURCE_REF); xml_rsc != NULL; xml_rsc = crm_next_same_xml(xml_rsc)) { xmlNode *xml_rsc_2 = NULL; xml_rsc_id = pcmk__xe_id(xml_rsc); rsc_1 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_1 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring colocation of set %s resource %s " "with set %s: No such resource", pcmk__xe_id(set1), xml_rsc_id, pcmk__xe_id(set2)); continue; } flags = pcmk__coloc_explicit | unpack_influence(id, rsc_1, influence_s); for (xml_rsc_2 = first_named_child(set2, PCMK_XE_RESOURCE_REF); xml_rsc_2 != NULL; xml_rsc_2 = crm_next_same_xml(xml_rsc_2)) { xml_rsc_id = pcmk__xe_id(xml_rsc_2); rsc_2 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_2 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring colocation of set %s resource " "%s with set %s resource %s: No such " "resource", pcmk__xe_id(set1), pcmk__xe_id(xml_rsc), pcmk__xe_id(set2), xml_rsc_id); continue; } pcmk__new_colocation(id, NULL, score, rsc_1, rsc_2, role_1, role_2, flags); } } } } static void unpack_simple_colocation(xmlNode *xml_obj, const char *id, const char *influence_s, pcmk_scheduler_t *scheduler) { int score_i = 0; uint32_t flags = pcmk__coloc_none; const char *score = crm_element_value(xml_obj, PCMK_XA_SCORE); const char *dependent_id = crm_element_value(xml_obj, PCMK_XA_RSC); const char *primary_id = crm_element_value(xml_obj, PCMK_XA_WITH_RSC); const char *dependent_role = crm_element_value(xml_obj, PCMK_XA_RSC_ROLE); const char *primary_role = crm_element_value(xml_obj, PCMK_XA_WITH_RSC_ROLE); const char *attr = crm_element_value(xml_obj, PCMK_XA_NODE_ATTRIBUTE); const char *primary_instance = NULL; const char *dependent_instance = NULL; pcmk_resource_t *primary = NULL; pcmk_resource_t *dependent = NULL; primary = pcmk__find_constraint_resource(scheduler->resources, primary_id); dependent = pcmk__find_constraint_resource(scheduler->resources, dependent_id); // @COMPAT: Deprecated since 2.1.5 primary_instance = crm_element_value(xml_obj, PCMK__XA_WITH_RSC_INSTANCE); dependent_instance = crm_element_value(xml_obj, PCMK__XA_RSC_INSTANCE); if (dependent_instance != NULL) { pcmk__warn_once(pcmk__wo_coloc_inst, "Support for " PCMK__XA_RSC_INSTANCE " is deprecated " "and will be removed in a future release"); } if (primary_instance != NULL) { pcmk__warn_once(pcmk__wo_coloc_inst, "Support for " PCMK__XA_WITH_RSC_INSTANCE " is " "deprecated and will be removed in a future release"); } if (dependent == NULL) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "does not exist", id, dependent_id); return; } else if (primary == NULL) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "does not exist", id, primary_id); return; } else if ((dependent_instance != NULL) && !pcmk__is_clone(dependent)) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "is not a clone but instance '%s' was requested", id, dependent_id, dependent_instance); return; } else if ((primary_instance != NULL) && !pcmk__is_clone(primary)) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "is not a clone but instance '%s' was requested", id, primary_id, primary_instance); return; } if (dependent_instance != NULL) { dependent = find_clone_instance(dependent, dependent_instance); if (dependent == NULL) { pcmk__config_warn("Ignoring constraint '%s' because resource '%s' " "does not have an instance '%s'", id, dependent_id, dependent_instance); return; } } if (primary_instance != NULL) { primary = find_clone_instance(primary, primary_instance); if (primary == NULL) { pcmk__config_warn("Ignoring constraint '%s' because resource '%s' " "does not have an instance '%s'", "'%s'", id, primary_id, primary_instance); return; } } if (pcmk__xe_attr_is_true(xml_obj, PCMK_XA_SYMMETRICAL)) { pcmk__config_warn("The colocation constraint " "'" PCMK_XA_SYMMETRICAL "' attribute has been " "removed"); } if (score) { score_i = char2score(score); } flags = pcmk__coloc_explicit | unpack_influence(id, dependent, influence_s); pcmk__new_colocation(id, attr, score_i, dependent, primary, dependent_role, primary_role, flags); } // \return Standard Pacemaker return code static int unpack_colocation_tags(xmlNode *xml_obj, xmlNode **expanded_xml, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *dependent_id = NULL; const char *primary_id = NULL; const char *dependent_role = NULL; const char *primary_role = NULL; pcmk_resource_t *dependent = NULL; pcmk_resource_t *primary = NULL; pcmk_tag_t *dependent_tag = NULL; pcmk_tag_t *primary_tag = NULL; xmlNode *dependent_set = NULL; xmlNode *primary_set = NULL; bool any_sets = false; *expanded_xml = NULL; CRM_CHECK(xml_obj != NULL, return EINVAL); id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("Ignoring <%s> constraint without " PCMK_XA_ID, xml_obj->name); return pcmk_rc_unpack_error; } // Check whether there are any resource sets with template or tag references *expanded_xml = pcmk__expand_tags_in_sets(xml_obj, scheduler); if (*expanded_xml != NULL) { crm_log_xml_trace(*expanded_xml, "Expanded " PCMK_XE_RSC_COLOCATION); return pcmk_rc_ok; } dependent_id = crm_element_value(xml_obj, PCMK_XA_RSC); primary_id = crm_element_value(xml_obj, PCMK_XA_WITH_RSC); if ((dependent_id == NULL) || (primary_id == NULL)) { return pcmk_rc_ok; } if (!pcmk__valid_resource_or_tag(scheduler, dependent_id, &dependent, &dependent_tag)) { pcmk__config_err("Ignoring constraint '%s' because '%s' is not a " "valid resource or tag", id, dependent_id); return pcmk_rc_unpack_error; } if (!pcmk__valid_resource_or_tag(scheduler, primary_id, &primary, &primary_tag)) { pcmk__config_err("Ignoring constraint '%s' because '%s' is not a " "valid resource or tag", id, primary_id); return pcmk_rc_unpack_error; } if ((dependent != NULL) && (primary != NULL)) { /* Neither side references any template/tag. */ return pcmk_rc_ok; } if ((dependent_tag != NULL) && (primary_tag != NULL)) { // A colocation constraint between two templates/tags makes no sense pcmk__config_err("Ignoring constraint '%s' because two templates or " "tags cannot be colocated", id); return pcmk_rc_unpack_error; } dependent_role = crm_element_value(xml_obj, PCMK_XA_RSC_ROLE); primary_role = crm_element_value(xml_obj, PCMK_XA_WITH_RSC_ROLE); *expanded_xml = copy_xml(xml_obj); /* Convert dependent's template/tag reference into constraint * PCMK_XE_RESOURCE_SET */ if (!pcmk__tag_to_set(*expanded_xml, &dependent_set, PCMK_XA_RSC, true, scheduler)) { free_xml(*expanded_xml); *expanded_xml = NULL; return pcmk_rc_unpack_error; } if (dependent_set != NULL) { if (dependent_role != NULL) { /* Move PCMK_XA_RSC_ROLE into converted PCMK_XE_RESOURCE_SET as * PCMK_XA_ROLE */ crm_xml_add(dependent_set, PCMK_XA_ROLE, dependent_role); xml_remove_prop(*expanded_xml, PCMK_XA_RSC_ROLE); } any_sets = true; } /* Convert primary's template/tag reference into constraint * PCMK_XE_RESOURCE_SET */ if (!pcmk__tag_to_set(*expanded_xml, &primary_set, PCMK_XA_WITH_RSC, true, scheduler)) { free_xml(*expanded_xml); *expanded_xml = NULL; return pcmk_rc_unpack_error; } if (primary_set != NULL) { if (primary_role != NULL) { /* Move PCMK_XA_WITH_RSC_ROLE into converted PCMK_XE_RESOURCE_SET as * PCMK_XA_ROLE */ crm_xml_add(primary_set, PCMK_XA_ROLE, primary_role); xml_remove_prop(*expanded_xml, PCMK_XA_WITH_RSC_ROLE); } any_sets = true; } if (any_sets) { crm_log_xml_trace(*expanded_xml, "Expanded " PCMK_XE_RSC_COLOCATION); } else { free_xml(*expanded_xml); *expanded_xml = NULL; } return pcmk_rc_ok; } /*! * \internal * \brief Parse a colocation constraint from XML into scheduler data * * \param[in,out] xml_obj Colocation constraint XML to unpack * \param[in,out] scheduler Scheduler data to add constraint to */ void pcmk__unpack_colocation(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { int score_i = 0; xmlNode *set = NULL; xmlNode *last = NULL; xmlNode *orig_xml = NULL; xmlNode *expanded_xml = NULL; const char *id = crm_element_value(xml_obj, PCMK_XA_ID); const char *score = NULL; const char *influence_s = NULL; if (pcmk__str_empty(id)) { pcmk__config_err("Ignoring " PCMK_XE_RSC_COLOCATION " without " CRM_ATTR_ID); return; } if (unpack_colocation_tags(xml_obj, &expanded_xml, scheduler) != pcmk_rc_ok) { return; } if (expanded_xml != NULL) { orig_xml = xml_obj; xml_obj = expanded_xml; } score = crm_element_value(xml_obj, PCMK_XA_SCORE); if (score != NULL) { score_i = char2score(score); } influence_s = crm_element_value(xml_obj, PCMK_XA_INFLUENCE); for (set = first_named_child(xml_obj, PCMK_XE_RESOURCE_SET); set != NULL; set = crm_next_same_xml(set)) { set = expand_idref(set, scheduler->input); if (set == NULL) { // Configuration error, message already logged if (expanded_xml != NULL) { free_xml(expanded_xml); } return; } if (pcmk__str_empty(pcmk__xe_id(set))) { pcmk__config_err("Ignoring " PCMK_XE_RESOURCE_SET " without " CRM_ATTR_ID); continue; } unpack_colocation_set(set, score_i, id, influence_s, scheduler); if (last != NULL) { colocate_rsc_sets(id, last, set, score_i, influence_s, scheduler); } last = set; } if (expanded_xml) { free_xml(expanded_xml); xml_obj = orig_xml; } if (last == NULL) { unpack_simple_colocation(xml_obj, id, influence_s, scheduler); } } /*! * \internal * \brief Make actions of a given type unrunnable for a given resource * * \param[in,out] rsc Resource whose actions should be blocked * \param[in] task Name of action to block * \param[in] reason Unrunnable start action causing the block */ static void mark_action_blocked(pcmk_resource_t *rsc, const char *task, const pcmk_resource_t *reason) { GList *iter = NULL; char *reason_text = crm_strdup_printf("colocation with %s", reason->id); for (iter = rsc->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = iter->data; if (pcmk_is_set(action->flags, pcmk_action_runnable) && pcmk__str_eq(action->task, task, pcmk__str_none)) { pcmk__clear_action_flags(action, pcmk_action_runnable); pe_action_set_reason(action, reason_text, false); pcmk__block_colocation_dependents(action); pcmk__update_action_for_orderings(action, rsc->cluster); } } // If parent resource can't perform an action, neither can any children for (iter = rsc->children; iter != NULL; iter = iter->next) { mark_action_blocked((pcmk_resource_t *) (iter->data), task, reason); } free(reason_text); } /*! * \internal * \brief If an action is unrunnable, block any relevant dependent actions * * If a given action is an unrunnable start or promote, block the start or * promote actions of resources colocated with it, as appropriate to the * colocations' configured roles. * * \param[in,out] action Action to check */ void pcmk__block_colocation_dependents(pcmk_action_t *action) { GList *iter = NULL; GList *colocations = NULL; pcmk_resource_t *rsc = NULL; bool is_start = false; if (pcmk_is_set(action->flags, pcmk_action_runnable)) { return; // Only unrunnable actions block dependents } is_start = pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_none); if (!is_start && !pcmk__str_eq(action->task, PCMK_ACTION_PROMOTE, pcmk__str_none)) { return; // Only unrunnable starts and promotes block dependents } CRM_ASSERT(action->rsc != NULL); // Start and promote are resource actions /* If this resource is part of a collective resource, dependents are blocked * only if all instances of the collective are unrunnable, so check the * collective resource. */ rsc = uber_parent(action->rsc); if (rsc->parent != NULL) { rsc = rsc->parent; // Bundle } // Colocation fails only if entire primary can't reach desired role for (iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = iter->data; pcmk_action_t *child_action = find_first_action(child->actions, NULL, action->task, NULL); if ((child_action == NULL) || pcmk_is_set(child_action->flags, pcmk_action_runnable)) { crm_trace("Not blocking %s colocation dependents because " "at least %s has runnable %s", rsc->id, child->id, action->task); return; // At least one child can reach desired role } } crm_trace("Blocking %s colocation dependents due to unrunnable %s %s", rsc->id, action->rsc->id, action->task); // Check each colocation where this resource is primary colocations = pcmk__with_this_colocations(rsc); for (iter = colocations; iter != NULL; iter = iter->next) { pcmk__colocation_t *colocation = iter->data; - if (colocation->score < INFINITY) { + if (colocation->score < PCMK_SCORE_INFINITY) { continue; // Only mandatory colocations block dependent } /* If the primary can't start, the dependent can't reach its colocated * role, regardless of what the primary or dependent colocation role is. * * If the primary can't be promoted, the dependent can't reach its * colocated role if the primary's colocation role is promoted. */ if (!is_start && (colocation->primary_role != pcmk_role_promoted)) { continue; } // Block the dependent from reaching its colocated role if (colocation->dependent_role == pcmk_role_promoted) { mark_action_blocked(colocation->dependent, PCMK_ACTION_PROMOTE, action->rsc); } else { mark_action_blocked(colocation->dependent, PCMK_ACTION_START, action->rsc); } } g_list_free(colocations); } /*! * \internal * \brief Get the resource to use for role comparisons * * A bundle replica includes a container and possibly an instance of the bundled * resource. The dependent in a "with bundle" colocation is colocated with a * particular bundle container. However, if the colocation includes a role, then * the role must be checked on the bundled resource instance inside the * container. The container itself will never be promoted; the bundled resource * may be. * * If the given resource is a bundle replica container, return the resource * inside it, if any. Otherwise, return the resource itself. * * \param[in] rsc Resource to check * * \return Resource to use for role comparisons */ static const pcmk_resource_t * get_resource_for_role(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk_rsc_replica_container)) { const pcmk_resource_t *child = pe__get_rsc_in_container(rsc); if (child != NULL) { return child; } } return rsc; } /*! * \internal * \brief Determine how a colocation constraint should affect a resource * * Colocation constraints have different effects at different points in the * scheduler sequence. Initially, they affect a resource's location; once that * is determined, then for promotable clones they can affect a resource * instance's role; after both are determined, the constraints no longer matter. * Given a specific colocation constraint, check what has been done so far to * determine what should be affected at the current point in the scheduler. * * \param[in] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint * \param[in] preview If true, pretend resources have already been assigned * * \return How colocation constraint should be applied at this point */ enum pcmk__coloc_affects pcmk__colocation_affects(const pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool preview) { const pcmk_resource_t *dependent_role_rsc = NULL; const pcmk_resource_t *primary_role_rsc = NULL; CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if (!preview && pcmk_is_set(primary->flags, pcmk_rsc_unassigned)) { // Primary resource has not been assigned yet, so we can't do anything return pcmk__coloc_affects_nothing; } dependent_role_rsc = get_resource_for_role(dependent); primary_role_rsc = get_resource_for_role(primary); if ((colocation->dependent_role >= pcmk_role_unpromoted) && (dependent_role_rsc->parent != NULL) && pcmk_is_set(dependent_role_rsc->parent->flags, pcmk_rsc_promotable) && !pcmk_is_set(dependent_role_rsc->flags, pcmk_rsc_unassigned)) { /* This is a colocation by role, and the dependent is a promotable clone * that has already been assigned, so the colocation should now affect * the role. */ return pcmk__coloc_affects_role; } if (!preview && !pcmk_is_set(dependent->flags, pcmk_rsc_unassigned)) { /* The dependent resource has already been through assignment, so the * constraint no longer has any effect. Log an error if a mandatory * colocation constraint has been violated. */ const pcmk_node_t *primary_node = primary->allocated_to; if (dependent->allocated_to == NULL) { crm_trace("Skipping colocation '%s': %s will not run anywhere", colocation->id, dependent->id); - } else if (colocation->score >= INFINITY) { + } else if (colocation->score >= PCMK_SCORE_INFINITY) { // Dependent resource must colocate with primary resource if (!pcmk__same_node(primary_node, dependent->allocated_to)) { pcmk__sched_err("%s must be colocated with %s but is not " "(%s vs. %s)", dependent->id, primary->id, pcmk__node_name(dependent->allocated_to), pcmk__node_name(primary_node)); } - } else if (colocation->score <= -CRM_SCORE_INFINITY) { + } else if (colocation->score <= -PCMK_SCORE_INFINITY) { // Dependent resource must anti-colocate with primary resource if (pcmk__same_node(dependent->allocated_to, primary_node)) { pcmk__sched_err("%s and %s must be anti-colocated but are " "assigned to the same node (%s)", dependent->id, primary->id, pcmk__node_name(primary_node)); } } return pcmk__coloc_affects_nothing; } if ((colocation->dependent_role != pcmk_role_unknown) && (colocation->dependent_role != dependent_role_rsc->next_role)) { crm_trace("Skipping %scolocation '%s': dependent limited to %s role " "but %s next role is %s", ((colocation->score < 0)? "anti-" : ""), colocation->id, pcmk_role_text(colocation->dependent_role), dependent_role_rsc->id, pcmk_role_text(dependent_role_rsc->next_role)); return pcmk__coloc_affects_nothing; } if ((colocation->primary_role != pcmk_role_unknown) && (colocation->primary_role != primary_role_rsc->next_role)) { crm_trace("Skipping %scolocation '%s': primary limited to %s role " "but %s next role is %s", ((colocation->score < 0)? "anti-" : ""), colocation->id, pcmk_role_text(colocation->primary_role), primary_role_rsc->id, pcmk_role_text(primary_role_rsc->next_role)); return pcmk__coloc_affects_nothing; } return pcmk__coloc_affects_location; } /*! * \internal * \brief Apply colocation to dependent for assignment purposes * * Update the allowed node scores of the dependent resource in a colocation, * for the purposes of assigning it to a node. * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint */ void pcmk__apply_coloc_to_scores(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation) { const char *attr = colocation->node_attribute; const char *value = NULL; GHashTable *work = NULL; GHashTableIter iter; pcmk_node_t *node = NULL; if (primary->allocated_to != NULL) { value = pcmk__colocation_node_attr(primary->allocated_to, attr, primary); } else if (colocation->score < 0) { // Nothing to do (anti-colocation with something that is not running) return; } work = pcmk__copy_node_table(dependent->allowed_nodes); g_hash_table_iter_init(&iter, work); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if (primary->allocated_to == NULL) { node->weight = pcmk__add_scores(-colocation->score, node->weight); pcmk__rsc_trace(dependent, "Applied %s to %s score on %s (now %s after " "subtracting %s because primary %s inactive)", colocation->id, dependent->id, pcmk__node_name(node), pcmk_readable_score(node->weight), pcmk_readable_score(colocation->score), primary->id); continue; } if (pcmk__str_eq(pcmk__colocation_node_attr(node, attr, dependent), value, pcmk__str_casei)) { /* Add colocation score only if optional (or minus infinity). A * mandatory colocation is a requirement rather than a preference, * so we don't need to consider it for relative assignment purposes. * The resource will simply be forbidden from running on the node if * the primary isn't active there (via the condition above). */ - if (colocation->score < CRM_SCORE_INFINITY) { + if (colocation->score < PCMK_SCORE_INFINITY) { node->weight = pcmk__add_scores(colocation->score, node->weight); pcmk__rsc_trace(dependent, "Applied %s to %s score on %s (now %s after " "adding %s)", colocation->id, dependent->id, pcmk__node_name(node), pcmk_readable_score(node->weight), pcmk_readable_score(colocation->score)); } continue; } - if (colocation->score >= CRM_SCORE_INFINITY) { + if (colocation->score >= PCMK_SCORE_INFINITY) { /* Only mandatory colocations are relevant when the colocation * attribute doesn't match, because an attribute not matching is not * a negative preference -- the colocation is simply relevant only * where it matches. */ - node->weight = -CRM_SCORE_INFINITY; + node->weight = -PCMK_SCORE_INFINITY; pcmk__rsc_trace(dependent, "Banned %s from %s because colocation %s attribute %s " "does not match", dependent->id, pcmk__node_name(node), colocation->id, attr); } } - if ((colocation->score <= -INFINITY) || (colocation->score >= INFINITY) + if ((colocation->score <= -PCMK_SCORE_INFINITY) + || (colocation->score >= PCMK_SCORE_INFINITY) || pcmk__any_node_available(work)) { g_hash_table_destroy(dependent->allowed_nodes); dependent->allowed_nodes = work; work = NULL; } else { pcmk__rsc_info(dependent, "%s: Rolling back scores from %s (no available nodes)", dependent->id, primary->id); } if (work != NULL) { g_hash_table_destroy(work); } } /*! * \internal * \brief Apply colocation to dependent for role purposes * * Update the priority of the dependent resource in a colocation, for the * purposes of selecting its role * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint */ void pcmk__apply_coloc_to_priority(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation) { const char *dependent_value = NULL; const char *primary_value = NULL; const char *attr = colocation->node_attribute; int score_multiplier = 1; const pcmk_resource_t *primary_role_rsc = NULL; CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if ((primary->allocated_to == NULL) || (dependent->allocated_to == NULL)) { return; } dependent_value = pcmk__colocation_node_attr(dependent->allocated_to, attr, dependent); primary_value = pcmk__colocation_node_attr(primary->allocated_to, attr, primary); primary_role_rsc = get_resource_for_role(primary); if (!pcmk__str_eq(dependent_value, primary_value, pcmk__str_casei)) { - if ((colocation->score == INFINITY) + if ((colocation->score == PCMK_SCORE_INFINITY) && (colocation->dependent_role == pcmk_role_promoted)) { - dependent->priority = -INFINITY; + dependent->priority = -PCMK_SCORE_INFINITY; } return; } if ((colocation->primary_role != pcmk_role_unknown) && (colocation->primary_role != primary_role_rsc->next_role)) { return; } if (colocation->dependent_role == pcmk_role_unpromoted) { score_multiplier = -1; } dependent->priority = pcmk__add_scores(score_multiplier * colocation->score, dependent->priority); pcmk__rsc_trace(dependent, "Applied %s to %s promotion priority (now %s after %s %s)", colocation->id, dependent->id, pcmk_readable_score(dependent->priority), ((score_multiplier == 1)? "adding" : "subtracting"), pcmk_readable_score(colocation->score)); } /*! * \internal * \brief Find score of highest-scored node that matches colocation attribute * * \param[in] rsc Resource whose allowed nodes should be searched * \param[in] attr Colocation attribute name (must not be NULL) * \param[in] value Colocation attribute value to require */ static int best_node_score_matching_attr(const pcmk_resource_t *rsc, const char *attr, const char *value) { GHashTableIter iter; pcmk_node_t *node = NULL; - int best_score = -INFINITY; + int best_score = -PCMK_SCORE_INFINITY; const char *best_node = NULL; // Find best allowed node with matching attribute g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if ((node->weight > best_score) && pcmk__node_available(node, false, false) && pcmk__str_eq(value, pcmk__colocation_node_attr(node, attr, rsc), pcmk__str_casei)) { best_score = node->weight; best_node = node->details->uname; } } if (!pcmk__str_eq(attr, CRM_ATTR_UNAME, pcmk__str_none)) { if (best_node == NULL) { crm_info("No allowed node for %s matches node attribute %s=%s", rsc->id, attr, value); } else { crm_info("Allowed node %s for %s had best score (%d) " "of those matching node attribute %s=%s", best_node, rsc->id, best_score, attr, value); } } return best_score; } /*! * \internal * \brief Check whether a resource is allowed only on a single node * * \param[in] rsc Resource to check * * \return \c true if \p rsc is allowed only on one node, otherwise \c false */ static bool allowed_on_one(const pcmk_resource_t *rsc) { GHashTableIter iter; pcmk_node_t *allowed_node = NULL; int allowed_nodes = 0; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &allowed_node)) { if ((allowed_node->weight >= 0) && (++allowed_nodes > 1)) { pcmk__rsc_trace(rsc, "%s is allowed on multiple nodes", rsc->id); return false; } } pcmk__rsc_trace(rsc, "%s is allowed %s", rsc->id, ((allowed_nodes == 1)? "on a single node" : "nowhere")); return (allowed_nodes == 1); } /*! * \internal * \brief Add resource's colocation matches to current node assignment scores * * For each node in a given table, if any of a given resource's allowed nodes * have a matching value for the colocation attribute, add the highest of those * nodes' scores to the node's score. * * \param[in,out] nodes Table of nodes with assignment scores so far * \param[in] source_rsc Resource whose node scores to add * \param[in] target_rsc Resource on whose behalf to update \p nodes * \param[in] colocation Original colocation constraint (used to get * configured primary resource's stickiness, and * to get colocation node attribute; pass NULL to * ignore stickiness and use default attribute) * \param[in] factor Factor by which to multiply scores being added * \param[in] only_positive Whether to add only positive scores */ static void add_node_scores_matching_attr(GHashTable *nodes, const pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const pcmk__colocation_t *colocation, float factor, bool only_positive) { GHashTableIter iter; pcmk_node_t *node = NULL; const char *attr = colocation->node_attribute; // Iterate through each node g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { float delta_f = 0; int delta = 0; int score = 0; int new_score = 0; const char *value = pcmk__colocation_node_attr(node, attr, target_rsc); score = best_node_score_matching_attr(source_rsc, attr, value); if ((factor < 0) && (score < 0)) { /* If the dependent is anti-colocated, we generally don't want the * primary to prefer nodes that the dependent avoids. That could * lead to unnecessary shuffling of the primary when the dependent * hits its migration threshold somewhere, for example. * * However, there are cases when it is desirable. If the dependent * can't run anywhere but where the primary is, it would be * worthwhile to move the primary for the sake of keeping the * dependent active. * * We can't know that exactly at this point since we don't know * where the primary will be assigned, but we can limit considering * the preference to when the dependent is allowed only on one node. * This is less than ideal for multiple reasons: * * - the dependent could be allowed on more than one node but have * anti-colocation primaries on each; * - the dependent could be a clone or bundle with multiple * instances, and the dependent as a whole is allowed on multiple * nodes but some instance still can't run * - the dependent has considered node-specific criteria such as * location constraints and stickiness by this point, but might * have other factors that end up disallowing a node * * but the alternative is making the primary move when it doesn't * need to. * * We also consider the primary's stickiness and influence, so the * user has some say in the matter. (This is the configured primary, * not a particular instance of the primary, but that doesn't matter * unless stickiness uses a rule to vary by node, and that seems * acceptable to ignore.) */ if ((colocation->primary->stickiness >= -score) || !pcmk__colocation_has_influence(colocation, NULL) || !allowed_on_one(colocation->dependent)) { crm_trace("%s: Filtering %d + %f * %d " "(double negative disallowed)", pcmk__node_name(node), node->weight, factor, score); continue; } } if (node->weight == INFINITY_HACK) { crm_trace("%s: Filtering %d + %f * %d (node was marked unusable)", pcmk__node_name(node), node->weight, factor, score); continue; } delta_f = factor * score; // Round the number; see http://c-faq.com/fp/round.html delta = (int) ((delta_f < 0)? (delta_f - 0.5) : (delta_f + 0.5)); /* Small factors can obliterate the small scores that are often actually * used in configurations. If the score and factor are nonzero, ensure * that the result is nonzero as well. */ if ((delta == 0) && (score != 0)) { if (factor > 0.0) { delta = 1; } else if (factor < 0.0) { delta = -1; } } new_score = pcmk__add_scores(delta, node->weight); if (only_positive && (new_score < 0) && (node->weight > 0)) { crm_trace("%s: Filtering %d + %f * %d = %d " "(negative disallowed, marking node unusable)", pcmk__node_name(node), node->weight, factor, score, new_score); node->weight = INFINITY_HACK; continue; } if (only_positive && (new_score < 0) && (node->weight == 0)) { crm_trace("%s: Filtering %d + %f * %d = %d (negative disallowed)", pcmk__node_name(node), node->weight, factor, score, new_score); continue; } crm_trace("%s: %d + %f * %d = %d", pcmk__node_name(node), node->weight, factor, score, new_score); node->weight = new_score; } } /*! * \internal * \brief Update nodes with scores of colocated resources' nodes * * Given a table of nodes and a resource, update the nodes' scores with the * scores of the best nodes matching the attribute used for each of the * resource's relevant colocations. * * \param[in,out] source_rsc Resource whose node scores to add * \param[in] target_rsc Resource on whose behalf to update \p *nodes * \param[in] log_id Resource ID for logs (if \c NULL, use * \p source_rsc ID) * \param[in,out] nodes Nodes to update (set initial contents to \c NULL * to copy allowed nodes from \p source_rsc) * \param[in] colocation Original colocation constraint (used to get * configured primary resource's stickiness, and * to get colocation node attribute; if \c NULL, * source_rsc's own matching node scores * will not be added, and \p *nodes must be \c NULL * as well) * \param[in] factor Incorporate scores multiplied by this factor * \param[in] flags Bitmask of enum pcmk__coloc_select values * * \note \c NULL \p target_rsc, \c NULL \p *nodes, \c NULL \p colocation, and * the \c pcmk__coloc_select_this_with flag are used together (and only by * \c cmp_resources()). * \note The caller remains responsible for freeing \p *nodes. * \note This is the shared implementation of * \c pcmk_assignment_methods_t:add_colocated_node_scores(). */ void pcmk__add_colocated_node_scores(pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags) { GHashTable *work = NULL; CRM_ASSERT((source_rsc != NULL) && (nodes != NULL) && ((colocation != NULL) || ((target_rsc == NULL) && (*nodes == NULL)))); if (log_id == NULL) { log_id = source_rsc->id; } // Avoid infinite recursion if (pcmk_is_set(source_rsc->flags, pcmk_rsc_updating_nodes)) { pcmk__rsc_info(source_rsc, "%s: Breaking dependency loop at %s", log_id, source_rsc->id); return; } pcmk__set_rsc_flags(source_rsc, pcmk_rsc_updating_nodes); if (*nodes == NULL) { work = pcmk__copy_node_table(source_rsc->allowed_nodes); target_rsc = source_rsc; } else { const bool pos = pcmk_is_set(flags, pcmk__coloc_select_nonnegative); pcmk__rsc_trace(source_rsc, "%s: Merging %s scores from %s (at %.6f)", log_id, (pos? "positive" : "all"), source_rsc->id, factor); work = pcmk__copy_node_table(*nodes); add_node_scores_matching_attr(work, source_rsc, target_rsc, colocation, factor, pos); } if (work == NULL) { pcmk__clear_rsc_flags(source_rsc, pcmk_rsc_updating_nodes); return; } if (pcmk__any_node_available(work)) { GList *colocations = NULL; if (pcmk_is_set(flags, pcmk__coloc_select_this_with)) { colocations = pcmk__this_with_colocations(source_rsc); pcmk__rsc_trace(source_rsc, "Checking additional %d optional '%s with' " "constraints", g_list_length(colocations), source_rsc->id); } else { colocations = pcmk__with_this_colocations(source_rsc); pcmk__rsc_trace(source_rsc, "Checking additional %d optional 'with %s' " "constraints", g_list_length(colocations), source_rsc->id); } flags |= pcmk__coloc_select_active; for (GList *iter = colocations; iter != NULL; iter = iter->next) { pcmk__colocation_t *constraint = iter->data; pcmk_resource_t *other = NULL; - float other_factor = factor * constraint->score / (float) INFINITY; + float other_factor = factor * constraint->score + / (float) PCMK_SCORE_INFINITY; if (pcmk_is_set(flags, pcmk__coloc_select_this_with)) { other = constraint->primary; } else if (!pcmk__colocation_has_influence(constraint, NULL)) { continue; } else { other = constraint->dependent; } pcmk__rsc_trace(source_rsc, "Optionally merging score of '%s' constraint " "(%s with %s)", constraint->id, constraint->dependent->id, constraint->primary->id); other->cmds->add_colocated_node_scores(other, target_rsc, log_id, &work, constraint, other_factor, flags); pe__show_node_scores(true, NULL, log_id, work, source_rsc->cluster); } g_list_free(colocations); } else if (pcmk_is_set(flags, pcmk__coloc_select_active)) { pcmk__rsc_info(source_rsc, "%s: Rolling back optional scores from %s", log_id, source_rsc->id); g_hash_table_destroy(work); pcmk__clear_rsc_flags(source_rsc, pcmk_rsc_updating_nodes); return; } if (pcmk_is_set(flags, pcmk__coloc_select_nonnegative)) { pcmk_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, work); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if (node->weight == INFINITY_HACK) { node->weight = 1; } } } if (*nodes != NULL) { g_hash_table_destroy(*nodes); } *nodes = work; pcmk__clear_rsc_flags(source_rsc, pcmk_rsc_updating_nodes); } /*! * \internal * \brief Apply a "with this" colocation to a resource's allowed node scores * * \param[in,out] data Colocation to apply * \param[in,out] user_data Resource being assigned */ void pcmk__add_dependent_scores(gpointer data, gpointer user_data) { pcmk__colocation_t *colocation = data; pcmk_resource_t *target_rsc = user_data; pcmk_resource_t *source_rsc = colocation->dependent; - const float factor = colocation->score / (float) INFINITY; + const float factor = colocation->score / (float) PCMK_SCORE_INFINITY; uint32_t flags = pcmk__coloc_select_active; if (!pcmk__colocation_has_influence(colocation, NULL)) { return; } if (target_rsc->variant == pcmk_rsc_variant_clone) { flags |= pcmk__coloc_select_nonnegative; } pcmk__rsc_trace(target_rsc, "%s: Incorporating attenuated %s assignment scores due " "to colocation %s", target_rsc->id, source_rsc->id, colocation->id); source_rsc->cmds->add_colocated_node_scores(source_rsc, target_rsc, source_rsc->id, &target_rsc->allowed_nodes, colocation, factor, flags); } /*! * \internal * \brief Exclude nodes from a dependent's node table if not in a given list * * Given a dependent resource in a colocation and a list of nodes where the * primary resource will run, set a node's score to \c -INFINITY in the * dependent's node table if not found in the primary nodes list. * * \param[in,out] dependent Dependent resource * \param[in] primary Primary resource (for logging only) * \param[in] colocation Colocation constraint (for logging only) * \param[in] primary_nodes List of nodes where the primary will have * unblocked instances in a suitable role * \param[in] merge_scores If \c true and a node is found in both \p table * and \p list, add the node's score in \p list to * the node's score in \p table */ void pcmk__colocation_intersect_nodes(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, const GList *primary_nodes, bool merge_scores) { GHashTableIter iter; pcmk_node_t *dependent_node = NULL; CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); g_hash_table_iter_init(&iter, dependent->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &dependent_node)) { const pcmk_node_t *primary_node = NULL; primary_node = pe_find_node_id(primary_nodes, dependent_node->details->id); if (primary_node == NULL) { - dependent_node->weight = -INFINITY; + dependent_node->weight = -PCMK_SCORE_INFINITY; pcmk__rsc_trace(dependent, "Banning %s from %s (no primary instance) for %s", dependent->id, pcmk__node_name(dependent_node), colocation->id); } else if (merge_scores) { dependent_node->weight = pcmk__add_scores(dependent_node->weight, primary_node->weight); pcmk__rsc_trace(dependent, "Added %s's score %s to %s's score for %s (now %s) " "for colocation %s", primary->id, pcmk_readable_score(primary_node->weight), dependent->id, pcmk__node_name(dependent_node), pcmk_readable_score(dependent_node->weight), colocation->id); } } } /*! * \internal * \brief Get all colocations affecting a resource as the primary * * \param[in] rsc Resource to get colocations for * * \return Newly allocated list of colocations affecting \p rsc as primary * * \note This is a convenience wrapper for the with_this_colocations() method. */ GList * pcmk__with_this_colocations(const pcmk_resource_t *rsc) { GList *list = NULL; rsc->cmds->with_this_colocations(rsc, rsc, &list); return list; } /*! * \internal * \brief Get all colocations affecting a resource as the dependent * * \param[in] rsc Resource to get colocations for * * \return Newly allocated list of colocations affecting \p rsc as dependent * * \note This is a convenience wrapper for the this_with_colocations() method. */ GList * pcmk__this_with_colocations(const pcmk_resource_t *rsc) { GList *list = NULL; rsc->cmds->this_with_colocations(rsc, rsc, &list); return list; } diff --git a/lib/pacemaker/pcmk_sched_group.c b/lib/pacemaker/pcmk_sched_group.c index 64fa361060..e2aede26e6 100644 --- a/lib/pacemaker/pcmk_sched_group.c +++ b/lib/pacemaker/pcmk_sched_group.c @@ -1,951 +1,952 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Assign a group resource to a node * * \param[in,out] rsc Group resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * \param[in] stop_if_fail If \c true and a child of \p rsc can't be * assigned to a node, set the child's next role to * stopped and update existing actions * * \return Node that \p rsc is assigned to, if assigned entirely to one node * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ pcmk_node_t * pcmk__group_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { pcmk_node_t *first_assigned_node = NULL; pcmk_resource_t *first_member = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)); if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { return rsc->allocated_to; // Assignment already done } if (pcmk_is_set(rsc->flags, pcmk_rsc_assigning)) { pcmk__rsc_debug(rsc, "Assignment dependency loop detected involving %s", rsc->id); return NULL; } if (rsc->children == NULL) { // No members to assign pcmk__clear_rsc_flags(rsc, pcmk_rsc_unassigned); return NULL; } pcmk__set_rsc_flags(rsc, pcmk_rsc_assigning); first_member = (pcmk_resource_t *) rsc->children->data; rsc->role = first_member->role; pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; pcmk_node_t *node = NULL; pcmk__rsc_trace(rsc, "Assigning group %s member %s", rsc->id, member->id); node = member->cmds->assign(member, prefer, stop_if_fail); if (first_assigned_node == NULL) { first_assigned_node = node; } } pe__set_next_role(rsc, first_member->next_role, "first group member"); pcmk__clear_rsc_flags(rsc, pcmk_rsc_assigning|pcmk_rsc_unassigned); if (!pe__group_flag_is_set(rsc, pcmk__group_colocated)) { return NULL; } return first_assigned_node; } /*! * \internal * \brief Create a pseudo-operation for a group as an ordering point * * \param[in,out] group Group resource to create action for * \param[in] action Action name * * \return Newly created pseudo-operation */ static pcmk_action_t * create_group_pseudo_op(pcmk_resource_t *group, const char *action) { pcmk_action_t *op = custom_action(group, pcmk__op_key(group->id, action, 0), action, NULL, TRUE, group->cluster); pcmk__set_action_flags(op, pcmk_action_pseudo|pcmk_action_runnable); return op; } /*! * \internal * \brief Create all actions needed for a given group resource * * \param[in,out] rsc Group resource to create actions for */ void pcmk__group_create_actions(pcmk_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)); pcmk__rsc_trace(rsc, "Creating actions for group %s", rsc->id); // Create actions for individual group members for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; member->cmds->create_actions(member); } // Create pseudo-actions for group itself to serve as ordering points create_group_pseudo_op(rsc, PCMK_ACTION_START); create_group_pseudo_op(rsc, PCMK_ACTION_RUNNING); create_group_pseudo_op(rsc, PCMK_ACTION_STOP); create_group_pseudo_op(rsc, PCMK_ACTION_STOPPED); if (crm_is_true(g_hash_table_lookup(rsc->meta, PCMK_META_PROMOTABLE))) { create_group_pseudo_op(rsc, PCMK_ACTION_DEMOTE); create_group_pseudo_op(rsc, PCMK_ACTION_DEMOTED); create_group_pseudo_op(rsc, PCMK_ACTION_PROMOTE); create_group_pseudo_op(rsc, PCMK_ACTION_PROMOTED); } } // User data for member_internal_constraints() struct member_data { // These could be derived from member but this avoids some function calls bool ordered; bool colocated; bool promotable; pcmk_resource_t *last_active; pcmk_resource_t *previous_member; }; /*! * \internal * \brief Create implicit constraints needed for a group member * * \param[in,out] data Group member to create implicit constraints for * \param[in,out] user_data Member data (struct member_data *) */ static void member_internal_constraints(gpointer data, gpointer user_data) { pcmk_resource_t *member = (pcmk_resource_t *) data; struct member_data *member_data = (struct member_data *) user_data; // For ordering demote vs demote or stop vs stop uint32_t down_flags = pcmk__ar_then_implies_first_graphed; // For ordering demote vs demoted or stop vs stopped uint32_t post_down_flags = pcmk__ar_first_implies_then_graphed; // Create the individual member's implicit constraints member->cmds->internal_constraints(member); if (member_data->previous_member == NULL) { // This is first member if (member_data->ordered) { pcmk__set_relation_flags(down_flags, pcmk__ar_ordered); post_down_flags = pcmk__ar_first_implies_then; } } else if (member_data->colocated) { uint32_t flags = pcmk__coloc_none; if (pcmk_is_set(member->flags, pcmk_rsc_critical)) { flags |= pcmk__coloc_influence; } // Colocate this member with the previous one - pcmk__new_colocation("#group-members", NULL, INFINITY, member, - member_data->previous_member, NULL, NULL, flags); + pcmk__new_colocation("#group-members", NULL, PCMK_SCORE_INFINITY, + member, member_data->previous_member, NULL, NULL, + flags); } if (member_data->promotable) { // Demote group -> demote member -> group is demoted pcmk__order_resource_actions(member->parent, PCMK_ACTION_DEMOTE, member, PCMK_ACTION_DEMOTE, down_flags); pcmk__order_resource_actions(member, PCMK_ACTION_DEMOTE, member->parent, PCMK_ACTION_DEMOTED, post_down_flags); // Promote group -> promote member -> group is promoted pcmk__order_resource_actions(member, PCMK_ACTION_PROMOTE, member->parent, PCMK_ACTION_PROMOTED, pcmk__ar_unrunnable_first_blocks |pcmk__ar_first_implies_then |pcmk__ar_first_implies_then_graphed); pcmk__order_resource_actions(member->parent, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pcmk__ar_then_implies_first_graphed); } // Stop group -> stop member -> group is stopped pcmk__order_stops(member->parent, member, down_flags); pcmk__order_resource_actions(member, PCMK_ACTION_STOP, member->parent, PCMK_ACTION_STOPPED, post_down_flags); // Start group -> start member -> group is started pcmk__order_starts(member->parent, member, pcmk__ar_then_implies_first_graphed); pcmk__order_resource_actions(member, PCMK_ACTION_START, member->parent, PCMK_ACTION_RUNNING, pcmk__ar_unrunnable_first_blocks |pcmk__ar_first_implies_then |pcmk__ar_first_implies_then_graphed); if (!member_data->ordered) { pcmk__order_starts(member->parent, member, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks |pcmk__ar_then_implies_first_graphed); if (member_data->promotable) { pcmk__order_resource_actions(member->parent, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks |pcmk__ar_then_implies_first_graphed); } } else if (member_data->previous_member == NULL) { pcmk__order_starts(member->parent, member, pcmk__ar_none); if (member_data->promotable) { pcmk__order_resource_actions(member->parent, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pcmk__ar_none); } } else { // Order this member relative to the previous one pcmk__order_starts(member_data->previous_member, member, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks); pcmk__order_stops(member, member_data->previous_member, pcmk__ar_ordered|pcmk__ar_intermediate_stop); /* In unusual circumstances (such as adding a new member to the middle * of a group with unmanaged later members), this member may be active * while the previous (new) member is inactive. In this situation, the * usual restart orderings will be irrelevant, so we need to order this * member's stop before the previous member's start. */ if ((member->running_on != NULL) && (member_data->previous_member->running_on == NULL)) { pcmk__order_resource_actions(member, PCMK_ACTION_STOP, member_data->previous_member, PCMK_ACTION_START, pcmk__ar_then_implies_first |pcmk__ar_unrunnable_first_blocks); } if (member_data->promotable) { pcmk__order_resource_actions(member_data->previous_member, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks); pcmk__order_resource_actions(member, PCMK_ACTION_DEMOTE, member_data->previous_member, PCMK_ACTION_DEMOTE, pcmk__ar_ordered); } } // Make sure partially active groups shut down in sequence if (member->running_on != NULL) { if (member_data->ordered && (member_data->previous_member != NULL) && (member_data->previous_member->running_on == NULL) && (member_data->last_active != NULL) && (member_data->last_active->running_on != NULL)) { pcmk__order_stops(member, member_data->last_active, pcmk__ar_ordered); } member_data->last_active = member; } member_data->previous_member = member; } /*! * \internal * \brief Create implicit constraints needed for a group resource * * \param[in,out] rsc Group resource to create implicit constraints for */ void pcmk__group_internal_constraints(pcmk_resource_t *rsc) { struct member_data member_data = { false, }; const pcmk_resource_t *top = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)); /* Order group pseudo-actions relative to each other for restarting: * stop group -> group is stopped -> start group -> group is started */ pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, rsc, PCMK_ACTION_STOPPED, pcmk__ar_unrunnable_first_blocks); pcmk__order_resource_actions(rsc, PCMK_ACTION_STOPPED, rsc, PCMK_ACTION_START, pcmk__ar_ordered); pcmk__order_resource_actions(rsc, PCMK_ACTION_START, rsc, PCMK_ACTION_RUNNING, pcmk__ar_unrunnable_first_blocks); top = pe__const_top_resource(rsc, false); member_data.ordered = pe__group_flag_is_set(rsc, pcmk__group_ordered); member_data.colocated = pe__group_flag_is_set(rsc, pcmk__group_colocated); member_data.promotable = pcmk_is_set(top->flags, pcmk_rsc_promotable); g_list_foreach(rsc->children, member_internal_constraints, &member_data); } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint for a group with some other resource, apply the * score to the dependent's allowed node scores (if we are still placing * resources) or priority (if we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent group resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply */ static void colocate_group_with(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation) { pcmk_resource_t *member = NULL; if (dependent->children == NULL) { return; } pcmk__rsc_trace(primary, "Processing %s (group %s with %s) for dependent", colocation->id, dependent->id, primary->id); if (pe__group_flag_is_set(dependent, pcmk__group_colocated)) { // Colocate first member (internal colocations will handle the rest) member = (pcmk_resource_t *) dependent->children->data; member->cmds->apply_coloc_score(member, primary, colocation, true); return; } - if (colocation->score >= INFINITY) { + if (colocation->score >= PCMK_SCORE_INFINITY) { pcmk__config_err("%s: Cannot perform mandatory colocation between " "non-colocated group and %s", dependent->id, primary->id); return; } // Colocate each member individually for (GList *iter = dependent->children; iter != NULL; iter = iter->next) { member = (pcmk_resource_t *) iter->data; member->cmds->apply_coloc_score(member, primary, colocation, true); } } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint for some other resource with a group, apply the * score to the dependent's allowed node scores (if we are still placing * resources) or priority (if we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary group resource in colocation * \param[in] colocation Colocation constraint to apply */ static void colocate_with_group(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation) { const pcmk_resource_t *member = NULL; pcmk__rsc_trace(primary, "Processing colocation %s (%s with group %s) for primary", colocation->id, dependent->id, primary->id); if (pcmk_is_set(primary->flags, pcmk_rsc_unassigned)) { return; } if (pe__group_flag_is_set(primary, pcmk__group_colocated)) { - if (colocation->score >= INFINITY) { + if (colocation->score >= PCMK_SCORE_INFINITY) { /* For mandatory colocations, the entire group must be assignable * (and in the specified role if any), so apply the colocation based * on the last member. */ member = pe__last_group_member(primary); } else if (primary->children != NULL) { /* For optional colocations, whether the group is partially or fully * up doesn't matter, so apply the colocation based on the first * member. */ member = (pcmk_resource_t *) primary->children->data; } if (member == NULL) { return; // Nothing to colocate with } member->cmds->apply_coloc_score(dependent, member, colocation, false); return; } - if (colocation->score >= INFINITY) { + if (colocation->score >= PCMK_SCORE_INFINITY) { pcmk__config_err("%s: Cannot perform mandatory colocation with" " non-colocated group %s", dependent->id, primary->id); return; } // Colocate dependent with each member individually for (const GList *iter = primary->children; iter != NULL; iter = iter->next) { member = iter->data; member->cmds->apply_coloc_score(dependent, member, colocation, false); } } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__group_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if (for_dependent) { colocate_group_with(dependent, primary, colocation); } else { // Method should only be called for primitive dependents CRM_ASSERT(dependent->variant == pcmk_rsc_variant_primitive); colocate_with_group(dependent, primary, colocation); } } /*! * \internal * \brief Return action flags for a given group resource action * * \param[in,out] action Group action to get flags for * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__group_action_flags(pcmk_action_t *action, const pcmk_node_t *node) { // Default flags for a group action uint32_t flags = pcmk_action_optional |pcmk_action_runnable |pcmk_action_pseudo; CRM_ASSERT(action != NULL); // Update flags considering each member's own flags for same action for (GList *iter = action->rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; // Check whether member has the same action enum action_tasks task = get_complex_task(member, action->task); const char *task_s = pcmk_action_text(task); pcmk_action_t *member_action = find_first_action(member->actions, NULL, task_s, node); if (member_action != NULL) { uint32_t member_flags = member->cmds->action_flags(member_action, node); // Group action is mandatory if any member action is if (pcmk_is_set(flags, pcmk_action_optional) && !pcmk_is_set(member_flags, pcmk_action_optional)) { pcmk__rsc_trace(action->rsc, "%s is mandatory because %s is", action->uuid, member_action->uuid); pcmk__clear_raw_action_flags(flags, "group action", pcmk_action_optional); pcmk__clear_action_flags(action, pcmk_action_optional); } // Group action is unrunnable if any member action is if (!pcmk__str_eq(task_s, action->task, pcmk__str_none) && pcmk_is_set(flags, pcmk_action_runnable) && !pcmk_is_set(member_flags, pcmk_action_runnable)) { pcmk__rsc_trace(action->rsc, "%s is unrunnable because %s is", action->uuid, member_action->uuid); pcmk__clear_raw_action_flags(flags, "group action", pcmk_action_runnable); pcmk__clear_action_flags(action, pcmk_action_runnable); } /* Group (pseudo-)actions other than stop or demote are unrunnable * unless every member will do it. */ } else if ((task != pcmk_action_stop) && (task != pcmk_action_demote)) { pcmk__rsc_trace(action->rsc, "%s is not runnable because %s will not %s", action->uuid, member->id, task_s); pcmk__clear_raw_action_flags(flags, "group action", pcmk_action_runnable); } } return flags; } /*! * \internal * \brief Update two actions according to an ordering between them * * Given information about an ordering of two actions, update the actions' flags * (and runnable_before members if appropriate) as appropriate for the ordering. * Effects may cascade to other orderings involving the actions as well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this node * (only used when interleaving instances) * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk_action_optional to affect only * mandatory actions, and pcmk_action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t pcmk__group_update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler) { uint32_t changed = pcmk__updated_none; // Group method can be called only on behalf of "then" action CRM_ASSERT((first != NULL) && (then != NULL) && (then->rsc != NULL) && (scheduler != NULL)); // Update the actions for the group itself changed |= pcmk__update_ordered_actions(first, then, node, flags, filter, type, scheduler); // Update the actions for each group member for (GList *iter = then->rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; pcmk_action_t *member_action = find_first_action(member->actions, NULL, then->task, node); if (member_action != NULL) { changed |= member->cmds->update_ordered_actions(first, member_action, node, flags, filter, type, scheduler); } } return changed; } /*! * \internal * \brief Apply a location constraint to a group's allowed node scores * * \param[in,out] rsc Group resource to apply constraint to * \param[in,out] location Location constraint to apply */ void pcmk__group_apply_location(pcmk_resource_t *rsc, pcmk__location_t *location) { GList *node_list_orig = NULL; GList *node_list_copy = NULL; bool reset_scores = true; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group) && (location != NULL)); node_list_orig = location->nodes; node_list_copy = pcmk__copy_node_list(node_list_orig, true); reset_scores = pe__group_flag_is_set(rsc, pcmk__group_colocated); // Apply the constraint for the group itself (updates node scores) pcmk__apply_location(rsc, location); // Apply the constraint for each member for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; member->cmds->apply_location(member, location); if (reset_scores) { /* The first member of colocated groups needs to use the original * node scores, but subsequent members should work on a copy, since * the first member's scores already incorporate theirs. */ reset_scores = false; location->nodes = node_list_copy; } } location->nodes = node_list_orig; g_list_free_full(node_list_copy, free); } // Group implementation of pcmk_assignment_methods_t:colocated_resources() GList * pcmk__group_colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs) { const pcmk_resource_t *member = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)); if (orig_rsc == NULL) { orig_rsc = rsc; } if (pe__group_flag_is_set(rsc, pcmk__group_colocated) || pcmk__is_clone(rsc->parent)) { /* This group has colocated members and/or is cloned -- either way, * add every child's colocated resources to the list. The first and last * members will include the group's own colocations. */ colocated_rscs = g_list_prepend(colocated_rscs, (gpointer) rsc); for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = (const pcmk_resource_t *) iter->data; colocated_rscs = member->cmds->colocated_resources(member, orig_rsc, colocated_rscs); } } else if (rsc->children != NULL) { /* This group's members are not colocated, and the group is not cloned, * so just add the group's own colocations to the list. */ colocated_rscs = pcmk__colocated_resources(rsc, orig_rsc, colocated_rscs); } return colocated_rscs; } // Group implementation of pcmk_assignment_methods_t:with_this_colocations() void pcmk__with_group_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group) && (orig_rsc != NULL) && (list != NULL)); // Ignore empty groups if (rsc->children == NULL) { return; } /* "With this" colocations are needed only for the group itself and for its * last member. (Previous members will chain via the group internal * colocations.) */ if ((orig_rsc != rsc) && (orig_rsc != pe__last_group_member(rsc))) { return; } pcmk__rsc_trace(rsc, "Adding 'with %s' colocations to list for %s", rsc->id, orig_rsc->id); // Add the group's own colocations pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); // If cloned, add any relevant colocations with the clone if (rsc->parent != NULL) { rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list); } if (!pe__group_flag_is_set(rsc, pcmk__group_colocated)) { // @COMPAT Non-colocated groups are deprecated return; } // Add explicit colocations with the group's (other) children for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *member = iter->data; if (member != orig_rsc) { member->cmds->with_this_colocations(member, orig_rsc, list); } } } // Group implementation of pcmk_assignment_methods_t:this_with_colocations() void pcmk__group_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *member = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group) && (orig_rsc != NULL) && (list != NULL)); // Ignore empty groups if (rsc->children == NULL) { return; } /* "This with" colocations are normally needed only for the group itself and * for its first member. */ if ((rsc == orig_rsc) || (orig_rsc == (const pcmk_resource_t *) rsc->children->data)) { pcmk__rsc_trace(rsc, "Adding '%s with' colocations to list for %s", rsc->id, orig_rsc->id); // Add the group's own colocations pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); // If cloned, add any relevant colocations involving the clone if (rsc->parent != NULL) { rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list); } if (!pe__group_flag_is_set(rsc, pcmk__group_colocated)) { // @COMPAT Non-colocated groups are deprecated return; } // Add explicit colocations involving the group's (other) children for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = iter->data; if (member != orig_rsc) { member->cmds->this_with_colocations(member, orig_rsc, list); } } return; } /* Later group members honor the group's colocations indirectly, due to the * internal group colocations that chain everything from the first member. * However, if an earlier group member is unmanaged, this chaining will not * happen, so the group's mandatory colocations must be explicitly added. */ for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = iter->data; if (orig_rsc == member) { break; // We've seen all earlier members, and none are unmanaged } if (!pcmk_is_set(member->flags, pcmk_rsc_managed)) { crm_trace("Adding mandatory '%s with' colocations to list for " "member %s because earlier member %s is unmanaged", rsc->id, orig_rsc->id, member->id); for (const GList *cons_iter = rsc->rsc_cons; cons_iter != NULL; cons_iter = cons_iter->next) { const pcmk__colocation_t *colocation = NULL; colocation = (const pcmk__colocation_t *) cons_iter->data; - if (colocation->score == INFINITY) { + if (colocation->score == PCMK_SCORE_INFINITY) { pcmk__add_this_with(list, colocation, orig_rsc); } } // @TODO Add mandatory (or all?) clone constraints if cloned break; } } } /*! * \internal * \brief Update nodes with scores of colocated resources' nodes * * Given a table of nodes and a resource, update the nodes' scores with the * scores of the best nodes matching the attribute used for each of the * resource's relevant colocations. * * \param[in,out] source_rsc Group resource whose node scores to add * \param[in] target_rsc Resource on whose behalf to update \p *nodes * \param[in] log_id Resource ID for logs (if \c NULL, use * \p source_rsc ID) * \param[in,out] nodes Nodes to update (set initial contents to \c NULL * to copy allowed nodes from \p source_rsc) * \param[in] colocation Original colocation constraint (used to get * configured primary resource's stickiness, and * to get colocation node attribute; if \c NULL, * source_rsc's own matching node scores will * not be added, and \p *nodes must be \c NULL as * well) * \param[in] factor Incorporate scores multiplied by this factor * \param[in] flags Bitmask of enum pcmk__coloc_select values * * \note \c NULL \p target_rsc, \c NULL \p *nodes, \c NULL \p colocation, and * the \c pcmk__coloc_select_this_with flag are used together (and only by * \c cmp_resources()). * \note The caller remains responsible for freeing \p *nodes. * \note This is the group implementation of * \c pcmk_assignment_methods_t:add_colocated_node_scores(). */ void pcmk__group_add_colocated_node_scores(pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags) { pcmk_resource_t *member = NULL; CRM_ASSERT((source_rsc != NULL) && (source_rsc->variant == pcmk_rsc_variant_group) && (nodes != NULL) && ((colocation != NULL) || ((target_rsc == NULL) && (*nodes == NULL)))); if (log_id == NULL) { log_id = source_rsc->id; } // Avoid infinite recursion if (pcmk_is_set(source_rsc->flags, pcmk_rsc_updating_nodes)) { pcmk__rsc_info(source_rsc, "%s: Breaking dependency loop at %s", log_id, source_rsc->id); return; } pcmk__set_rsc_flags(source_rsc, pcmk_rsc_updating_nodes); // Ignore empty groups (only possible with schema validation disabled) if (source_rsc->children == NULL) { return; } /* Refer the operation to the first or last member as appropriate. * * cmp_resources() is the only caller that passes a NULL nodes table, * and is also the only caller using pcmk__coloc_select_this_with. * For "this with" colocations, the last member will recursively incorporate * all the other members' "this with" colocations via the internal group * colocations (and via the first member, the group's own colocations). * * For "with this" colocations, the first member works similarly. */ if (*nodes == NULL) { member = pe__last_group_member(source_rsc); } else { member = source_rsc->children->data; } pcmk__rsc_trace(source_rsc, "%s: Merging scores from group %s using member %s " "(at %.6f)", log_id, source_rsc->id, member->id, factor); member->cmds->add_colocated_node_scores(member, target_rsc, log_id, nodes, colocation, factor, flags); pcmk__clear_rsc_flags(source_rsc, pcmk_rsc_updating_nodes); } // Group implementation of pcmk_assignment_methods_t:add_utilization() void pcmk__group_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { pcmk_resource_t *member = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group) && (orig_rsc != NULL) && (utilization != NULL)); if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { return; } pcmk__rsc_trace(orig_rsc, "%s: Adding group %s as colocated utilization", orig_rsc->id, rsc->id); if (pe__group_flag_is_set(rsc, pcmk__group_colocated) || pcmk__is_clone(rsc->parent)) { // Every group member will be on same node, so sum all members for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = (pcmk_resource_t *) iter->data; if (pcmk_is_set(member->flags, pcmk_rsc_unassigned) && (g_list_find(all_rscs, member) == NULL)) { member->cmds->add_utilization(member, orig_rsc, all_rscs, utilization); } } } else if (rsc->children != NULL) { // Just add first member's utilization member = (pcmk_resource_t *) rsc->children->data; if ((member != NULL) && pcmk_is_set(member->flags, pcmk_rsc_unassigned) && (g_list_find(all_rscs, member) == NULL)) { member->cmds->add_utilization(member, orig_rsc, all_rscs, utilization); } } } void pcmk__group_shutdown_lock(pcmk_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; member->cmds->shutdown_lock(member); } } diff --git a/lib/pacemaker/pcmk_sched_instances.c b/lib/pacemaker/pcmk_sched_instances.c index 544f1e3e0b..eba34deea7 100644 --- a/lib/pacemaker/pcmk_sched_instances.c +++ b/lib/pacemaker/pcmk_sched_instances.c @@ -1,1689 +1,1689 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ /* This file is intended for code usable with both clone instances and bundle * replica containers. */ #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Check whether a node is allowed to run an instance * * \param[in] instance Clone instance or bundle container to check * \param[in] node Node to check * \param[in] max_per_node Maximum number of instances allowed to run on a node * * \return true if \p node is allowed to run \p instance, otherwise false */ static bool can_run_instance(const pcmk_resource_t *instance, const pcmk_node_t *node, int max_per_node) { pcmk_node_t *allowed_node = NULL; if (pcmk_is_set(instance->flags, pcmk_rsc_removed)) { pcmk__rsc_trace(instance, "%s cannot run on %s: orphaned", instance->id, pcmk__node_name(node)); return false; } if (!pcmk__node_available(node, false, false)) { pcmk__rsc_trace(instance, "%s cannot run on %s: node cannot run resources", instance->id, pcmk__node_name(node)); return false; } allowed_node = pcmk__top_allowed_node(instance, node); if (allowed_node == NULL) { crm_warn("%s cannot run on %s: node not allowed", instance->id, pcmk__node_name(node)); return false; } if (allowed_node->weight < 0) { pcmk__rsc_trace(instance, "%s cannot run on %s: parent score is %s there", instance->id, pcmk__node_name(node), pcmk_readable_score(allowed_node->weight)); return false; } if (allowed_node->count >= max_per_node) { pcmk__rsc_trace(instance, "%s cannot run on %s: node already has %d instance%s", instance->id, pcmk__node_name(node), max_per_node, pcmk__plural_s(max_per_node)); return false; } pcmk__rsc_trace(instance, "%s can run on %s (%d already running)", instance->id, pcmk__node_name(node), allowed_node->count); return true; } /*! * \internal * \brief Ban a clone instance or bundle replica from unavailable allowed nodes * * \param[in,out] instance Clone instance or bundle replica to ban * \param[in] max_per_node Maximum instances allowed to run on a node */ static void ban_unavailable_allowed_nodes(pcmk_resource_t *instance, int max_per_node) { if (instance->allowed_nodes != NULL) { GHashTableIter iter; pcmk_node_t *node = NULL; g_hash_table_iter_init(&iter, instance->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (!can_run_instance(instance, node, max_per_node)) { pcmk__rsc_trace(instance, "Banning %s from unavailable node %s", instance->id, pcmk__node_name(node)); - node->weight = -INFINITY; + node->weight = -PCMK_SCORE_INFINITY; for (GList *child_iter = instance->children; child_iter != NULL; child_iter = child_iter->next) { pcmk_resource_t *child = child_iter->data; pcmk_node_t *child_node = NULL; child_node = g_hash_table_lookup(child->allowed_nodes, node->details->id); if (child_node != NULL) { pcmk__rsc_trace(instance, "Banning %s child %s " "from unavailable node %s", instance->id, child->id, pcmk__node_name(node)); - child_node->weight = -INFINITY; + child_node->weight = -PCMK_SCORE_INFINITY; } } } } } } /*! * \internal * \brief Create a hash table with a single node in it * * \param[in] node Node to copy into new table * * \return Newly created hash table containing a copy of \p node * \note The caller is responsible for freeing the result with * g_hash_table_destroy(). */ static GHashTable * new_node_table(pcmk_node_t *node) { GHashTable *table = pcmk__strkey_table(NULL, free); node = pe__copy_node(node); g_hash_table_insert(table, (gpointer) node->details->id, node); return table; } /*! * \internal * \brief Apply a resource's parent's colocation scores to a node table * * \param[in] rsc Resource whose colocations should be applied * \param[in,out] nodes Node table to apply colocations to */ static void apply_parent_colocations(const pcmk_resource_t *rsc, GHashTable **nodes) { GList *colocations = pcmk__this_with_colocations(rsc); for (const GList *iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *colocation = iter->data; pcmk_resource_t *other = colocation->primary; - float factor = colocation->score / (float) INFINITY; + float factor = colocation->score / (float) PCMK_SCORE_INFINITY; other->cmds->add_colocated_node_scores(other, rsc, rsc->id, nodes, colocation, factor, pcmk__coloc_select_default); } g_list_free(colocations); colocations = pcmk__with_this_colocations(rsc); for (const GList *iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *colocation = iter->data; pcmk_resource_t *other = colocation->dependent; - float factor = colocation->score / (float) INFINITY; + float factor = colocation->score / (float) PCMK_SCORE_INFINITY; if (!pcmk__colocation_has_influence(colocation, rsc)) { continue; } other->cmds->add_colocated_node_scores(other, rsc, rsc->id, nodes, colocation, factor, pcmk__coloc_select_nonnegative); } g_list_free(colocations); } /*! * \internal * \brief Compare clone or bundle instances based on colocation scores * * Determine the relative order in which two clone or bundle instances should be * assigned to nodes, considering the scores of colocation constraints directly * or indirectly involving them. * * \param[in] instance1 First instance to compare * \param[in] instance2 Second instance to compare * * \return A negative number if \p instance1 should be assigned first, * a positive number if \p instance2 should be assigned first, * or 0 if assignment order doesn't matter */ static int cmp_instance_by_colocation(const pcmk_resource_t *instance1, const pcmk_resource_t *instance2) { int rc = 0; pcmk_node_t *node1 = NULL; pcmk_node_t *node2 = NULL; pcmk_node_t *current_node1 = pcmk__current_node(instance1); pcmk_node_t *current_node2 = pcmk__current_node(instance2); GHashTable *colocated_scores1 = NULL; GHashTable *colocated_scores2 = NULL; CRM_ASSERT((instance1 != NULL) && (instance1->parent != NULL) && (instance2 != NULL) && (instance2->parent != NULL) && (current_node1 != NULL) && (current_node2 != NULL)); // Create node tables initialized with each node colocated_scores1 = new_node_table(current_node1); colocated_scores2 = new_node_table(current_node2); // Apply parental colocations apply_parent_colocations(instance1, &colocated_scores1); apply_parent_colocations(instance2, &colocated_scores2); // Find original nodes again, with scores updated for colocations node1 = g_hash_table_lookup(colocated_scores1, current_node1->details->id); node2 = g_hash_table_lookup(colocated_scores2, current_node2->details->id); // Compare nodes by updated scores if (node1->weight < node2->weight) { crm_trace("Assign %s (%d on %s) after %s (%d on %s)", instance1->id, node1->weight, pcmk__node_name(node1), instance2->id, node2->weight, pcmk__node_name(node2)); rc = 1; } else if (node1->weight > node2->weight) { crm_trace("Assign %s (%d on %s) before %s (%d on %s)", instance1->id, node1->weight, pcmk__node_name(node1), instance2->id, node2->weight, pcmk__node_name(node2)); rc = -1; } g_hash_table_destroy(colocated_scores1); g_hash_table_destroy(colocated_scores2); return rc; } /*! * \internal * \brief Check whether a resource or any of its children are failed * * \param[in] rsc Resource to check * * \return true if \p rsc or any of its children are failed, otherwise false */ static bool did_fail(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { return true; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { if (did_fail((const pcmk_resource_t *) iter->data)) { return true; } } return false; } /*! * \internal * \brief Check whether a node is allowed to run a resource * * \param[in] rsc Resource to check * \param[in,out] node Node to check (will be set NULL if not allowed) * * \return true if *node is either NULL or allowed for \p rsc, otherwise false */ static bool node_is_allowed(const pcmk_resource_t *rsc, pcmk_node_t **node) { if (*node != NULL) { pcmk_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes, (*node)->details->id); if ((allowed == NULL) || (allowed->weight < 0)) { pcmk__rsc_trace(rsc, "%s: current location (%s) is unavailable", rsc->id, pcmk__node_name(*node)); *node = NULL; return false; } } return true; } /*! * \internal * \brief Compare two clone or bundle instances' instance numbers * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a's instance number is lower, * a positive number if \p b's instance number is lower, * or 0 if their instance numbers are the same */ gint pcmk__cmp_instance_number(gconstpointer a, gconstpointer b) { const pcmk_resource_t *instance1 = (const pcmk_resource_t *) a; const pcmk_resource_t *instance2 = (const pcmk_resource_t *) b; char *div1 = NULL; char *div2 = NULL; CRM_ASSERT((instance1 != NULL) && (instance2 != NULL)); // Clone numbers are after a colon, bundle numbers after a dash div1 = strrchr(instance1->id, ':'); if (div1 == NULL) { div1 = strrchr(instance1->id, '-'); } div2 = strrchr(instance2->id, ':'); if (div2 == NULL) { div2 = strrchr(instance2->id, '-'); } CRM_ASSERT((div1 != NULL) && (div2 != NULL)); return (gint) (strtol(div1 + 1, NULL, 10) - strtol(div2 + 1, NULL, 10)); } /*! * \internal * \brief Compare clone or bundle instances according to assignment order * * Compare two clone or bundle instances according to the order they should be * assigned to nodes, preferring (in order): * * - Active instance that is less multiply active * - Instance that is not active on a disallowed node * - Instance with higher configured priority * - Active instance whose current node can run resources * - Active instance whose parent is allowed on current node * - Active instance whose current node has fewer other instances * - Active instance * - Instance that isn't failed * - Instance whose colocations result in higher score on current node * - Instance with lower ID in lexicographic order * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a should be assigned first, * a positive number if \p b should be assigned first, * or 0 if assignment order doesn't matter */ gint pcmk__cmp_instance(gconstpointer a, gconstpointer b) { int rc = 0; pcmk_node_t *node1 = NULL; pcmk_node_t *node2 = NULL; unsigned int nnodes1 = 0; unsigned int nnodes2 = 0; bool can1 = true; bool can2 = true; const pcmk_resource_t *instance1 = (const pcmk_resource_t *) a; const pcmk_resource_t *instance2 = (const pcmk_resource_t *) b; CRM_ASSERT((instance1 != NULL) && (instance2 != NULL)); node1 = instance1->fns->active_node(instance1, &nnodes1, NULL); node2 = instance2->fns->active_node(instance2, &nnodes2, NULL); /* If both instances are running and at least one is multiply * active, prefer instance that's running on fewer nodes. */ if ((nnodes1 > 0) && (nnodes2 > 0)) { if (nnodes1 < nnodes2) { crm_trace("Assign %s (active on %d) before %s (active on %d): " "less multiply active", instance1->id, nnodes1, instance2->id, nnodes2); return -1; } else if (nnodes1 > nnodes2) { crm_trace("Assign %s (active on %d) after %s (active on %d): " "more multiply active", instance1->id, nnodes1, instance2->id, nnodes2); return 1; } } /* An instance that is either inactive or active on an allowed node is * preferred over an instance that is active on a no-longer-allowed node. */ can1 = node_is_allowed(instance1, &node1); can2 = node_is_allowed(instance2, &node2); if (can1 && !can2) { crm_trace("Assign %s before %s: not active on a disallowed node", instance1->id, instance2->id); return -1; } else if (!can1 && can2) { crm_trace("Assign %s after %s: active on a disallowed node", instance1->id, instance2->id); return 1; } // Prefer instance with higher configured priority if (instance1->priority > instance2->priority) { crm_trace("Assign %s before %s: priority (%d > %d)", instance1->id, instance2->id, instance1->priority, instance2->priority); return -1; } else if (instance1->priority < instance2->priority) { crm_trace("Assign %s after %s: priority (%d < %d)", instance1->id, instance2->id, instance1->priority, instance2->priority); return 1; } // Prefer active instance if ((node1 == NULL) && (node2 == NULL)) { crm_trace("No assignment preference for %s vs. %s: inactive", instance1->id, instance2->id); return 0; } else if (node1 == NULL) { crm_trace("Assign %s after %s: active", instance1->id, instance2->id); return 1; } else if (node2 == NULL) { crm_trace("Assign %s before %s: active", instance1->id, instance2->id); return -1; } // Prefer instance whose current node can run resources can1 = pcmk__node_available(node1, false, false); can2 = pcmk__node_available(node2, false, false); if (can1 && !can2) { crm_trace("Assign %s before %s: current node can run resources", instance1->id, instance2->id); return -1; } else if (!can1 && can2) { crm_trace("Assign %s after %s: current node can't run resources", instance1->id, instance2->id); return 1; } // Prefer instance whose parent is allowed to run on instance's current node node1 = pcmk__top_allowed_node(instance1, node1); node2 = pcmk__top_allowed_node(instance2, node2); if ((node1 == NULL) && (node2 == NULL)) { crm_trace("No assignment preference for %s vs. %s: " "parent not allowed on either instance's current node", instance1->id, instance2->id); return 0; } else if (node1 == NULL) { crm_trace("Assign %s after %s: parent not allowed on current node", instance1->id, instance2->id); return 1; } else if (node2 == NULL) { crm_trace("Assign %s before %s: parent allowed on current node", instance1->id, instance2->id); return -1; } // Prefer instance whose current node is running fewer other instances if (node1->count < node2->count) { crm_trace("Assign %s before %s: fewer active instances on current node", instance1->id, instance2->id); return -1; } else if (node1->count > node2->count) { crm_trace("Assign %s after %s: more active instances on current node", instance1->id, instance2->id); return 1; } // Prefer instance that isn't failed can1 = did_fail(instance1); can2 = did_fail(instance2); if (!can1 && can2) { crm_trace("Assign %s before %s: not failed", instance1->id, instance2->id); return -1; } else if (can1 && !can2) { crm_trace("Assign %s after %s: failed", instance1->id, instance2->id); return 1; } // Prefer instance with higher cumulative colocation score on current node rc = cmp_instance_by_colocation(instance1, instance2); if (rc != 0) { return rc; } // Prefer instance with lower instance number rc = pcmk__cmp_instance_number(instance1, instance2); if (rc < 0) { crm_trace("Assign %s before %s: instance number", instance1->id, instance2->id); } else if (rc > 0) { crm_trace("Assign %s after %s: instance number", instance1->id, instance2->id); } else { crm_trace("No assignment preference for %s vs. %s", instance1->id, instance2->id); } return rc; } /*! * \internal * \brief Increment the parent's instance count after assigning an instance * * An instance's parent tracks how many instances have been assigned to each * node via its pcmk_node_t:count member. After assigning an instance to a node, * find the corresponding node in the parent's allowed table and increment it. * * \param[in,out] instance Instance whose parent to update * \param[in] assigned_to Node to which the instance was assigned */ static void increment_parent_count(pcmk_resource_t *instance, const pcmk_node_t *assigned_to) { pcmk_node_t *allowed = NULL; if (assigned_to == NULL) { return; } allowed = pcmk__top_allowed_node(instance, assigned_to); if (allowed == NULL) { /* The instance is allowed on the node, but its parent isn't. This * shouldn't be possible if the resource is managed, and we won't be * able to limit the number of instances assigned to the node. */ CRM_LOG_ASSERT(!pcmk_is_set(instance->flags, pcmk_rsc_managed)); } else { allowed->count++; } } /*! * \internal * \brief Assign an instance to a node * * \param[in,out] instance Clone instance or bundle replica container * \param[in] prefer If not NULL, attempt early assignment to this * node, if still the best choice; otherwise, * perform final assignment * \param[in] max_per_node Assign at most this many instances to one node * * \return Node to which \p instance is assigned */ static const pcmk_node_t * assign_instance(pcmk_resource_t *instance, const pcmk_node_t *prefer, int max_per_node) { pcmk_node_t *chosen = NULL; pcmk__rsc_trace(instance, "Assigning %s (preferring %s)", instance->id, ((prefer == NULL)? "no node" : prefer->details->uname)); if (pcmk_is_set(instance->flags, pcmk_rsc_assigning)) { pcmk__rsc_debug(instance, "Assignment loop detected involving %s colocations", instance->id); return NULL; } ban_unavailable_allowed_nodes(instance, max_per_node); // Failed early assignments are reversible (stop_if_fail=false) chosen = instance->cmds->assign(instance, prefer, (prefer == NULL)); increment_parent_count(instance, chosen); return chosen; } /*! * \internal * \brief Try to assign an instance to its current node early * * \param[in] rsc Clone or bundle being assigned (for logs only) * \param[in] instance Clone instance or bundle replica container * \param[in] current Instance's current node * \param[in] max_per_node Maximum number of instances per node * \param[in] available Number of instances still available for assignment * * \return \c true if \p instance was successfully assigned to its current node, * or \c false otherwise */ static bool assign_instance_early(const pcmk_resource_t *rsc, pcmk_resource_t *instance, const pcmk_node_t *current, int max_per_node, int available) { const pcmk_node_t *chosen = NULL; int reserved = 0; pcmk_resource_t *parent = instance->parent; GHashTable *allowed_orig = NULL; GHashTable *allowed_orig_parent = parent->allowed_nodes; const pcmk_node_t *allowed_node = NULL; pcmk__rsc_trace(instance, "Trying to assign %s to its current node %s", instance->id, pcmk__node_name(current)); allowed_node = g_hash_table_lookup(instance->allowed_nodes, current->details->id); if (!pcmk__node_available(allowed_node, true, false)) { pcmk__rsc_info(instance, "Not assigning %s to current node %s: unavailable", instance->id, pcmk__node_name(current)); return false; } /* On each iteration, if instance gets assigned to a node other than its * current one, we reserve one instance for the chosen node, unassign * instance, restore instance's original node tables, and try again. This * way, instances are proportionally assigned to nodes based on preferences, * but shuffling of specific instances is minimized. If a node will be * assigned instances at all, it preferentially receives instances that are * currently active there. * * parent->allowed_nodes tracks the number of instances assigned to each * node. If a node already has max_per_node instances assigned, * ban_unavailable_allowed_nodes() marks it as unavailable. * * In the end, we restore the original parent->allowed_nodes to undo the * changes to counts during tentative assignments. If we successfully * assigned instance to its current node, we increment that node's counter. */ // Back up the allowed node tables of instance and its children recursively pcmk__copy_node_tables(instance, &allowed_orig); // Update instances-per-node counts in a scratch table parent->allowed_nodes = pcmk__copy_node_table(parent->allowed_nodes); while (reserved < available) { chosen = assign_instance(instance, current, max_per_node); if (pcmk__same_node(chosen, current)) { // Successfully assigned to current node break; } // Assignment updates scores, so restore to original state pcmk__rsc_debug(instance, "Rolling back node scores for %s", instance->id); pcmk__restore_node_tables(instance, allowed_orig); if (chosen == NULL) { // Assignment failed, so give up pcmk__rsc_info(instance, "Not assigning %s to current node %s: unavailable", instance->id, pcmk__node_name(current)); pcmk__set_rsc_flags(instance, pcmk_rsc_unassigned); break; } // We prefer more strongly to assign an instance to the chosen node pcmk__rsc_debug(instance, "Not assigning %s to current node %s: %s is better", instance->id, pcmk__node_name(current), pcmk__node_name(chosen)); // Reserve one instance for the chosen node and try again if (++reserved >= available) { pcmk__rsc_info(instance, "Not assigning %s to current node %s: " "other assignments are more important", instance->id, pcmk__node_name(current)); } else { pcmk__rsc_debug(instance, "Reserved an instance of %s for %s. Retrying " "assignment of %s to %s", rsc->id, pcmk__node_name(chosen), instance->id, pcmk__node_name(current)); } // Clear this assignment (frees chosen); leave instance counts in parent pcmk__unassign_resource(instance); chosen = NULL; } g_hash_table_destroy(allowed_orig); // Restore original instances-per-node counts g_hash_table_destroy(parent->allowed_nodes); parent->allowed_nodes = allowed_orig_parent; if (chosen == NULL) { // Couldn't assign instance to current node return false; } pcmk__rsc_trace(instance, "Assigned %s to current node %s", instance->id, pcmk__node_name(current)); increment_parent_count(instance, chosen); return true; } /*! * \internal * \brief Reset the node counts of a resource's allowed nodes to zero * * \param[in,out] rsc Resource to reset * * \return Number of nodes that are available to run resources */ static unsigned int reset_allowed_node_counts(pcmk_resource_t *rsc) { unsigned int available_nodes = 0; pcmk_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { node->count = 0; if (pcmk__node_available(node, false, false)) { available_nodes++; } } return available_nodes; } /*! * \internal * \brief Check whether an instance has a preferred node * * \param[in] instance Clone instance or bundle replica container * \param[in] optimal_per_node Optimal number of instances per node * * \return Instance's current node if still available, otherwise NULL */ static const pcmk_node_t * preferred_node(const pcmk_resource_t *instance, int optimal_per_node) { const pcmk_node_t *node = NULL; const pcmk_node_t *parent_node = NULL; // Check whether instance is active, healthy, and not yet assigned if ((instance->running_on == NULL) || !pcmk_is_set(instance->flags, pcmk_rsc_unassigned) || pcmk_is_set(instance->flags, pcmk_rsc_failed)) { return NULL; } // Check whether instance's current node can run resources node = pcmk__current_node(instance); if (!pcmk__node_available(node, true, false)) { pcmk__rsc_trace(instance, "Not assigning %s to %s early (unavailable)", instance->id, pcmk__node_name(node)); return NULL; } // Check whether node already has optimal number of instances assigned parent_node = pcmk__top_allowed_node(instance, node); if ((parent_node != NULL) && (parent_node->count >= optimal_per_node)) { pcmk__rsc_trace(instance, "Not assigning %s to %s early " "(optimal instances already assigned)", instance->id, pcmk__node_name(node)); return NULL; } return node; } /*! * \internal * \brief Assign collective instances to nodes * * \param[in,out] collective Clone or bundle resource being assigned * \param[in,out] instances List of clone instances or bundle containers * \param[in] max_total Maximum instances to assign in total * \param[in] max_per_node Maximum instances to assign to any one node */ void pcmk__assign_instances(pcmk_resource_t *collective, GList *instances, int max_total, int max_per_node) { // Reuse node count to track number of assigned instances unsigned int available_nodes = reset_allowed_node_counts(collective); int optimal_per_node = 0; int assigned = 0; GList *iter = NULL; pcmk_resource_t *instance = NULL; const pcmk_node_t *current = NULL; if (available_nodes > 0) { optimal_per_node = max_total / available_nodes; } if (optimal_per_node < 1) { optimal_per_node = 1; } pcmk__rsc_debug(collective, "Assigning up to %d %s instance%s to up to %u node%s " "(at most %d per host, %d optimal)", max_total, collective->id, pcmk__plural_s(max_total), available_nodes, pcmk__plural_s(available_nodes), max_per_node, optimal_per_node); // Assign as many instances as possible to their current location for (iter = instances; (iter != NULL) && (assigned < max_total); iter = iter->next) { int available = max_total - assigned; instance = iter->data; if (!pcmk_is_set(instance->flags, pcmk_rsc_unassigned)) { continue; // Already assigned } current = preferred_node(instance, optimal_per_node); if ((current != NULL) && assign_instance_early(collective, instance, current, max_per_node, available)) { assigned++; } } pcmk__rsc_trace(collective, "Assigned %d of %d instance%s to current node", assigned, max_total, pcmk__plural_s(max_total)); for (iter = instances; iter != NULL; iter = iter->next) { instance = (pcmk_resource_t *) iter->data; if (!pcmk_is_set(instance->flags, pcmk_rsc_unassigned)) { continue; // Already assigned } if (instance->running_on != NULL) { current = pcmk__current_node(instance); if (pcmk__top_allowed_node(instance, current) == NULL) { const char *unmanaged = ""; if (!pcmk_is_set(instance->flags, pcmk_rsc_managed)) { unmanaged = "Unmanaged resource "; } crm_notice("%s%s is running on %s which is no longer allowed", unmanaged, instance->id, pcmk__node_name(current)); } } if (assigned >= max_total) { pcmk__rsc_debug(collective, "Not assigning %s because maximum %d instances " "already assigned", instance->id, max_total); - resource_location(instance, NULL, -INFINITY, + resource_location(instance, NULL, -PCMK_SCORE_INFINITY, "collective_limit_reached", collective->cluster); } else if (assign_instance(instance, NULL, max_per_node) != NULL) { assigned++; } } pcmk__rsc_debug(collective, "Assigned %d of %d possible instance%s of %s", assigned, max_total, pcmk__plural_s(max_total), collective->id); } enum instance_state { instance_starting = (1 << 0), instance_stopping = (1 << 1), /* This indicates that some instance is restarting. It's not the same as * instance_starting|instance_stopping, which would indicate that some * instance is starting, and some instance (not necessarily the same one) is * stopping. */ instance_restarting = (1 << 2), instance_active = (1 << 3), instance_all = instance_starting|instance_stopping |instance_restarting|instance_active, }; /*! * \internal * \brief Check whether an instance is active, starting, and/or stopping * * \param[in] instance Clone instance or bundle replica container * \param[in,out] state Whether any instance is starting, stopping, etc. */ static void check_instance_state(const pcmk_resource_t *instance, uint32_t *state) { const GList *iter = NULL; uint32_t instance_state = 0; // State of just this instance // No need to check further if all conditions have already been detected if (pcmk_all_flags_set(*state, instance_all)) { return; } // If instance is a collective (a cloned group), check its children instead if (instance->variant > pcmk_rsc_variant_primitive) { for (iter = instance->children; (iter != NULL) && !pcmk_all_flags_set(*state, instance_all); iter = iter->next) { check_instance_state((const pcmk_resource_t *) iter->data, state); } return; } // If we get here, instance is a primitive if (instance->running_on != NULL) { instance_state |= instance_active; } // Check each of the instance's actions for runnable start or stop for (iter = instance->actions; (iter != NULL) && !pcmk_all_flags_set(instance_state, instance_starting |instance_stopping); iter = iter->next) { const pcmk_action_t *action = (const pcmk_action_t *) iter->data; const bool optional = pcmk_is_set(action->flags, pcmk_action_optional); if (pcmk__str_eq(PCMK_ACTION_START, action->task, pcmk__str_none)) { if (!optional && pcmk_is_set(action->flags, pcmk_action_runnable)) { pcmk__rsc_trace(instance, "Instance is starting due to %s", action->uuid); instance_state |= instance_starting; } else { pcmk__rsc_trace(instance, "%s doesn't affect %s state (%s)", action->uuid, instance->id, (optional? "optional" : "unrunnable")); } } else if (pcmk__str_eq(PCMK_ACTION_STOP, action->task, pcmk__str_none)) { /* Only stop actions can be pseudo-actions for primitives. That * indicates that the node they are on is being fenced, so the stop * is implied rather than actually executed. */ if (!optional && pcmk_any_flags_set(action->flags, pcmk_action_pseudo |pcmk_action_runnable)) { pcmk__rsc_trace(instance, "Instance is stopping due to %s", action->uuid); instance_state |= instance_stopping; } else { pcmk__rsc_trace(instance, "%s doesn't affect %s state (%s)", action->uuid, instance->id, (optional? "optional" : "unrunnable")); } } } if (pcmk_all_flags_set(instance_state, instance_starting|instance_stopping)) { instance_state |= instance_restarting; } *state |= instance_state; } /*! * \internal * \brief Create actions for collective resource instances * * \param[in,out] collective Clone or bundle resource to create actions for * \param[in,out] instances List of clone instances or bundle containers */ void pcmk__create_instance_actions(pcmk_resource_t *collective, GList *instances) { uint32_t state = 0; pcmk_action_t *stop = NULL; pcmk_action_t *stopped = NULL; pcmk_action_t *start = NULL; pcmk_action_t *started = NULL; pcmk__rsc_trace(collective, "Creating collective instance actions for %s", collective->id); // Create actions for each instance appropriate to its variant for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; instance->cmds->create_actions(instance); check_instance_state(instance, &state); } // Create pseudo-actions for rsc start and started start = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_START, !pcmk_is_set(state, instance_starting), true); started = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_RUNNING, !pcmk_is_set(state, instance_starting), false); - started->priority = INFINITY; + started->priority = PCMK_SCORE_INFINITY; if (pcmk_any_flags_set(state, instance_active|instance_starting)) { pcmk__set_action_flags(started, pcmk_action_runnable); } // Create pseudo-actions for rsc stop and stopped stop = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_STOP, !pcmk_is_set(state, instance_stopping), true); stopped = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_STOPPED, !pcmk_is_set(state, instance_stopping), true); - stopped->priority = INFINITY; + stopped->priority = PCMK_SCORE_INFINITY; if (!pcmk_is_set(state, instance_restarting)) { pcmk__set_action_flags(stop, pcmk_action_migratable); } if (collective->variant == pcmk_rsc_variant_clone) { pe__create_clone_notif_pseudo_ops(collective, start, started, stop, stopped); } } /*! * \internal * \brief Get a list of clone instances or bundle replica containers * * \param[in] rsc Clone or bundle resource * * \return Clone instances if \p rsc is a clone, or a newly created list of * \p rsc's replica containers if \p rsc is a bundle * \note The caller must call free_instance_list() on the result when the list * is no longer needed. */ static inline GList * get_instance_list(const pcmk_resource_t *rsc) { if (rsc->variant == pcmk_rsc_variant_bundle) { return pe__bundle_containers(rsc); } else { return rsc->children; } } /*! * \internal * \brief Free any memory created by get_instance_list() * * \param[in] rsc Clone or bundle resource passed to get_instance_list() * \param[in,out] list Return value of get_instance_list() for \p rsc */ static inline void free_instance_list(const pcmk_resource_t *rsc, GList *list) { if (list != rsc->children) { g_list_free(list); } } /*! * \internal * \brief Check whether an instance is compatible with a role and node * * \param[in] instance Clone instance or bundle replica container * \param[in] node Instance must match this node * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return true if \p instance is compatible with \p node and \p role, * otherwise false */ bool pcmk__instance_matches(const pcmk_resource_t *instance, const pcmk_node_t *node, enum rsc_role_e role, bool current) { pcmk_node_t *instance_node = NULL; CRM_CHECK((instance != NULL) && (node != NULL), return false); if ((role != pcmk_role_unknown) && (role != instance->fns->state(instance, current))) { pcmk__rsc_trace(instance, "%s is not a compatible instance (role is not %s)", instance->id, pcmk_role_text(role)); return false; } if (!is_set_recursive(instance, pcmk_rsc_blocked, true)) { // We only want instances that haven't failed instance_node = instance->fns->location(instance, NULL, current); } if (instance_node == NULL) { pcmk__rsc_trace(instance, "%s is not a compatible instance " "(not assigned to a node)", instance->id); return false; } if (!pcmk__same_node(instance_node, node)) { pcmk__rsc_trace(instance, "%s is not a compatible instance " "(assigned to %s not %s)", instance->id, pcmk__node_name(instance_node), pcmk__node_name(node)); return false; } return true; } #define display_role(r) \ (((r) == pcmk_role_unknown)? "matching" : pcmk_role_text(r)) /*! * \internal * \brief Find an instance that matches a given resource by node and role * * \param[in] match_rsc Resource that instance must match (for logging only) * \param[in] rsc Clone or bundle resource to check for matching instance * \param[in] node Instance must match this node * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return \p rsc instance matching \p node and \p role if any, otherwise NULL */ static pcmk_resource_t * find_compatible_instance_on_node(const pcmk_resource_t *match_rsc, const pcmk_resource_t *rsc, const pcmk_node_t *node, enum rsc_role_e role, bool current) { GList *instances = NULL; instances = get_instance_list(rsc); for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; if (pcmk__instance_matches(instance, node, role, current)) { pcmk__rsc_trace(match_rsc, "Found %s %s instance %s compatible with %s on %s", display_role(role), rsc->id, instance->id, match_rsc->id, pcmk__node_name(node)); free_instance_list(rsc, instances); // Only frees list, not contents return instance; } } free_instance_list(rsc, instances); pcmk__rsc_trace(match_rsc, "No %s %s instance found compatible with %s on %s", display_role(role), rsc->id, match_rsc->id, pcmk__node_name(node)); return NULL; } /*! * \internal * \brief Find a clone instance or bundle container compatible with a resource * * \param[in] match_rsc Resource that instance must match * \param[in] rsc Clone or bundle resource to check for matching instance * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return Compatible (by \p role and \p match_rsc location) instance of \p rsc * if any, otherwise NULL */ pcmk_resource_t * pcmk__find_compatible_instance(const pcmk_resource_t *match_rsc, const pcmk_resource_t *rsc, enum rsc_role_e role, bool current) { pcmk_resource_t *instance = NULL; GList *nodes = NULL; const pcmk_node_t *node = NULL; // If match_rsc has a node, check only that node node = match_rsc->fns->location(match_rsc, NULL, current); if (node != NULL) { return find_compatible_instance_on_node(match_rsc, rsc, node, role, current); } // Otherwise check for an instance matching any of match_rsc's allowed nodes nodes = pcmk__sort_nodes(g_hash_table_get_values(match_rsc->allowed_nodes), NULL); for (GList *iter = nodes; (iter != NULL) && (instance == NULL); iter = iter->next) { instance = find_compatible_instance_on_node(match_rsc, rsc, (pcmk_node_t *) iter->data, role, current); } if (instance == NULL) { pcmk__rsc_debug(rsc, "No %s instance found compatible with %s", rsc->id, match_rsc->id); } g_list_free(nodes); return instance; } /*! * \internal * \brief Unassign an instance if mandatory ordering has no interleave match * * \param[in] first 'First' action in an ordering * \param[in] then 'Then' action in an ordering * \param[in,out] then_instance 'Then' instance that has no interleave match * \param[in] type Group of enum pcmk__action_relation_flags * \param[in] current If true, "then" action is stopped or demoted * * \return true if \p then_instance was unassigned, otherwise false */ static bool unassign_if_mandatory(const pcmk_action_t *first, const pcmk_action_t *then, pcmk_resource_t *then_instance, uint32_t type, bool current) { // Allow "then" instance to go down even without an interleave match if (current) { pcmk__rsc_trace(then->rsc, "%s has no instance to order before stopping " "or demoting %s", first->rsc->id, then_instance->id); /* If the "first" action must be runnable, but there is no "first" * instance, the "then" instance must not be allowed to come up. */ } else if (pcmk_any_flags_set(type, pcmk__ar_unrunnable_first_blocks |pcmk__ar_first_implies_then)) { pcmk__rsc_info(then->rsc, "Inhibiting %s from being active " "because there is no %s instance to interleave", then_instance->id, first->rsc->id); return pcmk__assign_resource(then_instance, NULL, true, true); } return false; } /*! * \internal * \brief Find first matching action for a clone instance or bundle container * * \param[in] action Action in an interleaved ordering * \param[in] instance Clone instance or bundle container being interleaved * \param[in] action_name Action to look for * \param[in] node If not NULL, require action to be on this node * \param[in] for_first If true, \p instance is the 'first' resource in the * ordering, otherwise it is the 'then' resource * * \return First action for \p instance (or in some cases if \p instance is a * bundle container, its containerized resource) that matches * \p action_name and \p node if any, otherwise NULL */ static pcmk_action_t * find_instance_action(const pcmk_action_t *action, const pcmk_resource_t *instance, const char *action_name, const pcmk_node_t *node, bool for_first) { const pcmk_resource_t *rsc = NULL; pcmk_action_t *matching_action = NULL; /* If instance is a bundle container, sometimes we should interleave the * action for the container itself, and sometimes for the containerized * resource. * * For example, given "start bundle A then bundle B", B likely requires the * service inside A's container to be active, rather than just the * container, so we should interleave the action for A's containerized * resource. On the other hand, it's possible B's container itself requires * something from A, so we should interleave the action for B's container. * * Essentially, for 'first', we should use the containerized resource for * everything except stop, and for 'then', we should use the container for * everything except promote and demote (which can only be performed on the * containerized resource). */ if ((for_first && !pcmk__str_any_of(action->task, PCMK_ACTION_STOP, PCMK_ACTION_STOPPED, NULL)) || (!for_first && pcmk__str_any_of(action->task, PCMK_ACTION_PROMOTE, PCMK_ACTION_PROMOTED, PCMK_ACTION_DEMOTE, PCMK_ACTION_DEMOTED, NULL))) { rsc = pe__get_rsc_in_container(instance); } if (rsc == NULL) { rsc = instance; // No containerized resource, use instance itself } else { node = NULL; // Containerized actions are on bundle-created guest } matching_action = find_first_action(rsc->actions, NULL, action_name, node); if (matching_action != NULL) { return matching_action; } if (pcmk_is_set(instance->flags, pcmk_rsc_removed) || pcmk__str_any_of(action_name, PCMK_ACTION_STOP, PCMK_ACTION_DEMOTE, NULL)) { crm_trace("No %s action found for %s%s", action_name, pcmk_is_set(instance->flags, pcmk_rsc_removed)? "orphan " : "", instance->id); } else { crm_err("No %s action found for %s to interleave (bug?)", action_name, instance->id); } return NULL; } /*! * \internal * \brief Get the original action name of a bundle or clone action * * Given an action for a bundle or clone, get the original action name, * mapping notify to the action being notified, and if the instances are * primitives, mapping completion actions to the action that was completed * (for example, stopped to stop). * * \param[in] action Clone or bundle action to check * * \return Original action name for \p action */ static const char * orig_action_name(const pcmk_action_t *action) { // Any instance will do const pcmk_resource_t *instance = action->rsc->children->data; char *action_type = NULL; const char *action_name = action->task; enum action_tasks orig_task = pcmk_action_unspecified; if (pcmk__strcase_any_of(action->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_NOTIFIED, NULL)) { // action->uuid is RSC_(confirmed-){pre,post}_notify_ACTION_INTERVAL CRM_CHECK(parse_op_key(action->uuid, NULL, &action_type, NULL), return pcmk_action_text(pcmk_action_unspecified)); action_name = strstr(action_type, "_notify_"); CRM_CHECK(action_name != NULL, return pcmk_action_text(pcmk_action_unspecified)); action_name += strlen("_notify_"); } orig_task = get_complex_task(instance, action_name); free(action_type); return pcmk_action_text(orig_task); } /*! * \internal * \brief Update two interleaved actions according to an ordering between them * * Given information about an ordering of two interleaved actions, update the * actions' flags (and runnable_before members if appropriate) as appropriate * for the ordering. Effects may cascade to other orderings involving the * actions as well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this node * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk_action_optional to affect only * mandatory actions, and pcmk_action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * * \return Group of enum pcmk__updated flags indicating what was updated */ static uint32_t update_interleaved_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t filter, uint32_t type) { GList *instances = NULL; uint32_t changed = pcmk__updated_none; const char *orig_first_task = orig_action_name(first); // Stops and demotes must be interleaved with instance on current node bool current = pcmk__ends_with(first->uuid, "_" PCMK_ACTION_STOPPED "_0") || pcmk__ends_with(first->uuid, "_" PCMK_ACTION_DEMOTED "_0"); // Update the specified actions for each "then" instance individually instances = get_instance_list(then->rsc); for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *first_instance = NULL; pcmk_resource_t *then_instance = iter->data; pcmk_action_t *first_action = NULL; pcmk_action_t *then_action = NULL; // Find a "first" instance to interleave with this "then" instance first_instance = pcmk__find_compatible_instance(then_instance, first->rsc, pcmk_role_unknown, current); if (first_instance == NULL) { // No instance can be interleaved if (unassign_if_mandatory(first, then, then_instance, type, current)) { pcmk__set_updated_flags(changed, first, pcmk__updated_then); } continue; } first_action = find_instance_action(first, first_instance, orig_first_task, node, true); if (first_action == NULL) { continue; } then_action = find_instance_action(then, then_instance, then->task, node, false); if (then_action == NULL) { continue; } if (order_actions(first_action, then_action, type)) { pcmk__set_updated_flags(changed, first, pcmk__updated_first|pcmk__updated_then); } changed |= then_instance->cmds->update_ordered_actions( first_action, then_action, node, first_instance->cmds->action_flags(first_action, node), filter, type, then->rsc->cluster); } free_instance_list(then->rsc, instances); return changed; } /*! * \internal * \brief Check whether two actions in an ordering can be interleaved * * \param[in] first 'First' action in the ordering * \param[in] then 'Then' action in the ordering * * \return true if \p first and \p then can be interleaved, otherwise false */ static bool can_interleave_actions(const pcmk_action_t *first, const pcmk_action_t *then) { bool interleave = false; pcmk_resource_t *rsc = NULL; if ((first->rsc == NULL) || (then->rsc == NULL)) { crm_trace("Not interleaving %s with %s: not resource actions", first->uuid, then->uuid); return false; } if (first->rsc == then->rsc) { crm_trace("Not interleaving %s with %s: same resource", first->uuid, then->uuid); return false; } if ((first->rsc->variant < pcmk_rsc_variant_clone) || (then->rsc->variant < pcmk_rsc_variant_clone)) { crm_trace("Not interleaving %s with %s: not clones or bundles", first->uuid, then->uuid); return false; } if (pcmk__ends_with(then->uuid, "_stop_0") || pcmk__ends_with(then->uuid, "_demote_0")) { rsc = first->rsc; } else { rsc = then->rsc; } interleave = crm_is_true(g_hash_table_lookup(rsc->meta, PCMK_META_INTERLEAVE)); pcmk__rsc_trace(rsc, "'%s then %s' will %sbe interleaved (based on %s)", first->uuid, then->uuid, (interleave? "" : "not "), rsc->id); return interleave; } /*! * \internal * \brief Update non-interleaved instance actions according to an ordering * * Given information about an ordering of two non-interleaved actions, update * the actions' flags (and runnable_before members if appropriate) as * appropriate for the ordering. Effects may cascade to other orderings * involving the actions as well. * * \param[in,out] instance Clone instance or bundle container * \param[in,out] first "First" action in ordering * \param[in] then "Then" action in ordering (for \p instance's parent) * \param[in] node If not NULL, limit scope of ordering to this node * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk_action_optional to affect only * mandatory actions, and pcmk_action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * * \return Group of enum pcmk__updated flags indicating what was updated */ static uint32_t update_noninterleaved_actions(pcmk_resource_t *instance, pcmk_action_t *first, const pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type) { pcmk_action_t *instance_action = NULL; uint32_t instance_flags = 0; uint32_t changed = pcmk__updated_none; // Check whether instance has an equivalent of "then" action instance_action = find_first_action(instance->actions, NULL, then->task, node); if (instance_action == NULL) { return changed; } // Check whether action is runnable instance_flags = instance->cmds->action_flags(instance_action, node); if (!pcmk_is_set(instance_flags, pcmk_action_runnable)) { return changed; } // If so, update actions for the instance changed = instance->cmds->update_ordered_actions(first, instance_action, node, flags, filter, type, instance->cluster); // Propagate any changes to later actions if (pcmk_is_set(changed, pcmk__updated_then)) { for (GList *after_iter = instance_action->actions_after; after_iter != NULL; after_iter = after_iter->next) { pcmk__related_action_t *after = after_iter->data; pcmk__update_action_for_orderings(after->action, instance->cluster); } } return changed; } /*! * \internal * \brief Update two actions according to an ordering between them * * Given information about an ordering of two clone or bundle actions, update * the actions' flags (and runnable_before members if appropriate) as * appropriate for the ordering. Effects may cascade to other orderings * involving the actions as well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this node * (only used when interleaving instances) * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk_action_optional to affect only * mandatory actions, and pcmk_action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t pcmk__instance_update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler) { CRM_ASSERT((first != NULL) && (then != NULL) && (scheduler != NULL)); if (then->rsc == NULL) { return pcmk__updated_none; } else if (can_interleave_actions(first, then)) { return update_interleaved_actions(first, then, node, filter, type); } else { uint32_t changed = pcmk__updated_none; GList *instances = get_instance_list(then->rsc); // Update actions for the clone or bundle resource itself changed |= pcmk__update_ordered_actions(first, then, node, flags, filter, type, scheduler); // Update the 'then' clone instances or bundle containers individually for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = iter->data; changed |= update_noninterleaved_actions(instance, first, then, node, flags, filter, type); } free_instance_list(then->rsc, instances); return changed; } } #define pe__clear_action_summary_flags(flags, action, flag) do { \ flags = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Action summary", action->rsc->id, \ flags, flag, #flag); \ } while (0) /*! * \internal * \brief Return action flags for a given clone or bundle action * * \param[in,out] action Action for a clone or bundle * \param[in] instances Clone instances or bundle containers * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__collective_action_flags(pcmk_action_t *action, const GList *instances, const pcmk_node_t *node) { bool any_runnable = false; const char *action_name = orig_action_name(action); // Set original assumptions (optional and runnable may be cleared below) uint32_t flags = pcmk_action_optional |pcmk_action_runnable |pcmk_action_pseudo; for (const GList *iter = instances; iter != NULL; iter = iter->next) { const pcmk_resource_t *instance = iter->data; const pcmk_node_t *instance_node = NULL; pcmk_action_t *instance_action = NULL; uint32_t instance_flags; // Node is relevant only to primitive instances if (instance->variant == pcmk_rsc_variant_primitive) { instance_node = node; } instance_action = find_first_action(instance->actions, NULL, action_name, instance_node); if (instance_action == NULL) { pcmk__rsc_trace(action->rsc, "%s has no %s action on %s", instance->id, action_name, pcmk__node_name(node)); continue; } pcmk__rsc_trace(action->rsc, "%s has %s for %s on %s", instance->id, instance_action->uuid, action_name, pcmk__node_name(node)); instance_flags = instance->cmds->action_flags(instance_action, node); // If any instance action is mandatory, so is the collective action if (pcmk_is_set(flags, pcmk_action_optional) && !pcmk_is_set(instance_flags, pcmk_action_optional)) { pcmk__rsc_trace(instance, "%s is mandatory because %s is", action->uuid, instance_action->uuid); pe__clear_action_summary_flags(flags, action, pcmk_action_optional); pcmk__clear_action_flags(action, pcmk_action_optional); } // If any instance action is runnable, so is the collective action if (pcmk_is_set(instance_flags, pcmk_action_runnable)) { any_runnable = true; } } if (!any_runnable) { pcmk__rsc_trace(action->rsc, "%s is not runnable because no instance can run %s", action->uuid, action_name); pe__clear_action_summary_flags(flags, action, pcmk_action_runnable); if (node == NULL) { pcmk__clear_action_flags(action, pcmk_action_runnable); } } return flags; } diff --git a/lib/pacemaker/pcmk_sched_location.c b/lib/pacemaker/pcmk_sched_location.c index 7db4182055..dd9884b6ac 100644 --- a/lib/pacemaker/pcmk_sched_location.c +++ b/lib/pacemaker/pcmk_sched_location.c @@ -1,731 +1,731 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include "libpacemaker_private.h" static int get_node_score(const char *rule, const char *score, bool raw, pcmk_node_t *node, pcmk_resource_t *rsc) { int score_f = 0; if (score == NULL) { pcmk__config_warn("Rule %s: no score specified (assuming 0)", rule); } else if (raw) { score_f = char2score(score); } else { const char *target = NULL; const char *attr_score = NULL; target = g_hash_table_lookup(rsc->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); attr_score = pcmk__node_attr(node, score, target, pcmk__rsc_node_current); if (attr_score == NULL) { crm_debug("Rule %s: %s did not have a value for %s", rule, pcmk__node_name(node), score); - score_f = -INFINITY; + score_f = -PCMK_SCORE_INFINITY; } else { crm_debug("Rule %s: %s had value %s for %s", rule, pcmk__node_name(node), attr_score, score); score_f = char2score(attr_score); } } return score_f; } /*! * \internal * \brief Parse a role configuration for a location constraint * * \param[in] role_spec Role specification * \param[out] role Where to store parsed role * * \return true if role specification is valid, otherwise false */ static bool parse_location_role(const char *role_spec, enum rsc_role_e *role) { if (role_spec == NULL) { *role = pcmk_role_unknown; return true; } *role = pcmk_parse_role(role_spec); switch (*role) { case pcmk_role_unknown: return false; case pcmk_role_started: case pcmk_role_unpromoted: /* Any promotable clone instance cannot be promoted without being in * the unpromoted role first. Therefore, any constraint for the * started or unpromoted role applies to every role. */ *role = pcmk_role_unknown; break; default: break; } return true; } /*! * \internal * \brief Generate a location constraint from a rule * * \param[in,out] rsc Resource that constraint is for * \param[in] rule_xml Rule XML (sub-element of location constraint) * \param[in] discovery Value of \c PCMK_XA_RESOURCE_DISCOVERY for * constraint * \param[out] next_change Where to set when rule evaluation will change * \param[in] re_match_data Regular expression submatches * * \return New location constraint if rule is valid, otherwise NULL */ static pcmk__location_t * generate_location_rule(pcmk_resource_t *rsc, xmlNode *rule_xml, const char *discovery, crm_time_t *next_change, pe_re_match_data_t *re_match_data) { const char *rule_id = NULL; const char *score = NULL; const char *boolean = NULL; const char *role_spec = NULL; GList *iter = NULL; GList *nodes = NULL; bool do_and = true; bool accept = true; bool raw_score = true; bool score_allocated = false; pcmk__location_t *location_rule = NULL; enum rsc_role_e role = pcmk_role_unknown; rule_xml = expand_idref(rule_xml, rsc->cluster->input); if (rule_xml == NULL) { return NULL; // Error already logged } rule_id = crm_element_value(rule_xml, PCMK_XA_ID); boolean = crm_element_value(rule_xml, PCMK_XA_BOOLEAN_OP); role_spec = crm_element_value(rule_xml, PCMK_XA_ROLE); if (parse_location_role(role_spec, &role)) { crm_trace("Setting rule %s role filter to %s", rule_id, role_spec); } else { pcmk__config_err("Ignoring rule %s: Invalid " PCMK_XA_ROLE " '%s'", rule_id, role_spec); return NULL; } crm_trace("Processing location constraint rule %s", rule_id); score = crm_element_value(rule_xml, PCMK_XA_SCORE); if (score == NULL) { score = crm_element_value(rule_xml, PCMK_XA_SCORE_ATTRIBUTE); if (score != NULL) { raw_score = false; } } if (pcmk__str_eq(boolean, PCMK_VALUE_OR, pcmk__str_casei)) { do_and = false; } else if (!pcmk__str_eq(boolean, PCMK_VALUE_AND, pcmk__str_null_matches|pcmk__str_casei)) { pcmk__config_warn("Location constraint rule %s has invalid " PCMK_XA_BOOLEAN_OP " value '%s', using default " "('" PCMK_VALUE_AND "')", rule_id, boolean); } location_rule = pcmk__new_location(rule_id, rsc, 0, discovery, NULL); if (location_rule == NULL) { return NULL; // Error already logged } location_rule->role_filter = role; if ((re_match_data != NULL) && (re_match_data->nregs > 0) && (re_match_data->pmatch[0].rm_so != -1) && !raw_score) { char *result = pe_expand_re_matches(score, re_match_data); if (result != NULL) { score = result; score_allocated = true; } } if (do_and) { nodes = pcmk__copy_node_list(rsc->cluster->nodes, true); for (iter = nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = iter->data; node->weight = get_node_score(rule_id, score, raw_score, node, rsc); } } for (iter = rsc->cluster->nodes; iter != NULL; iter = iter->next) { int score_f = 0; pcmk_node_t *node = iter->data; pe_match_data_t match_data = { .re = re_match_data, .params = pe_rsc_params(rsc, node, rsc->cluster), .meta = rsc->meta, }; accept = pe_test_rule(rule_xml, node->details->attrs, pcmk_role_unknown, rsc->cluster->now, next_change, &match_data); crm_trace("Rule %s %s on %s", pcmk__xe_id(rule_xml), (accept? "passed" : "failed"), pcmk__node_name(node)); score_f = get_node_score(rule_id, score, raw_score, node, rsc); if (accept) { pcmk_node_t *local = pe_find_node_id(nodes, node->details->id); if ((local == NULL) && do_and) { continue; } else if (local == NULL) { local = pe__copy_node(node); nodes = g_list_append(nodes, local); } if (!do_and) { local->weight = pcmk__add_scores(local->weight, score_f); } crm_trace("%s has score %s after %s", pcmk__node_name(node), pcmk_readable_score(local->weight), rule_id); } else if (do_and && !accept) { // Remove it pcmk_node_t *delete = pe_find_node_id(nodes, node->details->id); if (delete != NULL) { nodes = g_list_remove(nodes, delete); crm_trace("%s did not match", pcmk__node_name(node)); } free(delete); } } if (score_allocated) { free((char *)score); } location_rule->nodes = nodes; if (location_rule->nodes == NULL) { crm_trace("No matching nodes for location constraint rule %s", rule_id); return NULL; } else { crm_trace("Location constraint rule %s matched %d nodes", rule_id, g_list_length(location_rule->nodes)); } return location_rule; } static void unpack_rsc_location(xmlNode *xml_obj, pcmk_resource_t *rsc, const char *role_spec, const char *score, pe_re_match_data_t *re_match_data) { const char *rsc_id = crm_element_value(xml_obj, PCMK_XA_RSC); const char *id = crm_element_value(xml_obj, PCMK_XA_ID); const char *node = crm_element_value(xml_obj, PCMK_XE_NODE); const char *discovery = crm_element_value(xml_obj, PCMK_XA_RESOURCE_DISCOVERY); if (rsc == NULL) { pcmk__config_warn("Ignoring constraint '%s' because resource '%s' " "does not exist", id, rsc_id); return; } if (score == NULL) { score = crm_element_value(xml_obj, PCMK_XA_SCORE); } if ((node != NULL) && (score != NULL)) { int score_i = char2score(score); pcmk_node_t *match = pe_find_node(rsc->cluster->nodes, node); enum rsc_role_e role = pcmk_role_unknown; pcmk__location_t *location = NULL; if (!match) { return; } if (role_spec == NULL) { role_spec = crm_element_value(xml_obj, PCMK_XA_ROLE); } if (parse_location_role(role_spec, &role)) { crm_trace("Setting location constraint %s role filter: %s", id, role_spec); } else { /* @COMPAT The previous behavior of creating the constraint ignoring * the role is retained for now, but we should ignore the entire * constraint when we can break backward compatibility. */ pcmk__config_err("Ignoring role in constraint %s: " "Invalid value '%s'", id, role_spec); } location = pcmk__new_location(id, rsc, score_i, discovery, match); if (location == NULL) { return; // Error already logged } location->role_filter = role; } else { bool empty = true; crm_time_t *next_change = crm_time_new_undefined(); /* This loop is logically parallel to pe_evaluate_rules(), except * instead of checking whether any rule is active, we set up location * constraints for each active rule. */ for (xmlNode *rule_xml = first_named_child(xml_obj, PCMK_XE_RULE); rule_xml != NULL; rule_xml = crm_next_same_xml(rule_xml)) { empty = false; crm_trace("Unpacking %s/%s", id, pcmk__xe_id(rule_xml)); generate_location_rule(rsc, rule_xml, discovery, next_change, re_match_data); } if (empty) { pcmk__config_err("Ignoring constraint '%s' because it contains " "no rules", id); } /* If there is a point in the future when the evaluation of a rule will * change, make sure the scheduler is re-run by that time. */ if (crm_time_is_defined(next_change)) { time_t t = (time_t) crm_time_get_seconds_since_epoch(next_change); pe__update_recheck_time(t, rsc->cluster, "location rule evaluation"); } crm_time_free(next_change); } } static void unpack_simple_location(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { const char *id = crm_element_value(xml_obj, PCMK_XA_ID); const char *value = crm_element_value(xml_obj, PCMK_XA_RSC); if (value) { pcmk_resource_t *rsc; rsc = pcmk__find_constraint_resource(scheduler->resources, value); unpack_rsc_location(xml_obj, rsc, NULL, NULL, NULL); } value = crm_element_value(xml_obj, PCMK_XA_RSC_PATTERN); if (value) { regex_t *r_patt = calloc(1, sizeof(regex_t)); bool invert = false; if (value[0] == '!') { value++; invert = true; } if (regcomp(r_patt, value, REG_EXTENDED) != 0) { pcmk__config_err("Ignoring constraint '%s' because " PCMK_XA_RSC_PATTERN " has invalid value '%s'", id, value); free(r_patt); return; } for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *r = iter->data; int nregs = 0; regmatch_t *pmatch = NULL; int status; if (r_patt->re_nsub > 0) { nregs = r_patt->re_nsub + 1; } else { nregs = 1; } pmatch = calloc(nregs, sizeof(regmatch_t)); status = regexec(r_patt, r->id, nregs, pmatch, 0); if (!invert && (status == 0)) { pe_re_match_data_t re_match_data = { .string = r->id, .nregs = nregs, .pmatch = pmatch }; crm_debug("'%s' matched '%s' for %s", r->id, value, id); unpack_rsc_location(xml_obj, r, NULL, NULL, &re_match_data); } else if (invert && (status != 0)) { crm_debug("'%s' is an inverted match of '%s' for %s", r->id, value, id); unpack_rsc_location(xml_obj, r, NULL, NULL, NULL); } else { crm_trace("'%s' does not match '%s' for %s", r->id, value, id); } free(pmatch); } regfree(r_patt); free(r_patt); } } // \return Standard Pacemaker return code static int unpack_location_tags(xmlNode *xml_obj, xmlNode **expanded_xml, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *rsc_id = NULL; const char *state = NULL; pcmk_resource_t *rsc = NULL; pcmk_tag_t *tag = NULL; xmlNode *rsc_set = NULL; *expanded_xml = NULL; CRM_CHECK(xml_obj != NULL, return EINVAL); id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("Ignoring <%s> constraint without " PCMK_XA_ID, xml_obj->name); return pcmk_rc_unpack_error; } // Check whether there are any resource sets with template or tag references *expanded_xml = pcmk__expand_tags_in_sets(xml_obj, scheduler); if (*expanded_xml != NULL) { crm_log_xml_trace(*expanded_xml, "Expanded " PCMK_XE_RSC_LOCATION); return pcmk_rc_ok; } rsc_id = crm_element_value(xml_obj, PCMK_XA_RSC); if (rsc_id == NULL) { return pcmk_rc_ok; } if (!pcmk__valid_resource_or_tag(scheduler, rsc_id, &rsc, &tag)) { pcmk__config_err("Ignoring constraint '%s' because '%s' is not a " "valid resource or tag", id, rsc_id); return pcmk_rc_unpack_error; } else if (rsc != NULL) { // No template is referenced return pcmk_rc_ok; } state = crm_element_value(xml_obj, PCMK_XA_ROLE); *expanded_xml = copy_xml(xml_obj); /* Convert any template or tag reference into constraint * PCMK_XE_RESOURCE_SET */ if (!pcmk__tag_to_set(*expanded_xml, &rsc_set, PCMK_XA_RSC, false, scheduler)) { free_xml(*expanded_xml); *expanded_xml = NULL; return pcmk_rc_unpack_error; } if (rsc_set != NULL) { if (state != NULL) { /* Move PCMK_XA_RSC_ROLE into converted PCMK_XE_RESOURCE_SET as * PCMK_XA_ROLE attribute */ crm_xml_add(rsc_set, PCMK_XA_ROLE, state); xml_remove_prop(*expanded_xml, PCMK_XA_ROLE); } crm_log_xml_trace(*expanded_xml, "Expanded " PCMK_XE_RSC_LOCATION); } else { // No sets free_xml(*expanded_xml); *expanded_xml = NULL; } return pcmk_rc_ok; } // \return Standard Pacemaker return code static int unpack_location_set(xmlNode *location, xmlNode *set, pcmk_scheduler_t *scheduler) { xmlNode *xml_rsc = NULL; pcmk_resource_t *resource = NULL; const char *set_id; const char *role; const char *local_score; CRM_CHECK(set != NULL, return EINVAL); set_id = pcmk__xe_id(set); if (set_id == NULL) { pcmk__config_err("Ignoring " PCMK_XE_RESOURCE_SET " without " PCMK_XA_ID " in constraint '%s'", pcmk__s(pcmk__xe_id(location), "(missing ID)")); return pcmk_rc_unpack_error; } role = crm_element_value(set, PCMK_XA_ROLE); local_score = crm_element_value(set, PCMK_XA_SCORE); for (xml_rsc = first_named_child(set, PCMK_XE_RESOURCE_REF); xml_rsc != NULL; xml_rsc = crm_next_same_xml(xml_rsc)) { resource = pcmk__find_constraint_resource(scheduler->resources, pcmk__xe_id(xml_rsc)); if (resource == NULL) { pcmk__config_err("%s: No resource found for %s", set_id, pcmk__xe_id(xml_rsc)); return pcmk_rc_unpack_error; } unpack_rsc_location(location, resource, role, local_score, NULL); } return pcmk_rc_ok; } void pcmk__unpack_location(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { xmlNode *set = NULL; bool any_sets = false; xmlNode *orig_xml = NULL; xmlNode *expanded_xml = NULL; if (unpack_location_tags(xml_obj, &expanded_xml, scheduler) != pcmk_rc_ok) { return; } if (expanded_xml) { orig_xml = xml_obj; xml_obj = expanded_xml; } for (set = first_named_child(xml_obj, PCMK_XE_RESOURCE_SET); set != NULL; set = crm_next_same_xml(set)) { any_sets = true; set = expand_idref(set, scheduler->input); if ((set == NULL) // Configuration error, message already logged || (unpack_location_set(xml_obj, set, scheduler) != pcmk_rc_ok)) { if (expanded_xml) { free_xml(expanded_xml); } return; } } if (expanded_xml) { free_xml(expanded_xml); xml_obj = orig_xml; } if (!any_sets) { unpack_simple_location(xml_obj, scheduler); } } /*! * \internal * \brief Add a new location constraint to scheduler data * * \param[in] id XML ID of location constraint * \param[in,out] rsc Resource in location constraint * \param[in] node_score Constraint score * \param[in] discover_mode Resource discovery option for constraint * \param[in] node Node in constraint (or NULL if rule-based) * * \return Newly allocated location constraint * \note The result will be added to the cluster (via \p rsc) and should not be * freed separately. */ pcmk__location_t * pcmk__new_location(const char *id, pcmk_resource_t *rsc, int node_score, const char *discover_mode, pcmk_node_t *node) { pcmk__location_t *new_con = NULL; if (id == NULL) { pcmk__config_err("Invalid constraint: no ID specified"); return NULL; } else if (rsc == NULL) { pcmk__config_err("Invalid constraint %s: no resource specified", id); return NULL; } else if (node == NULL) { CRM_CHECK(node_score == 0, return NULL); } new_con = calloc(1, sizeof(pcmk__location_t)); if (new_con != NULL) { new_con->id = strdup(id); new_con->rsc = rsc; new_con->nodes = NULL; new_con->role_filter = pcmk_role_unknown; if (pcmk__str_eq(discover_mode, PCMK_VALUE_ALWAYS, pcmk__str_null_matches|pcmk__str_casei)) { new_con->discover_mode = pcmk_probe_always; } else if (pcmk__str_eq(discover_mode, PCMK_VALUE_NEVER, pcmk__str_casei)) { new_con->discover_mode = pcmk_probe_never; } else if (pcmk__str_eq(discover_mode, PCMK_VALUE_EXCLUSIVE, pcmk__str_casei)) { new_con->discover_mode = pcmk_probe_exclusive; rsc->exclusive_discover = TRUE; } else { pcmk__config_err("Invalid " PCMK_XA_RESOURCE_DISCOVERY " value %s " "in location constraint", discover_mode); } if (node != NULL) { pcmk_node_t *copy = pe__copy_node(node); copy->weight = node_score; new_con->nodes = g_list_prepend(NULL, copy); } rsc->cluster->placement_constraints = g_list_prepend( rsc->cluster->placement_constraints, new_con); rsc->rsc_location = g_list_prepend(rsc->rsc_location, new_con); } return new_con; } /*! * \internal * \brief Apply all location constraints * * \param[in,out] scheduler Scheduler data */ void pcmk__apply_locations(pcmk_scheduler_t *scheduler) { for (GList *iter = scheduler->placement_constraints; iter != NULL; iter = iter->next) { pcmk__location_t *location = iter->data; location->rsc->cmds->apply_location(location->rsc, location); } } /*! * \internal * \brief Apply a location constraint to a resource's allowed node scores * * \param[in,out] rsc Resource to apply constraint to * \param[in,out] location Location constraint to apply * * \note This does not consider the resource's children, so the resource's * apply_location() method should be used instead in most cases. */ void pcmk__apply_location(pcmk_resource_t *rsc, pcmk__location_t *location) { bool need_role = false; CRM_ASSERT((rsc != NULL) && (location != NULL)); // If a role was specified, ensure constraint is applicable need_role = (location->role_filter > pcmk_role_unknown); if (need_role && (location->role_filter != rsc->next_role)) { pcmk__rsc_trace(rsc, "Not applying %s to %s because role will be %s not %s", location->id, rsc->id, pcmk_role_text(rsc->next_role), pcmk_role_text(location->role_filter)); return; } if (location->nodes == NULL) { pcmk__rsc_trace(rsc, "Not applying %s to %s because no nodes match", location->id, rsc->id); return; } pcmk__rsc_trace(rsc, "Applying %s%s%s to %s", location->id, (need_role? " for role " : ""), (need_role? pcmk_role_text(location->role_filter) : ""), rsc->id); for (GList *iter = location->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = iter->data; pcmk_node_t *allowed_node = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (allowed_node == NULL) { pcmk__rsc_trace(rsc, "* = %d on %s", node->weight, pcmk__node_name(node)); allowed_node = pe__copy_node(node); g_hash_table_insert(rsc->allowed_nodes, (gpointer) allowed_node->details->id, allowed_node); } else { pcmk__rsc_trace(rsc, "* + %d on %s", node->weight, pcmk__node_name(node)); allowed_node->weight = pcmk__add_scores(allowed_node->weight, node->weight); } if (allowed_node->rsc_discover_mode < location->discover_mode) { if (location->discover_mode == pcmk_probe_exclusive) { rsc->exclusive_discover = TRUE; } /* exclusive > never > always... always is default */ allowed_node->rsc_discover_mode = location->discover_mode; } } } diff --git a/lib/pacemaker/pcmk_sched_nodes.c b/lib/pacemaker/pcmk_sched_nodes.c index b02d553ca0..bddd23ebd2 100644 --- a/lib/pacemaker/pcmk_sched_nodes.c +++ b/lib/pacemaker/pcmk_sched_nodes.c @@ -1,437 +1,437 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Check whether a node is available to run resources * * \param[in] node Node to check * \param[in] consider_score If true, consider a negative score unavailable * \param[in] consider_guest If true, consider a guest node unavailable whose * resource will not be active * * \return true if node is online and not shutting down, unclean, or in standby * or maintenance mode, otherwise false */ bool pcmk__node_available(const pcmk_node_t *node, bool consider_score, bool consider_guest) { if ((node == NULL) || (node->details == NULL) || !node->details->online || node->details->shutdown || node->details->unclean || node->details->standby || node->details->maintenance) { return false; } if (consider_score && (node->weight < 0)) { return false; } // @TODO Go through all callers to see which should set consider_guest if (consider_guest && pcmk__is_guest_or_bundle_node(node)) { pcmk_resource_t *guest = node->details->remote_rsc->container; if (guest->fns->location(guest, NULL, FALSE) == NULL) { return false; } } return true; } /*! * \internal * \brief Copy a hash table of node objects * * \param[in] nodes Hash table to copy * * \return New copy of nodes (or NULL if nodes is NULL) */ GHashTable * pcmk__copy_node_table(GHashTable *nodes) { GHashTable *new_table = NULL; GHashTableIter iter; pcmk_node_t *node = NULL; if (nodes == NULL) { return NULL; } new_table = pcmk__strkey_table(NULL, free); g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { pcmk_node_t *new_node = pe__copy_node(node); g_hash_table_insert(new_table, (gpointer) new_node->details->id, new_node); } return new_table; } /*! * \internal * \brief Free a table of node tables * * \param[in,out] data Table to free * * \note This is a \c GDestroyNotify wrapper for \c g_hash_table_destroy(). */ static void destroy_node_tables(gpointer data) { g_hash_table_destroy((GHashTable *) data); } /*! * \internal * \brief Recursively copy the node tables of a resource * * Build a hash table containing copies of the allowed nodes tables of \p rsc * and its entire tree of descendants. The key is the resource ID, and the value * is a copy of the resource's node table. * * \param[in] rsc Resource whose node table to copy * \param[in,out] copy Where to store the copied node tables * * \note \p *copy should be \c NULL for the top-level call. * \note The caller is responsible for freeing \p copy using * \c g_hash_table_destroy(). */ void pcmk__copy_node_tables(const pcmk_resource_t *rsc, GHashTable **copy) { CRM_ASSERT((rsc != NULL) && (copy != NULL)); if (*copy == NULL) { *copy = pcmk__strkey_table(NULL, destroy_node_tables); } g_hash_table_insert(*copy, rsc->id, pcmk__copy_node_table(rsc->allowed_nodes)); for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk__copy_node_tables((const pcmk_resource_t *) iter->data, copy); } } /*! * \internal * \brief Recursively restore the node tables of a resource from backup * * Given a hash table containing backup copies of the allowed nodes tables of * \p rsc and its entire tree of descendants, replace the resources' current * node tables with the backed-up copies. * * \param[in,out] rsc Resource whose node tables to restore * \param[in] backup Table of backup node tables (created by * \c pcmk__copy_node_tables()) * * \note This function frees the resources' current node tables. */ void pcmk__restore_node_tables(pcmk_resource_t *rsc, GHashTable *backup) { CRM_ASSERT((rsc != NULL) && (backup != NULL)); g_hash_table_destroy(rsc->allowed_nodes); // Copy to avoid danger with multiple restores rsc->allowed_nodes = g_hash_table_lookup(backup, rsc->id); rsc->allowed_nodes = pcmk__copy_node_table(rsc->allowed_nodes); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk__restore_node_tables((pcmk_resource_t *) iter->data, backup); } } /*! * \internal * \brief Copy a list of node objects * * \param[in] list List to copy * \param[in] reset Set copies' scores to 0 * * \return New list of shallow copies of nodes in original list */ GList * pcmk__copy_node_list(const GList *list, bool reset) { GList *result = NULL; for (const GList *iter = list; iter != NULL; iter = iter->next) { pcmk_node_t *new_node = NULL; pcmk_node_t *this_node = iter->data; new_node = pe__copy_node(this_node); if (reset) { new_node->weight = 0; } result = g_list_prepend(result, new_node); } return result; } /*! * \internal * \brief Compare two nodes for assignment preference * * Given two nodes, check which one is more preferred by assignment criteria * such as node score and utilization. * * \param[in] a First node to compare * \param[in] b Second node to compare * \param[in] data Node to prefer if all else equal * * \return -1 if \p a is preferred, +1 if \p b is preferred, or 0 if they are * equally preferred */ static gint compare_nodes(gconstpointer a, gconstpointer b, gpointer data) { const pcmk_node_t *node1 = (const pcmk_node_t *) a; const pcmk_node_t *node2 = (const pcmk_node_t *) b; const pcmk_node_t *preferred = (const pcmk_node_t *) data; - int node1_score = -INFINITY; - int node2_score = -INFINITY; + int node1_score = -PCMK_SCORE_INFINITY; + int node2_score = -PCMK_SCORE_INFINITY; int result = 0; if (a == NULL) { return 1; } if (b == NULL) { return -1; } // Compare node scores if (pcmk__node_available(node1, false, false)) { node1_score = node1->weight; } if (pcmk__node_available(node2, false, false)) { node2_score = node2->weight; } if (node1_score > node2_score) { crm_trace("%s before %s (score %d > %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1_score, node2_score); return -1; } if (node1_score < node2_score) { crm_trace("%s after %s (score %d < %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1_score, node2_score); return 1; } // If appropriate, compare node utilization if (pcmk__str_eq(node1->details->data_set->placement_strategy, PCMK_VALUE_MINIMAL, pcmk__str_casei)) { goto equal; } if (pcmk__str_eq(node1->details->data_set->placement_strategy, PCMK_VALUE_BALANCED, pcmk__str_casei)) { result = pcmk__compare_node_capacities(node1, node2); if (result < 0) { crm_trace("%s before %s (greater capacity by %d attributes)", pcmk__node_name(node1), pcmk__node_name(node2), result * -1); return -1; } else if (result > 0) { crm_trace("%s after %s (lower capacity by %d attributes)", pcmk__node_name(node1), pcmk__node_name(node2), result); return 1; } } // Compare number of resources already assigned to node if (node1->details->num_resources < node2->details->num_resources) { crm_trace("%s before %s (%d resources < %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1->details->num_resources, node2->details->num_resources); return -1; } else if (node1->details->num_resources > node2->details->num_resources) { crm_trace("%s after %s (%d resources > %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1->details->num_resources, node2->details->num_resources); return 1; } // Check whether one node is already running desired resource if (preferred != NULL) { if (pcmk__same_node(preferred, node1)) { crm_trace("%s before %s (preferred node)", pcmk__node_name(node1), pcmk__node_name(node2)); return -1; } else if (pcmk__same_node(preferred, node2)) { crm_trace("%s after %s (not preferred node)", pcmk__node_name(node1), pcmk__node_name(node2)); return 1; } } // If all else is equal, prefer node with lowest-sorting name equal: result = strcmp(node1->details->uname, node2->details->uname); if (result < 0) { crm_trace("%s before %s (name)", pcmk__node_name(node1), pcmk__node_name(node2)); return -1; } else if (result > 0) { crm_trace("%s after %s (name)", pcmk__node_name(node1), pcmk__node_name(node2)); return 1; } crm_trace("%s == %s", pcmk__node_name(node1), pcmk__node_name(node2)); return 0; } /*! * \internal * \brief Sort a list of nodes by assigment preference * * \param[in,out] nodes Node list to sort * \param[in] active_node Node where resource being assigned is active * * \return New head of sorted list */ GList * pcmk__sort_nodes(GList *nodes, pcmk_node_t *active_node) { return g_list_sort_with_data(nodes, compare_nodes, active_node); } /*! * \internal * \brief Check whether any node is available to run resources * * \param[in] nodes Nodes to check * * \return true if any node in \p nodes is available to run resources, * otherwise false */ bool pcmk__any_node_available(GHashTable *nodes) { GHashTableIter iter; const pcmk_node_t *node = NULL; if (nodes == NULL) { return false; } g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (pcmk__node_available(node, true, false)) { return true; } } return false; } /*! * \internal * \brief Apply node health values for all nodes in cluster * * \param[in,out] scheduler Scheduler data */ void pcmk__apply_node_health(pcmk_scheduler_t *scheduler) { int base_health = 0; enum pcmk__health_strategy strategy; const char *strategy_str = pcmk__cluster_option(scheduler->config_hash, PCMK_OPT_NODE_HEALTH_STRATEGY); strategy = pcmk__parse_health_strategy(strategy_str); if (strategy == pcmk__health_strategy_none) { return; } crm_info("Applying node health strategy '%s'", strategy_str); // The progressive strategy can use a base health score if (strategy == pcmk__health_strategy_progressive) { base_health = pe__health_score(PCMK_OPT_NODE_HEALTH_BASE, scheduler); } for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; int health = pe__sum_node_health_scores(node, base_health); // An overall health score of 0 has no effect if (health == 0) { continue; } crm_info("Overall system health of %s is %d", pcmk__node_name(node), health); // Use node health as a location score for each resource on the node for (GList *r = scheduler->resources; r != NULL; r = r->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) r->data; bool constrain = true; if (health < 0) { /* Negative health scores do not apply to resources with * PCMK_META_ALLOW_UNHEALTHY_NODES=true. */ constrain = !crm_is_true(g_hash_table_lookup(rsc->meta, PCMK_META_ALLOW_UNHEALTHY_NODES)); } if (constrain) { pcmk__new_location(strategy_str, rsc, health, NULL, node); } else { pcmk__rsc_trace(rsc, "%s is immune from health ban on %s", rsc->id, pcmk__node_name(node)); } } } } /*! * \internal * \brief Check for a node in a resource's parent's allowed nodes * * \param[in] rsc Resource whose parent should be checked * \param[in] node Node to check for * * \return Equivalent of \p node from \p rsc's parent's allowed nodes if any, * otherwise NULL */ pcmk_node_t * pcmk__top_allowed_node(const pcmk_resource_t *rsc, const pcmk_node_t *node) { GHashTable *allowed_nodes = NULL; if ((rsc == NULL) || (node == NULL)) { return NULL; } else if (rsc->parent == NULL) { allowed_nodes = rsc->allowed_nodes; } else { allowed_nodes = rsc->parent->allowed_nodes; } return g_hash_table_lookup(allowed_nodes, node->details->id); } diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c index 10a9457479..75c9e31a2c 100644 --- a/lib/pacemaker/pcmk_sched_primitive.c +++ b/lib/pacemaker/pcmk_sched_primitive.c @@ -1,1679 +1,1680 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include // uint8_t, uint32_t #include #include #include "libpacemaker_private.h" static void stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void start_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); #define RSC_ROLE_MAX (pcmk_role_promoted + 1) static enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = { /* This array lists the immediate next role when transitioning from one role * to a target role. For example, when going from Stopped to Promoted, the * next role is Unpromoted, because the resource must be started before it * can be promoted. The current state then becomes Started, which is fed * into this array again, giving a next role of Promoted. * * Current role Immediate next role Final target role * ------------ ------------------- ----------------- */ /* Unknown */ { pcmk_role_unknown, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_stopped, /* Started */ pcmk_role_stopped, /* Unpromoted */ pcmk_role_stopped, /* Promoted */ }, /* Stopped */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_started, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_unpromoted, /* Promoted */ }, /* Started */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_started, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, /* Unpromoted */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_stopped, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, /* Promoted */ { pcmk_role_stopped, /* Unknown */ pcmk_role_unpromoted, /* Stopped */ pcmk_role_unpromoted, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, }; /*! * \internal * \brief Function to schedule actions needed for a role change * * \param[in,out] rsc Resource whose role is changing * \param[in,out] node Node where resource will be in its next role * \param[in] optional Whether scheduled actions should be optional */ typedef void (*rsc_transition_fn)(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static rsc_transition_fn rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = { /* This array lists the function needed to transition directly from one role * to another. NULL indicates that nothing is needed. * * Current role Transition function Next role * ------------ ------------------- ---------- */ /* Unknown */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ assert_role_error, /* Started */ assert_role_error, /* Unpromoted */ assert_role_error, /* Promoted */ }, /* Stopped */ { assert_role_error, /* Unknown */ NULL, /* Stopped */ start_resource, /* Started */ start_resource, /* Unpromoted */ assert_role_error, /* Promoted */ }, /* Started */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ NULL, /* Started */ NULL, /* Unpromoted */ promote_resource, /* Promoted */ }, /* Unpromoted */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ stop_resource, /* Started */ NULL, /* Unpromoted */ promote_resource, /* Promoted */ }, /* Promoted */ { assert_role_error, /* Unknown */ demote_resource, /* Stopped */ demote_resource, /* Started */ demote_resource, /* Unpromoted */ NULL, /* Promoted */ }, }; /*! * \internal * \brief Get a list of a resource's allowed nodes sorted by node score * * \param[in] rsc Resource to check * * \return List of allowed nodes sorted by node score */ static GList * sorted_allowed_nodes(const pcmk_resource_t *rsc) { if (rsc->allowed_nodes != NULL) { GList *nodes = g_hash_table_get_values(rsc->allowed_nodes); if (nodes != NULL) { return pcmk__sort_nodes(nodes, pcmk__current_node(rsc)); } } return NULL; } /*! * \internal * \brief Assign a resource to its best allowed node, if possible * * \param[in,out] rsc Resource to choose a node for * \param[in] prefer If not \c NULL, prefer this node when all else * equal * \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a * node, set next role to stopped and update * existing actions * * \return true if \p rsc could be assigned to a node, otherwise false * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ static bool assign_best_node(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { GList *nodes = NULL; pcmk_node_t *chosen = NULL; pcmk_node_t *best = NULL; const pcmk_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc); if (prefer == NULL) { prefer = most_free_node; } if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { // We've already finished assignment of resources to nodes return rsc->allocated_to != NULL; } // Sort allowed nodes by score nodes = sorted_allowed_nodes(rsc); if (nodes != NULL) { best = (pcmk_node_t *) nodes->data; // First node has best score } if ((prefer != NULL) && (nodes != NULL)) { // Get the allowed node version of prefer chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id); if (chosen == NULL) { pcmk__rsc_trace(rsc, "Preferred node %s for %s was unknown", pcmk__node_name(prefer), rsc->id); /* Favor the preferred node as long as its score is at least as good as * the best allowed node's. * * An alternative would be to favor the preferred node even if the best * node is better, when the best node's score is less than INFINITY. */ } else if (chosen->weight < best->weight) { pcmk__rsc_trace(rsc, "Preferred node %s for %s was unsuitable", pcmk__node_name(chosen), rsc->id); chosen = NULL; } else if (!pcmk__node_available(chosen, true, false)) { pcmk__rsc_trace(rsc, "Preferred node %s for %s was unavailable", pcmk__node_name(chosen), rsc->id); chosen = NULL; } else { pcmk__rsc_trace(rsc, "Chose preferred node %s for %s " "(ignoring %d candidates)", pcmk__node_name(chosen), rsc->id, g_list_length(nodes)); } } if ((chosen == NULL) && (best != NULL)) { /* Either there is no preferred node, or the preferred node is not * suitable, but another node is allowed to run the resource. */ chosen = best; if (!pcmk__is_unique_clone(rsc->parent) && (chosen->weight > 0) // Zero not acceptable && pcmk__node_available(chosen, false, false)) { /* If the resource is already running on a node, prefer that node if * it is just as good as the chosen node. * * We don't do this for unique clone instances, because * pcmk__assign_instances() has already assigned instances to their * running nodes when appropriate, and if we get here, we don't want * remaining unassigned instances to prefer a node that's already * running another instance. */ pcmk_node_t *running = pcmk__current_node(rsc); if (running == NULL) { // Nothing to do } else if (!pcmk__node_available(running, true, false)) { pcmk__rsc_trace(rsc, "Current node for %s (%s) can't run resources", rsc->id, pcmk__node_name(running)); } else { int nodes_with_best_score = 1; for (GList *iter = nodes->next; iter; iter = iter->next) { pcmk_node_t *allowed = (pcmk_node_t *) iter->data; if (allowed->weight != chosen->weight) { // The nodes are sorted by score, so no more are equal break; } if (pcmk__same_node(allowed, running)) { // Scores are equal, so prefer the current node chosen = allowed; } nodes_with_best_score++; } if (nodes_with_best_score > 1) { uint8_t log_level = LOG_INFO; - if (chosen->weight >= INFINITY) { + if (chosen->weight >= PCMK_SCORE_INFINITY) { log_level = LOG_WARNING; } do_crm_log(log_level, "Chose %s for %s from %d nodes with score %s", pcmk__node_name(chosen), rsc->id, nodes_with_best_score, pcmk_readable_score(chosen->weight)); } } } pcmk__rsc_trace(rsc, "Chose %s for %s from %d candidates", pcmk__node_name(chosen), rsc->id, g_list_length(nodes)); } pcmk__assign_resource(rsc, chosen, false, stop_if_fail); g_list_free(nodes); return rsc->allocated_to != NULL; } /*! * \internal * \brief Apply a "this with" colocation to a node's allowed node scores * * \param[in,out] colocation Colocation to apply * \param[in,out] rsc Resource being assigned */ static void apply_this_with(pcmk__colocation_t *colocation, pcmk_resource_t *rsc) { GHashTable *archive = NULL; pcmk_resource_t *other = colocation->primary; // In certain cases, we will need to revert the node scores if ((colocation->dependent_role >= pcmk_role_promoted) - || ((colocation->score < 0) && (colocation->score > -INFINITY))) { + || ((colocation->score < 0) + && (colocation->score > -PCMK_SCORE_INFINITY))) { archive = pcmk__copy_node_table(rsc->allowed_nodes); } if (pcmk_is_set(other->flags, pcmk_rsc_unassigned)) { pcmk__rsc_trace(rsc, "%s: Assigning colocation %s primary %s first" "(score=%d role=%s)", rsc->id, colocation->id, other->id, colocation->score, pcmk_role_text(colocation->dependent_role)); other->cmds->assign(other, NULL, true); } // Apply the colocation score to this resource's allowed node scores rsc->cmds->apply_coloc_score(rsc, other, colocation, true); if ((archive != NULL) && !pcmk__any_node_available(rsc->allowed_nodes)) { pcmk__rsc_info(rsc, "%s: Reverting scores from colocation with %s " "because no nodes allowed", rsc->id, other->id); g_hash_table_destroy(rsc->allowed_nodes); rsc->allowed_nodes = archive; archive = NULL; } if (archive != NULL) { g_hash_table_destroy(archive); } } /*! * \internal * \brief Update a Pacemaker Remote node once its connection has been assigned * * \param[in] connection Connection resource that has been assigned */ static void remote_connection_assigned(const pcmk_resource_t *connection) { pcmk_node_t *remote_node = pe_find_node(connection->cluster->nodes, connection->id); CRM_CHECK(remote_node != NULL, return); if ((connection->allocated_to != NULL) && (connection->next_role != pcmk_role_stopped)) { crm_trace("Pacemaker Remote node %s will be online", remote_node->details->id); remote_node->details->online = TRUE; if (remote_node->details->unseen) { // Avoid unnecessary fence, since we will attempt connection remote_node->details->unclean = FALSE; } } else { crm_trace("Pacemaker Remote node %s will be shut down " "(%sassigned connection's next role is %s)", remote_node->details->id, ((connection->allocated_to == NULL)? "un" : ""), pcmk_role_text(connection->next_role)); remote_node->details->shutdown = TRUE; } } /*! * \internal * \brief Assign a primitive resource to a node * * \param[in,out] rsc Resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a * node, set next role to stopped and update * existing actions * * \return Node that \p rsc is assigned to, if assigned entirely to one node * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ pcmk_node_t * pcmk__primitive_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { GList *this_with_colocations = NULL; GList *with_this_colocations = NULL; GList *iter = NULL; pcmk__colocation_t *colocation = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); // Never assign a child without parent being assigned first if ((rsc->parent != NULL) && !pcmk_is_set(rsc->parent->flags, pcmk_rsc_assigning)) { pcmk__rsc_debug(rsc, "%s: Assigning parent %s first", rsc->id, rsc->parent->id); rsc->parent->cmds->assign(rsc->parent, prefer, stop_if_fail); } if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { // Assignment has already been done const char *node_name = "no node"; if (rsc->allocated_to != NULL) { node_name = pcmk__node_name(rsc->allocated_to); } pcmk__rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name); return rsc->allocated_to; } // Ensure we detect assignment loops if (pcmk_is_set(rsc->flags, pcmk_rsc_assigning)) { pcmk__rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id); return NULL; } pcmk__set_rsc_flags(rsc, pcmk_rsc_assigning); pe__show_node_scores(true, rsc, "Pre-assignment", rsc->allowed_nodes, rsc->cluster); this_with_colocations = pcmk__this_with_colocations(rsc); with_this_colocations = pcmk__with_this_colocations(rsc); // Apply mandatory colocations first, to satisfy as many as possible for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; - if ((colocation->score <= -CRM_SCORE_INFINITY) - || (colocation->score >= CRM_SCORE_INFINITY)) { + if ((colocation->score <= -PCMK_SCORE_INFINITY) + || (colocation->score >= PCMK_SCORE_INFINITY)) { apply_this_with(colocation, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; - if ((colocation->score <= -CRM_SCORE_INFINITY) - || (colocation->score >= CRM_SCORE_INFINITY)) { + if ((colocation->score <= -PCMK_SCORE_INFINITY) + || (colocation->score >= PCMK_SCORE_INFINITY)) { pcmk__add_dependent_scores(colocation, rsc); } } pe__show_node_scores(true, rsc, "Mandatory-colocations", rsc->allowed_nodes, rsc->cluster); // Then apply optional colocations for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; - if ((colocation->score > -CRM_SCORE_INFINITY) - && (colocation->score < CRM_SCORE_INFINITY)) { + if ((colocation->score > -PCMK_SCORE_INFINITY) + && (colocation->score < PCMK_SCORE_INFINITY)) { apply_this_with(colocation, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; - if ((colocation->score > -CRM_SCORE_INFINITY) - && (colocation->score < CRM_SCORE_INFINITY)) { + if ((colocation->score > -PCMK_SCORE_INFINITY) + && (colocation->score < PCMK_SCORE_INFINITY)) { pcmk__add_dependent_scores(colocation, rsc); } } g_list_free(this_with_colocations); g_list_free(with_this_colocations); if (rsc->next_role == pcmk_role_stopped) { pcmk__rsc_trace(rsc, "Banning %s from all nodes because it will be stopped", rsc->id); - resource_location(rsc, NULL, -INFINITY, PCMK_META_TARGET_ROLE, - rsc->cluster); + resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, + PCMK_META_TARGET_ROLE, rsc->cluster); } else if ((rsc->next_role > rsc->role) && !pcmk_is_set(rsc->cluster->flags, pcmk_sched_quorate) && (rsc->cluster->no_quorum_policy == pcmk_no_quorum_freeze)) { crm_notice("Resource %s cannot be elevated from %s to %s due to " PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE, rsc->id, pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role)); pe__set_next_role(rsc, rsc->role, PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE); } pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); // Unmanage resource if fencing is enabled but no device is configured if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled) && !pcmk_is_set(rsc->cluster->flags, pcmk_sched_have_fencing)) { pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed); } if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { // Unmanaged resources stay on their current node const char *reason = NULL; pcmk_node_t *assign_to = NULL; pe__set_next_role(rsc, rsc->role, "unmanaged"); assign_to = pcmk__current_node(rsc); if (assign_to == NULL) { reason = "inactive"; } else if (rsc->role == pcmk_role_promoted) { reason = "promoted"; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { reason = "failed"; } else { reason = "active"; } pcmk__rsc_info(rsc, "Unmanaged resource %s assigned to %s: %s", rsc->id, (assign_to? assign_to->details->uname : "no node"), reason); pcmk__assign_resource(rsc, assign_to, true, stop_if_fail); } else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_stop_all)) { // Must stop at some point, but be consistent with stop_if_fail if (stop_if_fail) { pcmk__rsc_debug(rsc, "Forcing %s to stop: " PCMK_OPT_STOP_ALL_RESOURCES, rsc->id); } pcmk__assign_resource(rsc, NULL, true, stop_if_fail); } else if (!assign_best_node(rsc, prefer, stop_if_fail)) { // Assignment failed if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { pcmk__rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id); } else if ((rsc->running_on != NULL) && stop_if_fail) { pcmk__rsc_info(rsc, "Stopping orphan resource %s", rsc->id); } } pcmk__clear_rsc_flags(rsc, pcmk_rsc_assigning); if (rsc->is_remote_node) { remote_connection_assigned(rsc); } return rsc->allocated_to; } /*! * \internal * \brief Schedule actions to bring resource down and back to current role * * \param[in,out] rsc Resource to restart * \param[in,out] current Node that resource should be brought down on * \param[in] need_stop Whether the resource must be stopped * \param[in] need_promote Whether the resource must be promoted * * \return Role that resource would have after scheduled actions are taken */ static void schedule_restart_actions(pcmk_resource_t *rsc, pcmk_node_t *current, bool need_stop, bool need_promote) { enum rsc_role_e role = rsc->role; enum rsc_role_e next_role; rsc_transition_fn fn = NULL; pcmk__set_rsc_flags(rsc, pcmk_rsc_restarting); // Bring resource down to a stop on its current node while (role != pcmk_role_stopped) { next_role = rsc_state_matrix[role][pcmk_role_stopped]; pcmk__rsc_trace(rsc, "Creating %s action to take %s down from %s to %s", (need_stop? "required" : "optional"), rsc->id, pcmk_role_text(role), pcmk_role_text(next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, current, !need_stop); role = next_role; } // Bring resource up to its next role on its next node while ((rsc->role <= rsc->next_role) && (role != rsc->role) && !pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) { bool required = need_stop; next_role = rsc_state_matrix[role][rsc->role]; if ((next_role == pcmk_role_promoted) && need_promote) { required = true; } pcmk__rsc_trace(rsc, "Creating %s action to take %s up from %s to %s", (required? "required" : "optional"), rsc->id, pcmk_role_text(role), pcmk_role_text(next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, rsc->allocated_to, !required); role = next_role; } pcmk__clear_rsc_flags(rsc, pcmk_rsc_restarting); } /*! * \internal * \brief If a resource's next role is not explicitly specified, set a default * * \param[in,out] rsc Resource to set next role for * * \return "explicit" if next role was explicitly set, otherwise "implicit" */ static const char * set_default_next_role(pcmk_resource_t *rsc) { if (rsc->next_role != pcmk_role_unknown) { return "explicit"; } if (rsc->allocated_to == NULL) { pe__set_next_role(rsc, pcmk_role_stopped, "assignment"); } else { pe__set_next_role(rsc, pcmk_role_started, "assignment"); } return "implicit"; } /*! * \internal * \brief Create an action to represent an already pending start * * \param[in,out] rsc Resource to create start action for */ static void create_pending_start(pcmk_resource_t *rsc) { pcmk_action_t *start = NULL; pcmk__rsc_trace(rsc, "Creating action for %s to represent already pending start", rsc->id); start = start_action(rsc, rsc->allocated_to, TRUE); pcmk__set_action_flags(start, pcmk_action_always_in_graph); } /*! * \internal * \brief Schedule actions needed to take a resource to its next role * * \param[in,out] rsc Resource to schedule actions for */ static void schedule_role_transition_actions(pcmk_resource_t *rsc) { enum rsc_role_e role = rsc->role; while (role != rsc->next_role) { enum rsc_role_e next_role = rsc_state_matrix[role][rsc->next_role]; rsc_transition_fn fn = NULL; pcmk__rsc_trace(rsc, "Creating action to take %s from %s to %s " "(ending at %s)", rsc->id, pcmk_role_text(role), pcmk_role_text(next_role), pcmk_role_text(rsc->next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, rsc->allocated_to, false); role = next_role; } } /*! * \internal * \brief Create all actions needed for a given primitive resource * * \param[in,out] rsc Primitive resource to create actions for */ void pcmk__primitive_create_actions(pcmk_resource_t *rsc) { bool need_stop = false; bool need_promote = false; bool is_moving = false; bool allow_migrate = false; bool multiply_active = false; pcmk_node_t *current = NULL; unsigned int num_all_active = 0; unsigned int num_clean_active = 0; const char *next_role_source = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); next_role_source = set_default_next_role(rsc); pcmk__rsc_trace(rsc, "Creating all actions for %s transition from %s to %s " "(%s) on %s", rsc->id, pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role), next_role_source, pcmk__node_name(rsc->allocated_to)); current = rsc->fns->active_node(rsc, &num_all_active, &num_clean_active); g_list_foreach(rsc->dangling_migrations, pcmk__abort_dangling_migration, rsc); if ((current != NULL) && (rsc->allocated_to != NULL) && !pcmk__same_node(current, rsc->allocated_to) && (rsc->next_role >= pcmk_role_started)) { pcmk__rsc_trace(rsc, "Moving %s from %s to %s", rsc->id, pcmk__node_name(current), pcmk__node_name(rsc->allocated_to)); is_moving = true; allow_migrate = pcmk__rsc_can_migrate(rsc, current); // This is needed even if migrating (though I'm not sure why ...) need_stop = true; } // Check whether resource is partially migrated and/or multiply active if ((rsc->partial_migration_source != NULL) && (rsc->partial_migration_target != NULL) && allow_migrate && (num_all_active == 2) && pcmk__same_node(current, rsc->partial_migration_source) && pcmk__same_node(rsc->allocated_to, rsc->partial_migration_target)) { /* A partial migration is in progress, and the migration target remains * the same as when the migration began. */ pcmk__rsc_trace(rsc, "Partial migration of %s from %s to %s will continue", rsc->id, pcmk__node_name(rsc->partial_migration_source), pcmk__node_name(rsc->partial_migration_target)); } else if ((rsc->partial_migration_source != NULL) || (rsc->partial_migration_target != NULL)) { // A partial migration is in progress but can't be continued if (num_all_active > 2) { // The resource is migrating *and* multiply active! crm_notice("Forcing recovery of %s because it is migrating " "from %s to %s and possibly active elsewhere", rsc->id, pcmk__node_name(rsc->partial_migration_source), pcmk__node_name(rsc->partial_migration_target)); } else { // The migration source or target isn't available crm_notice("Forcing recovery of %s because it can no longer " "migrate from %s to %s", rsc->id, pcmk__node_name(rsc->partial_migration_source), pcmk__node_name(rsc->partial_migration_target)); } need_stop = true; rsc->partial_migration_source = rsc->partial_migration_target = NULL; allow_migrate = false; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) { multiply_active = (num_all_active > 1); } else { /* If a resource has PCMK_META_REQUIRES set to PCMK_VALUE_NOTHING or * PCMK_VALUE_QUORUM, don't consider it active on unclean nodes (similar * to how all resources behave when PCMK_OPT_STONITH_ENABLED is false). * We can start such resources elsewhere before fencing completes, and * if we considered the resource active on the failed node, we would * attempt recovery for being active on multiple nodes. */ multiply_active = (num_clean_active > 1); } if (multiply_active) { const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS); // Resource was (possibly) incorrectly multiply active pcmk__sched_err("%s resource %s might be active on %u nodes (%s)", pcmk__s(class, "Untyped"), rsc->id, num_all_active, pcmk_multiply_active_text(rsc->recovery_type)); crm_notice("For more information, see \"What are multiply active " "resources?\" at " "https://projects.clusterlabs.org/w/clusterlabs/faq/"); switch (rsc->recovery_type) { case pcmk_multiply_active_restart: need_stop = true; break; case pcmk_multiply_active_unexpected: need_stop = true; // stop_resource() will skip expected node pcmk__set_rsc_flags(rsc, pcmk_rsc_stop_unexpected); break; default: break; } } else { pcmk__clear_rsc_flags(rsc, pcmk_rsc_stop_unexpected); } if (pcmk_is_set(rsc->flags, pcmk_rsc_start_pending)) { create_pending_start(rsc); } if (is_moving) { // Remaining tests are only for resources staying where they are } else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { if (pcmk_is_set(rsc->flags, pcmk_rsc_stop_if_failed)) { need_stop = true; pcmk__rsc_trace(rsc, "Recovering %s", rsc->id); } else { pcmk__rsc_trace(rsc, "Recovering %s by demotion", rsc->id); if (rsc->next_role == pcmk_role_promoted) { need_promote = true; } } } else if (pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) { pcmk__rsc_trace(rsc, "Blocking further actions on %s", rsc->id); need_stop = true; } else if ((rsc->role > pcmk_role_started) && (current != NULL) && (rsc->allocated_to != NULL)) { pcmk_action_t *start = NULL; pcmk__rsc_trace(rsc, "Creating start action for promoted resource %s", rsc->id); start = start_action(rsc, rsc->allocated_to, TRUE); if (!pcmk_is_set(start->flags, pcmk_action_optional)) { // Recovery of a promoted resource pcmk__rsc_trace(rsc, "%s restart is required for recovery", rsc->id); need_stop = true; } } // Create any actions needed to bring resource down and back up to same role schedule_restart_actions(rsc, current, need_stop, need_promote); // Create any actions needed to take resource from this role to the next schedule_role_transition_actions(rsc); pcmk__create_recurring_actions(rsc); if (allow_migrate) { pcmk__create_migration_actions(rsc, current); } } /*! * \internal * \brief Ban a resource from any allowed nodes that are Pacemaker Remote nodes * * \param[in] rsc Resource to check */ static void rsc_avoids_remote_nodes(const pcmk_resource_t *rsc) { GHashTableIter iter; pcmk_node_t *node = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (node->details->remote_rsc != NULL) { - node->weight = -INFINITY; + node->weight = -PCMK_SCORE_INFINITY; } } } /*! * \internal * \brief Return allowed nodes as (possibly sorted) list * * Convert a resource's hash table of allowed nodes to a list. If printing to * stdout, sort the list, to keep action ID numbers consistent for regression * test output (while avoiding the performance hit on a live cluster). * * \param[in] rsc Resource to check for allowed nodes * * \return List of resource's allowed nodes * \note Callers should take care not to rely on the list being sorted. */ static GList * allowed_nodes_as_list(const pcmk_resource_t *rsc) { GList *allowed_nodes = NULL; if (rsc->allowed_nodes) { allowed_nodes = g_hash_table_get_values(rsc->allowed_nodes); } if (!pcmk__is_daemon) { allowed_nodes = g_list_sort(allowed_nodes, pe__cmp_node_name); } return allowed_nodes; } /*! * \internal * \brief Create implicit constraints needed for a primitive resource * * \param[in,out] rsc Primitive resource to create implicit constraints for */ void pcmk__primitive_internal_constraints(pcmk_resource_t *rsc) { GList *allowed_nodes = NULL; bool check_unfencing = false; bool check_utilization = false; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk__rsc_trace(rsc, "Skipping implicit constraints for unmanaged resource " "%s", rsc->id); return; } // Whether resource requires unfencing check_unfencing = !pcmk_is_set(rsc->flags, pcmk_rsc_fence_device) && pcmk_is_set(rsc->cluster->flags, pcmk_sched_enable_unfencing) && pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing); // Whether a non-default placement strategy is used check_utilization = (g_hash_table_size(rsc->utilization) > 0) && !pcmk__str_eq(rsc->cluster->placement_strategy, PCMK_VALUE_DEFAULT, pcmk__str_casei); // Order stops before starts (i.e. restart) pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered |pcmk__ar_first_implies_then |pcmk__ar_intermediate_stop, rsc->cluster); // Promotable ordering: demote before stop, start before promote if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk_rsc_promotable) || (rsc->role > pcmk_role_unpromoted)) { pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_DEMOTE, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, pcmk__ar_promoted_then_implies_first, rsc->cluster); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_PROMOTE, 0), NULL, pcmk__ar_unrunnable_first_blocks, rsc->cluster); } // Don't clear resource history if probing on same node pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0), NULL, pcmk__ar_if_on_same_node|pcmk__ar_then_cancels_first, rsc->cluster); // Certain checks need allowed nodes if (check_unfencing || check_utilization || (rsc->container != NULL)) { allowed_nodes = allowed_nodes_as_list(rsc); } if (check_unfencing) { g_list_foreach(allowed_nodes, pcmk__order_restart_vs_unfence, rsc); } if (check_utilization) { pcmk__create_utilization_constraints(rsc, allowed_nodes); } if (rsc->container != NULL) { pcmk_resource_t *remote_rsc = NULL; if (rsc->is_remote_node) { // rsc is the implicit remote connection for a guest or bundle node /* Guest resources are not allowed to run on Pacemaker Remote nodes, * to avoid nesting remotes. However, bundles are allowed. */ if (!pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) { rsc_avoids_remote_nodes(rsc->container); } /* If someone cleans up a guest or bundle node's container, we will * likely schedule a (re-)probe of the container and recovery of the * connection. Order the connection stop after the container probe, * so that if we detect the container running, we will trigger a new * transition and avoid the unnecessary recovery. */ pcmk__order_resource_actions(rsc->container, PCMK_ACTION_MONITOR, rsc, PCMK_ACTION_STOP, pcmk__ar_ordered); /* A user can specify that a resource must start on a Pacemaker Remote * node by explicitly configuring it with the container=NODENAME * meta-attribute. This is of questionable merit, since location * constraints can accomplish the same thing. But we support it, so here * we check whether a resource (that is not itself a remote connection) * has container set to a remote node or guest node resource. */ } else if (rsc->container->is_remote_node) { remote_rsc = rsc->container; } else { remote_rsc = pe__resource_contains_guest_node(rsc->cluster, rsc->container); } if (remote_rsc != NULL) { /* Force the resource on the Pacemaker Remote node instead of * colocating the resource with the container resource. */ for (GList *item = allowed_nodes; item; item = item->next) { pcmk_node_t *node = item->data; if (node->details->remote_rsc != remote_rsc) { - node->weight = -INFINITY; + node->weight = -PCMK_SCORE_INFINITY; } } } else { /* This resource is either a filler for a container that does NOT * represent a Pacemaker Remote node, or a Pacemaker Remote * connection resource for a guest node or bundle. */ int score; crm_trace("Order and colocate %s relative to its container %s", rsc->id, rsc->container->id); pcmk__new_ordering(rsc->container, pcmk__op_key(rsc->container->id, PCMK_ACTION_START, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks, rsc->cluster); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, rsc->container, pcmk__op_key(rsc->container->id, PCMK_ACTION_STOP, 0), NULL, pcmk__ar_then_implies_first, rsc->cluster); if (pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) { score = 10000; /* Highly preferred but not essential */ } else { - score = INFINITY; /* Force them to run on the same host */ + score = PCMK_SCORE_INFINITY; // Force to run on same host } pcmk__new_colocation("#resource-with-container", NULL, score, rsc, rsc->container, NULL, NULL, pcmk__coloc_influence); } } if (rsc->is_remote_node || pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) { /* Remote connections and fencing devices are not allowed to run on * Pacemaker Remote nodes */ rsc_avoids_remote_nodes(rsc); } g_list_free(allowed_nodes); } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__primitive_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { enum pcmk__coloc_affects filter_results; CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if (for_dependent) { // Always process on behalf of primary resource primary->cmds->apply_coloc_score(dependent, primary, colocation, false); return; } filter_results = pcmk__colocation_affects(dependent, primary, colocation, false); pcmk__rsc_trace(dependent, "%s %s with %s (%s, score=%d, filter=%d)", ((colocation->score > 0)? "Colocating" : "Anti-colocating"), dependent->id, primary->id, colocation->id, colocation->score, filter_results); switch (filter_results) { case pcmk__coloc_affects_role: pcmk__apply_coloc_to_priority(dependent, primary, colocation); break; case pcmk__coloc_affects_location: pcmk__apply_coloc_to_scores(dependent, primary, colocation); break; default: // pcmk__coloc_affects_nothing return; } } /* Primitive implementation of * pcmk_assignment_methods_t:with_this_colocations() */ void pcmk__with_primitive_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (list != NULL)); if (rsc == orig_rsc) { /* For the resource itself, add all of its own colocations and relevant * colocations from its parent (if any). */ pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); if (rsc->parent != NULL) { rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list); } } else { // For an ancestor, add only explicitly configured constraints for (GList *iter = rsc->rsc_cons_lhs; iter != NULL; iter = iter->next) { pcmk__colocation_t *colocation = iter->data; if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) { pcmk__add_with_this(list, colocation, orig_rsc); } } } } /* Primitive implementation of * pcmk_assignment_methods_t:this_with_colocations() */ void pcmk__primitive_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (list != NULL)); if (rsc == orig_rsc) { /* For the resource itself, add all of its own colocations and relevant * colocations from its parent (if any). */ pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); if (rsc->parent != NULL) { rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list); } } else { // For an ancestor, add only explicitly configured constraints for (GList *iter = rsc->rsc_cons; iter != NULL; iter = iter->next) { pcmk__colocation_t *colocation = iter->data; if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) { pcmk__add_this_with(list, colocation, orig_rsc); } } } } /*! * \internal * \brief Return action flags for a given primitive resource action * * \param[in,out] action Action to get flags for * \param[in] node If not NULL, limit effects to this node (ignored) * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__primitive_action_flags(pcmk_action_t *action, const pcmk_node_t *node) { CRM_ASSERT(action != NULL); return (uint32_t) action->flags; } /*! * \internal * \brief Check whether a node is a multiply active resource's expected node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return \c true if \p rsc is multiply active with * \c PCMK_META_MULTIPLE_ACTIVE set to \c stop_unexpected, and \p node * is the node where it will remain active * \note This assumes that the resource's next role cannot be changed to stopped * after this is called, which should be reasonable if status has already * been unpacked and resources have been assigned to nodes. */ static bool is_expected_node(const pcmk_resource_t *rsc, const pcmk_node_t *node) { return pcmk_all_flags_set(rsc->flags, pcmk_rsc_stop_unexpected|pcmk_rsc_restarting) && (rsc->next_role > pcmk_role_stopped) && pcmk__same_node(rsc->allocated_to, node); } /*! * \internal * \brief Schedule actions needed to stop a resource wherever it is active * * \param[in,out] rsc Resource being stopped * \param[in] node Node where resource is being stopped (ignored) * \param[in] optional Whether actions should be optional */ static void stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *current = (pcmk_node_t *) iter->data; pcmk_action_t *stop = NULL; if (is_expected_node(rsc, current)) { /* We are scheduling restart actions for a multiply active resource * with PCMK_META_MULTIPLE_ACTIVE=stop_unexpected, and this is where * it should not be stopped. */ pcmk__rsc_trace(rsc, "Skipping stop of multiply active resource %s " "on expected node %s", rsc->id, pcmk__node_name(current)); continue; } if (rsc->partial_migration_target != NULL) { // Continue migration if node originally was and remains target if (pcmk__same_node(current, rsc->partial_migration_target) && pcmk__same_node(current, rsc->allocated_to)) { pcmk__rsc_trace(rsc, "Skipping stop of %s on %s " "because partial migration there will continue", rsc->id, pcmk__node_name(current)); continue; } else { pcmk__rsc_trace(rsc, "Forcing stop of %s on %s " "because migration target changed", rsc->id, pcmk__node_name(current)); optional = false; } } pcmk__rsc_trace(rsc, "Scheduling stop of %s on %s", rsc->id, pcmk__node_name(current)); stop = stop_action(rsc, current, optional); if (rsc->allocated_to == NULL) { pe_action_set_reason(stop, "node availability", true); } else if (pcmk_all_flags_set(rsc->flags, pcmk_rsc_restarting |pcmk_rsc_stop_unexpected)) { /* We are stopping a multiply active resource on a node that is * not its expected node, and we are still scheduling restart * actions, so the stop is for being multiply active. */ pe_action_set_reason(stop, "being multiply active", true); } if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk__clear_action_flags(stop, pcmk_action_runnable); } if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_remove_after_stop)) { pcmk__schedule_cleanup(rsc, current, optional); } if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing)) { pcmk_action_t *unfence = pe_fence_op(current, PCMK_ACTION_ON, true, NULL, false, rsc->cluster); order_actions(stop, unfence, pcmk__ar_then_implies_first); if (!pcmk__node_unfenced(current)) { pcmk__sched_err("Stopping %s until %s can be unfenced", rsc->id, pcmk__node_name(current)); } } } } /*! * \internal * \brief Schedule actions needed to start a resource on a node * * \param[in,out] rsc Resource being started * \param[in,out] node Node where resource should be started * \param[in] optional Whether actions should be optional */ static void start_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { pcmk_action_t *start = NULL; CRM_ASSERT(node != NULL); pcmk__rsc_trace(rsc, "Scheduling %s start of %s on %s (score %d)", (optional? "optional" : "required"), rsc->id, pcmk__node_name(node), node->weight); start = start_action(rsc, node, TRUE); pcmk__order_vs_unfence(rsc, node, start, pcmk__ar_first_implies_then); if (pcmk_is_set(start->flags, pcmk_action_runnable) && !optional) { pcmk__clear_action_flags(start, pcmk_action_optional); } if (is_expected_node(rsc, node)) { /* This could be a problem if the start becomes necessary for other * reasons later. */ pcmk__rsc_trace(rsc, "Start of multiply active resouce %s " "on expected node %s will be a pseudo-action", rsc->id, pcmk__node_name(node)); pcmk__set_action_flags(start, pcmk_action_pseudo); } } /*! * \internal * \brief Schedule actions needed to promote a resource on a node * * \param[in,out] rsc Resource being promoted * \param[in] node Node where resource should be promoted * \param[in] optional Whether actions should be optional */ static void promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { GList *iter = NULL; GList *action_list = NULL; bool runnable = true; CRM_ASSERT(node != NULL); // Any start must be runnable for promotion to be runnable action_list = pe__resource_actions(rsc, node, PCMK_ACTION_START, true); for (iter = action_list; iter != NULL; iter = iter->next) { pcmk_action_t *start = (pcmk_action_t *) iter->data; if (!pcmk_is_set(start->flags, pcmk_action_runnable)) { runnable = false; } } g_list_free(action_list); if (runnable) { pcmk_action_t *promote = promote_action(rsc, node, optional); pcmk__rsc_trace(rsc, "Scheduling %s promotion of %s on %s", (optional? "optional" : "required"), rsc->id, pcmk__node_name(node)); if (is_expected_node(rsc, node)) { /* This could be a problem if the promote becomes necessary for * other reasons later. */ pcmk__rsc_trace(rsc, "Promotion of multiply active resouce %s " "on expected node %s will be a pseudo-action", rsc->id, pcmk__node_name(node)); pcmk__set_action_flags(promote, pcmk_action_pseudo); } } else { pcmk__rsc_trace(rsc, "Not promoting %s on %s: start unrunnable", rsc->id, pcmk__node_name(node)); action_list = pe__resource_actions(rsc, node, PCMK_ACTION_PROMOTE, true); for (iter = action_list; iter != NULL; iter = iter->next) { pcmk_action_t *promote = (pcmk_action_t *) iter->data; pcmk__clear_action_flags(promote, pcmk_action_runnable); } g_list_free(action_list); } } /*! * \internal * \brief Schedule actions needed to demote a resource wherever it is active * * \param[in,out] rsc Resource being demoted * \param[in] node Node where resource should be demoted (ignored) * \param[in] optional Whether actions should be optional */ static void demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { /* Since this will only be called for a primitive (possibly as an instance * of a collective resource), the resource is multiply active if it is * running on more than one node, so we want to demote on all of them as * part of recovery, regardless of which one is the desired node. */ for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *current = (pcmk_node_t *) iter->data; if (is_expected_node(rsc, current)) { pcmk__rsc_trace(rsc, "Skipping demote of multiply active resource %s " "on expected node %s", rsc->id, pcmk__node_name(current)); } else { pcmk__rsc_trace(rsc, "Scheduling %s demotion of %s on %s", (optional? "optional" : "required"), rsc->id, pcmk__node_name(current)); demote_action(rsc, current, optional); } } } static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { CRM_ASSERT(false); } /*! * \internal * \brief Schedule cleanup of a resource * * \param[in,out] rsc Resource to clean up * \param[in] node Node to clean up on * \param[in] optional Whether clean-up should be optional */ void pcmk__schedule_cleanup(pcmk_resource_t *rsc, const pcmk_node_t *node, bool optional) { /* If the cleanup is required, its orderings are optional, because they're * relevant only if both actions are required. Conversely, if the cleanup is * optional, the orderings make the then action required if the first action * becomes required. */ uint32_t flag = optional? pcmk__ar_first_implies_then : pcmk__ar_ordered; CRM_CHECK((rsc != NULL) && (node != NULL), return); if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: resource failed", rsc->id, pcmk__node_name(node)); return; } if (node->details->unclean || !node->details->online) { pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: node unavailable", rsc->id, pcmk__node_name(node)); return; } crm_notice("Scheduling clean-up of %s on %s", rsc->id, pcmk__node_name(node)); delete_action(rsc, node, optional); // stop -> clean-up -> start pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, rsc, PCMK_ACTION_DELETE, flag); pcmk__order_resource_actions(rsc, PCMK_ACTION_DELETE, rsc, PCMK_ACTION_START, flag); } /*! * \internal * \brief Add primitive meta-attributes relevant to graph actions to XML * * \param[in] rsc Primitive resource whose meta-attributes should be added * \param[in,out] xml Transition graph action attributes XML to add to */ void pcmk__primitive_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml) { char *name = NULL; char *value = NULL; const pcmk_resource_t *parent = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (xml != NULL)); /* Clone instance numbers get set internally as meta-attributes, and are * needed in the transition graph (for example, to tell unique clone * instances apart). */ value = g_hash_table_lookup(rsc->meta, PCMK__META_CLONE); if (value != NULL) { name = crm_meta_name(PCMK__META_CLONE); crm_xml_add(xml, name, value); free(name); } // Not sure if this one is really needed ... value = g_hash_table_lookup(rsc->meta, PCMK_META_REMOTE_NODE); if (value != NULL) { name = crm_meta_name(PCMK_META_REMOTE_NODE); crm_xml_add(xml, name, value); free(name); } /* The container meta-attribute can be set on the primitive itself or one of * its parents (for example, a group inside a container resource), so check * them all, and keep the highest one found. */ for (parent = rsc; parent != NULL; parent = parent->parent) { if (parent->container != NULL) { crm_xml_add(xml, CRM_META "_" PCMK__META_CONTAINER, parent->container->id); } } /* Bundle replica children will get their external-ip set internally as a * meta-attribute. The graph action needs it, but under a different naming * convention than other meta-attributes. */ value = g_hash_table_lookup(rsc->meta, "external-ip"); if (value != NULL) { crm_xml_add(xml, "pcmk_external_ip", value); } } // Primitive implementation of pcmk_assignment_methods_t:add_utilization() void pcmk__primitive_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (orig_rsc != NULL) && (utilization != NULL)); if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { return; } pcmk__rsc_trace(orig_rsc, "%s: Adding primitive %s as colocated utilization", orig_rsc->id, rsc->id); pcmk__release_node_capacity(utilization, rsc); } /*! * \internal * \brief Get epoch time of node's shutdown attribute (or now if none) * * \param[in,out] node Node to check * * \return Epoch time corresponding to shutdown attribute if set or now if not */ static time_t shutdown_time(pcmk_node_t *node) { const char *shutdown = pcmk__node_attr(node, PCMK__NODE_ATTR_SHUTDOWN, NULL, pcmk__rsc_node_current); time_t result = 0; if (shutdown != NULL) { long long result_ll; if (pcmk__scan_ll(shutdown, &result_ll, 0LL) == pcmk_rc_ok) { result = (time_t) result_ll; } } return (result == 0)? get_effective_time(node->details->data_set) : result; } /*! * \internal * \brief Ban a resource from a node if it's not locked to the node * * \param[in] data Node to check * \param[in,out] user_data Resource to check */ static void ban_if_not_locked(gpointer data, gpointer user_data) { const pcmk_node_t *node = (const pcmk_node_t *) data; pcmk_resource_t *rsc = (pcmk_resource_t *) user_data; if (strcmp(node->details->uname, rsc->lock_node->details->uname) != 0) { - resource_location(rsc, node, -CRM_SCORE_INFINITY, + resource_location(rsc, node, -PCMK_SCORE_INFINITY, PCMK_OPT_SHUTDOWN_LOCK, rsc->cluster); } } // Primitive implementation of pcmk_assignment_methods_t:shutdown_lock() void pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc) { const char *class = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); class = crm_element_value(rsc->xml, PCMK_XA_CLASS); // Fence devices and remote connections can't be locked if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches) || rsc->is_remote_node) { return; } if (rsc->lock_node != NULL) { // The lock was obtained from resource history if (rsc->running_on != NULL) { /* The resource was started elsewhere even though it is now * considered locked. This shouldn't be possible, but as a * failsafe, we don't want to disturb the resource now. */ pcmk__rsc_info(rsc, "Cancelling shutdown lock " "because %s is already active", rsc->id); pe__clear_resource_history(rsc, rsc->lock_node); rsc->lock_node = NULL; rsc->lock_time = 0; } // Only a resource active on exactly one node can be locked } else if (pcmk__list_of_1(rsc->running_on)) { pcmk_node_t *node = rsc->running_on->data; if (node->details->shutdown) { if (node->details->unclean) { pcmk__rsc_debug(rsc, "Not locking %s to unclean %s for shutdown", rsc->id, pcmk__node_name(node)); } else { rsc->lock_node = node; rsc->lock_time = shutdown_time(node); } } } if (rsc->lock_node == NULL) { // No lock needed return; } if (rsc->cluster->shutdown_lock > 0) { time_t lock_expiration = rsc->lock_time + rsc->cluster->shutdown_lock; pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)", rsc->id, pcmk__node_name(rsc->lock_node), (long long) lock_expiration); pe__update_recheck_time(++lock_expiration, rsc->cluster, "shutdown lock expiration"); } else { pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown", rsc->id, pcmk__node_name(rsc->lock_node)); } // If resource is locked to one node, ban it from all other nodes g_list_foreach(rsc->cluster->nodes, ban_if_not_locked, rsc); } diff --git a/lib/pacemaker/pcmk_sched_promotable.c b/lib/pacemaker/pcmk_sched_promotable.c index 71ecea9ab2..fb8bb9a205 100644 --- a/lib/pacemaker/pcmk_sched_promotable.c +++ b/lib/pacemaker/pcmk_sched_promotable.c @@ -1,1313 +1,1313 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Add implicit promotion ordering for a promotable instance * * \param[in,out] clone Clone resource * \param[in,out] child Instance of \p clone being ordered * \param[in,out] last Previous instance ordered (NULL if \p child is first) */ static void order_instance_promotion(pcmk_resource_t *clone, pcmk_resource_t *child, pcmk_resource_t *last) { // "Promote clone" -> promote instance -> "clone promoted" pcmk__order_resource_actions(clone, PCMK_ACTION_PROMOTE, child, PCMK_ACTION_PROMOTE, pcmk__ar_ordered); pcmk__order_resource_actions(child, PCMK_ACTION_PROMOTE, clone, PCMK_ACTION_PROMOTED, pcmk__ar_ordered); // If clone is ordered, order this instance relative to last if ((last != NULL) && pe__clone_is_ordered(clone)) { pcmk__order_resource_actions(last, PCMK_ACTION_PROMOTE, child, PCMK_ACTION_PROMOTE, pcmk__ar_ordered); } } /*! * \internal * \brief Add implicit demotion ordering for a promotable instance * * \param[in,out] clone Clone resource * \param[in,out] child Instance of \p clone being ordered * \param[in] last Previous instance ordered (NULL if \p child is first) */ static void order_instance_demotion(pcmk_resource_t *clone, pcmk_resource_t *child, pcmk_resource_t *last) { // "Demote clone" -> demote instance -> "clone demoted" pcmk__order_resource_actions(clone, PCMK_ACTION_DEMOTE, child, PCMK_ACTION_DEMOTE, pcmk__ar_then_implies_first_graphed); pcmk__order_resource_actions(child, PCMK_ACTION_DEMOTE, clone, PCMK_ACTION_DEMOTED, pcmk__ar_first_implies_then_graphed); // If clone is ordered, order this instance relative to last if ((last != NULL) && pe__clone_is_ordered(clone)) { pcmk__order_resource_actions(child, PCMK_ACTION_DEMOTE, last, PCMK_ACTION_DEMOTE, pcmk__ar_ordered); } } /*! * \internal * \brief Check whether an instance will be promoted or demoted * * \param[in] rsc Instance to check * \param[out] demoting If \p rsc will be demoted, this will be set to true * \param[out] promoting If \p rsc will be promoted, this will be set to true */ static void check_for_role_change(const pcmk_resource_t *rsc, bool *demoting, bool *promoting) { const GList *iter = NULL; // If this is a cloned group, check group members recursively if (rsc->children != NULL) { for (iter = rsc->children; iter != NULL; iter = iter->next) { check_for_role_change((const pcmk_resource_t *) iter->data, demoting, promoting); } return; } for (iter = rsc->actions; iter != NULL; iter = iter->next) { const pcmk_action_t *action = (const pcmk_action_t *) iter->data; if (*promoting && *demoting) { return; } else if (pcmk_is_set(action->flags, pcmk_action_optional)) { continue; } else if (pcmk__str_eq(PCMK_ACTION_DEMOTE, action->task, pcmk__str_none)) { *demoting = true; } else if (pcmk__str_eq(PCMK_ACTION_PROMOTE, action->task, pcmk__str_none)) { *promoting = true; } } } /*! * \internal * \brief Add promoted-role location constraint scores to an instance's priority * * Adjust a promotable clone instance's promotion priority by the scores of any * location constraints in a list that are both limited to the promoted role and * for the node where the instance will be placed. * * \param[in,out] child Promotable clone instance * \param[in] location_constraints List of location constraints to apply * \param[in] chosen Node where \p child will be placed */ static void apply_promoted_locations(pcmk_resource_t *child, const GList *location_constraints, const pcmk_node_t *chosen) { for (const GList *iter = location_constraints; iter; iter = iter->next) { const pcmk__location_t *location = iter->data; const pcmk_node_t *constraint_node = NULL; if (location->role_filter == pcmk_role_promoted) { constraint_node = pe_find_node_id(location->nodes, chosen->details->id); } if (constraint_node != NULL) { int new_priority = pcmk__add_scores(child->priority, constraint_node->weight); pcmk__rsc_trace(child, "Applying location %s to %s promotion priority on " "%s: %s + %s = %s", location->id, child->id, pcmk__node_name(constraint_node), pcmk_readable_score(child->priority), pcmk_readable_score(constraint_node->weight), pcmk_readable_score(new_priority)); child->priority = new_priority; } } } /*! * \internal * \brief Get the node that an instance will be promoted on * * \param[in] rsc Promotable clone instance to check * * \return Node that \p rsc will be promoted on, or NULL if none */ static pcmk_node_t * node_to_be_promoted_on(const pcmk_resource_t *rsc) { pcmk_node_t *node = NULL; pcmk_node_t *local_node = NULL; const pcmk_resource_t *parent = NULL; // If this is a cloned group, bail if any group member can't be promoted for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) iter->data; if (node_to_be_promoted_on(child) == NULL) { pcmk__rsc_trace(rsc, "%s can't be promoted because member %s can't", rsc->id, child->id); return NULL; } } node = rsc->fns->location(rsc, NULL, FALSE); if (node == NULL) { pcmk__rsc_trace(rsc, "%s can't be promoted because it won't be active", rsc->id); return NULL; } else if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { if (rsc->fns->state(rsc, TRUE) == pcmk_role_promoted) { crm_notice("Unmanaged instance %s will be left promoted on %s", rsc->id, pcmk__node_name(node)); } else { pcmk__rsc_trace(rsc, "%s can't be promoted because it is unmanaged", rsc->id); return NULL; } } else if (rsc->priority < 0) { pcmk__rsc_trace(rsc, "%s can't be promoted because its promotion priority " "%d is negative", rsc->id, rsc->priority); return NULL; } else if (!pcmk__node_available(node, false, true)) { pcmk__rsc_trace(rsc, "%s can't be promoted because %s can't run resources", rsc->id, pcmk__node_name(node)); return NULL; } parent = pe__const_top_resource(rsc, false); local_node = g_hash_table_lookup(parent->allowed_nodes, node->details->id); if (local_node == NULL) { /* It should not be possible for the scheduler to have assigned the * instance to a node where its parent is not allowed, but it's good to * have a fail-safe. */ if (pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk__sched_err("%s can't be promoted because %s is not allowed " "on %s (scheduler bug?)", rsc->id, parent->id, pcmk__node_name(node)); } // else the instance is unmanaged and already promoted return NULL; } else if ((local_node->count >= pe__clone_promoted_node_max(parent)) && pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk__rsc_trace(rsc, "%s can't be promoted because %s has " "maximum promoted instances already", rsc->id, pcmk__node_name(node)); return NULL; } return local_node; } /*! * \internal * \brief Compare two promotable clone instances by promotion priority * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a has higher promotion priority, * a positive number if \p b has higher promotion priority, * or 0 if promotion priorities are equal */ static gint cmp_promotable_instance(gconstpointer a, gconstpointer b) { const pcmk_resource_t *rsc1 = (const pcmk_resource_t *) a; const pcmk_resource_t *rsc2 = (const pcmk_resource_t *) b; enum rsc_role_e role1 = pcmk_role_unknown; enum rsc_role_e role2 = pcmk_role_unknown; CRM_ASSERT((rsc1 != NULL) && (rsc2 != NULL)); // Check sort index set by pcmk__set_instance_roles() if (rsc1->sort_index > rsc2->sort_index) { pcmk__rsc_trace(rsc1, "%s has higher promotion priority than %s " "(sort index %d > %d)", rsc1->id, rsc2->id, rsc1->sort_index, rsc2->sort_index); return -1; } else if (rsc1->sort_index < rsc2->sort_index) { pcmk__rsc_trace(rsc1, "%s has lower promotion priority than %s " "(sort index %d < %d)", rsc1->id, rsc2->id, rsc1->sort_index, rsc2->sort_index); return 1; } // If those are the same, prefer instance whose current role is higher role1 = rsc1->fns->state(rsc1, TRUE); role2 = rsc2->fns->state(rsc2, TRUE); if (role1 > role2) { pcmk__rsc_trace(rsc1, "%s has higher promotion priority than %s " "(higher current role)", rsc1->id, rsc2->id); return -1; } else if (role1 < role2) { pcmk__rsc_trace(rsc1, "%s has lower promotion priority than %s " "(lower current role)", rsc1->id, rsc2->id); return 1; } // Finally, do normal clone instance sorting return pcmk__cmp_instance(a, b); } /*! * \internal * \brief Add a promotable clone instance's sort index to its node's score * * Add a promotable clone instance's sort index (which sums its promotion * preferences and scores of relevant location constraints for the promoted * role) to the node score of the instance's assigned node. * * \param[in] data Promotable clone instance * \param[in,out] user_data Clone parent of \p data */ static void add_sort_index_to_node_score(gpointer data, gpointer user_data) { const pcmk_resource_t *child = (const pcmk_resource_t *) data; pcmk_resource_t *clone = (pcmk_resource_t *) user_data; pcmk_node_t *node = NULL; const pcmk_node_t *chosen = NULL; if (child->sort_index < 0) { pcmk__rsc_trace(clone, "Not adding sort index of %s: negative", child->id); return; } chosen = child->fns->location(child, NULL, FALSE); if (chosen == NULL) { pcmk__rsc_trace(clone, "Not adding sort index of %s: inactive", child->id); return; } node = g_hash_table_lookup(clone->allowed_nodes, chosen->details->id); CRM_ASSERT(node != NULL); node->weight = pcmk__add_scores(child->sort_index, node->weight); pcmk__rsc_trace(clone, "Added cumulative priority of %s (%s) to score on %s " "(now %s)", child->id, pcmk_readable_score(child->sort_index), pcmk__node_name(node), pcmk_readable_score(node->weight)); } /*! * \internal * \brief Apply colocation to dependent's node scores if for promoted role * * \param[in,out] data Colocation constraint to apply * \param[in,out] user_data Promotable clone that is constraint's dependent */ static void apply_coloc_to_dependent(gpointer data, gpointer user_data) { pcmk__colocation_t *colocation = data; pcmk_resource_t *clone = user_data; pcmk_resource_t *primary = colocation->primary; uint32_t flags = pcmk__coloc_select_default; - float factor = colocation->score / (float) INFINITY; + float factor = colocation->score / (float) PCMK_SCORE_INFINITY; if (colocation->dependent_role != pcmk_role_promoted) { return; } - if (colocation->score < INFINITY) { + if (colocation->score < PCMK_SCORE_INFINITY) { flags = pcmk__coloc_select_active; } pcmk__rsc_trace(clone, "Applying colocation %s (promoted %s with %s) @%s", colocation->id, colocation->dependent->id, colocation->primary->id, pcmk_readable_score(colocation->score)); primary->cmds->add_colocated_node_scores(primary, clone, clone->id, &clone->allowed_nodes, colocation, factor, flags); } /*! * \internal * \brief Apply colocation to primary's node scores if for promoted role * * \param[in,out] data Colocation constraint to apply * \param[in,out] user_data Promotable clone that is constraint's primary */ static void apply_coloc_to_primary(gpointer data, gpointer user_data) { pcmk__colocation_t *colocation = data; pcmk_resource_t *clone = user_data; pcmk_resource_t *dependent = colocation->dependent; - const float factor = colocation->score / (float) INFINITY; + const float factor = colocation->score / (float) PCMK_SCORE_INFINITY; const uint32_t flags = pcmk__coloc_select_active |pcmk__coloc_select_nonnegative; if ((colocation->primary_role != pcmk_role_promoted) || !pcmk__colocation_has_influence(colocation, NULL)) { return; } pcmk__rsc_trace(clone, "Applying colocation %s (%s with promoted %s) @%s", colocation->id, colocation->dependent->id, colocation->primary->id, pcmk_readable_score(colocation->score)); dependent->cmds->add_colocated_node_scores(dependent, clone, clone->id, &clone->allowed_nodes, colocation, factor, flags); } /*! * \internal * \brief Set clone instance's sort index to its node's score * * \param[in,out] data Promotable clone instance * \param[in] user_data Parent clone of \p data */ static void set_sort_index_to_node_score(gpointer data, gpointer user_data) { pcmk_resource_t *child = (pcmk_resource_t *) data; const pcmk_resource_t *clone = (const pcmk_resource_t *) user_data; pcmk_node_t *chosen = child->fns->location(child, NULL, FALSE); if (!pcmk_is_set(child->flags, pcmk_rsc_managed) && (child->next_role == pcmk_role_promoted)) { - child->sort_index = INFINITY; + child->sort_index = PCMK_SCORE_INFINITY; pcmk__rsc_trace(clone, "Final sort index for %s is INFINITY " "(unmanaged promoted)", child->id); } else if ((chosen == NULL) || (child->sort_index < 0)) { pcmk__rsc_trace(clone, "Final sort index for %s is %d (ignoring node score)", child->id, child->sort_index); } else { const pcmk_node_t *node = g_hash_table_lookup(clone->allowed_nodes, chosen->details->id); CRM_ASSERT(node != NULL); child->sort_index = node->weight; pcmk__rsc_trace(clone, "Adding scores for %s: final sort index for %s is %d", clone->id, child->id, child->sort_index); } } /*! * \internal * \brief Sort a promotable clone's instances by descending promotion priority * * \param[in,out] clone Promotable clone to sort */ static void sort_promotable_instances(pcmk_resource_t *clone) { GList *colocations = NULL; if (pe__set_clone_flag(clone, pcmk__clone_promotion_constrained) == pcmk_rc_already) { return; } pcmk__set_rsc_flags(clone, pcmk_rsc_updating_nodes); for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) iter->data; pcmk__rsc_trace(clone, "Adding scores for %s: initial sort index for %s is %d", clone->id, child->id, child->sort_index); } pe__show_node_scores(true, clone, "Before", clone->allowed_nodes, clone->cluster); g_list_foreach(clone->children, add_sort_index_to_node_score, clone); colocations = pcmk__this_with_colocations(clone); g_list_foreach(colocations, apply_coloc_to_dependent, clone); g_list_free(colocations); colocations = pcmk__with_this_colocations(clone); g_list_foreach(colocations, apply_coloc_to_primary, clone); g_list_free(colocations); // Ban resource from all nodes if it needs a ticket but doesn't have it pcmk__require_promotion_tickets(clone); pe__show_node_scores(true, clone, "After", clone->allowed_nodes, clone->cluster); // Reset sort indexes to final node scores g_list_foreach(clone->children, set_sort_index_to_node_score, clone); // Finally, sort instances in descending order of promotion priority clone->children = g_list_sort(clone->children, cmp_promotable_instance); pcmk__clear_rsc_flags(clone, pcmk_rsc_updating_nodes); } /*! * \internal * \brief Find the active instance (if any) of an anonymous clone on a node * * \param[in] clone Anonymous clone to check * \param[in] id Instance ID (without instance number) to check * \param[in] node Node to check * * \return */ static pcmk_resource_t * find_active_anon_instance(const pcmk_resource_t *clone, const char *id, const pcmk_node_t *node) { for (GList *iter = clone->children; iter; iter = iter->next) { pcmk_resource_t *child = iter->data; pcmk_resource_t *active = NULL; // Use ->find_rsc() in case this is a cloned group active = clone->fns->find_rsc(child, id, node, pcmk_rsc_match_clone_only |pcmk_rsc_match_current_node); if (active != NULL) { return active; } } return NULL; } /* * \brief Check whether an anonymous clone instance is known on a node * * \param[in] clone Anonymous clone to check * \param[in] id Instance ID (without instance number) to check * \param[in] node Node to check * * \return true if \p id instance of \p clone is known on \p node, * otherwise false */ static bool anonymous_known_on(const pcmk_resource_t *clone, const char *id, const pcmk_node_t *node) { for (GList *iter = clone->children; iter; iter = iter->next) { pcmk_resource_t *child = iter->data; /* Use ->find_rsc() because this might be a cloned group, and knowing * that other members of the group are known here implies nothing. */ child = clone->fns->find_rsc(child, id, NULL, pcmk_rsc_match_clone_only); CRM_LOG_ASSERT(child != NULL); if (child != NULL) { if (g_hash_table_lookup(child->known_on, node->details->id)) { return true; } } } return false; } /*! * \internal * \brief Check whether a node is allowed to run a resource * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p node is allowed to run \p rsc, otherwise false */ static bool is_allowed(const pcmk_resource_t *rsc, const pcmk_node_t *node) { pcmk_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); return (allowed != NULL) && (allowed->weight >= 0); } /*! * \brief Check whether a clone instance's promotion score should be considered * * \param[in] rsc Promotable clone instance to check * \param[in] node Node where score would be applied * * \return true if \p rsc's promotion score should be considered on \p node, * otherwise false */ static bool promotion_score_applies(const pcmk_resource_t *rsc, const pcmk_node_t *node) { char *id = clone_strip(rsc->id); const pcmk_resource_t *parent = pe__const_top_resource(rsc, false); pcmk_resource_t *active = NULL; const char *reason = "allowed"; // Some checks apply only to anonymous clone instances if (!pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { // If instance is active on the node, its score definitely applies active = find_active_anon_instance(parent, id, node); if (active == rsc) { reason = "active"; goto check_allowed; } /* If *no* instance is active on this node, this instance's score will * count if it has been probed on this node. */ if ((active == NULL) && anonymous_known_on(parent, id, node)) { reason = "probed"; goto check_allowed; } } /* If this clone's status is unknown on *all* nodes (e.g. cluster startup), * take all instances' scores into account, to make sure we use any * permanent promotion scores. */ if ((rsc->running_on == NULL) && (g_hash_table_size(rsc->known_on) == 0)) { reason = "none probed"; goto check_allowed; } /* Otherwise, we've probed and/or started the resource *somewhere*, so * consider promotion scores on nodes where we know the status. */ if ((g_hash_table_lookup(rsc->known_on, node->details->id) != NULL) || (pe_find_node_id(rsc->running_on, node->details->id) != NULL)) { reason = "known"; } else { pcmk__rsc_trace(rsc, "Ignoring %s promotion score (for %s) on %s: " "not probed", rsc->id, id, pcmk__node_name(node)); free(id); return false; } check_allowed: if (is_allowed(rsc, node)) { pcmk__rsc_trace(rsc, "Counting %s promotion score (for %s) on %s: %s", rsc->id, id, pcmk__node_name(node), reason); free(id); return true; } pcmk__rsc_trace(rsc, "Ignoring %s promotion score (for %s) on %s: not allowed", rsc->id, id, pcmk__node_name(node)); free(id); return false; } /*! * \internal * \brief Get the value of a promotion score node attribute * * \param[in] rsc Promotable clone instance to get promotion score for * \param[in] node Node to get promotion score for * \param[in] name Resource name to use in promotion score attribute name * * \return Value of promotion score node attribute for \p rsc on \p node */ static const char * promotion_attr_value(const pcmk_resource_t *rsc, const pcmk_node_t *node, const char *name) { char *attr_name = NULL; const char *attr_value = NULL; const char *target = NULL; enum pcmk__rsc_node node_type = pcmk__rsc_node_assigned; if (pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { // Not assigned yet node_type = pcmk__rsc_node_current; } target = g_hash_table_lookup(rsc->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); attr_name = pcmk_promotion_score_name(name); attr_value = pcmk__node_attr(node, attr_name, target, node_type); free(attr_name); return attr_value; } /*! * \internal * \brief Get the promotion score for a clone instance on a node * * \param[in] rsc Promotable clone instance to get score for * \param[in] node Node to get score for * \param[out] is_default If non-NULL, will be set true if no score available * * \return Promotion score for \p rsc on \p node (or 0 if none) */ static int promotion_score(const pcmk_resource_t *rsc, const pcmk_node_t *node, bool *is_default) { char *name = NULL; const char *attr_value = NULL; if (is_default != NULL) { *is_default = true; } CRM_CHECK((rsc != NULL) && (node != NULL), return 0); /* If this is an instance of a cloned group, the promotion score is the sum * of all members' promotion scores. */ if (rsc->children != NULL) { int score = 0; for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *child = (const pcmk_resource_t *) iter->data; bool child_default = false; int child_score = promotion_score(child, node, &child_default); if (!child_default && (is_default != NULL)) { *is_default = false; } score += child_score; } return score; } if (!promotion_score_applies(rsc, node)) { return 0; } /* For the promotion score attribute name, use the name the resource is * known as in resource history, since that's what crm_attribute --promotion * would have used. */ name = (rsc->clone_name == NULL)? rsc->id : rsc->clone_name; attr_value = promotion_attr_value(rsc, node, name); if (attr_value != NULL) { pcmk__rsc_trace(rsc, "Promotion score for %s on %s = %s", name, pcmk__node_name(node), pcmk__s(attr_value, "(unset)")); } else if (!pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { /* If we don't have any resource history yet, we won't have clone_name. * In that case, for anonymous clones, try the resource name without * any instance number. */ name = clone_strip(rsc->id); if (strcmp(rsc->id, name) != 0) { attr_value = promotion_attr_value(rsc, node, name); pcmk__rsc_trace(rsc, "Promotion score for %s on %s (for %s) = %s", name, pcmk__node_name(node), rsc->id, pcmk__s(attr_value, "(unset)")); } free(name); } if (attr_value == NULL) { return 0; } if (is_default != NULL) { *is_default = false; } return char2score(attr_value); } /*! * \internal * \brief Include promotion scores in instances' node scores and priorities * * \param[in,out] rsc Promotable clone resource to update */ void pcmk__add_promotion_scores(pcmk_resource_t *rsc) { if (pe__set_clone_flag(rsc, pcmk__clone_promotion_added) == pcmk_rc_already) { return; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) iter->data; GHashTableIter iter; pcmk_node_t *node = NULL; int score, new_score; g_hash_table_iter_init(&iter, child_rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (!pcmk__node_available(node, false, false)) { /* This node will never be promoted, so don't apply the * promotion score, as that may lead to clone shuffling. */ continue; } score = promotion_score(child_rsc, node, NULL); if (score > 0) { new_score = pcmk__add_scores(node->weight, score); if (new_score != node->weight) { // Could remain INFINITY node->weight = new_score; pcmk__rsc_trace(rsc, "Added %s promotion priority (%s) to score " "on %s (now %s)", child_rsc->id, pcmk_readable_score(score), pcmk__node_name(node), pcmk_readable_score(new_score)); } } if (score > child_rsc->priority) { pcmk__rsc_trace(rsc, "Updating %s priority to promotion score " "(%d->%d)", child_rsc->id, child_rsc->priority, score); child_rsc->priority = score; } } } } /*! * \internal * \brief If a resource's current role is started, change it to unpromoted * * \param[in,out] data Resource to update * \param[in] user_data Ignored */ static void set_current_role_unpromoted(void *data, void *user_data) { pcmk_resource_t *rsc = (pcmk_resource_t *) data; if (rsc->role == pcmk_role_started) { // Promotable clones should use unpromoted role instead of started rsc->role = pcmk_role_unpromoted; } g_list_foreach(rsc->children, set_current_role_unpromoted, NULL); } /*! * \internal * \brief Set a resource's next role to unpromoted (or stopped if unassigned) * * \param[in,out] data Resource to update * \param[in] user_data Ignored */ static void set_next_role_unpromoted(void *data, void *user_data) { pcmk_resource_t *rsc = (pcmk_resource_t *) data; GList *assigned = NULL; rsc->fns->location(rsc, &assigned, FALSE); if (assigned == NULL) { pe__set_next_role(rsc, pcmk_role_stopped, "stopped instance"); } else { pe__set_next_role(rsc, pcmk_role_unpromoted, "unpromoted instance"); g_list_free(assigned); } g_list_foreach(rsc->children, set_next_role_unpromoted, NULL); } /*! * \internal * \brief Set a resource's next role to promoted if not already set * * \param[in,out] data Resource to update * \param[in] user_data Ignored */ static void set_next_role_promoted(void *data, gpointer user_data) { pcmk_resource_t *rsc = (pcmk_resource_t *) data; if (rsc->next_role == pcmk_role_unknown) { pe__set_next_role(rsc, pcmk_role_promoted, "promoted instance"); } g_list_foreach(rsc->children, set_next_role_promoted, NULL); } /*! * \internal * \brief Show instance's promotion score on node where it will be active * * \param[in,out] instance Promotable clone instance to show */ static void show_promotion_score(pcmk_resource_t *instance) { pcmk_node_t *chosen = instance->fns->location(instance, NULL, FALSE); if (pcmk_is_set(instance->cluster->flags, pcmk_sched_output_scores) && !pcmk__is_daemon && (instance->cluster->priv != NULL)) { pcmk__output_t *out = instance->cluster->priv; out->message(out, "promotion-score", instance, chosen, pcmk_readable_score(instance->sort_index)); } else { pcmk__rsc_debug(pe__const_top_resource(instance, false), "%s promotion score on %s: sort=%s priority=%s", instance->id, ((chosen == NULL)? "none" : pcmk__node_name(chosen)), pcmk_readable_score(instance->sort_index), pcmk_readable_score(instance->priority)); } } /*! * \internal * \brief Set a clone instance's promotion priority * * \param[in,out] data Promotable clone instance to update * \param[in] user_data Instance's parent clone */ static void set_instance_priority(gpointer data, gpointer user_data) { pcmk_resource_t *instance = (pcmk_resource_t *) data; const pcmk_resource_t *clone = (const pcmk_resource_t *) user_data; const pcmk_node_t *chosen = NULL; enum rsc_role_e next_role = pcmk_role_unknown; GList *list = NULL; pcmk__rsc_trace(clone, "Assigning priority for %s: %s", instance->id, pcmk_role_text(instance->next_role)); if (instance->fns->state(instance, TRUE) == pcmk_role_started) { set_current_role_unpromoted(instance, NULL); } // Only an instance that will be active can be promoted chosen = instance->fns->location(instance, &list, FALSE); if (pcmk__list_of_multiple(list)) { pcmk__config_err("Cannot promote non-colocated child %s", instance->id); } g_list_free(list); if (chosen == NULL) { return; } next_role = instance->fns->state(instance, FALSE); switch (next_role) { case pcmk_role_started: case pcmk_role_unknown: // Set instance priority to its promotion score (or -1 if none) { bool is_default = false; instance->priority = promotion_score(instance, chosen, &is_default); if (is_default) { /* Default to -1 if no value is set. This allows instances * eligible for promotion to be specified based solely on * PCMK_XE_RSC_LOCATION constraints, but prevents any * instance from being promoted if neither a constraint nor * a promotion score is present. */ instance->priority = -1; } } break; case pcmk_role_unpromoted: case pcmk_role_stopped: // Instance can't be promoted - instance->priority = -INFINITY; + instance->priority = -PCMK_SCORE_INFINITY; break; case pcmk_role_promoted: // Nothing needed (re-creating actions after scheduling fencing) break; default: CRM_CHECK(FALSE, crm_err("Unknown resource role %d for %s", next_role, instance->id)); } // Add relevant location constraint scores for promoted role apply_promoted_locations(instance, instance->rsc_location, chosen); apply_promoted_locations(instance, clone->rsc_location, chosen); // Consider instance's role-based colocations with other resources list = pcmk__this_with_colocations(instance); for (GList *iter = list; iter != NULL; iter = iter->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) iter->data; instance->cmds->apply_coloc_score(instance, cons->primary, cons, true); } g_list_free(list); instance->sort_index = instance->priority; if (next_role == pcmk_role_promoted) { - instance->sort_index = INFINITY; + instance->sort_index = PCMK_SCORE_INFINITY; } pcmk__rsc_trace(clone, "Assigning %s priority = %d", instance->id, instance->priority); } /*! * \internal * \brief Set a promotable clone instance's role * * \param[in,out] data Promotable clone instance to update * \param[in,out] user_data Pointer to count of instances chosen for promotion */ static void set_instance_role(gpointer data, gpointer user_data) { pcmk_resource_t *instance = (pcmk_resource_t *) data; int *count = (int *) user_data; const pcmk_resource_t *clone = pe__const_top_resource(instance, false); pcmk_node_t *chosen = NULL; show_promotion_score(instance); if (instance->sort_index < 0) { pcmk__rsc_trace(clone, "Not supposed to promote instance %s", instance->id); } else if ((*count < pe__clone_promoted_max(instance)) || !pcmk_is_set(clone->flags, pcmk_rsc_managed)) { chosen = node_to_be_promoted_on(instance); } if (chosen == NULL) { set_next_role_unpromoted(instance, NULL); return; } if ((instance->role < pcmk_role_promoted) && !pcmk_is_set(instance->cluster->flags, pcmk_sched_quorate) && (instance->cluster->no_quorum_policy == pcmk_no_quorum_freeze)) { crm_notice("Clone instance %s cannot be promoted without quorum", instance->id); set_next_role_unpromoted(instance, NULL); return; } chosen->count++; pcmk__rsc_info(clone, "Choosing %s (%s) on %s for promotion", instance->id, pcmk_role_text(instance->role), pcmk__node_name(chosen)); set_next_role_promoted(instance, NULL); (*count)++; } /*! * \internal * \brief Set roles for all instances of a promotable clone * * \param[in,out] rsc Promotable clone resource to update */ void pcmk__set_instance_roles(pcmk_resource_t *rsc) { int promoted = 0; GHashTableIter iter; pcmk_node_t *node = NULL; // Repurpose count to track the number of promoted instances assigned g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { node->count = 0; } // Set instances' promotion priorities and sort by highest priority first g_list_foreach(rsc->children, set_instance_priority, rsc); sort_promotable_instances(rsc); // Choose the first N eligible instances to be promoted g_list_foreach(rsc->children, set_instance_role, &promoted); pcmk__rsc_info(rsc, "%s: Promoted %d instances of a possible %d", rsc->id, promoted, pe__clone_promoted_max(rsc)); } /*! * * \internal * \brief Create actions for promotable clone instances * * \param[in,out] clone Promotable clone to create actions for * \param[out] any_promoting Will be set true if any instance is promoting * \param[out] any_demoting Will be set true if any instance is demoting */ static void create_promotable_instance_actions(pcmk_resource_t *clone, bool *any_promoting, bool *any_demoting) { for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; instance->cmds->create_actions(instance); check_for_role_change(instance, any_demoting, any_promoting); } } /*! * \internal * \brief Reset each promotable instance's resource priority * * Reset the priority of each instance of a promotable clone to the clone's * priority (after promotion actions are scheduled, when instance priorities * were repurposed as promotion scores). * * \param[in,out] clone Promotable clone to reset */ static void reset_instance_priorities(pcmk_resource_t *clone) { for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; instance->priority = clone->priority; } } /*! * \internal * \brief Create actions specific to promotable clones * * \param[in,out] clone Promotable clone to create actions for */ void pcmk__create_promotable_actions(pcmk_resource_t *clone) { bool any_promoting = false; bool any_demoting = false; // Create actions for each clone instance individually create_promotable_instance_actions(clone, &any_promoting, &any_demoting); // Create pseudo-actions for clone as a whole pe__create_promotable_pseudo_ops(clone, any_promoting, any_demoting); // Undo our temporary repurposing of resource priority for instances reset_instance_priorities(clone); } /*! * \internal * \brief Create internal orderings for a promotable clone's instances * * \param[in,out] clone Promotable clone instance to order */ void pcmk__order_promotable_instances(pcmk_resource_t *clone) { pcmk_resource_t *previous = NULL; // Needed for ordered clones pcmk__promotable_restart_ordering(clone); for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; // Demote before promote pcmk__order_resource_actions(instance, PCMK_ACTION_DEMOTE, instance, PCMK_ACTION_PROMOTE, pcmk__ar_ordered); order_instance_promotion(clone, instance, previous); order_instance_demotion(clone, instance, previous); previous = instance; } } /*! * \internal * \brief Update dependent's allowed nodes for colocation with promotable * * \param[in,out] dependent Dependent resource to update * \param[in] primary Primary resource * \param[in] primary_node Node where an instance of the primary will be * \param[in] colocation Colocation constraint to apply */ static void update_dependent_allowed_nodes(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk_node_t *primary_node, const pcmk__colocation_t *colocation) { GHashTableIter iter; pcmk_node_t *node = NULL; const char *primary_value = NULL; const char *attr = colocation->node_attribute; - if (colocation->score >= INFINITY) { + if (colocation->score >= PCMK_SCORE_INFINITY) { return; // Colocation is mandatory, so allowed node scores don't matter } primary_value = pcmk__colocation_node_attr(primary_node, attr, primary); pcmk__rsc_trace(colocation->primary, "Applying %s (%s with %s on %s by %s @%d) to %s", colocation->id, colocation->dependent->id, colocation->primary->id, pcmk__node_name(primary_node), attr, colocation->score, dependent->id); g_hash_table_iter_init(&iter, dependent->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { const char *dependent_value = pcmk__colocation_node_attr(node, attr, dependent); if (pcmk__str_eq(primary_value, dependent_value, pcmk__str_casei)) { node->weight = pcmk__add_scores(node->weight, colocation->score); pcmk__rsc_trace(colocation->primary, "Added %s score (%s) to %s (now %s)", colocation->id, pcmk_readable_score(colocation->score), pcmk__node_name(node), pcmk_readable_score(node->weight)); } } } /*! * \brief Update dependent for a colocation with a promotable clone * * \param[in] primary Primary resource in the colocation * \param[in,out] dependent Dependent resource in the colocation * \param[in] colocation Colocation constraint to apply */ void pcmk__update_dependent_with_promotable(const pcmk_resource_t *primary, pcmk_resource_t *dependent, const pcmk__colocation_t *colocation) { GList *affected_nodes = NULL; /* Build a list of all nodes where an instance of the primary will be, and * (for optional colocations) update the dependent's allowed node scores for * each one. */ for (GList *iter = primary->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; pcmk_node_t *node = instance->fns->location(instance, NULL, FALSE); if (node == NULL) { continue; } if (instance->fns->state(instance, FALSE) == colocation->primary_role) { update_dependent_allowed_nodes(dependent, primary, node, colocation); affected_nodes = g_list_prepend(affected_nodes, node); } } /* For mandatory colocations, add the primary's node score to the * dependent's node score for each affected node, and ban the dependent * from all other nodes. * * However, skip this for promoted-with-promoted colocations, otherwise * inactive dependent instances can't start (in the unpromoted role). */ - if ((colocation->score >= INFINITY) + if ((colocation->score >= PCMK_SCORE_INFINITY) && ((colocation->dependent_role != pcmk_role_promoted) || (colocation->primary_role != pcmk_role_promoted))) { pcmk__rsc_trace(colocation->primary, "Applying %s (mandatory %s with %s) to %s", colocation->id, colocation->dependent->id, colocation->primary->id, dependent->id); pcmk__colocation_intersect_nodes(dependent, primary, colocation, affected_nodes, true); } g_list_free(affected_nodes); } /*! * \internal * \brief Update dependent priority for colocation with promotable * * \param[in] primary Primary resource in the colocation * \param[in,out] dependent Dependent resource in the colocation * \param[in] colocation Colocation constraint to apply */ void pcmk__update_promotable_dependent_priority(const pcmk_resource_t *primary, pcmk_resource_t *dependent, const pcmk__colocation_t *colocation) { pcmk_resource_t *primary_instance = NULL; // Look for a primary instance where dependent will be primary_instance = pcmk__find_compatible_instance(dependent, primary, colocation->primary_role, false); if (primary_instance != NULL) { // Add primary instance's priority to dependent's int new_priority = pcmk__add_scores(dependent->priority, colocation->score); pcmk__rsc_trace(colocation->primary, "Applying %s (%s with %s) to %s priority " "(%s + %s = %s)", colocation->id, colocation->dependent->id, colocation->primary->id, dependent->id, pcmk_readable_score(dependent->priority), pcmk_readable_score(colocation->score), pcmk_readable_score(new_priority)); dependent->priority = new_priority; - } else if (colocation->score >= INFINITY) { + } else if (colocation->score >= PCMK_SCORE_INFINITY) { // Mandatory colocation, but primary won't be here pcmk__rsc_trace(colocation->primary, "Applying %s (%s with %s) to %s: can't be promoted", colocation->id, colocation->dependent->id, colocation->primary->id, dependent->id); - dependent->priority = -INFINITY; + dependent->priority = -PCMK_SCORE_INFINITY; } } diff --git a/lib/pacemaker/pcmk_sched_resource.c b/lib/pacemaker/pcmk_sched_resource.c index bc21169f96..0791994f8f 100644 --- a/lib/pacemaker/pcmk_sched_resource.c +++ b/lib/pacemaker/pcmk_sched_resource.c @@ -1,774 +1,774 @@ /* * Copyright 2014-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include "libpacemaker_private.h" // Resource assignment methods by resource variant static pcmk_assignment_methods_t assignment_methods[] = { { pcmk__primitive_assign, pcmk__primitive_create_actions, pcmk__probe_rsc_on_node, pcmk__primitive_internal_constraints, pcmk__primitive_apply_coloc_score, pcmk__colocated_resources, pcmk__with_primitive_colocations, pcmk__primitive_with_colocations, pcmk__add_colocated_node_scores, pcmk__apply_location, pcmk__primitive_action_flags, pcmk__update_ordered_actions, pcmk__output_resource_actions, pcmk__add_rsc_actions_to_graph, pcmk__primitive_add_graph_meta, pcmk__primitive_add_utilization, pcmk__primitive_shutdown_lock, }, { pcmk__group_assign, pcmk__group_create_actions, pcmk__probe_rsc_on_node, pcmk__group_internal_constraints, pcmk__group_apply_coloc_score, pcmk__group_colocated_resources, pcmk__with_group_colocations, pcmk__group_with_colocations, pcmk__group_add_colocated_node_scores, pcmk__group_apply_location, pcmk__group_action_flags, pcmk__group_update_ordered_actions, pcmk__output_resource_actions, pcmk__add_rsc_actions_to_graph, pcmk__noop_add_graph_meta, pcmk__group_add_utilization, pcmk__group_shutdown_lock, }, { pcmk__clone_assign, pcmk__clone_create_actions, pcmk__clone_create_probe, pcmk__clone_internal_constraints, pcmk__clone_apply_coloc_score, pcmk__colocated_resources, pcmk__with_clone_colocations, pcmk__clone_with_colocations, pcmk__add_colocated_node_scores, pcmk__clone_apply_location, pcmk__clone_action_flags, pcmk__instance_update_ordered_actions, pcmk__output_resource_actions, pcmk__clone_add_actions_to_graph, pcmk__clone_add_graph_meta, pcmk__clone_add_utilization, pcmk__clone_shutdown_lock, }, { pcmk__bundle_assign, pcmk__bundle_create_actions, pcmk__bundle_create_probe, pcmk__bundle_internal_constraints, pcmk__bundle_apply_coloc_score, pcmk__colocated_resources, pcmk__with_bundle_colocations, pcmk__bundle_with_colocations, pcmk__add_colocated_node_scores, pcmk__bundle_apply_location, pcmk__bundle_action_flags, pcmk__instance_update_ordered_actions, pcmk__output_bundle_actions, pcmk__bundle_add_actions_to_graph, pcmk__noop_add_graph_meta, pcmk__bundle_add_utilization, pcmk__bundle_shutdown_lock, } }; /*! * \internal * \brief Check whether a resource's agent standard, provider, or type changed * * \param[in,out] rsc Resource to check * \param[in,out] node Node needing unfencing if agent changed * \param[in] rsc_entry XML with previously known agent information * \param[in] active_on_node Whether \p rsc is active on \p node * * \return true if agent for \p rsc changed, otherwise false */ bool pcmk__rsc_agent_changed(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *rsc_entry, bool active_on_node) { bool changed = false; const char *attr_list[] = { PCMK_XA_TYPE, PCMK_XA_CLASS, PCMK_XA_PROVIDER, }; for (int i = 0; i < PCMK__NELEM(attr_list); i++) { const char *value = crm_element_value(rsc->xml, attr_list[i]); const char *old_value = crm_element_value(rsc_entry, attr_list[i]); if (!pcmk__str_eq(value, old_value, pcmk__str_none)) { changed = true; trigger_unfencing(rsc, node, "Device definition changed", NULL, rsc->cluster); if (active_on_node) { crm_notice("Forcing restart of %s on %s " "because %s changed from '%s' to '%s'", rsc->id, pcmk__node_name(node), attr_list[i], pcmk__s(old_value, ""), pcmk__s(value, "")); } } } if (changed && active_on_node) { // Make sure the resource is restarted custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, node, FALSE, rsc->cluster); pcmk__set_rsc_flags(rsc, pcmk_rsc_start_pending); } return changed; } /*! * \internal * \brief Add resource (and any matching children) to list if it matches ID * * \param[in] result List to add resource to * \param[in] rsc Resource to check * \param[in] id ID to match * * \return (Possibly new) head of list */ static GList * add_rsc_if_matching(GList *result, pcmk_resource_t *rsc, const char *id) { if ((strcmp(rsc->id, id) == 0) || ((rsc->clone_name != NULL) && (strcmp(rsc->clone_name, id) == 0))) { result = g_list_prepend(result, rsc); } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) iter->data; result = add_rsc_if_matching(result, child, id); } return result; } /*! * \internal * \brief Find all resources matching a given ID by either ID or clone name * * \param[in] id Resource ID to check * \param[in] scheduler Scheduler data * * \return List of all resources that match \p id * \note The caller is responsible for freeing the return value with * g_list_free(). */ GList * pcmk__rscs_matching_id(const char *id, const pcmk_scheduler_t *scheduler) { GList *result = NULL; CRM_CHECK((id != NULL) && (scheduler != NULL), return NULL); for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { result = add_rsc_if_matching(result, (pcmk_resource_t *) iter->data, id); } return result; } /*! * \internal * \brief Set the variant-appropriate assignment methods for a resource * * \param[in,out] data Resource to set assignment methods for * \param[in] user_data Ignored */ static void set_assignment_methods_for_rsc(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; rsc->cmds = &assignment_methods[rsc->variant]; g_list_foreach(rsc->children, set_assignment_methods_for_rsc, NULL); } /*! * \internal * \brief Set the variant-appropriate assignment methods for all resources * * \param[in,out] scheduler Scheduler data */ void pcmk__set_assignment_methods(pcmk_scheduler_t *scheduler) { g_list_foreach(scheduler->resources, set_assignment_methods_for_rsc, NULL); } /*! * \internal * \brief Wrapper for colocated_resources() method for readability * * \param[in] rsc Resource to add to colocated list * \param[in] orig_rsc Resource originally requested * \param[in,out] list Pointer to list to add to * * \return (Possibly new) head of list */ static inline void add_colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { *list = rsc->cmds->colocated_resources(rsc, orig_rsc, *list); } // Shared implementation of pcmk_assignment_methods_t:colocated_resources() GList * pcmk__colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs) { const GList *iter = NULL; GList *colocations = NULL; if (orig_rsc == NULL) { orig_rsc = rsc; } if ((rsc == NULL) || (g_list_find(colocated_rscs, rsc) != NULL)) { return colocated_rscs; } pcmk__rsc_trace(orig_rsc, "%s is in colocation chain with %s", rsc->id, orig_rsc->id); colocated_rscs = g_list_prepend(colocated_rscs, (gpointer) rsc); // Follow colocations where this resource is the dependent resource colocations = pcmk__this_with_colocations(rsc); for (iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *constraint = iter->data; const pcmk_resource_t *primary = constraint->primary; if (primary == orig_rsc) { continue; // Break colocation loop } - if ((constraint->score == INFINITY) && + if ((constraint->score == PCMK_SCORE_INFINITY) && (pcmk__colocation_affects(rsc, primary, constraint, true) == pcmk__coloc_affects_location)) { add_colocated_resources(primary, orig_rsc, &colocated_rscs); } } g_list_free(colocations); // Follow colocations where this resource is the primary resource colocations = pcmk__with_this_colocations(rsc); for (iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *constraint = iter->data; const pcmk_resource_t *dependent = constraint->dependent; if (dependent == orig_rsc) { continue; // Break colocation loop } if (pcmk__is_clone(rsc) && !pcmk__is_clone(dependent)) { continue; // We can't be sure whether dependent will be colocated } - if ((constraint->score == INFINITY) && + if ((constraint->score == PCMK_SCORE_INFINITY) && (pcmk__colocation_affects(dependent, rsc, constraint, true) == pcmk__coloc_affects_location)) { add_colocated_resources(dependent, orig_rsc, &colocated_rscs); } } g_list_free(colocations); return colocated_rscs; } // No-op function for variants that don't need to implement add_graph_meta() void pcmk__noop_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml) { } /*! * \internal * \brief Output a summary of scheduled actions for a resource * * \param[in,out] rsc Resource to output actions for */ void pcmk__output_resource_actions(pcmk_resource_t *rsc) { pcmk_node_t *next = NULL; pcmk_node_t *current = NULL; pcmk__output_t *out = NULL; CRM_ASSERT(rsc != NULL); out = rsc->cluster->priv; if (rsc->children != NULL) { for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) iter->data; child->cmds->output_actions(child); } return; } next = rsc->allocated_to; if (rsc->running_on) { current = pcmk__current_node(rsc); if (rsc->role == pcmk_role_stopped) { /* This can occur when resources are being recovered because * the current role can change in pcmk__primitive_create_actions() */ rsc->role = pcmk_role_started; } } if ((current == NULL) && pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { /* Don't log stopped orphans */ return; } out->message(out, "rsc-action", rsc, current, next); } /*! * \internal * \brief Add a resource to a node's list of assigned resources * * \param[in,out] node Node to add resource to * \param[in] rsc Resource to add */ static inline void add_assigned_resource(pcmk_node_t *node, pcmk_resource_t *rsc) { node->details->allocated_rsc = g_list_prepend(node->details->allocated_rsc, rsc); } /*! * \internal * \brief Assign a specified resource (of any variant) to a node * * Assign a specified resource and its children (if any) to a specified node, if * the node can run the resource (or unconditionally, if \p force is true). Mark * the resources as no longer provisional. * * If a resource can't be assigned (or \p node is \c NULL), unassign any * previous assignment. If \p stop_if_fail is \c true, set next role to stopped * and update any existing actions scheduled for the resource. * * \param[in,out] rsc Resource to assign * \param[in,out] node Node to assign \p rsc to * \param[in] force If true, assign to \p node even if unavailable * \param[in] stop_if_fail If \c true and either \p rsc can't be assigned * or \p chosen is \c NULL, set next role to * stopped and update existing actions (if \p rsc * is not a primitive, this applies to its * primitive descendants instead) * * \return \c true if the assignment of \p rsc changed, or \c false otherwise * * \note Assigning a resource to the NULL node using this function is different * from calling pcmk__unassign_resource(), in that it may also update any * actions created for the resource. * \note The \c pcmk_assignment_methods_t:assign() method is preferred, unless * a resource should be assigned to the \c NULL node or every resource in * a tree should be assigned to the same node. * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ bool pcmk__assign_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool force, bool stop_if_fail) { bool changed = false; CRM_ASSERT(rsc != NULL); if (rsc->children != NULL) { for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child_rsc = iter->data; changed |= pcmk__assign_resource(child_rsc, node, force, stop_if_fail); } return changed; } // Assigning a primitive if (!force && (node != NULL) && ((node->weight < 0) // Allow graph to assume that guest node connections will come up || (!pcmk__node_available(node, true, false) && !pcmk__is_guest_or_bundle_node(node)))) { pcmk__rsc_debug(rsc, "All nodes for resource %s are unavailable, unclean or " "shutting down (%s can%s run resources, with score %s)", rsc->id, pcmk__node_name(node), (pcmk__node_available(node, true, false)? "" : "not"), pcmk_readable_score(node->weight)); if (stop_if_fail) { pe__set_next_role(rsc, pcmk_role_stopped, "node availability"); } node = NULL; } if (rsc->allocated_to != NULL) { changed = !pcmk__same_node(rsc->allocated_to, node); } else { changed = (node != NULL); } pcmk__unassign_resource(rsc); pcmk__clear_rsc_flags(rsc, pcmk_rsc_unassigned); if (node == NULL) { char *rc_stopped = NULL; pcmk__rsc_debug(rsc, "Could not assign %s to a node", rsc->id); if (!stop_if_fail) { return changed; } pe__set_next_role(rsc, pcmk_role_stopped, "unable to assign"); for (GList *iter = rsc->actions; iter != NULL; iter = iter->next) { pcmk_action_t *op = (pcmk_action_t *) iter->data; pcmk__rsc_debug(rsc, "Updating %s for %s assignment failure", op->uuid, rsc->id); if (pcmk__str_eq(op->task, PCMK_ACTION_STOP, pcmk__str_none)) { pcmk__clear_action_flags(op, pcmk_action_optional); } else if (pcmk__str_eq(op->task, PCMK_ACTION_START, pcmk__str_none)) { pcmk__clear_action_flags(op, pcmk_action_runnable); } else { // Cancel recurring actions, unless for stopped state const char *interval_ms_s = NULL; const char *target_rc_s = NULL; interval_ms_s = g_hash_table_lookup(op->meta, PCMK_META_INTERVAL); target_rc_s = g_hash_table_lookup(op->meta, PCMK__META_OP_TARGET_RC); if (rc_stopped == NULL) { rc_stopped = pcmk__itoa(PCMK_OCF_NOT_RUNNING); } if (!pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches) && !pcmk__str_eq(rc_stopped, target_rc_s, pcmk__str_none)) { pcmk__clear_action_flags(op, pcmk_action_runnable); } } } free(rc_stopped); return changed; } pcmk__rsc_debug(rsc, "Assigning %s to %s", rsc->id, pcmk__node_name(node)); rsc->allocated_to = pe__copy_node(node); add_assigned_resource(node, rsc); node->details->num_resources++; node->count++; pcmk__consume_node_capacity(node->details->utilization, rsc); if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_show_utilization)) { pcmk__output_t *out = rsc->cluster->priv; out->message(out, "resource-util", rsc, node, __func__); } return changed; } /*! * \internal * \brief Remove any node assignment from a specified resource and its children * * If a specified resource has been assigned to a node, remove that assignment * and mark the resource as provisional again. * * \param[in,out] rsc Resource to unassign * * \note This function is called recursively on \p rsc and its children. */ void pcmk__unassign_resource(pcmk_resource_t *rsc) { pcmk_node_t *old = rsc->allocated_to; if (old == NULL) { crm_info("Unassigning %s", rsc->id); } else { crm_info("Unassigning %s from %s", rsc->id, pcmk__node_name(old)); } pcmk__set_rsc_flags(rsc, pcmk_rsc_unassigned); if (rsc->children == NULL) { if (old == NULL) { return; } rsc->allocated_to = NULL; /* We're going to free the pcmk_node_t, but its details member is shared * and will remain, so update that appropriately first. */ old->details->allocated_rsc = g_list_remove(old->details->allocated_rsc, rsc); old->details->num_resources--; pcmk__release_node_capacity(old->details->utilization, rsc); free(old); return; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk__unassign_resource((pcmk_resource_t *) iter->data); } } /*! * \internal * \brief Check whether a resource has reached its migration threshold on a node * * \param[in,out] rsc Resource to check * \param[in] node Node to check * \param[out] failed If threshold has been reached, this will be set to * resource that failed (possibly a parent of \p rsc) * * \return true if the migration threshold has been reached, false otherwise */ bool pcmk__threshold_reached(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_resource_t **failed) { int fail_count, remaining_tries; pcmk_resource_t *rsc_to_ban = rsc; // Migration threshold of 0 means never force away if (rsc->migration_threshold == 0) { return false; } // If we're ignoring failures, also ignore the migration threshold if (pcmk_is_set(rsc->flags, pcmk_rsc_ignore_failure)) { return false; } // If there are no failures, there's no need to force away fail_count = pe_get_failcount(node, rsc, NULL, pcmk__fc_effective|pcmk__fc_fillers, NULL); if (fail_count <= 0) { return false; } // If failed resource is anonymous clone instance, we'll force clone away if (!pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { rsc_to_ban = uber_parent(rsc); } // How many more times recovery will be tried on this node remaining_tries = rsc->migration_threshold - fail_count; if (remaining_tries <= 0) { pcmk__sched_warn("%s cannot run on %s due to reaching migration " "threshold (clean up resource to allow again)" CRM_XS " failures=%d " PCMK_META_MIGRATION_THRESHOLD "=%d", rsc_to_ban->id, pcmk__node_name(node), fail_count, rsc->migration_threshold); if (failed != NULL) { *failed = rsc_to_ban; } return true; } crm_info("%s can fail %d more time%s on " "%s before reaching migration threshold (%d)", rsc_to_ban->id, remaining_tries, pcmk__plural_s(remaining_tries), pcmk__node_name(node), rsc->migration_threshold); return false; } /*! * \internal * \brief Get a node's score * * \param[in] node Node with ID to check * \param[in] nodes List of nodes to look for \p node score in * * \return Node's score, or -INFINITY if not found */ static int get_node_score(const pcmk_node_t *node, GHashTable *nodes) { pcmk_node_t *found_node = NULL; if ((node != NULL) && (nodes != NULL)) { found_node = g_hash_table_lookup(nodes, node->details->id); } - return (found_node == NULL)? -INFINITY : found_node->weight; + return (found_node == NULL)? -PCMK_SCORE_INFINITY : found_node->weight; } /*! * \internal * \brief Compare two resources according to which should be assigned first * * \param[in] a First resource to compare * \param[in] b Second resource to compare * \param[in] data Sorted list of all nodes in cluster * * \return -1 if \p a should be assigned before \b, 0 if they are equal, * or +1 if \p a should be assigned after \b */ static gint cmp_resources(gconstpointer a, gconstpointer b, gpointer data) { /* GLib insists that this function require gconstpointer arguments, but we * make a small, temporary change to each argument (setting the * pe_rsc_merging flag) during comparison */ pcmk_resource_t *resource1 = (pcmk_resource_t *) a; pcmk_resource_t *resource2 = (pcmk_resource_t *) b; const GList *nodes = data; int rc = 0; - int r1_score = -INFINITY; - int r2_score = -INFINITY; + int r1_score = -PCMK_SCORE_INFINITY; + int r2_score = -PCMK_SCORE_INFINITY; pcmk_node_t *r1_node = NULL; pcmk_node_t *r2_node = NULL; GHashTable *r1_nodes = NULL; GHashTable *r2_nodes = NULL; const char *reason = NULL; // Resources with highest priority should be assigned first reason = "priority"; r1_score = resource1->priority; r2_score = resource2->priority; if (r1_score > r2_score) { rc = -1; goto done; } if (r1_score < r2_score) { rc = 1; goto done; } // We need nodes to make any other useful comparisons reason = "no node list"; if (nodes == NULL) { goto done; } // Calculate and log node scores resource1->cmds->add_colocated_node_scores(resource1, NULL, resource1->id, &r1_nodes, NULL, 1, pcmk__coloc_select_this_with); resource2->cmds->add_colocated_node_scores(resource2, NULL, resource2->id, &r2_nodes, NULL, 1, pcmk__coloc_select_this_with); pe__show_node_scores(true, NULL, resource1->id, r1_nodes, resource1->cluster); pe__show_node_scores(true, NULL, resource2->id, r2_nodes, resource2->cluster); // The resource with highest score on its current node goes first reason = "current location"; if (resource1->running_on != NULL) { r1_node = pcmk__current_node(resource1); } if (resource2->running_on != NULL) { r2_node = pcmk__current_node(resource2); } r1_score = get_node_score(r1_node, r1_nodes); r2_score = get_node_score(r2_node, r2_nodes); if (r1_score > r2_score) { rc = -1; goto done; } if (r1_score < r2_score) { rc = 1; goto done; } // Otherwise a higher score on any node will do reason = "score"; for (const GList *iter = nodes; iter != NULL; iter = iter->next) { const pcmk_node_t *node = (const pcmk_node_t *) iter->data; r1_score = get_node_score(node, r1_nodes); r2_score = get_node_score(node, r2_nodes); if (r1_score > r2_score) { rc = -1; goto done; } if (r1_score < r2_score) { rc = 1; goto done; } } done: crm_trace("%s (%d)%s%s %c %s (%d)%s%s: %s", resource1->id, r1_score, ((r1_node == NULL)? "" : " on "), ((r1_node == NULL)? "" : r1_node->details->id), ((rc < 0)? '>' : ((rc > 0)? '<' : '=')), resource2->id, r2_score, ((r2_node == NULL)? "" : " on "), ((r2_node == NULL)? "" : r2_node->details->id), reason); if (r1_nodes != NULL) { g_hash_table_destroy(r1_nodes); } if (r2_nodes != NULL) { g_hash_table_destroy(r2_nodes); } return rc; } /*! * \internal * \brief Sort resources in the order they should be assigned to nodes * * \param[in,out] scheduler Scheduler data */ void pcmk__sort_resources(pcmk_scheduler_t *scheduler) { GList *nodes = g_list_copy(scheduler->nodes); nodes = pcmk__sort_nodes(nodes, NULL); scheduler->resources = g_list_sort_with_data(scheduler->resources, cmp_resources, nodes); g_list_free(nodes); } diff --git a/lib/pacemaker/pcmk_sched_tickets.c b/lib/pacemaker/pcmk_sched_tickets.c index 0eb6e8513c..38f0eb09f2 100644 --- a/lib/pacemaker/pcmk_sched_tickets.c +++ b/lib/pacemaker/pcmk_sched_tickets.c @@ -1,532 +1,532 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include "libpacemaker_private.h" enum loss_ticket_policy { loss_ticket_stop, loss_ticket_demote, loss_ticket_fence, loss_ticket_freeze }; typedef struct { const char *id; pcmk_resource_t *rsc; pcmk_ticket_t *ticket; enum loss_ticket_policy loss_policy; int role; } rsc_ticket_t; /*! * \brief Check whether a ticket constraint matches a resource by role * * \param[in] rsc_ticket Ticket constraint * \param[in] rsc Resource to compare with ticket * * \param[in] true if constraint has no role or resource's role matches * constraint's, otherwise false */ static bool ticket_role_matches(const pcmk_resource_t *rsc, const rsc_ticket_t *rsc_ticket) { if ((rsc_ticket->role == pcmk_role_unknown) || (rsc_ticket->role == rsc->role)) { return true; } pcmk__rsc_trace(rsc, "Skipping constraint: \"%s\" state filter", pcmk_role_text(rsc_ticket->role)); return false; } /*! * \brief Create location constraints and fencing as needed for a ticket * * \param[in,out] rsc Resource affected by ticket * \param[in] rsc_ticket Ticket */ static void constraints_for_ticket(pcmk_resource_t *rsc, const rsc_ticket_t *rsc_ticket) { GList *iter = NULL; CRM_CHECK((rsc != NULL) && (rsc_ticket != NULL), return); if (rsc_ticket->ticket->granted && !rsc_ticket->ticket->standby) { return; } if (rsc->children) { pcmk__rsc_trace(rsc, "Processing ticket dependencies from %s", rsc->id); for (iter = rsc->children; iter != NULL; iter = iter->next) { constraints_for_ticket((pcmk_resource_t *) iter->data, rsc_ticket); } return; } pcmk__rsc_trace(rsc, "%s: Processing ticket dependency on %s (%s, %s)", rsc->id, rsc_ticket->ticket->id, rsc_ticket->id, pcmk_role_text(rsc_ticket->role)); if (!rsc_ticket->ticket->granted && (rsc->running_on != NULL)) { switch (rsc_ticket->loss_policy) { case loss_ticket_stop: - resource_location(rsc, NULL, -INFINITY, "__loss_of_ticket__", - rsc->cluster); + resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, + "__loss_of_ticket__", rsc->cluster); break; case loss_ticket_demote: // Promotion score will be set to -INFINITY in promotion_order() if (rsc_ticket->role != pcmk_role_promoted) { - resource_location(rsc, NULL, -INFINITY, + resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__loss_of_ticket__", rsc->cluster); } break; case loss_ticket_fence: if (!ticket_role_matches(rsc, rsc_ticket)) { return; } - resource_location(rsc, NULL, -INFINITY, "__loss_of_ticket__", - rsc->cluster); + resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, + "__loss_of_ticket__", rsc->cluster); for (iter = rsc->running_on; iter != NULL; iter = iter->next) { pe_fence_node(rsc->cluster, (pcmk_node_t *) iter->data, "deadman ticket was lost", FALSE); } break; case loss_ticket_freeze: if (!ticket_role_matches(rsc, rsc_ticket)) { return; } if (rsc->running_on != NULL) { pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed); pcmk__set_rsc_flags(rsc, pcmk_rsc_blocked); } break; } } else if (!rsc_ticket->ticket->granted) { if ((rsc_ticket->role != pcmk_role_promoted) || (rsc_ticket->loss_policy == loss_ticket_stop)) { - resource_location(rsc, NULL, -INFINITY, "__no_ticket__", - rsc->cluster); + resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, + "__no_ticket__", rsc->cluster); } } else if (rsc_ticket->ticket->standby) { if ((rsc_ticket->role != pcmk_role_promoted) || (rsc_ticket->loss_policy == loss_ticket_stop)) { - resource_location(rsc, NULL, -INFINITY, "__ticket_standby__", - rsc->cluster); + resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, + "__ticket_standby__", rsc->cluster); } } } static void rsc_ticket_new(const char *id, pcmk_resource_t *rsc, pcmk_ticket_t *ticket, const char *state, const char *loss_policy) { rsc_ticket_t *new_rsc_ticket = NULL; if (rsc == NULL) { pcmk__config_err("Ignoring ticket '%s' because resource " "does not exist", id); return; } new_rsc_ticket = calloc(1, sizeof(rsc_ticket_t)); if (new_rsc_ticket == NULL) { return; } if (pcmk__str_eq(state, PCMK__ROLE_STARTED, pcmk__str_null_matches|pcmk__str_casei)) { state = PCMK__ROLE_UNKNOWN; } new_rsc_ticket->id = id; new_rsc_ticket->ticket = ticket; new_rsc_ticket->rsc = rsc; new_rsc_ticket->role = pcmk_parse_role(state); if (pcmk__str_eq(loss_policy, PCMK_VALUE_FENCE, pcmk__str_casei)) { if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { new_rsc_ticket->loss_policy = loss_ticket_fence; } else { pcmk__config_err("Resetting '" PCMK_XA_LOSS_POLICY "' " "for ticket '%s' to '" PCMK_VALUE_STOP "' " "because fencing is not configured", ticket->id); loss_policy = PCMK_VALUE_STOP; } } if (new_rsc_ticket->loss_policy == loss_ticket_fence) { crm_debug("On loss of ticket '%s': Fence the nodes running %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); } else if (pcmk__str_eq(loss_policy, PCMK_VALUE_FREEZE, pcmk__str_casei)) { crm_debug("On loss of ticket '%s': Freeze %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); new_rsc_ticket->loss_policy = loss_ticket_freeze; } else if (pcmk__str_eq(loss_policy, PCMK_VALUE_DEMOTE, pcmk__str_casei)) { crm_debug("On loss of ticket '%s': Demote %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); new_rsc_ticket->loss_policy = loss_ticket_demote; } else if (pcmk__str_eq(loss_policy, PCMK_VALUE_STOP, pcmk__str_casei)) { crm_debug("On loss of ticket '%s': Stop %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); new_rsc_ticket->loss_policy = loss_ticket_stop; } else { if (new_rsc_ticket->role == pcmk_role_promoted) { crm_debug("On loss of ticket '%s': Default to demote %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); new_rsc_ticket->loss_policy = loss_ticket_demote; } else { crm_debug("On loss of ticket '%s': Default to stop %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); new_rsc_ticket->loss_policy = loss_ticket_stop; } } pcmk__rsc_trace(rsc, "%s (%s) ==> %s", rsc->id, pcmk_role_text(new_rsc_ticket->role), ticket->id); rsc->rsc_tickets = g_list_append(rsc->rsc_tickets, new_rsc_ticket); rsc->cluster->ticket_constraints = g_list_append( rsc->cluster->ticket_constraints, new_rsc_ticket); if (!(new_rsc_ticket->ticket->granted) || new_rsc_ticket->ticket->standby) { constraints_for_ticket(rsc, new_rsc_ticket); } } // \return Standard Pacemaker return code static int unpack_rsc_ticket_set(xmlNode *set, pcmk_ticket_t *ticket, const char *loss_policy, pcmk_scheduler_t *scheduler) { const char *set_id = NULL; const char *role = NULL; CRM_CHECK(set != NULL, return EINVAL); CRM_CHECK(ticket != NULL, return EINVAL); set_id = pcmk__xe_id(set); if (set_id == NULL) { pcmk__config_err("Ignoring <" PCMK_XE_RESOURCE_SET "> without " PCMK_XA_ID); return pcmk_rc_unpack_error; } role = crm_element_value(set, PCMK_XA_ROLE); for (xmlNode *xml_rsc = first_named_child(set, PCMK_XE_RESOURCE_REF); xml_rsc != NULL; xml_rsc = crm_next_same_xml(xml_rsc)) { pcmk_resource_t *resource = NULL; resource = pcmk__find_constraint_resource(scheduler->resources, pcmk__xe_id(xml_rsc)); if (resource == NULL) { pcmk__config_err("%s: No resource found for %s", set_id, pcmk__xe_id(xml_rsc)); return pcmk_rc_unpack_error; } pcmk__rsc_trace(resource, "Resource '%s' depends on ticket '%s'", resource->id, ticket->id); rsc_ticket_new(set_id, resource, ticket, role, loss_policy); } return pcmk_rc_ok; } static void unpack_simple_rsc_ticket(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *ticket_str = crm_element_value(xml_obj, PCMK_XA_TICKET); const char *loss_policy = crm_element_value(xml_obj, PCMK_XA_LOSS_POLICY); pcmk_ticket_t *ticket = NULL; const char *rsc_id = crm_element_value(xml_obj, PCMK_XA_RSC); const char *state = crm_element_value(xml_obj, PCMK_XA_RSC_ROLE); // @COMPAT: Deprecated since 2.1.5 const char *instance = crm_element_value(xml_obj, PCMK__XA_RSC_INSTANCE); pcmk_resource_t *rsc = NULL; if (instance != NULL) { pcmk__warn_once(pcmk__wo_coloc_inst, "Support for " PCMK__XA_RSC_INSTANCE " is deprecated " "and will be removed in a future release"); } CRM_CHECK(xml_obj != NULL, return); id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("Ignoring <%s> constraint without " PCMK_XA_ID, xml_obj->name); return; } if (ticket_str == NULL) { pcmk__config_err("Ignoring constraint '%s' without ticket specified", id); return; } else { ticket = g_hash_table_lookup(scheduler->tickets, ticket_str); } if (ticket == NULL) { pcmk__config_err("Ignoring constraint '%s' because ticket '%s' " "does not exist", id, ticket_str); return; } if (rsc_id == NULL) { pcmk__config_err("Ignoring constraint '%s' without resource", id); return; } else { rsc = pcmk__find_constraint_resource(scheduler->resources, rsc_id); } if (rsc == NULL) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "does not exist", id, rsc_id); return; } else if ((instance != NULL) && !pcmk__is_clone(rsc)) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "is not a clone but instance '%s' was requested", id, rsc_id, instance); return; } if (instance != NULL) { rsc = find_clone_instance(rsc, instance); if (rsc == NULL) { pcmk__config_warn("Ignoring constraint '%s' because resource '%s' " "does not have an instance '%s'", "'%s'", id, rsc_id, instance); return; } } rsc_ticket_new(id, rsc, ticket, state, loss_policy); } // \return Standard Pacemaker return code static int unpack_rsc_ticket_tags(xmlNode *xml_obj, xmlNode **expanded_xml, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *rsc_id = NULL; const char *state = NULL; pcmk_resource_t *rsc = NULL; pcmk_tag_t *tag = NULL; xmlNode *rsc_set = NULL; *expanded_xml = NULL; CRM_CHECK(xml_obj != NULL, return EINVAL); id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("Ignoring <%s> constraint without " PCMK_XA_ID, xml_obj->name); return pcmk_rc_unpack_error; } // Check whether there are any resource sets with template or tag references *expanded_xml = pcmk__expand_tags_in_sets(xml_obj, scheduler); if (*expanded_xml != NULL) { crm_log_xml_trace(*expanded_xml, "Expanded rsc_ticket"); return pcmk_rc_ok; } rsc_id = crm_element_value(xml_obj, PCMK_XA_RSC); if (rsc_id == NULL) { return pcmk_rc_ok; } if (!pcmk__valid_resource_or_tag(scheduler, rsc_id, &rsc, &tag)) { pcmk__config_err("Ignoring constraint '%s' because '%s' is not a " "valid resource or tag", id, rsc_id); return pcmk_rc_unpack_error; } else if (rsc != NULL) { // No template or tag is referenced return pcmk_rc_ok; } state = crm_element_value(xml_obj, PCMK_XA_RSC_ROLE); *expanded_xml = copy_xml(xml_obj); /* Convert any template or tag reference in "rsc" into ticket * PCMK_XE_RESOURCE_SET */ if (!pcmk__tag_to_set(*expanded_xml, &rsc_set, PCMK_XA_RSC, false, scheduler)) { free_xml(*expanded_xml); *expanded_xml = NULL; return pcmk_rc_unpack_error; } if (rsc_set != NULL) { if (state != NULL) { /* Move PCMK_XA_RSC_ROLE into converted PCMK_XE_RESOURCE_SET as a * PCMK_XA_ROLE attribute */ crm_xml_add(rsc_set, PCMK_XA_ROLE, state); xml_remove_prop(*expanded_xml, PCMK_XA_RSC_ROLE); } } else { free_xml(*expanded_xml); *expanded_xml = NULL; } return pcmk_rc_ok; } void pcmk__unpack_rsc_ticket(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { xmlNode *set = NULL; bool any_sets = false; const char *id = NULL; const char *ticket_str = NULL; pcmk_ticket_t *ticket = NULL; xmlNode *orig_xml = NULL; xmlNode *expanded_xml = NULL; CRM_CHECK(xml_obj != NULL, return); id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("Ignoring <%s> constraint without " PCMK_XA_ID, xml_obj->name); return; } if (scheduler->tickets == NULL) { scheduler->tickets = pcmk__strkey_table(free, destroy_ticket); } ticket_str = crm_element_value(xml_obj, PCMK_XA_TICKET); if (ticket_str == NULL) { pcmk__config_err("Ignoring constraint '%s' without ticket", id); return; } else { ticket = g_hash_table_lookup(scheduler->tickets, ticket_str); } if (ticket == NULL) { ticket = ticket_new(ticket_str, scheduler); if (ticket == NULL) { return; } } if (unpack_rsc_ticket_tags(xml_obj, &expanded_xml, scheduler) != pcmk_rc_ok) { return; } if (expanded_xml != NULL) { orig_xml = xml_obj; xml_obj = expanded_xml; } for (set = first_named_child(xml_obj, PCMK_XE_RESOURCE_SET); set != NULL; set = crm_next_same_xml(set)) { const char *loss_policy = NULL; any_sets = true; set = expand_idref(set, scheduler->input); loss_policy = crm_element_value(xml_obj, PCMK_XA_LOSS_POLICY); if ((set == NULL) // Configuration error, message already logged || (unpack_rsc_ticket_set(set, ticket, loss_policy, scheduler) != pcmk_rc_ok)) { if (expanded_xml != NULL) { free_xml(expanded_xml); } return; } } if (expanded_xml) { free_xml(expanded_xml); xml_obj = orig_xml; } if (!any_sets) { unpack_simple_rsc_ticket(xml_obj, scheduler); } } /*! * \internal * \brief Ban resource from a node if it doesn't have a promotion ticket * * If a resource has tickets for the promoted role, and the ticket is either not * granted or set to standby, then ban the resource from all nodes. * * \param[in,out] rsc Resource to check */ void pcmk__require_promotion_tickets(pcmk_resource_t *rsc) { for (GList *item = rsc->rsc_tickets; item != NULL; item = item->next) { rsc_ticket_t *rsc_ticket = (rsc_ticket_t *) item->data; if ((rsc_ticket->role == pcmk_role_promoted) && (!rsc_ticket->ticket->granted || rsc_ticket->ticket->standby)) { - resource_location(rsc, NULL, -INFINITY, + resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__stateful_without_ticket__", rsc->cluster); } } } diff --git a/lib/pacemaker/pcmk_sched_utilization.c b/lib/pacemaker/pcmk_sched_utilization.c index 86c6772fd6..05e3cf6bae 100644 --- a/lib/pacemaker/pcmk_sched_utilization.c +++ b/lib/pacemaker/pcmk_sched_utilization.c @@ -1,469 +1,469 @@ /* * Copyright 2014-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Get integer utilization from a string * * \param[in] s String representation of a node utilization value * * \return Integer equivalent of \p s * \todo It would make sense to restrict utilization values to nonnegative * integers, but the documentation just says "integers" and we didn't * restrict them initially, so for backward compatibility, allow any * integer. */ static int utilization_value(const char *s) { int value = 0; if ((s != NULL) && (pcmk__scan_min_int(s, &value, INT_MIN) == EINVAL)) { pcmk__config_warn("Using 0 for utilization instead of " "invalid value '%s'", value); value = 0; } return value; } /* * Functions for comparing node capacities */ struct compare_data { const pcmk_node_t *node1; const pcmk_node_t *node2; bool node2_only; int result; }; /*! * \internal * \brief Compare a single utilization attribute for two nodes * * Compare one utilization attribute for two nodes, decrementing the result if * the first node has greater capacity, and incrementing it if the second node * has greater capacity. * * \param[in] key Utilization attribute name to compare * \param[in] value Utilization attribute value to compare * \param[in,out] user_data Comparison data (as struct compare_data*) */ static void compare_utilization_value(gpointer key, gpointer value, gpointer user_data) { int node1_capacity = 0; int node2_capacity = 0; struct compare_data *data = user_data; const char *node2_value = NULL; if (data->node2_only) { if (g_hash_table_lookup(data->node1->details->utilization, key)) { return; // We've already compared this attribute } } else { node1_capacity = utilization_value((const char *) value); } node2_value = g_hash_table_lookup(data->node2->details->utilization, key); node2_capacity = utilization_value(node2_value); if (node1_capacity > node2_capacity) { data->result--; } else if (node1_capacity < node2_capacity) { data->result++; } } /*! * \internal * \brief Compare utilization capacities of two nodes * * \param[in] node1 First node to compare * \param[in] node2 Second node to compare * * \return Negative integer if node1 has more free capacity, * 0 if the capacities are equal, or a positive integer * if node2 has more free capacity */ int pcmk__compare_node_capacities(const pcmk_node_t *node1, const pcmk_node_t *node2) { struct compare_data data = { .node1 = node1, .node2 = node2, .node2_only = false, .result = 0, }; // Compare utilization values that node1 and maybe node2 have g_hash_table_foreach(node1->details->utilization, compare_utilization_value, &data); // Compare utilization values that only node2 has data.node2_only = true; g_hash_table_foreach(node2->details->utilization, compare_utilization_value, &data); return data.result; } /* * Functions for updating node capacities */ struct calculate_data { GHashTable *current_utilization; bool plus; }; /*! * \internal * \brief Update a single utilization attribute with a new value * * \param[in] key Name of utilization attribute to update * \param[in] value Value to add or substract * \param[in,out] user_data Calculation data (as struct calculate_data *) */ static void update_utilization_value(gpointer key, gpointer value, gpointer user_data) { int result = 0; const char *current = NULL; struct calculate_data *data = user_data; current = g_hash_table_lookup(data->current_utilization, key); if (data->plus) { result = utilization_value(current) + utilization_value(value); } else if (current) { result = utilization_value(current) - utilization_value(value); } g_hash_table_replace(data->current_utilization, strdup(key), pcmk__itoa(result)); } /*! * \internal * \brief Subtract a resource's utilization from node capacity * * \param[in,out] current_utilization Current node utilization attributes * \param[in] rsc Resource with utilization to subtract */ void pcmk__consume_node_capacity(GHashTable *current_utilization, const pcmk_resource_t *rsc) { struct calculate_data data = { .current_utilization = current_utilization, .plus = false, }; g_hash_table_foreach(rsc->utilization, update_utilization_value, &data); } /*! * \internal * \brief Add a resource's utilization to node capacity * * \param[in,out] current_utilization Current node utilization attributes * \param[in] rsc Resource with utilization to add */ void pcmk__release_node_capacity(GHashTable *current_utilization, const pcmk_resource_t *rsc) { struct calculate_data data = { .current_utilization = current_utilization, .plus = true, }; g_hash_table_foreach(rsc->utilization, update_utilization_value, &data); } /* * Functions for checking for sufficient node capacity */ struct capacity_data { const pcmk_node_t *node; const char *rsc_id; bool is_enough; }; /*! * \internal * \brief Check whether a single utilization attribute has sufficient capacity * * \param[in] key Name of utilization attribute to check * \param[in] value Amount of utilization required * \param[in,out] user_data Capacity data (as struct capacity_data *) */ static void check_capacity(gpointer key, gpointer value, gpointer user_data) { int required = 0; int remaining = 0; const char *node_value_s = NULL; struct capacity_data *data = user_data; node_value_s = g_hash_table_lookup(data->node->details->utilization, key); required = utilization_value(value); remaining = utilization_value(node_value_s); if (required > remaining) { crm_debug("Remaining capacity for %s on %s (%d) is insufficient " "for resource %s usage (%d)", (const char *) key, pcmk__node_name(data->node), remaining, data->rsc_id, required); data->is_enough = false; } } /*! * \internal * \brief Check whether a node has sufficient capacity for a resource * * \param[in] node Node to check * \param[in] rsc_id ID of resource to check (for debug logs only) * \param[in] utilization Required utilization amounts * * \return true if node has sufficient capacity for resource, otherwise false */ static bool have_enough_capacity(const pcmk_node_t *node, const char *rsc_id, GHashTable *utilization) { struct capacity_data data = { .node = node, .rsc_id = rsc_id, .is_enough = true, }; g_hash_table_foreach(utilization, check_capacity, &data); return data.is_enough; } /*! * \internal * \brief Sum the utilization requirements of a list of resources * * \param[in] orig_rsc Resource being assigned (for logging purposes) * \param[in] rscs Resources whose utilization should be summed * * \return Newly allocated hash table with sum of all utilization values * \note It is the caller's responsibility to free the return value using * g_hash_table_destroy(). */ static GHashTable * sum_resource_utilization(const pcmk_resource_t *orig_rsc, GList *rscs) { GHashTable *utilization = pcmk__strkey_table(free, free); for (GList *iter = rscs; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; rsc->cmds->add_utilization(rsc, orig_rsc, rscs, utilization); } return utilization; } /*! * \internal * \brief Ban resource from nodes with insufficient utilization capacity * * \param[in,out] rsc Resource to check * * \return Allowed node for \p rsc with most spare capacity, if there are no * nodes with enough capacity for \p rsc and all its colocated resources */ const pcmk_node_t * pcmk__ban_insufficient_capacity(pcmk_resource_t *rsc) { bool any_capable = false; char *rscs_id = NULL; pcmk_node_t *node = NULL; const pcmk_node_t *most_capable_node = NULL; GList *colocated_rscs = NULL; GHashTable *unassigned_utilization = NULL; GHashTableIter iter; CRM_CHECK(rsc != NULL, return NULL); // The default placement strategy ignores utilization if (pcmk__str_eq(rsc->cluster->placement_strategy, PCMK_VALUE_DEFAULT, pcmk__str_casei)) { return NULL; } // Check whether any resources are colocated with this one colocated_rscs = rsc->cmds->colocated_resources(rsc, NULL, NULL); if (colocated_rscs == NULL) { return NULL; } rscs_id = crm_strdup_printf("%s and its colocated resources", rsc->id); // If rsc isn't in the list, add it so we include its utilization if (g_list_find(colocated_rscs, rsc) == NULL) { colocated_rscs = g_list_append(colocated_rscs, rsc); } // Sum utilization of colocated resources that haven't been assigned yet unassigned_utilization = sum_resource_utilization(rsc, colocated_rscs); // Check whether any node has enough capacity for all the resources g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (!pcmk__node_available(node, true, false)) { continue; } if (have_enough_capacity(node, rscs_id, unassigned_utilization)) { any_capable = true; } // Keep track of node with most free capacity if ((most_capable_node == NULL) || (pcmk__compare_node_capacities(node, most_capable_node) < 0)) { most_capable_node = node; } } if (any_capable) { // If so, ban resource from any node with insufficient capacity g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (pcmk__node_available(node, true, false) && !have_enough_capacity(node, rscs_id, unassigned_utilization)) { pcmk__rsc_debug(rsc, "%s does not have enough capacity for %s", pcmk__node_name(node), rscs_id); - resource_location(rsc, node, -INFINITY, "__limit_utilization__", - rsc->cluster); + resource_location(rsc, node, -PCMK_SCORE_INFINITY, + "__limit_utilization__", rsc->cluster); } } most_capable_node = NULL; } else { // Otherwise, ban from nodes with insufficient capacity for rsc alone g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (pcmk__node_available(node, true, false) && !have_enough_capacity(node, rsc->id, rsc->utilization)) { pcmk__rsc_debug(rsc, "%s does not have enough capacity for %s", pcmk__node_name(node), rsc->id); - resource_location(rsc, node, -INFINITY, "__limit_utilization__", - rsc->cluster); + resource_location(rsc, node, -PCMK_SCORE_INFINITY, + "__limit_utilization__", rsc->cluster); } } } g_hash_table_destroy(unassigned_utilization); g_list_free(colocated_rscs); free(rscs_id); pe__show_node_scores(true, rsc, "Post-utilization", rsc->allowed_nodes, rsc->cluster); return most_capable_node; } /*! * \internal * \brief Create a new load_stopped pseudo-op for a node * * \param[in,out] node Node to create op for * * \return Newly created load_stopped op */ static pcmk_action_t * new_load_stopped_op(pcmk_node_t *node) { char *load_stopped_task = crm_strdup_printf(PCMK_ACTION_LOAD_STOPPED "_%s", node->details->uname); pcmk_action_t *load_stopped = get_pseudo_op(load_stopped_task, node->details->data_set); if (load_stopped->node == NULL) { load_stopped->node = pe__copy_node(node); pcmk__clear_action_flags(load_stopped, pcmk_action_optional); } free(load_stopped_task); return load_stopped; } /*! * \internal * \brief Create utilization-related internal constraints for a resource * * \param[in,out] rsc Resource to create constraints for * \param[in] allowed_nodes List of allowed next nodes for \p rsc */ void pcmk__create_utilization_constraints(pcmk_resource_t *rsc, const GList *allowed_nodes) { const GList *iter = NULL; pcmk_action_t *load_stopped = NULL; pcmk__rsc_trace(rsc, "Creating utilization constraints for %s - strategy: %s", rsc->id, rsc->cluster->placement_strategy); // "stop rsc then load_stopped" constraints for current nodes for (iter = rsc->running_on; iter != NULL; iter = iter->next) { load_stopped = new_load_stopped_op(iter->data); pcmk__new_ordering(rsc, stop_key(rsc), NULL, NULL, NULL, load_stopped, pcmk__ar_if_on_same_node_or_target, rsc->cluster); } // "load_stopped then start/migrate_to rsc" constraints for allowed nodes for (iter = allowed_nodes; iter; iter = iter->next) { load_stopped = new_load_stopped_op(iter->data); pcmk__new_ordering(NULL, NULL, load_stopped, rsc, start_key(rsc), NULL, pcmk__ar_if_on_same_node_or_target, rsc->cluster); pcmk__new_ordering(NULL, NULL, load_stopped, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_TO, 0), NULL, pcmk__ar_if_on_same_node_or_target, rsc->cluster); } } /*! * \internal * \brief Output node capacities if enabled * * \param[in] desc Prefix for output * \param[in,out] scheduler Scheduler data */ void pcmk__show_node_capacities(const char *desc, pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) { return; } for (const GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { const pcmk_node_t *node = (const pcmk_node_t *) iter->data; pcmk__output_t *out = scheduler->priv; out->message(out, "node-capacity", node, desc); } } diff --git a/lib/pacemaker/pcmk_scheduler.c b/lib/pacemaker/pcmk_scheduler.c index d6da630a8c..5674b98a3c 100644 --- a/lib/pacemaker/pcmk_scheduler.c +++ b/lib/pacemaker/pcmk_scheduler.c @@ -1,821 +1,821 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include "libpacemaker_private.h" CRM_TRACE_INIT_DATA(pacemaker); /*! * \internal * \brief Do deferred action checks after assignment * * When unpacking the resource history, the scheduler checks for resource * configurations that have changed since an action was run. However, at that * time, bundles using the REMOTE_CONTAINER_HACK don't have their final * parameter information, so instead they add a deferred check to a list. This * function processes one entry in that list. * * \param[in,out] rsc Resource that action history is for * \param[in,out] node Node that action history is for * \param[in] rsc_op Action history entry * \param[in] check Type of deferred check to do */ static void check_params(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *rsc_op, enum pcmk__check_parameters check) { const char *reason = NULL; pcmk__op_digest_t *digest_data = NULL; switch (check) { case pcmk__check_active: if (pcmk__check_action_config(rsc, node, rsc_op) && pe_get_failcount(node, rsc, NULL, pcmk__fc_effective, NULL)) { reason = "action definition changed"; } break; case pcmk__check_last_failure: digest_data = rsc_action_digest_cmp(rsc, rsc_op, node, rsc->cluster); switch (digest_data->rc) { case pcmk__digest_unknown: crm_trace("Resource %s history entry %s on %s has " "no digest to compare", rsc->id, pcmk__xe_id(rsc_op), node->details->id); break; case pcmk__digest_match: break; default: reason = "resource parameters have changed"; break; } break; } if (reason != NULL) { pe__clear_failcount(rsc, node, reason, rsc->cluster); } } /*! * \internal * \brief Check whether a resource has failcount clearing scheduled on a node * * \param[in] node Node to check * \param[in] rsc Resource to check * * \return true if \p rsc has failcount clearing scheduled on \p node, * otherwise false */ static bool failcount_clear_action_exists(const pcmk_node_t *node, const pcmk_resource_t *rsc) { GList *list = pe__resource_actions(rsc, node, PCMK_ACTION_CLEAR_FAILCOUNT, TRUE); if (list != NULL) { g_list_free(list); return true; } return false; } /*! * \internal * \brief Ban a resource from a node if it reached its failure threshold there * * \param[in,out] data Resource to check failure threshold for * \param[in] user_data Node to check resource on */ static void check_failure_threshold(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; const pcmk_node_t *node = user_data; // If this is a collective resource, apply recursively to children instead if (rsc->children != NULL) { g_list_foreach(rsc->children, check_failure_threshold, user_data); return; } if (!failcount_clear_action_exists(node, rsc)) { /* Don't force the resource away from this node due to a failcount * that's going to be cleared. * * @TODO Failcount clearing can be scheduled in * pcmk__handle_rsc_config_changes() via process_rsc_history(), or in * schedule_resource_actions() via check_params(). This runs well before * then, so it cannot detect those, meaning we might check the migration * threshold when we shouldn't. Worst case, we stop or move the * resource, then move it back in the next transition. */ pcmk_resource_t *failed = NULL; if (pcmk__threshold_reached(rsc, node, &failed)) { - resource_location(failed, node, -INFINITY, "__fail_limit__", - rsc->cluster); + resource_location(failed, node, -PCMK_SCORE_INFINITY, + "__fail_limit__", rsc->cluster); } } } /*! * \internal * \brief If resource has exclusive discovery, ban node if not allowed * * Location constraints have a PCMK_XA_RESOURCE_DISCOVERY option that allows * users to specify where probes are done for the affected resource. If this is * set to \c exclusive, probes will only be done on nodes listed in exclusive * constraints. This function bans the resource from the node if the node is not * listed. * * \param[in,out] data Resource to check * \param[in] user_data Node to check resource on */ static void apply_exclusive_discovery(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; const pcmk_node_t *node = user_data; if (rsc->exclusive_discover || pe__const_top_resource(rsc, false)->exclusive_discover) { pcmk_node_t *match = NULL; // If this is a collective resource, apply recursively to children g_list_foreach(rsc->children, apply_exclusive_discovery, user_data); match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if ((match != NULL) && (match->rsc_discover_mode != pcmk_probe_exclusive)) { - match->weight = -INFINITY; + match->weight = -PCMK_SCORE_INFINITY; } } } /*! * \internal * \brief Apply stickiness to a resource if appropriate * * \param[in,out] data Resource to check for stickiness * \param[in] user_data Ignored */ static void apply_stickiness(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; pcmk_node_t *node = NULL; // If this is a collective resource, apply recursively to children instead if (rsc->children != NULL) { g_list_foreach(rsc->children, apply_stickiness, NULL); return; } /* A resource is sticky if it is managed, has stickiness configured, and is * active on a single node. */ if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed) || (rsc->stickiness < 1) || !pcmk__list_of_1(rsc->running_on)) { return; } node = rsc->running_on->data; /* In a symmetric cluster, stickiness can always be used. In an * asymmetric cluster, we have to check whether the resource is still * allowed on the node, so we don't keep the resource somewhere it is no * longer explicitly enabled. */ if (!pcmk_is_set(rsc->cluster->flags, pcmk_sched_symmetric_cluster) && (g_hash_table_lookup(rsc->allowed_nodes, node->details->id) == NULL)) { pcmk__rsc_debug(rsc, "Ignoring %s stickiness because the cluster is " "asymmetric and %s is not explicitly allowed", rsc->id, pcmk__node_name(node)); return; } pcmk__rsc_debug(rsc, "Resource %s has %d stickiness on %s", rsc->id, rsc->stickiness, pcmk__node_name(node)); resource_location(rsc, node, rsc->stickiness, "stickiness", rsc->cluster); } /*! * \internal * \brief Apply shutdown locks for all resources as appropriate * * \param[in,out] scheduler Scheduler data */ static void apply_shutdown_locks(pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { return; } for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; rsc->cmds->shutdown_lock(rsc); } } /*! * \internal * \brief Calculate the number of available nodes in the cluster * * \param[in,out] scheduler Scheduler data */ static void count_available_nodes(pcmk_scheduler_t *scheduler) { if (pcmk_is_set(scheduler->flags, pcmk_sched_no_compat)) { return; } // @COMPAT for API backward compatibility only (cluster does not use value) for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; if ((node != NULL) && (node->weight >= 0) && node->details->online && (node->details->type != node_ping)) { scheduler->max_valid_nodes++; } } crm_trace("Online node count: %d", scheduler->max_valid_nodes); } /* * \internal * \brief Apply node-specific scheduling criteria * * After the CIB has been unpacked, process node-specific scheduling criteria * including shutdown locks, location constraints, resource stickiness, * migration thresholds, and exclusive resource discovery. */ static void apply_node_criteria(pcmk_scheduler_t *scheduler) { crm_trace("Applying node-specific scheduling criteria"); apply_shutdown_locks(scheduler); count_available_nodes(scheduler); pcmk__apply_locations(scheduler); g_list_foreach(scheduler->resources, apply_stickiness, NULL); for (GList *node_iter = scheduler->nodes; node_iter != NULL; node_iter = node_iter->next) { for (GList *rsc_iter = scheduler->resources; rsc_iter != NULL; rsc_iter = rsc_iter->next) { check_failure_threshold(rsc_iter->data, node_iter->data); apply_exclusive_discovery(rsc_iter->data, node_iter->data); } } } /*! * \internal * \brief Assign resources to nodes * * \param[in,out] scheduler Scheduler data */ static void assign_resources(pcmk_scheduler_t *scheduler) { GList *iter = NULL; crm_trace("Assigning resources to nodes"); if (!pcmk__str_eq(scheduler->placement_strategy, PCMK_VALUE_DEFAULT, pcmk__str_casei)) { pcmk__sort_resources(scheduler); } pcmk__show_node_capacities("Original", scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_have_remote_nodes)) { /* Assign remote connection resources first (which will also assign any * colocation dependencies). If the connection is migrating, always * prefer the partial migration target. */ for (iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; if (rsc->is_remote_node) { pcmk__rsc_trace(rsc, "Assigning remote connection resource '%s'", rsc->id); rsc->cmds->assign(rsc, rsc->partial_migration_target, true); } } } /* now do the rest of the resources */ for (iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; if (!rsc->is_remote_node) { pcmk__rsc_trace(rsc, "Assigning %s resource '%s'", rsc->xml->name, rsc->id); rsc->cmds->assign(rsc, NULL, true); } } pcmk__show_node_capacities("Remaining", scheduler); } /*! * \internal * \brief Schedule fail count clearing on online nodes if resource is orphaned * * \param[in,out] data Resource to check * \param[in] user_data Ignored */ static void clear_failcounts_if_orphaned(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { return; } crm_trace("Clear fail counts for orphaned resource %s", rsc->id); /* There's no need to recurse into rsc->children because those * should just be unassigned clone instances. */ for (GList *iter = rsc->cluster->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; pcmk_action_t *clear_op = NULL; if (!node->details->online) { continue; } if (pe_get_failcount(node, rsc, NULL, pcmk__fc_effective, NULL) == 0) { continue; } clear_op = pe__clear_failcount(rsc, node, "it is orphaned", rsc->cluster); /* We can't use order_action_then_stop() here because its * pcmk__ar_guest_allowed breaks things */ pcmk__new_ordering(clear_op->rsc, NULL, clear_op, rsc, stop_key(rsc), NULL, pcmk__ar_ordered, rsc->cluster); } } /*! * \internal * \brief Schedule any resource actions needed * * \param[in,out] scheduler Scheduler data */ static void schedule_resource_actions(pcmk_scheduler_t *scheduler) { // Process deferred action checks pe__foreach_param_check(scheduler, check_params); pe__free_param_checks(scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_probe_resources)) { crm_trace("Scheduling probes"); pcmk__schedule_probes(scheduler); } if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { g_list_foreach(scheduler->resources, clear_failcounts_if_orphaned, NULL); } crm_trace("Scheduling resource actions"); for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; rsc->cmds->create_actions(rsc); } } /*! * \internal * \brief Check whether a resource or any of its descendants are managed * * \param[in] rsc Resource to check * * \return true if resource or any descendant is managed, otherwise false */ static bool is_managed(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { return true; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { if (is_managed((pcmk_resource_t *) iter->data)) { return true; } } return false; } /*! * \internal * \brief Check whether any resources in the cluster are managed * * \param[in] scheduler Scheduler data * * \return true if any resource is managed, otherwise false */ static bool any_managed_resources(const pcmk_scheduler_t *scheduler) { for (const GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { if (is_managed((const pcmk_resource_t *) iter->data)) { return true; } } return false; } /*! * \internal * \brief Check whether a node requires fencing * * \param[in] node Node to check * \param[in] have_managed Whether any resource in cluster is managed * * \return true if \p node should be fenced, otherwise false */ static bool needs_fencing(const pcmk_node_t *node, bool have_managed) { return have_managed && node->details->unclean && pe_can_fence(node->details->data_set, node); } /*! * \internal * \brief Check whether a node requires shutdown * * \param[in] node Node to check * * \return true if \p node should be shut down, otherwise false */ static bool needs_shutdown(const pcmk_node_t *node) { if (pcmk__is_pacemaker_remote_node(node)) { /* Do not send shutdown actions for Pacemaker Remote nodes. * @TODO We might come up with a good use for this in the future. */ return false; } return node->details->online && node->details->shutdown; } /*! * \internal * \brief Track and order non-DC fencing * * \param[in,out] list List of existing non-DC fencing actions * \param[in,out] action Fencing action to prepend to \p list * \param[in] scheduler Scheduler data * * \return (Possibly new) head of \p list */ static GList * add_nondc_fencing(GList *list, pcmk_action_t *action, const pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing) && (list != NULL)) { /* Concurrent fencing is disabled, so order each non-DC * fencing in a chain. If there is any DC fencing or * shutdown, it will be ordered after the last action in the * chain later. */ order_actions((pcmk_action_t *) list->data, action, pcmk__ar_ordered); } return g_list_prepend(list, action); } /*! * \internal * \brief Schedule a node for fencing * * \param[in,out] node Node that requires fencing */ static pcmk_action_t * schedule_fencing(pcmk_node_t *node) { pcmk_action_t *fencing = pe_fence_op(node, NULL, FALSE, "node is unclean", FALSE, node->details->data_set); pcmk__sched_warn("Scheduling node %s for fencing", pcmk__node_name(node)); pcmk__order_vs_fence(fencing, node->details->data_set); return fencing; } /*! * \internal * \brief Create and order node fencing and shutdown actions * * \param[in,out] scheduler Scheduler data */ static void schedule_fencing_and_shutdowns(pcmk_scheduler_t *scheduler) { pcmk_action_t *dc_down = NULL; bool integrity_lost = false; bool have_managed = any_managed_resources(scheduler); GList *fencing_ops = NULL; GList *shutdown_ops = NULL; crm_trace("Scheduling fencing and shutdowns as needed"); if (!have_managed) { crm_notice("No fencing will be done until there are resources " "to manage"); } // Check each node for whether it needs fencing or shutdown for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; pcmk_action_t *fencing = NULL; /* Guest nodes are "fenced" by recovering their container resource, * so handle them separately. */ if (pcmk__is_guest_or_bundle_node(node)) { if (node->details->remote_requires_reset && have_managed && pe_can_fence(scheduler, node)) { pcmk__fence_guest(node); } continue; } if (needs_fencing(node, have_managed)) { fencing = schedule_fencing(node); // Track DC and non-DC fence actions separately if (node->details->is_dc) { dc_down = fencing; } else { fencing_ops = add_nondc_fencing(fencing_ops, fencing, scheduler); } } else if (needs_shutdown(node)) { pcmk_action_t *down_op = pcmk__new_shutdown_action(node); // Track DC and non-DC shutdown actions separately if (node->details->is_dc) { dc_down = down_op; } else { shutdown_ops = g_list_prepend(shutdown_ops, down_op); } } if ((fencing == NULL) && node->details->unclean) { integrity_lost = true; pcmk__config_warn("Node %s is unclean but cannot be fenced", pcmk__node_name(node)); } } if (integrity_lost) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("Resource functionality and data integrity " "cannot be guaranteed (configure, enable, " "and test fencing to correct this)"); } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) { crm_notice("Unclean nodes will not be fenced until quorum is " "attained or " PCMK_OPT_NO_QUORUM_POLICY " is set to " PCMK_VALUE_IGNORE); } } if (dc_down != NULL) { /* Order any non-DC shutdowns before any DC shutdown, to avoid repeated * DC elections. However, we don't want to order non-DC shutdowns before * a DC *fencing*, because even though we don't want a node that's * shutting down to become DC, the DC fencing could be ordered before a * clone stop that's also ordered before the shutdowns, thus leading to * a graph loop. */ if (pcmk__str_eq(dc_down->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { pcmk__order_after_each(dc_down, shutdown_ops); } // Order any non-DC fencing before any DC fencing or shutdown if (pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing)) { /* With concurrent fencing, order each non-DC fencing action * separately before any DC fencing or shutdown. */ pcmk__order_after_each(dc_down, fencing_ops); } else if (fencing_ops != NULL) { /* Without concurrent fencing, the non-DC fencing actions are * already ordered relative to each other, so we just need to order * the DC fencing after the last action in the chain (which is the * first item in the list). */ order_actions((pcmk_action_t *) fencing_ops->data, dc_down, pcmk__ar_ordered); } } g_list_free(fencing_ops); g_list_free(shutdown_ops); } static void log_resource_details(pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv; GList *all = NULL; /* Due to the `crm_mon --node=` feature, out->message() for all the * resource-related messages expects a list of nodes that we are allowed to * output information for. Here, we create a wildcard to match all nodes. */ all = g_list_prepend(all, (gpointer) "*"); for (GList *item = scheduler->resources; item != NULL; item = item->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) item->data; // Log all resources except inactive orphans if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed) || (rsc->role != pcmk_role_stopped)) { out->message(out, crm_map_element_name(rsc->xml), 0UL, rsc, all, all); } } g_list_free(all); } static void log_all_actions(pcmk_scheduler_t *scheduler) { /* This only ever outputs to the log, so ignore whatever output object was * previously set and just log instead. */ pcmk__output_t *prev_out = scheduler->priv; pcmk__output_t *out = NULL; if (pcmk__log_output_new(&out) != pcmk_rc_ok) { return; } pe__register_messages(out); pcmk__register_lib_messages(out); pcmk__output_set_log_level(out, LOG_NOTICE); scheduler->priv = out; out->begin_list(out, NULL, NULL, "Actions"); pcmk__output_actions(scheduler); out->end_list(out); out->finish(out, CRM_EX_OK, true, NULL); pcmk__output_free(out); scheduler->priv = prev_out; } /*! * \internal * \brief Log all required but unrunnable actions at trace level * * \param[in] scheduler Scheduler data */ static void log_unrunnable_actions(const pcmk_scheduler_t *scheduler) { const uint64_t flags = pcmk_action_optional |pcmk_action_runnable |pcmk_action_pseudo; crm_trace("Required but unrunnable actions:"); for (const GList *iter = scheduler->actions; iter != NULL; iter = iter->next) { const pcmk_action_t *action = (const pcmk_action_t *) iter->data; if (!pcmk_any_flags_set(action->flags, flags)) { pcmk__log_action("\t", action, true); } } } /*! * \internal * \brief Unpack the CIB for scheduling * * \param[in,out] cib CIB XML to unpack (may be NULL if already unpacked) * \param[in] flags Scheduler flags to set in addition to defaults * \param[in,out] scheduler Scheduler data */ static void unpack_cib(xmlNode *cib, unsigned long long flags, pcmk_scheduler_t *scheduler) { const char* localhost_save = NULL; if (pcmk_is_set(scheduler->flags, pcmk_sched_have_status)) { crm_trace("Reusing previously calculated cluster status"); pcmk__set_scheduler_flags(scheduler, flags); return; } if (scheduler->localhost) { localhost_save = scheduler->localhost; } CRM_ASSERT(cib != NULL); crm_trace("Calculating cluster status"); /* This will zero the entire struct without freeing anything first, so * callers should never call pcmk__schedule_actions() with a populated data * set unless pcmk_sched_have_status is set (i.e. cluster_status() was * previously called, whether directly or via pcmk__schedule_actions()). */ set_working_set_defaults(scheduler); if (localhost_save) { scheduler->localhost = localhost_save; } pcmk__set_scheduler_flags(scheduler, flags); scheduler->input = cib; cluster_status(scheduler); // Sets pcmk_sched_have_status } /*! * \internal * \brief Run the scheduler for a given CIB * * \param[in,out] cib CIB XML to use as scheduler input * \param[in] flags Scheduler flags to set in addition to defaults * \param[in,out] scheduler Scheduler data */ void pcmk__schedule_actions(xmlNode *cib, unsigned long long flags, pcmk_scheduler_t *scheduler) { unpack_cib(cib, flags, scheduler); pcmk__set_assignment_methods(scheduler); pcmk__apply_node_health(scheduler); pcmk__unpack_constraints(scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_validate_only)) { return; } if (!pcmk_is_set(scheduler->flags, pcmk_sched_location_only) && pcmk__is_daemon) { log_resource_details(scheduler); } apply_node_criteria(scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { return; } pcmk__create_internal_constraints(scheduler); pcmk__handle_rsc_config_changes(scheduler); assign_resources(scheduler); schedule_resource_actions(scheduler); /* Remote ordering constraints need to happen prior to calculating fencing * because it is one more place we can mark nodes as needing fencing. */ pcmk__order_remote_connection_actions(scheduler); schedule_fencing_and_shutdowns(scheduler); pcmk__apply_orderings(scheduler); log_all_actions(scheduler); pcmk__create_graph(scheduler); if (get_crm_log_level() == LOG_TRACE) { log_unrunnable_actions(scheduler); } } diff --git a/lib/pacemaker/pcmk_simulate.c b/lib/pacemaker/pcmk_simulate.c index 39b2ad93c5..42f9fb32e1 100644 --- a/lib/pacemaker/pcmk_simulate.c +++ b/lib/pacemaker/pcmk_simulate.c @@ -1,1008 +1,1008 @@ /* * Copyright 2021-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include "libpacemaker_private.h" static pcmk__output_t *out = NULL; static cib_t *fake_cib = NULL; static GList *fake_resource_list = NULL; static const GList *fake_op_fail_list = NULL; static void set_effective_date(pcmk_scheduler_t *scheduler, bool print_original, const char *use_date); /*! * \internal * \brief Create an action name for use in a dot graph * * \param[in] action Action to create name for * \param[in] verbose If true, add action ID to name * * \return Newly allocated string with action name * \note It is the caller's responsibility to free the result. */ static char * create_action_name(const pcmk_action_t *action, bool verbose) { char *action_name = NULL; const char *prefix = ""; const char *action_host = NULL; const char *clone_name = NULL; const char *task = action->task; if (action->node != NULL) { action_host = action->node->details->uname; } else if (!pcmk_is_set(action->flags, pcmk_action_pseudo)) { action_host = ""; } if (pcmk__str_eq(action->task, PCMK_ACTION_CANCEL, pcmk__str_none)) { prefix = "Cancel "; task = action->cancel_task; } if (action->rsc != NULL) { clone_name = action->rsc->clone_name; } if (clone_name != NULL) { char *key = NULL; guint interval_ms = 0; if (pcmk__guint_from_hash(action->meta, PCMK_META_INTERVAL, 0, &interval_ms) != pcmk_rc_ok) { interval_ms = 0; } if (pcmk__strcase_any_of(action->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_NOTIFIED, NULL)) { const char *n_type = g_hash_table_lookup(action->meta, "notify_key_type"); const char *n_task = g_hash_table_lookup(action->meta, "notify_key_operation"); CRM_ASSERT(n_type != NULL); CRM_ASSERT(n_task != NULL); key = pcmk__notify_key(clone_name, n_type, n_task); } else { key = pcmk__op_key(clone_name, task, interval_ms); } if (action_host != NULL) { action_name = crm_strdup_printf("%s%s %s", prefix, key, action_host); } else { action_name = crm_strdup_printf("%s%s", prefix, key); } free(key); } else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { const char *op = g_hash_table_lookup(action->meta, PCMK__META_STONITH_ACTION); action_name = crm_strdup_printf("%s%s '%s' %s", prefix, action->task, op, action_host); } else if (action->rsc && action_host) { action_name = crm_strdup_printf("%s%s %s", prefix, action->uuid, action_host); } else if (action_host) { action_name = crm_strdup_printf("%s%s %s", prefix, action->task, action_host); } else { action_name = crm_strdup_printf("%s", action->uuid); } if (verbose) { char *with_id = crm_strdup_printf("%s (%d)", action_name, action->id); free(action_name); action_name = with_id; } return action_name; } /*! * \internal * \brief Display the status of a cluster * * \param[in,out] scheduler Scheduler data * \param[in] show_opts How to modify display (as pcmk_show_opt_e flags) * \param[in] section_opts Sections to display (as pcmk_section_e flags) * \param[in] title What to use as list title * \param[in] print_spacer Whether to display a spacer first */ static void print_cluster_status(pcmk_scheduler_t *scheduler, uint32_t show_opts, uint32_t section_opts, const char *title, bool print_spacer) { pcmk__output_t *out = scheduler->priv; GList *all = NULL; crm_exit_t stonith_rc = 0; enum pcmk_pacemakerd_state state = pcmk_pacemakerd_state_invalid; section_opts |= pcmk_section_nodes | pcmk_section_resources; show_opts |= pcmk_show_inactive_rscs | pcmk_show_failed_detail; all = g_list_prepend(all, (gpointer) "*"); PCMK__OUTPUT_SPACER_IF(out, print_spacer); out->begin_list(out, NULL, NULL, "%s", title); out->message(out, "cluster-status", scheduler, state, stonith_rc, NULL, pcmk__fence_history_none, section_opts, show_opts, NULL, all, all); out->end_list(out); g_list_free(all); } /*! * \internal * \brief Display a summary of all actions scheduled in a transition * * \param[in,out] scheduler Scheduler data (fully scheduled) * \param[in] print_spacer Whether to display a spacer first */ static void print_transition_summary(pcmk_scheduler_t *scheduler, bool print_spacer) { pcmk__output_t *out = scheduler->priv; PCMK__OUTPUT_SPACER_IF(out, print_spacer); out->begin_list(out, NULL, NULL, "Transition Summary"); pcmk__output_actions(scheduler); out->end_list(out); } /*! * \internal * \brief Reset scheduler input, output, date, and flags * * \param[in,out] scheduler Scheduler data * \param[in] input What to set as cluster input * \param[in] out What to set as cluster output object * \param[in] use_date What to set as cluster's current timestamp * \param[in] flags Group of enum pcmk_scheduler_flags to set */ static void reset(pcmk_scheduler_t *scheduler, xmlNodePtr input, pcmk__output_t *out, const char *use_date, unsigned int flags) { scheduler->input = input; scheduler->priv = out; set_effective_date(scheduler, true, use_date); if (pcmk_is_set(flags, pcmk_sim_sanitized)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_sanitized); } if (pcmk_is_set(flags, pcmk_sim_show_scores)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_output_scores); } if (pcmk_is_set(flags, pcmk_sim_show_utilization)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_show_utilization); } } /*! * \brief Write out a file in dot(1) format describing the actions that will * be taken by the scheduler in response to an input CIB file. * * \param[in,out] scheduler Scheduler data * \param[in] dot_file The filename to write * \param[in] all_actions Write all actions, even those that are optional * or are on unmanaged resources * \param[in] verbose Add extra information, such as action IDs, to the * output * * \return Standard Pacemaker return code */ static int write_sim_dotfile(pcmk_scheduler_t *scheduler, const char *dot_file, bool all_actions, bool verbose) { GList *iter = NULL; FILE *dot_strm = fopen(dot_file, "w"); if (dot_strm == NULL) { return errno; } fprintf(dot_strm, " digraph \"g\" {\n"); for (iter = scheduler->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = (pcmk_action_t *) iter->data; const char *style = "dashed"; const char *font = "black"; const char *color = "black"; char *action_name = create_action_name(action, verbose); if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { font = "orange"; } if (pcmk_is_set(action->flags, pcmk_action_added_to_graph)) { style = PCMK__VALUE_BOLD; color = "green"; } else if ((action->rsc != NULL) && !pcmk_is_set(action->rsc->flags, pcmk_rsc_managed)) { color = "red"; font = "purple"; if (!all_actions) { goto do_not_write; } } else if (pcmk_is_set(action->flags, pcmk_action_optional)) { color = "blue"; if (!all_actions) { goto do_not_write; } } else { color = "red"; CRM_LOG_ASSERT(!pcmk_is_set(action->flags, pcmk_action_runnable)); } pcmk__set_action_flags(action, pcmk_action_added_to_graph); fprintf(dot_strm, "\"%s\" [ style=%s color=\"%s\" fontcolor=\"%s\"]\n", action_name, style, color, font); do_not_write: free(action_name); } for (iter = scheduler->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = (pcmk_action_t *) iter->data; for (GList *before_iter = action->actions_before; before_iter != NULL; before_iter = before_iter->next) { pcmk__related_action_t *before = before_iter->data; char *before_name = NULL; char *after_name = NULL; const char *style = "dashed"; bool optional = true; if (before->state == pe_link_dumped) { optional = false; style = PCMK__VALUE_BOLD; } else if ((uint32_t) before->type == pcmk__ar_none) { continue; } else if (pcmk_is_set(before->action->flags, pcmk_action_added_to_graph) && pcmk_is_set(action->flags, pcmk_action_added_to_graph) && (uint32_t) before->type != pcmk__ar_if_on_same_node_or_target) { optional = false; } if (all_actions || !optional) { before_name = create_action_name(before->action, verbose); after_name = create_action_name(action, verbose); fprintf(dot_strm, "\"%s\" -> \"%s\" [ style = %s]\n", before_name, after_name, style); free(before_name); free(after_name); } } } fprintf(dot_strm, "}\n"); fflush(dot_strm); fclose(dot_strm); return pcmk_rc_ok; } /*! * \brief Profile the configuration updates and scheduler actions in a single * CIB file, printing the profiling timings. * * \note \p scheduler->priv must have been set to a valid \p pcmk__output_t * object before this function is called. * * \param[in] xml_file The CIB file to profile * \param[in] repeat Number of times to run * \param[in,out] scheduler Scheduler data * \param[in] use_date The date to set the cluster's time to (may be NULL) */ static void profile_file(const char *xml_file, long long repeat, pcmk_scheduler_t *scheduler, const char *use_date) { pcmk__output_t *out = scheduler->priv; xmlNode *cib_object = NULL; clock_t start = 0; clock_t end; unsigned long long scheduler_flags = pcmk_sched_no_compat; CRM_ASSERT(out != NULL); cib_object = filename2xml(xml_file); start = clock(); if (pcmk_find_cib_element(cib_object, PCMK_XE_STATUS) == NULL) { create_xml_node(cib_object, PCMK_XE_STATUS); } if (cli_config_update(&cib_object, NULL, FALSE) == FALSE) { free_xml(cib_object); return; } if (validate_xml(cib_object, NULL, FALSE) != TRUE) { free_xml(cib_object); return; } if (pcmk_is_set(scheduler->flags, pcmk_sched_output_scores)) { scheduler_flags |= pcmk_sched_output_scores; } if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) { scheduler_flags |= pcmk_sched_show_utilization; } for (int i = 0; i < repeat; ++i) { xmlNode *input = (repeat == 1)? cib_object : copy_xml(cib_object); scheduler->input = input; set_effective_date(scheduler, false, use_date); pcmk__schedule_actions(input, scheduler_flags, scheduler); pe_reset_working_set(scheduler); } end = clock(); out->message(out, "profile", xml_file, start, end); } void pcmk__profile_dir(const char *dir, long long repeat, pcmk_scheduler_t *scheduler, const char *use_date) { pcmk__output_t *out = scheduler->priv; struct dirent **namelist; int file_num = scandir(dir, &namelist, 0, alphasort); CRM_ASSERT(out != NULL); if (file_num > 0) { struct stat prop; char buffer[FILENAME_MAX]; out->begin_list(out, NULL, NULL, "Timings"); while (file_num--) { if ('.' == namelist[file_num]->d_name[0]) { free(namelist[file_num]); continue; } else if (!pcmk__ends_with_ext(namelist[file_num]->d_name, ".xml")) { free(namelist[file_num]); continue; } snprintf(buffer, sizeof(buffer), "%s/%s", dir, namelist[file_num]->d_name); if (stat(buffer, &prop) == 0 && S_ISREG(prop.st_mode)) { profile_file(buffer, repeat, scheduler, use_date); } free(namelist[file_num]); } free(namelist); out->end_list(out); } } /*! * \brief Set the date of the cluster, either to the value given by * \p use_date, or to the \c PCMK_XA_EXECUTION_DATE value in the CIB. * * \note \p scheduler->priv must have been set to a valid \p pcmk__output_t * object before this function is called. * * \param[in,out] scheduler Scheduler data * \param[in] print_original If \p true, the \c PCMK_XA_EXECUTION_DATE * should also be printed * \param[in] use_date The date to set the cluster's time to * (may be NULL) */ static void set_effective_date(pcmk_scheduler_t *scheduler, bool print_original, const char *use_date) { pcmk__output_t *out = scheduler->priv; time_t original_date = 0; CRM_ASSERT(out != NULL); crm_element_value_epoch(scheduler->input, PCMK_XA_EXECUTION_DATE, &original_date); if (use_date) { scheduler->now = crm_time_new(use_date); out->info(out, "Setting effective cluster time: %s", use_date); crm_time_log(LOG_NOTICE, "Pretending 'now' is", scheduler->now, crm_time_log_date | crm_time_log_timeofday); } else if (original_date != 0) { scheduler->now = pcmk__copy_timet(original_date); if (print_original) { char *when = crm_time_as_string(scheduler->now, crm_time_log_date|crm_time_log_timeofday); out->info(out, "Using the original execution date of: %s", when); free(when); } } } /*! * \internal * \brief Simulate successfully executing a pseudo-action in a graph * * \param[in,out] graph Graph to update with pseudo-action result * \param[in,out] action Pseudo-action to simulate executing * * \return Standard Pacemaker return code */ static int simulate_pseudo_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *node = crm_element_value(action->xml, PCMK__META_ON_NODE); const char *task = crm_element_value(action->xml, PCMK__XA_OPERATION_KEY); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); out->message(out, "inject-pseudo-action", node, task); pcmk__update_graph(graph, action); return pcmk_rc_ok; } /*! * \internal * \brief Simulate executing a resource action in a graph * * \param[in,out] graph Graph to update with resource action result * \param[in,out] action Resource action to simulate executing * * \return Standard Pacemaker return code */ static int simulate_resource_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { int rc; lrmd_event_data_t *op = NULL; int target_outcome = PCMK_OCF_OK; const char *rtype = NULL; const char *rclass = NULL; const char *resource = NULL; const char *rprovider = NULL; const char *resource_config_name = NULL; const char *operation = crm_element_value(action->xml, PCMK_XA_OPERATION); const char *target_rc_s = crm_meta_value(action->params, PCMK__META_OP_TARGET_RC); xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; xmlNode *action_rsc = first_named_child(action->xml, PCMK_XE_PRIMITIVE); char *node = crm_element_value_copy(action->xml, PCMK__META_ON_NODE); char *uuid = NULL; const char *router_node = crm_element_value(action->xml, PCMK__XA_ROUTER_NODE); // Certain actions don't need to be displayed or history entries if (pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_none)) { crm_debug("No history injection for %s op on %s", operation, node); goto done; // Confirm action and update graph } if (action_rsc == NULL) { // Shouldn't be possible crm_log_xml_err(action->xml, "Bad"); free(node); return EPROTO; } /* A resource might be known by different names in the configuration and in * the action (for example, a clone instance). Grab the configuration name * (which is preferred when writing history), and if necessary, the instance * name. */ resource_config_name = crm_element_value(action_rsc, PCMK_XA_ID); if (resource_config_name == NULL) { // Shouldn't be possible crm_log_xml_err(action->xml, "No ID"); free(node); return EPROTO; } resource = resource_config_name; if (pe_find_resource(fake_resource_list, resource) == NULL) { const char *longname = crm_element_value(action_rsc, PCMK__XA_LONG_ID); if ((longname != NULL) && (pe_find_resource(fake_resource_list, longname) != NULL)) { resource = longname; } } // Certain actions need to be displayed but don't need history entries if (pcmk__strcase_any_of(operation, PCMK_ACTION_DELETE, PCMK_ACTION_META_DATA, NULL)) { out->message(out, "inject-rsc-action", resource, operation, node, (guint) 0); goto done; // Confirm action and update graph } rclass = crm_element_value(action_rsc, PCMK_XA_CLASS); rtype = crm_element_value(action_rsc, PCMK_XA_TYPE); rprovider = crm_element_value(action_rsc, PCMK_XA_PROVIDER); pcmk__scan_min_int(target_rc_s, &target_outcome, 0); CRM_ASSERT(fake_cib->cmds->query(fake_cib, NULL, NULL, cib_sync_call|cib_scope_local) == pcmk_ok); // Ensure the action node is in the CIB uuid = crm_element_value_copy(action->xml, PCMK__META_ON_NODE_UUID); cib_node = pcmk__inject_node(fake_cib, node, ((router_node == NULL)? uuid: node)); free(uuid); CRM_ASSERT(cib_node != NULL); // Add a history entry for the action cib_resource = pcmk__inject_resource_history(out, cib_node, resource, resource_config_name, rclass, rtype, rprovider); if (cib_resource == NULL) { crm_err("Could not simulate action %d history for resource %s", action->id, resource); free(node); free_xml(cib_node); return EINVAL; } // Simulate and display an executor event for the action result op = pcmk__event_from_graph_action(cib_resource, action, PCMK_EXEC_DONE, target_outcome, "User-injected result"); out->message(out, "inject-rsc-action", resource, op->op_type, node, op->interval_ms); // Check whether action is in a list of desired simulated failures for (const GList *iter = fake_op_fail_list; iter != NULL; iter = iter->next) { const char *spec = (const char *) iter->data; char *key = NULL; const char *match_name = NULL; // Allow user to specify anonymous clone with or without instance number key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = resource; } free(key); // If not found, try the resource's name in the configuration if ((match_name == NULL) && (strcmp(resource, resource_config_name) != 0)) { key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource_config_name, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = resource_config_name; } free(key); } if (match_name == NULL) { continue; // This failed action entry doesn't match } // ${match_name}_${task}_${interval_in_ms}@${node}=${rc} rc = sscanf(spec, "%*[^=]=%d", (int *) &op->rc); if (rc != 1) { out->err(out, "Invalid failed operation '%s' " "(result code must be integer)", spec); continue; // Keep checking other list entries } out->info(out, "Pretending action %d failed with rc=%d", action->id, op->rc); pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); - graph->abort_priority = INFINITY; + graph->abort_priority = PCMK_SCORE_INFINITY; pcmk__inject_failcount(out, cib_node, match_name, op->op_type, op->interval_ms, op->rc); break; } pcmk__inject_action_result(cib_resource, op, target_outcome); lrmd_free_event(op); rc = fake_cib->cmds->modify(fake_cib, PCMK_XE_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); done: free(node); free_xml(cib_node); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); pcmk__update_graph(graph, action); return pcmk_rc_ok; } /*! * \internal * \brief Simulate successfully executing a cluster action * * \param[in,out] graph Graph to update with action result * \param[in,out] action Cluster action to simulate * * \return Standard Pacemaker return code */ static int simulate_cluster_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *node = crm_element_value(action->xml, PCMK__META_ON_NODE); const char *task = crm_element_value(action->xml, PCMK_XA_OPERATION); xmlNode *rsc = first_named_child(action->xml, PCMK_XE_PRIMITIVE); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); out->message(out, "inject-cluster-action", node, task, rsc); pcmk__update_graph(graph, action); return pcmk_rc_ok; } /*! * \internal * \brief Simulate successfully executing a fencing action * * \param[in,out] graph Graph to update with action result * \param[in,out] action Fencing action to simulate * * \return Standard Pacemaker return code */ static int simulate_fencing_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *op = crm_meta_value(action->params, PCMK__META_STONITH_ACTION); char *target = crm_element_value_copy(action->xml, PCMK__META_ON_NODE); out->message(out, "inject-fencing-action", target, op); if (!pcmk__str_eq(op, PCMK_ACTION_ON, pcmk__str_casei)) { int rc = pcmk_ok; GString *xpath = g_string_sized_new(512); // Set node state to offline xmlNode *cib_node = pcmk__inject_node_state_change(fake_cib, target, false); CRM_ASSERT(cib_node != NULL); crm_xml_add(cib_node, PCMK_XA_CRM_DEBUG_ORIGIN, __func__); rc = fake_cib->cmds->replace(fake_cib, PCMK_XE_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); // Simulate controller clearing node's resource history and attributes pcmk__g_strcat(xpath, "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='", target, "']/" PCMK__XE_LRM, NULL); fake_cib->cmds->remove(fake_cib, (const char *) xpath->str, NULL, cib_xpath|cib_sync_call|cib_scope_local); g_string_truncate(xpath, 0); pcmk__g_strcat(xpath, "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='", target, "']" "/" PCMK__XE_TRANSIENT_ATTRIBUTES, NULL); fake_cib->cmds->remove(fake_cib, (const char *) xpath->str, NULL, cib_xpath|cib_sync_call|cib_scope_local); free_xml(cib_node); g_string_free(xpath, TRUE); } pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); pcmk__update_graph(graph, action); free(target); return pcmk_rc_ok; } enum pcmk__graph_status pcmk__simulate_transition(pcmk_scheduler_t *scheduler, cib_t *cib, const GList *op_fail_list) { pcmk__graph_t *transition = NULL; enum pcmk__graph_status graph_rc; pcmk__graph_functions_t simulation_fns = { simulate_pseudo_action, simulate_resource_action, simulate_cluster_action, simulate_fencing_action, }; out = scheduler->priv; fake_cib = cib; fake_op_fail_list = op_fail_list; if (!out->is_quiet(out)) { out->begin_list(out, NULL, NULL, "Executing Cluster Transition"); } pcmk__set_graph_functions(&simulation_fns); transition = pcmk__unpack_graph(scheduler->graph, crm_system_name); pcmk__log_graph(LOG_DEBUG, transition); fake_resource_list = scheduler->resources; do { graph_rc = pcmk__execute_graph(transition); } while (graph_rc == pcmk__graph_active); fake_resource_list = NULL; if (graph_rc != pcmk__graph_complete) { out->err(out, "Transition failed: %s", pcmk__graph_status2text(graph_rc)); pcmk__log_graph(LOG_ERR, transition); out->err(out, "An invalid transition was produced"); } pcmk__free_graph(transition); if (!out->is_quiet(out)) { // If not quiet, we'll need the resulting CIB for later display xmlNode *cib_object = NULL; int rc = fake_cib->cmds->query(fake_cib, NULL, &cib_object, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); pe_reset_working_set(scheduler); scheduler->input = cib_object; out->end_list(out); } return graph_rc; } int pcmk__simulate(pcmk_scheduler_t *scheduler, pcmk__output_t *out, const pcmk_injections_t *injections, unsigned int flags, uint32_t section_opts, const char *use_date, const char *input_file, const char *graph_file, const char *dot_file) { int printed = pcmk_rc_no_output; int rc = pcmk_rc_ok; xmlNodePtr input = NULL; cib_t *cib = NULL; rc = cib__signon_query(out, &cib, &input); if (rc != pcmk_rc_ok) { goto simulate_done; } reset(scheduler, input, out, use_date, flags); cluster_status(scheduler); if ((cib->variant == cib_native) && pcmk_is_set(section_opts, pcmk_section_times)) { if (pcmk__our_nodename == NULL) { // Currently used only in the times section pcmk__query_node_name(out, 0, &pcmk__our_nodename, 0); } scheduler->localhost = pcmk__our_nodename; } if (!out->is_quiet(out)) { const bool show_pending = pcmk_is_set(flags, pcmk_sim_show_pending); if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) { printed = out->message(out, "maint-mode", scheduler->flags); } if (scheduler->disabled_resources || scheduler->blocked_resources) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); printed = out->info(out, "%d of %d resource instances DISABLED and " "%d BLOCKED from further action due to failure", scheduler->disabled_resources, scheduler->ninstances, scheduler->blocked_resources); } /* Most formatted output headers use caps for each word, but this one * only has the first word capitalized for compatibility with pcs. */ print_cluster_status(scheduler, (show_pending? pcmk_show_pending : 0), section_opts, "Current cluster status", (printed == pcmk_rc_ok)); printed = pcmk_rc_ok; } // If the user requested any injections, handle them if ((injections->node_down != NULL) || (injections->node_fail != NULL) || (injections->node_up != NULL) || (injections->op_inject != NULL) || (injections->ticket_activate != NULL) || (injections->ticket_grant != NULL) || (injections->ticket_revoke != NULL) || (injections->ticket_standby != NULL) || (injections->watchdog != NULL)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); pcmk__inject_scheduler_input(scheduler, cib, injections); printed = pcmk_rc_ok; rc = cib->cmds->query(cib, NULL, &input, cib_sync_call); if (rc != pcmk_rc_ok) { rc = pcmk_legacy2rc(rc); goto simulate_done; } cleanup_calculations(scheduler); reset(scheduler, input, out, use_date, flags); cluster_status(scheduler); } if (input_file != NULL) { rc = write_xml_file(input, input_file, FALSE); if (rc < 0) { rc = pcmk_legacy2rc(rc); goto simulate_done; } } if (pcmk_any_flags_set(flags, pcmk_sim_process | pcmk_sim_simulate)) { pcmk__output_t *logger_out = NULL; unsigned long long scheduler_flags = pcmk_sched_no_compat; if (pcmk_is_set(scheduler->flags, pcmk_sched_output_scores)) { scheduler_flags |= pcmk_sched_output_scores; } if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) { scheduler_flags |= pcmk_sched_show_utilization; } if (pcmk_all_flags_set(scheduler->flags, pcmk_sched_output_scores |pcmk_sched_show_utilization)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Assignment Scores and Utilization Information"); printed = pcmk_rc_ok; } else if (pcmk_is_set(scheduler->flags, pcmk_sched_output_scores)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Assignment Scores"); printed = pcmk_rc_ok; } else if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Utilization Information"); printed = pcmk_rc_ok; } else { rc = pcmk__log_output_new(&logger_out); if (rc != pcmk_rc_ok) { goto simulate_done; } pe__register_messages(logger_out); pcmk__register_lib_messages(logger_out); scheduler->priv = logger_out; } pcmk__schedule_actions(input, scheduler_flags, scheduler); if (logger_out == NULL) { out->end_list(out); } else { logger_out->finish(logger_out, CRM_EX_OK, true, NULL); pcmk__output_free(logger_out); scheduler->priv = out; } input = NULL; /* Don't try and free it twice */ if (graph_file != NULL) { rc = write_xml_file(scheduler->graph, graph_file, FALSE); if (rc < 0) { rc = pcmk_rc_graph_error; goto simulate_done; } } if (dot_file != NULL) { rc = write_sim_dotfile(scheduler, dot_file, pcmk_is_set(flags, pcmk_sim_all_actions), pcmk_is_set(flags, pcmk_sim_verbose)); if (rc != pcmk_rc_ok) { rc = pcmk_rc_dot_error; goto simulate_done; } } if (!out->is_quiet(out)) { print_transition_summary(scheduler, printed == pcmk_rc_ok); } } rc = pcmk_rc_ok; if (!pcmk_is_set(flags, pcmk_sim_simulate)) { goto simulate_done; } PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); if (pcmk__simulate_transition(scheduler, cib, injections->op_fail) != pcmk__graph_complete) { rc = pcmk_rc_invalid_transition; } if (out->is_quiet(out)) { goto simulate_done; } set_effective_date(scheduler, true, use_date); if (pcmk_is_set(flags, pcmk_sim_show_scores)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_output_scores); } if (pcmk_is_set(flags, pcmk_sim_show_utilization)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_show_utilization); } cluster_status(scheduler); print_cluster_status(scheduler, 0, section_opts, "Revised Cluster Status", true); simulate_done: cib__clean_up_connection(&cib); return rc; } int pcmk_simulate(xmlNodePtr *xml, pcmk_scheduler_t *scheduler, const pcmk_injections_t *injections, unsigned int flags, unsigned int section_opts, const char *use_date, const char *input_file, const char *graph_file, const char *dot_file) { pcmk__output_t *out = NULL; int rc = pcmk_rc_ok; rc = pcmk__xml_output_new(&out, xml); if (rc != pcmk_rc_ok) { return rc; } pe__register_messages(out); pcmk__register_lib_messages(out); rc = pcmk__simulate(scheduler, out, injections, flags, section_opts, use_date, input_file, graph_file, dot_file); pcmk__xml_output_finish(out, pcmk_rc2exitc(rc), xml); return rc; } diff --git a/lib/pengine/bundle.c b/lib/pengine/bundle.c index a6dfac994b..589ea63ad3 100644 --- a/lib/pengine/bundle.c +++ b/lib/pengine/bundle.c @@ -1,2255 +1,2255 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include enum pe__bundle_mount_flags { pe__bundle_mount_none = 0x00, // mount instance-specific subdirectory rather than source directly pe__bundle_mount_subdir = 0x01 }; typedef struct { char *source; char *target; char *options; uint32_t flags; // bitmask of pe__bundle_mount_flags } pe__bundle_mount_t; typedef struct { char *source; char *target; } pe__bundle_port_t; enum pe__container_agent { PE__CONTAINER_AGENT_UNKNOWN, PE__CONTAINER_AGENT_DOCKER, PE__CONTAINER_AGENT_RKT, PE__CONTAINER_AGENT_PODMAN, }; #define PE__CONTAINER_AGENT_UNKNOWN_S "unknown" #define PE__CONTAINER_AGENT_DOCKER_S "docker" #define PE__CONTAINER_AGENT_RKT_S "rkt" #define PE__CONTAINER_AGENT_PODMAN_S "podman" typedef struct pe__bundle_variant_data_s { int promoted_max; int nreplicas; int nreplicas_per_host; char *prefix; char *image; const char *ip_last; char *host_network; char *host_netmask; char *control_port; char *container_network; char *ip_range_start; gboolean add_host; gchar *container_host_options; char *container_command; char *launcher_options; const char *attribute_target; pcmk_resource_t *child; GList *replicas; // pcmk__bundle_replica_t * GList *ports; // pe__bundle_port_t * GList *mounts; // pe__bundle_mount_t * enum pe__container_agent agent_type; } pe__bundle_variant_data_t; #define get_bundle_variant_data(data, rsc) \ CRM_ASSERT(rsc != NULL); \ CRM_ASSERT(rsc->variant == pcmk_rsc_variant_bundle); \ CRM_ASSERT(rsc->variant_opaque != NULL); \ data = (pe__bundle_variant_data_t *) rsc->variant_opaque; /*! * \internal * \brief Get maximum number of bundle replicas allowed to run * * \param[in] rsc Bundle or bundled resource to check * * \return Maximum replicas for bundle corresponding to \p rsc */ int pe__bundle_max(const pcmk_resource_t *rsc) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, pe__const_top_resource(rsc, true)); return bundle_data->nreplicas; } /*! * \internal * \brief Get the resource inside a bundle * * \param[in] bundle Bundle to check * * \return Resource inside \p bundle if any, otherwise NULL */ pcmk_resource_t * pe__bundled_resource(const pcmk_resource_t *rsc) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, pe__const_top_resource(rsc, true)); return bundle_data->child; } /*! * \internal * \brief Get containerized resource corresponding to a given bundle container * * \param[in] instance Collective instance that might be a bundle container * * \return Bundled resource instance inside \p instance if it is a bundle * container instance, otherwise NULL */ const pcmk_resource_t * pe__get_rsc_in_container(const pcmk_resource_t *instance) { const pe__bundle_variant_data_t *data = NULL; const pcmk_resource_t *top = pe__const_top_resource(instance, true); if ((top == NULL) || (top->variant != pcmk_rsc_variant_bundle)) { return NULL; } get_bundle_variant_data(data, top); for (const GList *iter = data->replicas; iter != NULL; iter = iter->next) { const pcmk__bundle_replica_t *replica = iter->data; if (instance == replica->container) { return replica->child; } } return NULL; } /*! * \internal * \brief Check whether a given node is created by a bundle * * \param[in] bundle Bundle resource to check * \param[in] node Node to check * * \return true if \p node is an instance of \p bundle, otherwise false */ bool pe__node_is_bundle_instance(const pcmk_resource_t *bundle, const pcmk_node_t *node) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; if (pcmk__same_node(node, replica->node)) { return true; } } return false; } /*! * \internal * \brief Get the container of a bundle's first replica * * \param[in] bundle Bundle resource to get container for * * \return Container resource from first replica of \p bundle if any, * otherwise NULL */ pcmk_resource_t * pe__first_container(const pcmk_resource_t *bundle) { const pe__bundle_variant_data_t *bundle_data = NULL; const pcmk__bundle_replica_t *replica = NULL; get_bundle_variant_data(bundle_data, bundle); if (bundle_data->replicas == NULL) { return NULL; } replica = bundle_data->replicas->data; return replica->container; } /*! * \internal * \brief Iterate over bundle replicas * * \param[in,out] bundle Bundle to iterate over * \param[in] fn Function to call for each replica (its return value * indicates whether to continue iterating) * \param[in,out] user_data Pointer to pass to \p fn */ void pe__foreach_bundle_replica(pcmk_resource_t *bundle, bool (*fn)(pcmk__bundle_replica_t *, void *), void *user_data) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { if (!fn((pcmk__bundle_replica_t *) iter->data, user_data)) { break; } } } /*! * \internal * \brief Iterate over const bundle replicas * * \param[in] bundle Bundle to iterate over * \param[in] fn Function to call for each replica (its return value * indicates whether to continue iterating) * \param[in,out] user_data Pointer to pass to \p fn */ void pe__foreach_const_bundle_replica(const pcmk_resource_t *bundle, bool (*fn)(const pcmk__bundle_replica_t *, void *), void *user_data) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (const GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { if (!fn((const pcmk__bundle_replica_t *) iter->data, user_data)) { break; } } } static char * next_ip(const char *last_ip) { unsigned int oct1 = 0; unsigned int oct2 = 0; unsigned int oct3 = 0; unsigned int oct4 = 0; int rc = sscanf(last_ip, "%u.%u.%u.%u", &oct1, &oct2, &oct3, &oct4); if (rc != 4) { /*@ TODO check for IPv6 */ return NULL; } else if (oct3 > 253) { return NULL; } else if (oct4 > 253) { ++oct3; oct4 = 1; } else { ++oct4; } return crm_strdup_printf("%u.%u.%u.%u", oct1, oct2, oct3, oct4); } static void allocate_ip(pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica, GString *buffer) { if(data->ip_range_start == NULL) { return; } else if(data->ip_last) { replica->ipaddr = next_ip(data->ip_last); } else { replica->ipaddr = strdup(data->ip_range_start); } data->ip_last = replica->ipaddr; switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: if (data->add_host) { g_string_append_printf(buffer, " --add-host=%s-%d:%s", data->prefix, replica->offset, replica->ipaddr); } else { g_string_append_printf(buffer, " --hosts-entry=%s=%s-%d", replica->ipaddr, data->prefix, replica->offset); } break; case PE__CONTAINER_AGENT_RKT: g_string_append_printf(buffer, " --hosts-entry=%s=%s-%d", replica->ipaddr, data->prefix, replica->offset); break; default: // PE__CONTAINER_AGENT_UNKNOWN break; } } static xmlNode * create_resource(const char *name, const char *provider, const char *kind) { xmlNode *rsc = create_xml_node(NULL, PCMK_XE_PRIMITIVE); crm_xml_add(rsc, PCMK_XA_ID, name); crm_xml_add(rsc, PCMK_XA_CLASS, PCMK_RESOURCE_CLASS_OCF); crm_xml_add(rsc, PCMK_XA_PROVIDER, provider); crm_xml_add(rsc, PCMK_XA_TYPE, kind); return rsc; } /*! * \internal * \brief Check whether cluster can manage resource inside container * * \param[in,out] data Container variant data * * \return TRUE if networking configuration is acceptable, FALSE otherwise * * \note The resource is manageable if an IP range or control port has been * specified. If a control port is used without an IP range, replicas per * host must be 1. */ static bool valid_network(pe__bundle_variant_data_t *data) { if(data->ip_range_start) { return TRUE; } if(data->control_port) { if(data->nreplicas_per_host > 1) { pcmk__config_err("Specifying the '" PCMK_XA_CONTROL_PORT "' for %s " "requires '" PCMK_XA_REPLICAS_PER_HOST "=1'", data->prefix); data->nreplicas_per_host = 1; // @TODO to be sure: // pcmk__clear_rsc_flags(rsc, pcmk_rsc_unique); } return TRUE; } return FALSE; } static int create_ip_resource(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { if(data->ip_range_start) { char *id = NULL; xmlNode *xml_ip = NULL; xmlNode *xml_obj = NULL; id = crm_strdup_printf("%s-ip-%s", data->prefix, replica->ipaddr); crm_xml_sanitize_id(id); xml_ip = create_resource(id, "heartbeat", "IPaddr2"); free(id); xml_obj = create_xml_node(xml_ip, PCMK_XE_INSTANCE_ATTRIBUTES); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "ip", replica->ipaddr); if(data->host_network) { crm_create_nvpair_xml(xml_obj, NULL, "nic", data->host_network); } if(data->host_netmask) { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", data->host_netmask); } else { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", "32"); } xml_obj = create_xml_node(xml_ip, PCMK_XE_OPERATIONS); crm_create_op_xml(xml_obj, pcmk__xe_id(xml_ip), PCMK_ACTION_MONITOR, "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (pe__unpack_resource(xml_ip, &replica->ip, parent, parent->cluster) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } parent->children = g_list_append(parent->children, replica->ip); } return pcmk_rc_ok; } static const char* container_agent_str(enum pe__container_agent t) { switch (t) { case PE__CONTAINER_AGENT_DOCKER: return PE__CONTAINER_AGENT_DOCKER_S; case PE__CONTAINER_AGENT_RKT: return PE__CONTAINER_AGENT_RKT_S; case PE__CONTAINER_AGENT_PODMAN: return PE__CONTAINER_AGENT_PODMAN_S; default: // PE__CONTAINER_AGENT_UNKNOWN break; } return PE__CONTAINER_AGENT_UNKNOWN_S; } static int create_container_resource(pcmk_resource_t *parent, const pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { char *id = NULL; xmlNode *xml_container = NULL; xmlNode *xml_obj = NULL; // Agent-specific const char *hostname_opt = NULL; const char *env_opt = NULL; const char *agent_str = NULL; int volid = 0; // rkt-only GString *buffer = NULL; GString *dbuffer = NULL; // Where syntax differences are drop-in replacements, set them now switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: hostname_opt = "-h "; env_opt = "-e "; break; case PE__CONTAINER_AGENT_RKT: hostname_opt = "--hostname="; env_opt = "--environment="; break; default: // PE__CONTAINER_AGENT_UNKNOWN return pcmk_rc_unpack_error; } agent_str = container_agent_str(data->agent_type); buffer = g_string_sized_new(4096); id = crm_strdup_printf("%s-%s-%d", data->prefix, agent_str, replica->offset); crm_xml_sanitize_id(id); xml_container = create_resource(id, "heartbeat", agent_str); free(id); xml_obj = create_xml_node(xml_container, PCMK_XE_INSTANCE_ATTRIBUTES); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "image", data->image); crm_create_nvpair_xml(xml_obj, NULL, "allow_pull", PCMK_VALUE_TRUE); crm_create_nvpair_xml(xml_obj, NULL, "force_kill", PCMK_VALUE_FALSE); crm_create_nvpair_xml(xml_obj, NULL, "reuse", PCMK_VALUE_FALSE); if (data->agent_type == PE__CONTAINER_AGENT_DOCKER) { g_string_append(buffer, " --restart=no"); } /* Set a container hostname only if we have an IP to map it to. The user can * set -h or --uts=host themselves if they want a nicer name for logs, but * this makes applications happy who need their hostname to match the IP * they bind to. */ if (data->ip_range_start != NULL) { g_string_append_printf(buffer, " %s%s-%d", hostname_opt, data->prefix, replica->offset); } pcmk__g_strcat(buffer, " ", env_opt, "PCMK_stderr=1", NULL); if (data->container_network != NULL) { pcmk__g_strcat(buffer, " --net=", data->container_network, NULL); } if (data->control_port != NULL) { pcmk__g_strcat(buffer, " ", env_opt, "PCMK_" PCMK__ENV_REMOTE_PORT "=", data->control_port, NULL); } else { g_string_append_printf(buffer, " %sPCMK_" PCMK__ENV_REMOTE_PORT "=%d", env_opt, DEFAULT_REMOTE_PORT); } for (GList *iter = data->mounts; iter != NULL; iter = iter->next) { pe__bundle_mount_t *mount = (pe__bundle_mount_t *) iter->data; char *source = NULL; if (pcmk_is_set(mount->flags, pe__bundle_mount_subdir)) { source = crm_strdup_printf("%s/%s-%d", mount->source, data->prefix, replica->offset); pcmk__add_separated_word(&dbuffer, 1024, source, ","); } switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: pcmk__g_strcat(buffer, " -v ", pcmk__s(source, mount->source), ":", mount->target, NULL); if (mount->options != NULL) { pcmk__g_strcat(buffer, ":", mount->options, NULL); } break; case PE__CONTAINER_AGENT_RKT: g_string_append_printf(buffer, " --volume vol%d,kind=host," "source=%s%s%s " "--mount volume=vol%d,target=%s", volid, pcmk__s(source, mount->source), (mount->options != NULL)? "," : "", pcmk__s(mount->options, ""), volid, mount->target); volid++; break; default: break; } free(source); } for (GList *iter = data->ports; iter != NULL; iter = iter->next) { pe__bundle_port_t *port = (pe__bundle_port_t *) iter->data; switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: if (replica->ipaddr != NULL) { pcmk__g_strcat(buffer, " -p ", replica->ipaddr, ":", port->source, ":", port->target, NULL); } else if (!pcmk__str_eq(data->container_network, PCMK_VALUE_HOST, pcmk__str_none)) { // No need to do port mapping if net == host pcmk__g_strcat(buffer, " -p ", port->source, ":", port->target, NULL); } break; case PE__CONTAINER_AGENT_RKT: if (replica->ipaddr != NULL) { pcmk__g_strcat(buffer, " --port=", port->target, ":", replica->ipaddr, ":", port->source, NULL); } else { pcmk__g_strcat(buffer, " --port=", port->target, ":", port->source, NULL); } break; default: break; } } /* @COMPAT: We should use pcmk__add_word() here, but we can't yet, because * it would cause restarts during rolling upgrades. * * In a previous version of the container resource creation logic, if * data->launcher_options is not NULL, we append * (" %s", data->launcher_options) even if data->launcher_options is an * empty string. Likewise for data->container_host_options. Using * * pcmk__add_word(buffer, 0, data->launcher_options) * * removes that extra trailing space, causing a resource definition change. */ if (data->launcher_options != NULL) { pcmk__g_strcat(buffer, " ", data->launcher_options, NULL); } if (data->container_host_options != NULL) { pcmk__g_strcat(buffer, " ", data->container_host_options, NULL); } crm_create_nvpair_xml(xml_obj, NULL, "run_opts", (const char *) buffer->str); g_string_free(buffer, TRUE); crm_create_nvpair_xml(xml_obj, NULL, "mount_points", (dbuffer != NULL)? (const char *) dbuffer->str : ""); if (dbuffer != NULL) { g_string_free(dbuffer, TRUE); } if (replica->child != NULL) { if (data->container_command != NULL) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } else { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", SBIN_DIR "/pacemaker-remoted"); } /* TODO: Allow users to specify their own? * * We just want to know if the container is alive; we'll monitor the * child independently. */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); #if 0 /* @TODO Consider supporting the use case where we can start and stop * resources, but not proxy local commands (such as setting node * attributes), by running the local executor in stand-alone mode. * However, this would probably be better done via ACLs as with other * Pacemaker Remote nodes. */ } else if ((child != NULL) && data->untrusted) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", CRM_DAEMON_DIR "/pacemaker-execd"); crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", CRM_DAEMON_DIR "/pacemaker/cts-exec-helper -c poke"); #endif } else { if (data->container_command != NULL) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } /* TODO: Allow users to specify their own? * * We don't know what's in the container, so we just want to know if it * is alive. */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); } xml_obj = create_xml_node(xml_container, PCMK_XE_OPERATIONS); crm_create_op_xml(xml_obj, pcmk__xe_id(xml_container), PCMK_ACTION_MONITOR, "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (pe__unpack_resource(xml_container, &replica->container, parent, parent->cluster) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } pcmk__set_rsc_flags(replica->container, pcmk_rsc_replica_container); parent->children = g_list_append(parent->children, replica->container); return pcmk_rc_ok; } /*! * \brief Ban a node from a resource's (and its children's) allowed nodes list * * \param[in,out] rsc Resource to modify * \param[in] uname Name of node to ban */ static void disallow_node(pcmk_resource_t *rsc, const char *uname) { gpointer match = g_hash_table_lookup(rsc->allowed_nodes, uname); if (match) { - ((pcmk_node_t *) match)->weight = -INFINITY; + ((pcmk_node_t *) match)->weight = -PCMK_SCORE_INFINITY; ((pcmk_node_t *) match)->rsc_discover_mode = pcmk_probe_never; } if (rsc->children) { g_list_foreach(rsc->children, (GFunc) disallow_node, (gpointer) uname); } } static int create_remote_resource(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { if (replica->child && valid_network(data)) { GHashTableIter gIter; pcmk_node_t *node = NULL; xmlNode *xml_remote = NULL; char *id = crm_strdup_printf("%s-%d", data->prefix, replica->offset); char *port_s = NULL; const char *uname = NULL; const char *connect_name = NULL; if (pe_find_resource(parent->cluster->resources, id) != NULL) { free(id); // The biggest hammer we have id = crm_strdup_printf("pcmk-internal-%s-remote-%d", replica->child->id, replica->offset); //@TODO return error instead of asserting? CRM_ASSERT(pe_find_resource(parent->cluster->resources, id) == NULL); } /* REMOTE_CONTAINER_HACK: Using "#uname" as the server name when the * connection does not have its own IP is a magic string that we use to * support nested remotes (i.e. a bundle running on a remote node). */ connect_name = (replica->ipaddr? replica->ipaddr : "#uname"); if (data->control_port == NULL) { port_s = pcmk__itoa(DEFAULT_REMOTE_PORT); } /* This sets replica->container as replica->remote's container, which is * similar to what happens with guest nodes. This is how the scheduler * knows that the bundle node is fenced by recovering the container, and * that remote should be ordered relative to the container. */ xml_remote = pe_create_remote_xml(NULL, id, replica->container->id, NULL, NULL, NULL, connect_name, (data->control_port? data->control_port : port_s)); free(port_s); /* Abandon our created ID, and pull the copy from the XML, because we * need something that will get freed during scheduler data cleanup to * use as the node ID and uname. */ free(id); id = NULL; uname = pcmk__xe_id(xml_remote); /* Ensure a node has been created for the guest (it may have already * been, if it has a permanent node attribute), and ensure its weight is * -INFINITY so no other resources can run on it. */ node = pe_find_node(parent->cluster->nodes, uname); if (node == NULL) { node = pe_create_node(uname, uname, PCMK_VALUE_REMOTE, PCMK_VALUE_MINUS_INFINITY, parent->cluster); } else { - node->weight = -INFINITY; + node->weight = -PCMK_SCORE_INFINITY; } node->rsc_discover_mode = pcmk_probe_never; /* unpack_remote_nodes() ensures that each remote node and guest node * has a pcmk_node_t entry. Ideally, it would do the same for bundle * nodes. Unfortunately, a bundle has to be mostly unpacked before it's * obvious what nodes will be needed, so we do it just above. * * Worse, that means that the node may have been utilized while * unpacking other resources, without our weight correction. The most * likely place for this to happen is when pe__unpack_resource() calls * resource_location() to set a default score in symmetric clusters. * This adds a node *copy* to each resource's allowed nodes, and these * copies will have the wrong weight. * * As a hacky workaround, fix those copies here. * * @TODO Possible alternative: ensure bundles are unpacked before other * resources, so the weight is correct before any copies are made. */ g_list_foreach(parent->cluster->resources, (GFunc) disallow_node, (gpointer) uname); replica->node = pe__copy_node(node); replica->node->weight = 500; replica->node->rsc_discover_mode = pcmk_probe_exclusive; /* Ensure the node shows up as allowed and with the correct discovery set */ if (replica->child->allowed_nodes != NULL) { g_hash_table_destroy(replica->child->allowed_nodes); } replica->child->allowed_nodes = pcmk__strkey_table(NULL, free); g_hash_table_insert(replica->child->allowed_nodes, (gpointer) replica->node->details->id, pe__copy_node(replica->node)); { pcmk_node_t *copy = pe__copy_node(replica->node); - copy->weight = -INFINITY; + copy->weight = -PCMK_SCORE_INFINITY; g_hash_table_insert(replica->child->parent->allowed_nodes, (gpointer) replica->node->details->id, copy); } if (pe__unpack_resource(xml_remote, &replica->remote, parent, parent->cluster) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } g_hash_table_iter_init(&gIter, replica->remote->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&node)) { if (pcmk__is_pacemaker_remote_node(node)) { /* Remote resources can only run on 'normal' cluster node */ - node->weight = -INFINITY; + node->weight = -PCMK_SCORE_INFINITY; } } replica->node->details->remote_rsc = replica->remote; // Ensure pcmk__is_guest_or_bundle_node() functions correctly replica->remote->container = replica->container; /* A bundle's #kind is closer to "container" (guest node) than the * "remote" set by pe_create_node(). */ pcmk__insert_dup(replica->node->details->attrs, CRM_ATTR_KIND, "container"); /* One effect of this is that setup_container() will add * replica->remote to replica->container's fillers, which will make * pe__resource_contains_guest_node() true for replica->container. * * replica->child does NOT get added to replica->container's fillers. * The only noticeable effect if it did would be for its fail count to * be taken into account when checking replica->container's migration * threshold. */ parent->children = g_list_append(parent->children, replica->remote); } return pcmk_rc_ok; } static int create_replica_resources(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { int rc = pcmk_rc_ok; rc = create_container_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } rc = create_ip_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } rc = create_remote_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } if ((replica->child != NULL) && (replica->ipaddr != NULL)) { pcmk__insert_meta(replica->child, "external-ip", replica->ipaddr); } if (replica->remote != NULL) { /* * Allow the remote connection resource to be allocated to a * different node than the one on which the container is active. * * This makes it possible to have Pacemaker Remote nodes running * containers with pacemaker-remoted inside in order to start * services inside those containers. */ pcmk__set_rsc_flags(replica->remote, pcmk_rsc_remote_nesting_allowed); } return rc; } static void mount_add(pe__bundle_variant_data_t *bundle_data, const char *source, const char *target, const char *options, uint32_t flags) { pe__bundle_mount_t *mount = calloc(1, sizeof(pe__bundle_mount_t)); CRM_ASSERT(mount != NULL); mount->source = strdup(source); mount->target = strdup(target); pcmk__str_update(&mount->options, options); mount->flags = flags; bundle_data->mounts = g_list_append(bundle_data->mounts, mount); } static void mount_free(pe__bundle_mount_t *mount) { free(mount->source); free(mount->target); free(mount->options); free(mount); } static void port_free(pe__bundle_port_t *port) { free(port->source); free(port->target); free(port); } static pcmk__bundle_replica_t * replica_for_remote(pcmk_resource_t *remote) { pcmk_resource_t *top = remote; pe__bundle_variant_data_t *bundle_data = NULL; if (top == NULL) { return NULL; } while (top->parent != NULL) { top = top->parent; } get_bundle_variant_data(bundle_data, top); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; if (replica->remote == remote) { return replica; } } CRM_LOG_ASSERT(FALSE); return NULL; } bool pe__bundle_needs_remote_name(pcmk_resource_t *rsc) { const char *value; GHashTable *params = NULL; if (rsc == NULL) { return false; } // Use NULL node since pcmk__bundle_expand() uses that to set value params = pe_rsc_params(rsc, NULL, rsc->cluster); value = g_hash_table_lookup(params, PCMK_REMOTE_RA_ADDR); return pcmk__str_eq(value, "#uname", pcmk__str_casei) && xml_contains_remote_node(rsc->xml); } const char * pe__add_bundle_remote_name(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler, xmlNode *xml, const char *field) { // REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside pcmk_node_t *node = NULL; pcmk__bundle_replica_t *replica = NULL; if (!pe__bundle_needs_remote_name(rsc)) { return NULL; } replica = replica_for_remote(rsc); if (replica == NULL) { return NULL; } node = replica->container->allocated_to; if (node == NULL) { /* If it won't be running anywhere after the * transition, go with where it's running now. */ node = pcmk__current_node(replica->container); } if(node == NULL) { crm_trace("Cannot determine address for bundle connection %s", rsc->id); return NULL; } crm_trace("Setting address for bundle connection %s to bundle host %s", rsc->id, pcmk__node_name(node)); if(xml != NULL && field != NULL) { crm_xml_add(xml, field, node->details->uname); } return node->details->uname; } #define pe__set_bundle_mount_flags(mount_xml, flags, flags_to_set) do { \ flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Bundle mount", pcmk__xe_id(mount_xml), \ flags, (flags_to_set), #flags_to_set); \ } while (0) gboolean pe__unpack_bundle(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { const char *value = NULL; xmlNode *xml_obj = NULL; const xmlNode *xml_child = NULL; xmlNode *xml_resource = NULL; pe__bundle_variant_data_t *bundle_data = NULL; bool need_log_mount = TRUE; CRM_ASSERT(rsc != NULL); pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id); bundle_data = calloc(1, sizeof(pe__bundle_variant_data_t)); rsc->variant_opaque = bundle_data; bundle_data->prefix = strdup(rsc->id); xml_obj = first_named_child(rsc->xml, PCMK_XE_DOCKER); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_DOCKER; } else { xml_obj = first_named_child(rsc->xml, PCMK__XE_RKT); if (xml_obj != NULL) { pcmk__warn_once(pcmk__wo_rkt, "Support for " PCMK__XE_RKT " in bundles " "(such as %s) is deprecated and will be " "removed in a future release", rsc->id); bundle_data->agent_type = PE__CONTAINER_AGENT_RKT; } else { xml_obj = first_named_child(rsc->xml, PCMK_XE_PODMAN); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_PODMAN; } else { return FALSE; } } } // Use 0 for default, minimum, and invalid PCMK_XA_PROMOTED_MAX value = crm_element_value(xml_obj, PCMK_XA_PROMOTED_MAX); if (value == NULL) { // @COMPAT deprecated since 2.0.0 value = crm_element_value(xml_obj, PCMK__XA_PROMOTED_MAX_LEGACY); } pcmk__scan_min_int(value, &bundle_data->promoted_max, 0); /* Default replicas to PCMK_XA_PROMOTED_MAX if it was specified and 1 * otherwise */ value = crm_element_value(xml_obj, PCMK_XA_REPLICAS); if ((value == NULL) && (bundle_data->promoted_max > 0)) { bundle_data->nreplicas = bundle_data->promoted_max; } else { pcmk__scan_min_int(value, &bundle_data->nreplicas, 1); } /* * Communication between containers on the same host via the * floating IPs only works if the container is started with: * --userland-proxy=false --ip-masq=false */ value = crm_element_value(xml_obj, PCMK_XA_REPLICAS_PER_HOST); pcmk__scan_min_int(value, &bundle_data->nreplicas_per_host, 1); if (bundle_data->nreplicas_per_host == 1) { pcmk__clear_rsc_flags(rsc, pcmk_rsc_unique); } bundle_data->container_command = crm_element_value_copy(xml_obj, PCMK_XA_RUN_COMMAND); bundle_data->launcher_options = crm_element_value_copy(xml_obj, PCMK_XA_OPTIONS); bundle_data->image = crm_element_value_copy(xml_obj, PCMK_XA_IMAGE); bundle_data->container_network = crm_element_value_copy(xml_obj, PCMK_XA_NETWORK); xml_obj = first_named_child(rsc->xml, PCMK_XE_NETWORK); if(xml_obj) { bundle_data->ip_range_start = crm_element_value_copy(xml_obj, PCMK_XA_IP_RANGE_START); bundle_data->host_netmask = crm_element_value_copy(xml_obj, PCMK_XA_HOST_NETMASK); bundle_data->host_network = crm_element_value_copy(xml_obj, PCMK_XA_HOST_INTERFACE); bundle_data->control_port = crm_element_value_copy(xml_obj, PCMK_XA_CONTROL_PORT); value = crm_element_value(xml_obj, PCMK_XA_ADD_HOST); if (crm_str_to_boolean(value, &bundle_data->add_host) != 1) { bundle_data->add_host = TRUE; } for (xml_child = first_named_child(xml_obj, PCMK_XE_PORT_MAPPING); xml_child != NULL; xml_child = crm_next_same_xml(xml_child)) { pe__bundle_port_t *port = calloc(1, sizeof(pe__bundle_port_t)); port->source = crm_element_value_copy(xml_child, PCMK_XA_PORT); if(port->source == NULL) { port->source = crm_element_value_copy(xml_child, PCMK_XA_RANGE); } else { port->target = crm_element_value_copy(xml_child, PCMK_XA_INTERNAL_PORT); } if(port->source != NULL && strlen(port->source) > 0) { if(port->target == NULL) { port->target = strdup(port->source); } bundle_data->ports = g_list_append(bundle_data->ports, port); } else { pcmk__config_err("Invalid " PCMK_XA_PORT " directive %s", pcmk__xe_id(xml_child)); port_free(port); } } } xml_obj = first_named_child(rsc->xml, PCMK_XE_STORAGE); for (xml_child = first_named_child(xml_obj, PCMK_XE_STORAGE_MAPPING); xml_child != NULL; xml_child = crm_next_same_xml(xml_child)) { const char *source = crm_element_value(xml_child, PCMK_XA_SOURCE_DIR); const char *target = crm_element_value(xml_child, PCMK_XA_TARGET_DIR); const char *options = crm_element_value(xml_child, PCMK_XA_OPTIONS); int flags = pe__bundle_mount_none; if (source == NULL) { source = crm_element_value(xml_child, PCMK_XA_SOURCE_DIR_ROOT); pe__set_bundle_mount_flags(xml_child, flags, pe__bundle_mount_subdir); } if (source && target) { mount_add(bundle_data, source, target, options, flags); if (strcmp(target, "/var/log") == 0) { need_log_mount = FALSE; } } else { pcmk__config_err("Invalid mount directive %s", pcmk__xe_id(xml_child)); } } xml_obj = first_named_child(rsc->xml, PCMK_XE_PRIMITIVE); if (xml_obj && valid_network(bundle_data)) { char *value = NULL; xmlNode *xml_set = NULL; xml_resource = create_xml_node(NULL, PCMK_XE_CLONE); /* @COMPAT We no longer use the tag, but we need to keep it as * part of the resource name, so that bundles don't restart in a rolling * upgrade. (It also avoids needing to change regression tests.) */ crm_xml_set_id(xml_resource, "%s-%s", bundle_data->prefix, (bundle_data->promoted_max? "master" : (const char *)xml_resource->name)); xml_set = create_xml_node(xml_resource, PCMK_XE_META_ATTRIBUTES); crm_xml_set_id(xml_set, "%s-%s-meta", bundle_data->prefix, xml_resource->name); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_ORDERED, PCMK_VALUE_TRUE); value = pcmk__itoa(bundle_data->nreplicas); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_CLONE_MAX, value); free(value); value = pcmk__itoa(bundle_data->nreplicas_per_host); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_CLONE_NODE_MAX, value); free(value); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_GLOBALLY_UNIQUE, pcmk__btoa(bundle_data->nreplicas_per_host > 1)); if (bundle_data->promoted_max) { crm_create_nvpair_xml(xml_set, NULL, PCMK_META_PROMOTABLE, PCMK_VALUE_TRUE); value = pcmk__itoa(bundle_data->promoted_max); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_PROMOTED_MAX, value); free(value); } //crm_xml_add(xml_obj, PCMK_XA_ID, bundle_data->prefix); add_node_copy(xml_resource, xml_obj); } else if(xml_obj) { pcmk__config_err("Cannot control %s inside %s without either " PCMK_XA_IP_RANGE_START " or " PCMK_XA_CONTROL_PORT, rsc->id, pcmk__xe_id(xml_obj)); return FALSE; } if(xml_resource) { int lpc = 0; GList *childIter = NULL; pe__bundle_port_t *port = NULL; GString *buffer = NULL; if (pe__unpack_resource(xml_resource, &(bundle_data->child), rsc, scheduler) != pcmk_rc_ok) { return FALSE; } /* Currently, we always map the default authentication key location * into the same location inside the container. * * Ideally, we would respect the host's PCMK_authkey_location, but: * - it may be different on different nodes; * - the actual connection will do extra checking to make sure the key * file exists and is readable, that we can't do here on the DC * - tools such as crm_resource and crm_simulate may not have the same * environment variables as the cluster, causing operation digests to * differ * * Always using the default location inside the container is fine, * because we control the pacemaker_remote environment, and it avoids * having to pass another environment variable to the container. * * @TODO A better solution may be to have only pacemaker_remote use the * environment variable, and have the cluster nodes use a new * cluster option for key location. This would introduce the limitation * of the location being the same on all cluster nodes, but that's * reasonable. */ mount_add(bundle_data, DEFAULT_REMOTE_KEY_LOCATION, DEFAULT_REMOTE_KEY_LOCATION, NULL, pe__bundle_mount_none); if (need_log_mount) { mount_add(bundle_data, CRM_BUNDLE_DIR, "/var/log", NULL, pe__bundle_mount_subdir); } port = calloc(1, sizeof(pe__bundle_port_t)); if(bundle_data->control_port) { port->source = strdup(bundle_data->control_port); } else { /* If we wanted to respect PCMK_remote_port, we could use * crm_default_remote_port() here and elsewhere in this file instead * of DEFAULT_REMOTE_PORT. * * However, it gains nothing, since we control both the container * environment and the connection resource parameters, and the user * can use a different port if desired by setting * PCMK_XA_CONTROL_PORT. */ port->source = pcmk__itoa(DEFAULT_REMOTE_PORT); } port->target = strdup(port->source); bundle_data->ports = g_list_append(bundle_data->ports, port); buffer = g_string_sized_new(1024); for (childIter = bundle_data->child->children; childIter != NULL; childIter = childIter->next) { pcmk__bundle_replica_t *replica = NULL; replica = calloc(1, sizeof(pcmk__bundle_replica_t)); replica->child = childIter->data; replica->child->exclusive_discover = TRUE; replica->offset = lpc++; // Ensure the child's notify gets set based on the underlying primitive's value if (pcmk_is_set(replica->child->flags, pcmk_rsc_notify)) { pcmk__set_rsc_flags(bundle_data->child, pcmk_rsc_notify); } allocate_ip(bundle_data, replica, buffer); bundle_data->replicas = g_list_append(bundle_data->replicas, replica); bundle_data->attribute_target = g_hash_table_lookup(replica->child->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); } bundle_data->container_host_options = g_string_free(buffer, FALSE); if (bundle_data->attribute_target) { pcmk__insert_dup(rsc->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET, bundle_data->attribute_target); pcmk__insert_dup(bundle_data->child->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET, bundle_data->attribute_target); } } else { // Just a naked container, no pacemaker-remote GString *buffer = g_string_sized_new(1024); for (int lpc = 0; lpc < bundle_data->nreplicas; lpc++) { pcmk__bundle_replica_t *replica = NULL; replica = calloc(1, sizeof(pcmk__bundle_replica_t)); replica->offset = lpc; allocate_ip(bundle_data, replica, buffer); bundle_data->replicas = g_list_append(bundle_data->replicas, replica); } bundle_data->container_host_options = g_string_free(buffer, FALSE); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; if (create_replica_resources(rsc, bundle_data, replica) != pcmk_rc_ok) { pcmk__config_err("Failed unpacking resource %s", rsc->id); rsc->fns->free(rsc); return FALSE; } /* Utilization needs special handling for bundles. It makes no sense for * the inner primitive to have utilization, because it is tied * one-to-one to the guest node created by the container resource -- and * there's no way to set capacities for that guest node anyway. * * What the user really wants is to configure utilization for the * container. However, the schema only allows utilization for * primitives, and the container resource is implicit anyway, so the * user can *only* configure utilization for the inner primitive. If * they do, move the primitive's utilization values to the container. * * @TODO This means that bundles without an inner primitive can't have * utilization. An alternative might be to allow utilization values in * the top-level bundle XML in the schema, and copy those to each * container. */ if (replica->child != NULL) { GHashTable *empty = replica->container->utilization; replica->container->utilization = replica->child->utilization; replica->child->utilization = empty; } } if (bundle_data->child) { rsc->children = g_list_append(rsc->children, bundle_data->child); } return TRUE; } static int replica_resource_active(pcmk_resource_t *rsc, gboolean all) { if (rsc) { gboolean child_active = rsc->fns->active(rsc, all); if (child_active && !all) { return TRUE; } else if (!child_active && all) { return FALSE; } } return -1; } gboolean pe__bundle_active(pcmk_resource_t *rsc, gboolean all) { pe__bundle_variant_data_t *bundle_data = NULL; GList *iter = NULL; get_bundle_variant_data(bundle_data, rsc); for (iter = bundle_data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; int rsc_active; rsc_active = replica_resource_active(replica->ip, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->child, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->container, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->remote, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } } /* If "all" is TRUE, we've already checked that no resources were inactive, * so return TRUE; if "all" is FALSE, we didn't find any active resources, * so return FALSE. */ return all; } /*! * \internal * \brief Find the bundle replica corresponding to a given node * * \param[in] bundle Top-level bundle resource * \param[in] node Node to search for * * \return Bundle replica if found, NULL otherwise */ pcmk_resource_t * pe__find_bundle_replica(const pcmk_resource_t *bundle, const pcmk_node_t *node) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_ASSERT(bundle && node); get_bundle_variant_data(bundle_data, bundle); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica && replica->node); if (pcmk__same_node(replica->node, node)) { return replica->child; } } return NULL; } /*! * \internal * \deprecated This function will be removed in a future release */ static void print_rsc_in_list(pcmk_resource_t *rsc, const char *pre_text, long options, void *print_data) { if (rsc != NULL) { if (options & pe_print_html) { status_print("
  • "); } rsc->fns->print(rsc, pre_text, options, print_data); if (options & pe_print_html) { status_print("
    • \n"); } } } /*! * \internal * \deprecated This function will be removed in a future release */ static void bundle_print_xml(pcmk_resource_t *rsc, const char *pre_text, long options, void *print_data) { pe__bundle_variant_data_t *bundle_data = NULL; char *child_text = NULL; CRM_CHECK(rsc != NULL, return); if (pre_text == NULL) { pre_text = ""; } child_text = crm_strdup_printf("%s ", pre_text); get_bundle_variant_data(bundle_data, rsc); status_print("%sid); status_print("type=\"%s\" ", container_agent_str(bundle_data->agent_type)); status_print("image=\"%s\" ", bundle_data->image); status_print("unique=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_unique)); status_print("managed=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_managed)); status_print("failed=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_failed)); status_print(">\n"); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); status_print("%s \n", pre_text, replica->offset); print_rsc_in_list(replica->ip, child_text, options, print_data); print_rsc_in_list(replica->child, child_text, options, print_data); print_rsc_in_list(replica->container, child_text, options, print_data); print_rsc_in_list(replica->remote, child_text, options, print_data); status_print("%s \n", pre_text); } status_print("%s\n", pre_text); free(child_text); } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_xml(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean printed_header = FALSE; gboolean print_everything = TRUE; const char *desc = NULL; CRM_ASSERT(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; char *id = NULL; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, print_everything); if (!print_everything && !print_ip && !print_child && !print_ctnr && !print_remote) { continue; } if (!printed_header) { const char *type = container_agent_str(bundle_data->agent_type); const char *unique = pcmk__flag_text(rsc->flags, pcmk_rsc_unique); const char *maintenance = pcmk__flag_text(rsc->flags, pcmk_rsc_maintenance); const char *managed = pcmk__flag_text(rsc->flags, pcmk_rsc_managed); const char *failed = pcmk__flag_text(rsc->flags, pcmk_rsc_failed); printed_header = TRUE; desc = pe__resource_description(rsc, show_opts); rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_BUNDLE, 8, PCMK_XA_ID, rsc->id, PCMK_XA_TYPE, type, PCMK_XA_IMAGE, bundle_data->image, PCMK_XA_UNIQUE, unique, PCMK_XA_MAINTENANCE, maintenance, PCMK_XA_MANAGED, managed, PCMK_XA_FAILED, failed, PCMK_XA_DESCRIPTION, desc); CRM_ASSERT(rc == pcmk_rc_ok); } id = pcmk__itoa(replica->offset); rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_REPLICA, 1, PCMK_XA_ID, id); free(id); CRM_ASSERT(rc == pcmk_rc_ok); if (print_ip) { out->message(out, crm_map_element_name(replica->ip->xml), show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), show_opts, replica->remote, only_node, only_rsc); } pcmk__output_xml_pop_parent(out); // replica } if (printed_header) { pcmk__output_xml_pop_parent(out); // bundle } return rc; } static void pe__bundle_replica_output_html(pcmk__output_t *out, pcmk__bundle_replica_t *replica, pcmk_node_t *node, uint32_t show_opts) { pcmk_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } pe__common_output_html(out, rsc, buffer, node, show_opts); } /*! * \internal * \brief Get a string describing a resource's unmanaged state or lack thereof * * \param[in] rsc Resource to describe * * \return A string indicating that a resource is in maintenance mode or * otherwise unmanaged, or an empty string otherwise */ static const char * get_unmanaged_str(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk_rsc_maintenance)) { return " (maintenance)"; } if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { return " (unmanaged)"; } return ""; } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_html(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const char *desc = NULL; pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; CRM_ASSERT(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); desc = pe__resource_description(rsc, show_opts); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, print_everything); if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) || (print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) { /* The text output messages used below require pe_print_implicit to * be set to do anything. */ uint32_t new_show_opts = show_opts | pcmk_show_implicit_rscs; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); if (pcmk__list_of_multiple(bundle_data->replicas)) { out->begin_list(out, NULL, NULL, "Replica[%d]", replica->offset); } if (print_ip) { out->message(out, crm_map_element_name(replica->ip->xml), new_show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), new_show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), new_show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), new_show_opts, replica->remote, only_node, only_rsc); } if (pcmk__list_of_multiple(bundle_data->replicas)) { out->end_list(out); } } else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) { continue; } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); pe__bundle_replica_output_html(out, replica, pcmk__current_node(replica->container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } static void pe__bundle_replica_output_text(pcmk__output_t *out, pcmk__bundle_replica_t *replica, pcmk_node_t *node, uint32_t show_opts) { const pcmk_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } pe__common_output_text(out, rsc, buffer, node, show_opts); } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_text(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const char *desc = NULL; pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; desc = pe__resource_description(rsc, show_opts); get_bundle_variant_data(bundle_data, rsc); CRM_ASSERT(rsc != NULL); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, print_everything); if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) || (print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) { /* The text output messages used below require pe_print_implicit to * be set to do anything. */ uint32_t new_show_opts = show_opts | pcmk_show_implicit_rscs; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); if (pcmk__list_of_multiple(bundle_data->replicas)) { out->list_item(out, NULL, "Replica[%d]", replica->offset); } out->begin_list(out, NULL, NULL, NULL); if (print_ip) { out->message(out, crm_map_element_name(replica->ip->xml), new_show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), new_show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), new_show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), new_show_opts, replica->remote, only_node, only_rsc); } out->end_list(out); } else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) { continue; } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); pe__bundle_replica_output_text(out, replica, pcmk__current_node(replica->container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } /*! * \internal * \deprecated This function will be removed in a future release */ static void print_bundle_replica(pcmk__bundle_replica_t *replica, const char *pre_text, long options, void *print_data) { pcmk_node_t *node = NULL; pcmk_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } node = pcmk__current_node(replica->container); common_print(rsc, pre_text, buffer, node, options, print_data); } /*! * \internal * \deprecated This function will be removed in a future release */ void pe__print_bundle(pcmk_resource_t *rsc, const char *pre_text, long options, void *print_data) { pe__bundle_variant_data_t *bundle_data = NULL; char *child_text = NULL; CRM_CHECK(rsc != NULL, return); if (options & pe_print_xml) { bundle_print_xml(rsc, pre_text, options, print_data); return; } get_bundle_variant_data(bundle_data, rsc); if (pre_text == NULL) { pre_text = " "; } status_print("%sContainer bundle%s: %s [%s]%s%s\n", pre_text, ((bundle_data->nreplicas > 1)? " set" : ""), rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", pcmk_is_set(rsc->flags, pcmk_rsc_managed)? "" : " (unmanaged)"); if (options & pe_print_html) { status_print("
    \n
      \n"); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (options & pe_print_html) { status_print("
    • "); } if (pcmk_is_set(options, pe_print_implicit)) { child_text = crm_strdup_printf(" %s", pre_text); if (pcmk__list_of_multiple(bundle_data->replicas)) { status_print(" %sReplica[%d]\n", pre_text, replica->offset); } if (options & pe_print_html) { status_print("
      \n
        \n"); } print_rsc_in_list(replica->ip, child_text, options, print_data); print_rsc_in_list(replica->container, child_text, options, print_data); print_rsc_in_list(replica->remote, child_text, options, print_data); print_rsc_in_list(replica->child, child_text, options, print_data); if (options & pe_print_html) { status_print("
      \n"); } } else { child_text = crm_strdup_printf("%s ", pre_text); print_bundle_replica(replica, child_text, options, print_data); } free(child_text); if (options & pe_print_html) { status_print("
      • \n"); } } if (options & pe_print_html) { status_print("
    \n"); } } static void free_bundle_replica(pcmk__bundle_replica_t *replica) { if (replica == NULL) { return; } if (replica->node) { free(replica->node); replica->node = NULL; } if (replica->ip) { free_xml(replica->ip->xml); replica->ip->xml = NULL; replica->ip->fns->free(replica->ip); replica->ip = NULL; } if (replica->container) { free_xml(replica->container->xml); replica->container->xml = NULL; replica->container->fns->free(replica->container); replica->container = NULL; } if (replica->remote) { free_xml(replica->remote->xml); replica->remote->xml = NULL; replica->remote->fns->free(replica->remote); replica->remote = NULL; } free(replica->ipaddr); free(replica); } void pe__free_bundle(pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); pcmk__rsc_trace(rsc, "Freeing %s", rsc->id); free(bundle_data->prefix); free(bundle_data->image); free(bundle_data->control_port); free(bundle_data->host_network); free(bundle_data->host_netmask); free(bundle_data->ip_range_start); free(bundle_data->container_network); free(bundle_data->launcher_options); free(bundle_data->container_command); g_free(bundle_data->container_host_options); g_list_free_full(bundle_data->replicas, (GDestroyNotify) free_bundle_replica); g_list_free_full(bundle_data->mounts, (GDestroyNotify)mount_free); g_list_free_full(bundle_data->ports, (GDestroyNotify)port_free); g_list_free(rsc->children); if(bundle_data->child) { free_xml(bundle_data->child->xml); bundle_data->child->xml = NULL; bundle_data->child->fns->free(bundle_data->child); } common_free(rsc); } enum rsc_role_e pe__bundle_resource_state(const pcmk_resource_t *rsc, gboolean current) { enum rsc_role_e container_role = pcmk_role_unknown; return container_role; } /*! * \brief Get the number of configured replicas in a bundle * * \param[in] rsc Bundle resource * * \return Number of configured replicas, or 0 on error */ int pe_bundle_replicas(const pcmk_resource_t *rsc) { if ((rsc == NULL) || (rsc->variant != pcmk_rsc_variant_bundle)) { return 0; } else { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); return bundle_data->nreplicas; } } void pe__count_bundle(pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); for (GList *item = bundle_data->replicas; item != NULL; item = item->next) { pcmk__bundle_replica_t *replica = item->data; if (replica->ip) { replica->ip->fns->count(replica->ip); } if (replica->child) { replica->child->fns->count(replica->child); } if (replica->container) { replica->container->fns->count(replica->container); } if (replica->remote) { replica->remote->fns->count(replica->remote); } } } gboolean pe__bundle_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent) { gboolean passes = FALSE; pe__bundle_variant_data_t *bundle_data = NULL; if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches)) { passes = TRUE; } else { get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; if (replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, FALSE)) { passes = TRUE; break; } else if (replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, FALSE)) { passes = TRUE; break; } else if (!replica->container->fns->is_filtered(replica->container, only_rsc, FALSE)) { passes = TRUE; break; } else if (replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, FALSE)) { passes = TRUE; break; } } } return !passes; } /*! * \internal * \brief Get a list of a bundle's containers * * \param[in] bundle Bundle resource * * \return Newly created list of \p bundle's containers * \note It is the caller's responsibility to free the result with * g_list_free(). */ GList * pe__bundle_containers(const pcmk_resource_t *bundle) { GList *containers = NULL; const pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, bundle); for (GList *iter = data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; containers = g_list_append(containers, replica->container); } return containers; } // Bundle implementation of pcmk_rsc_methods_t:active_node() pcmk_node_t * pe__bundle_active_node(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean) { pcmk_node_t *active = NULL; pcmk_node_t *node = NULL; pcmk_resource_t *container = NULL; GList *containers = NULL; GList *iter = NULL; GHashTable *nodes = NULL; const pe__bundle_variant_data_t *data = NULL; if (count_all != NULL) { *count_all = 0; } if (count_clean != NULL) { *count_clean = 0; } if (rsc == NULL) { return NULL; } /* For the purposes of this method, we only care about where the bundle's * containers are active, so build a list of active containers. */ get_bundle_variant_data(data, rsc); for (iter = data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; if (replica->container->running_on != NULL) { containers = g_list_append(containers, replica->container); } } if (containers == NULL) { return NULL; } /* If the bundle has only a single active container, just use that * container's method. If live migration is ever supported for bundle * containers, this will allow us to prefer the migration source when there * is only one container and it is migrating. For now, this just lets us * avoid creating the nodes table. */ if (pcmk__list_of_1(containers)) { container = containers->data; node = container->fns->active_node(container, count_all, count_clean); g_list_free(containers); return node; } // Add all containers' active nodes to a hash table (for uniqueness) nodes = g_hash_table_new(NULL, NULL); for (iter = containers; iter != NULL; iter = iter->next) { container = iter->data; for (GList *node_iter = container->running_on; node_iter != NULL; node_iter = node_iter->next) { node = node_iter->data; // If insert returns true, we haven't counted this node yet if (g_hash_table_insert(nodes, (gpointer) node->details, (gpointer) node) && !pe__count_active_node(rsc, node, &active, count_all, count_clean)) { goto done; } } } done: g_list_free(containers); g_hash_table_destroy(nodes); return active; } /*! * \internal * \brief Get maximum bundle resource instances per node * * \param[in] rsc Bundle resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int pe__bundle_max_per_node(const pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); CRM_ASSERT(bundle_data->nreplicas_per_host >= 0); return (unsigned int) bundle_data->nreplicas_per_host; } diff --git a/lib/pengine/clone.c b/lib/pengine/clone.c index 54c3bb09c7..d4f8dfbcdf 100644 --- a/lib/pengine/clone.c +++ b/lib/pengine/clone.c @@ -1,1540 +1,1540 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #ifdef PCMK__COMPAT_2_0 #define PROMOTED_INSTANCES PCMK__ROLE_PROMOTED_LEGACY "s" #define UNPROMOTED_INSTANCES PCMK__ROLE_UNPROMOTED_LEGACY "s" #else #define PROMOTED_INSTANCES PCMK__ROLE_PROMOTED #define UNPROMOTED_INSTANCES PCMK__ROLE_UNPROMOTED #endif typedef struct clone_variant_data_s { int clone_max; int clone_node_max; int promoted_max; int promoted_node_max; int total_clones; uint32_t flags; // Group of enum pcmk__clone_flags notify_data_t *stop_notify; notify_data_t *start_notify; notify_data_t *demote_notify; notify_data_t *promote_notify; xmlNode *xml_obj_child; } clone_variant_data_t; #define get_clone_variant_data(data, rsc) \ CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_clone)); \ data = (clone_variant_data_t *) rsc->variant_opaque; /*! * \internal * \brief Return the maximum number of clone instances allowed to be run * * \param[in] clone Clone or clone instance to check * * \return Maximum instances for \p clone */ int pe__clone_max(const pcmk_resource_t *clone) { const clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, pe__const_top_resource(clone, false)); return clone_data->clone_max; } /*! * \internal * \brief Return the maximum number of clone instances allowed per node * * \param[in] clone Promotable clone or clone instance to check * * \return Maximum allowed instances per node for \p clone */ int pe__clone_node_max(const pcmk_resource_t *clone) { const clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, pe__const_top_resource(clone, false)); return clone_data->clone_node_max; } /*! * \internal * \brief Return the maximum number of clone instances allowed to be promoted * * \param[in] clone Promotable clone or clone instance to check * * \return Maximum promoted instances for \p clone */ int pe__clone_promoted_max(const pcmk_resource_t *clone) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, pe__const_top_resource(clone, false)); return clone_data->promoted_max; } /*! * \internal * \brief Return the maximum number of clone instances allowed to be promoted * * \param[in] clone Promotable clone or clone instance to check * * \return Maximum promoted instances for \p clone */ int pe__clone_promoted_node_max(const pcmk_resource_t *clone) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, pe__const_top_resource(clone, false)); return clone_data->promoted_node_max; } static GList * sorted_hash_table_values(GHashTable *table) { GList *retval = NULL; GHashTableIter iter; gpointer key, value; g_hash_table_iter_init(&iter, table); while (g_hash_table_iter_next(&iter, &key, &value)) { if (!g_list_find_custom(retval, value, (GCompareFunc) strcmp)) { retval = g_list_prepend(retval, (char *) value); } } retval = g_list_sort(retval, (GCompareFunc) strcmp); return retval; } static GList * nodes_with_status(GHashTable *table, const char *status) { GList *retval = NULL; GHashTableIter iter; gpointer key, value; g_hash_table_iter_init(&iter, table); while (g_hash_table_iter_next(&iter, &key, &value)) { if (!strcmp((char *) value, status)) { retval = g_list_prepend(retval, key); } } retval = g_list_sort(retval, (GCompareFunc) pcmk__numeric_strcasecmp); return retval; } static GString * node_list_to_str(const GList *list) { GString *retval = NULL; for (const GList *iter = list; iter != NULL; iter = iter->next) { pcmk__add_word(&retval, 1024, (const char *) iter->data); } return retval; } static void clone_header(pcmk__output_t *out, int *rc, const pcmk_resource_t *rsc, clone_variant_data_t *clone_data, const char *desc) { GString *attrs = NULL; if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { pcmk__add_separated_word(&attrs, 64, "promotable", ", "); } if (pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { pcmk__add_separated_word(&attrs, 64, "unique", ", "); } if (pe__resource_is_disabled(rsc)) { pcmk__add_separated_word(&attrs, 64, "disabled", ", "); } if (pcmk_is_set(rsc->flags, pcmk_rsc_maintenance)) { pcmk__add_separated_word(&attrs, 64, "maintenance", ", "); } else if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk__add_separated_word(&attrs, 64, "unmanaged", ", "); } if (attrs != NULL) { PCMK__OUTPUT_LIST_HEADER(out, FALSE, *rc, "Clone Set: %s [%s] (%s)%s%s%s", rsc->id, pcmk__xe_id(clone_data->xml_obj_child), (const char *) attrs->str, desc ? " (" : "", desc ? desc : "", desc ? ")" : ""); g_string_free(attrs, TRUE); } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, *rc, "Clone Set: %s [%s]%s%s%s", rsc->id, pcmk__xe_id(clone_data->xml_obj_child), desc ? " (" : "", desc ? desc : "", desc ? ")" : ""); } } void pe__force_anon(const char *standard, pcmk_resource_t *rsc, const char *rid, pcmk_scheduler_t *scheduler) { if (pcmk__is_clone(rsc)) { clone_variant_data_t *clone_data = rsc->variant_opaque; pcmk__config_warn("Ignoring " PCMK_META_GLOBALLY_UNIQUE " for %s " "because %s resources such as %s can be used only as " "anonymous clones", rsc->id, standard, rid); clone_data->clone_node_max = 1; clone_data->clone_max = QB_MIN(clone_data->clone_max, g_list_length(scheduler->nodes)); } } pcmk_resource_t * find_clone_instance(const pcmk_resource_t *rsc, const char *sub_id) { char *child_id = NULL; pcmk_resource_t *child = NULL; const char *child_base = NULL; clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); child_base = pcmk__xe_id(clone_data->xml_obj_child); child_id = crm_strdup_printf("%s:%s", child_base, sub_id); child = pe_find_resource(rsc->children, child_id); free(child_id); return child; } pcmk_resource_t * pe__create_clone_child(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { gboolean as_orphan = FALSE; char *inc_num = NULL; char *inc_max = NULL; pcmk_resource_t *child_rsc = NULL; xmlNode *child_copy = NULL; clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); CRM_CHECK(clone_data->xml_obj_child != NULL, return FALSE); if (clone_data->total_clones >= clone_data->clone_max) { // If we've already used all available instances, this is an orphan as_orphan = TRUE; } // Allocate instance numbers in numerical order (starting at 0) inc_num = pcmk__itoa(clone_data->total_clones); inc_max = pcmk__itoa(clone_data->clone_max); child_copy = copy_xml(clone_data->xml_obj_child); crm_xml_add(child_copy, PCMK__META_CLONE, inc_num); if (pe__unpack_resource(child_copy, &child_rsc, rsc, scheduler) != pcmk_rc_ok) { goto bail; } /* child_rsc->globally_unique = rsc->globally_unique; */ CRM_ASSERT(child_rsc); clone_data->total_clones += 1; pcmk__rsc_trace(child_rsc, "Setting clone attributes for: %s", child_rsc->id); rsc->children = g_list_append(rsc->children, child_rsc); if (as_orphan) { pe__set_resource_flags_recursive(child_rsc, pcmk_rsc_removed); } pcmk__insert_meta(child_rsc, PCMK_META_CLONE_MAX, inc_max); pcmk__rsc_trace(rsc, "Added %s instance %s", rsc->id, child_rsc->id); bail: free(inc_num); free(inc_max); return child_rsc; } /*! * \internal * \brief Unpack a nonnegative integer value from a resource meta-attribute * * \param[in] rsc Resource with meta-attribute * \param[in] meta_name Name of meta-attribute to unpack * \param[in] deprecated_name If not NULL, try unpacking this * if \p meta_name is unset * \param[in] default_value Value to use if unset * * \return Integer parsed from resource's specified meta-attribute if a valid * nonnegative integer, \p default_value if unset, or 0 if invalid */ static int unpack_meta_int(const pcmk_resource_t *rsc, const char *meta_name, const char *deprecated_name, int default_value) { int integer = default_value; const char *value = g_hash_table_lookup(rsc->meta, meta_name); if ((value == NULL) && (deprecated_name != NULL)) { value = g_hash_table_lookup(rsc->meta, deprecated_name); } if (value != NULL) { pcmk__scan_min_int(value, &integer, 0); } return integer; } gboolean clone_unpack(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { int lpc = 0; xmlNode *a_child = NULL; xmlNode *xml_obj = rsc->xml; clone_variant_data_t *clone_data = NULL; pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id); clone_data = calloc(1, sizeof(clone_variant_data_t)); rsc->variant_opaque = clone_data; if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { // Use 1 as default but 0 for minimum and invalid // @COMPAT PCMK__META_PROMOTED_MAX_LEGACY deprecated since 2.0.0 clone_data->promoted_max = unpack_meta_int(rsc, PCMK_META_PROMOTED_MAX, PCMK__META_PROMOTED_MAX_LEGACY, 1); // Use 1 as default but 0 for minimum and invalid // @COMPAT PCMK__META_PROMOTED_NODE_MAX_LEGACY deprecated since 2.0.0 clone_data->promoted_node_max = unpack_meta_int(rsc, PCMK_META_PROMOTED_NODE_MAX, PCMK__META_PROMOTED_NODE_MAX_LEGACY, 1); } // Implied by calloc() /* clone_data->xml_obj_child = NULL; */ // Use 1 as default but 0 for minimum and invalid clone_data->clone_node_max = unpack_meta_int(rsc, PCMK_META_CLONE_NODE_MAX, NULL, 1); /* Use number of nodes (but always at least 1, which is handy for crm_verify * for a CIB without nodes) as default, but 0 for minimum and invalid */ clone_data->clone_max = unpack_meta_int(rsc, PCMK_META_CLONE_MAX, NULL, QB_MAX(1, g_list_length(scheduler->nodes))); if (crm_is_true(g_hash_table_lookup(rsc->meta, PCMK_META_ORDERED))) { clone_data->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Clone", rsc->id, clone_data->flags, pcmk__clone_ordered, "pcmk__clone_ordered"); } if (!pcmk_is_set(rsc->flags, pcmk_rsc_unique) && (clone_data->clone_node_max > 1)) { pcmk__config_err("Ignoring " PCMK_META_CLONE_NODE_MAX " of %d for %s " "because anonymous clones support only one instance " "per node", clone_data->clone_node_max, rsc->id); clone_data->clone_node_max = 1; } pcmk__rsc_trace(rsc, "Options for %s", rsc->id); pcmk__rsc_trace(rsc, "\tClone max: %d", clone_data->clone_max); pcmk__rsc_trace(rsc, "\tClone node max: %d", clone_data->clone_node_max); pcmk__rsc_trace(rsc, "\tClone is unique: %s", pcmk__flag_text(rsc->flags, pcmk_rsc_unique)); pcmk__rsc_trace(rsc, "\tClone is promotable: %s", pcmk__flag_text(rsc->flags, pcmk_rsc_promotable)); // Clones may contain a single group or primitive for (a_child = pcmk__xe_first_child(xml_obj); a_child != NULL; a_child = pcmk__xe_next(a_child)) { if (pcmk__str_any_of((const char *) a_child->name, PCMK_XE_PRIMITIVE, PCMK_XE_GROUP, NULL)) { clone_data->xml_obj_child = a_child; break; } } if (clone_data->xml_obj_child == NULL) { pcmk__config_err("%s has nothing to clone", rsc->id); return FALSE; } /* * Make clones ever so slightly sticky by default * * This helps ensure clone instances are not shuffled around the cluster * for no benefit in situations when pre-allocation is not appropriate */ if (g_hash_table_lookup(rsc->meta, PCMK_META_RESOURCE_STICKINESS) == NULL) { pcmk__insert_meta(rsc, PCMK_META_RESOURCE_STICKINESS, "1"); } /* This ensures that the PCMK_META_GLOBALLY_UNIQUE value always exists for * children to inherit when being unpacked, as well as in resource agents' * environment. */ pcmk__insert_meta(rsc, PCMK_META_GLOBALLY_UNIQUE, pcmk__flag_text(rsc->flags, pcmk_rsc_unique)); if (clone_data->clone_max <= 0) { /* Create one child instance so that unpack_find_resource() will hook up * any orphans up to the parent correctly. */ if (pe__create_clone_child(rsc, scheduler) == NULL) { return FALSE; } } else { // Create a child instance for each available instance number for (lpc = 0; lpc < clone_data->clone_max; lpc++) { if (pe__create_clone_child(rsc, scheduler) == NULL) { return FALSE; } } } pcmk__rsc_trace(rsc, "Added %d children to resource %s...", clone_data->clone_max, rsc->id); return TRUE; } gboolean clone_active(pcmk_resource_t * rsc, gboolean all) { GList *gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; gboolean child_active = child_rsc->fns->active(child_rsc, all); if (all == FALSE && child_active) { return TRUE; } else if (all && child_active == FALSE) { return FALSE; } } if (all) { return TRUE; } else { return FALSE; } } /*! * \internal * \deprecated This function will be removed in a future release */ static void short_print(const char *list, const char *prefix, const char *type, const char *suffix, long options, void *print_data) { if(suffix == NULL) { suffix = ""; } if (!pcmk__str_empty(list)) { if (options & pe_print_html) { status_print("
  • "); } status_print("%s%s: [ %s ]%s", prefix, type, list, suffix); if (options & pe_print_html) { status_print("
    • \n"); } else if (options & pe_print_suppres_nl) { /* nothing */ } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print("\n"); } } } static const char * configured_role_str(pcmk_resource_t * rsc) { const char *target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); if ((target_role == NULL) && rsc->children && rsc->children->data) { pcmk_resource_t *instance = rsc->children->data; // Any instance will do target_role = g_hash_table_lookup(instance->meta, PCMK_META_TARGET_ROLE); } return target_role; } static enum rsc_role_e configured_role(pcmk_resource_t *rsc) { enum rsc_role_e role = pcmk_role_unknown; const char *target_role = configured_role_str(rsc); if (target_role != NULL) { role = pcmk_parse_role(target_role); if (role == pcmk_role_unknown) { pcmk__config_err("Invalid " PCMK_META_TARGET_ROLE " for resource %s", rsc->id); } } return role; } /*! * \internal * \deprecated This function will be removed in a future release */ static void clone_print_xml(pcmk_resource_t *rsc, const char *pre_text, long options, void *print_data) { char *child_text = crm_strdup_printf("%s ", pre_text); const char *target_role = configured_role_str(rsc); GList *gIter = rsc->children; status_print("%sid); status_print("multi_state=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_promotable)); status_print("unique=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_unique)); status_print("managed=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_managed)); status_print("failed=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_failed)); status_print("failure_ignored=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_ignore_failure)); if (target_role) { status_print("target_role=\"%s\" ", target_role); } status_print(">\n"); for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; child_rsc->fns->print(child_rsc, child_text, options, print_data); } status_print("%s\n", pre_text); free(child_text); } bool is_set_recursive(const pcmk_resource_t *rsc, long long flag, bool any) { GList *gIter; bool all = !any; if (pcmk_is_set(rsc->flags, flag)) { if(any) { return TRUE; } } else if(all) { return FALSE; } for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) { if(is_set_recursive(gIter->data, flag, any)) { if(any) { return TRUE; } } else if(all) { return FALSE; } } if(all) { return TRUE; } return FALSE; } /*! * \internal * \deprecated This function will be removed in a future release */ void clone_print(pcmk_resource_t *rsc, const char *pre_text, long options, void *print_data) { GString *list_text = NULL; char *child_text = NULL; GString *stopped_list = NULL; GList *promoted_list = NULL; GList *started_list = NULL; GList *gIter = rsc->children; clone_variant_data_t *clone_data = NULL; int active_instances = 0; if (pre_text == NULL) { pre_text = " "; } if (options & pe_print_xml) { clone_print_xml(rsc, pre_text, options, print_data); return; } get_clone_variant_data(clone_data, rsc); child_text = crm_strdup_printf("%s ", pre_text); status_print("%sClone Set: %s [%s]%s%s%s", pcmk__s(pre_text, ""), rsc->id, pcmk__xe_id(clone_data->xml_obj_child), pcmk_is_set(rsc->flags, pcmk_rsc_promotable)? " (promotable)" : "", pcmk_is_set(rsc->flags, pcmk_rsc_unique)? " (unique)" : "", pcmk_is_set(rsc->flags, pcmk_rsc_managed)? "" : " (unmanaged)"); if (options & pe_print_html) { status_print("\n
      \n"); } else if ((options & pe_print_log) == 0) { status_print("\n"); } for (; gIter != NULL; gIter = gIter->next) { gboolean print_full = FALSE; pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; gboolean partially_active = child_rsc->fns->active(child_rsc, FALSE); if (options & pe_print_clone_details) { print_full = TRUE; } if (pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { // Print individual instance when unique (except stopped orphans) if (partially_active || !pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { print_full = TRUE; } // Everything else in this block is for anonymous clones } else if (pcmk_is_set(options, pe_print_pending) && (child_rsc->pending_task != NULL) && strcmp(child_rsc->pending_task, "probe")) { // Print individual instance when non-probe action is pending print_full = TRUE; } else if (partially_active == FALSE) { // List stopped instances when requested (except orphans) if (!pcmk_is_set(child_rsc->flags, pcmk_rsc_removed) && !pcmk_is_set(options, pe_print_clone_active)) { pcmk__add_word(&stopped_list, 1024, child_rsc->id); } } else if (is_set_recursive(child_rsc, pcmk_rsc_removed, TRUE) || !is_set_recursive(child_rsc, pcmk_rsc_managed, FALSE) || is_set_recursive(child_rsc, pcmk_rsc_failed, TRUE)) { // Print individual instance when active orphaned/unmanaged/failed print_full = TRUE; } else if (child_rsc->fns->active(child_rsc, TRUE)) { // Instance of fully active anonymous clone pcmk_node_t *location = NULL; location = child_rsc->fns->location(child_rsc, NULL, TRUE); if (location) { // Instance is active on a single node enum rsc_role_e a_role = child_rsc->fns->state(child_rsc, TRUE); if (location->details->online == FALSE && location->details->unclean) { print_full = TRUE; } else if (a_role > pcmk_role_unpromoted) { promoted_list = g_list_append(promoted_list, location); } else { started_list = g_list_append(started_list, location); } } else { /* uncolocated group - bleh */ print_full = TRUE; } } else { // Instance of partially active anonymous clone print_full = TRUE; } if (print_full) { if (options & pe_print_html) { status_print("
    • \n"); } child_rsc->fns->print(child_rsc, child_text, options, print_data); if (options & pe_print_html) { status_print("
      • \n"); } } } /* Promoted */ promoted_list = g_list_sort(promoted_list, pe__cmp_node_name); for (gIter = promoted_list; gIter; gIter = gIter->next) { pcmk_node_t *host = gIter->data; pcmk__add_word(&list_text, 1024, host->details->uname); active_instances++; } if (list_text != NULL) { short_print((const char *) list_text->str, child_text, PROMOTED_INSTANCES, NULL, options, print_data); g_string_truncate(list_text, 0); } g_list_free(promoted_list); /* Started/Unpromoted */ started_list = g_list_sort(started_list, pe__cmp_node_name); for (gIter = started_list; gIter; gIter = gIter->next) { pcmk_node_t *host = gIter->data; pcmk__add_word(&list_text, 1024, host->details->uname); active_instances++; } if (list_text != NULL) { if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { enum rsc_role_e role = configured_role(rsc); if (role == pcmk_role_unpromoted) { short_print((const char *) list_text->str, child_text, UNPROMOTED_INSTANCES " (" PCMK_META_TARGET_ROLE ")", NULL, options, print_data); } else { short_print((const char *) list_text->str, child_text, UNPROMOTED_INSTANCES, NULL, options, print_data); } } else { short_print((const char *) list_text->str, child_text, "Started", NULL, options, print_data); } } g_list_free(started_list); if (!pcmk_is_set(options, pe_print_clone_active)) { const char *state = "Stopped"; enum rsc_role_e role = configured_role(rsc); if (role == pcmk_role_stopped) { state = "Stopped (disabled)"; } if (!pcmk_is_set(rsc->flags, pcmk_rsc_unique) && (clone_data->clone_max > active_instances)) { GList *nIter; GList *list = g_hash_table_get_values(rsc->allowed_nodes); /* Custom stopped list for non-unique clones */ if (stopped_list != NULL) { g_string_truncate(stopped_list, 0); } if (list == NULL) { /* Clusters with PCMK_OPT_SYMMETRIC_CLUSTER=false haven't * calculated allowed_nodes yet. If we've not probed for them * yet, the Stopped list will be empty. */ list = g_hash_table_get_values(rsc->known_on); } list = g_list_sort(list, pe__cmp_node_name); for (nIter = list; nIter != NULL; nIter = nIter->next) { pcmk_node_t *node = (pcmk_node_t *) nIter->data; if (pe_find_node(rsc->running_on, node->details->uname) == NULL) { pcmk__add_word(&stopped_list, 1024, node->details->uname); } } g_list_free(list); } if (stopped_list != NULL) { short_print((const char *) stopped_list->str, child_text, state, NULL, options, print_data); } } if (options & pe_print_html) { status_print("
    \n"); } if (list_text != NULL) { g_string_free(list_text, TRUE); } if (stopped_list != NULL) { g_string_free(stopped_list, TRUE); } free(child_text); } PCMK__OUTPUT_ARGS("clone", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__clone_xml(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); GList *gIter = rsc->children; GList *all = NULL; int rc = pcmk_rc_no_output; gboolean printed_header = FALSE; gboolean print_everything = TRUE; if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) || (strstr(rsc->id, ":") != NULL && pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches)); all = g_list_prepend(all, (gpointer) "*"); for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; if (pcmk__rsc_filtered_by_node(child_rsc, only_node)) { continue; } if (child_rsc->fns->is_filtered(child_rsc, only_rsc, print_everything)) { continue; } if (!printed_header) { const char *multi_state = pcmk__flag_text(rsc->flags, pcmk_rsc_promotable); const char *unique = pcmk__flag_text(rsc->flags, pcmk_rsc_unique); const char *maintenance = pcmk__flag_text(rsc->flags, pcmk_rsc_maintenance); const char *managed = pcmk__flag_text(rsc->flags, pcmk_rsc_managed); const char *disabled = pcmk__btoa(pe__resource_is_disabled(rsc)); const char *failed = pcmk__flag_text(rsc->flags, pcmk_rsc_failed); const char *ignored = pcmk__flag_text(rsc->flags, pcmk_rsc_ignore_failure); const char *target_role = configured_role_str(rsc); const char *desc = pe__resource_description(rsc, show_opts); printed_header = TRUE; rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_CLONE, 10, PCMK_XA_ID, rsc->id, PCMK_XA_MULTI_STATE, multi_state, PCMK_XA_UNIQUE, unique, PCMK_XA_MAINTENANCE, maintenance, PCMK_XA_MANAGED, managed, PCMK_XA_DISABLED, disabled, PCMK_XA_FAILED, failed, PCMK_XA_FAILURE_IGNORED, ignored, PCMK_XA_TARGET_ROLE, target_role, PCMK_XA_DESCRIPTION, desc); CRM_ASSERT(rc == pcmk_rc_ok); } out->message(out, crm_map_element_name(child_rsc->xml), show_opts, child_rsc, only_node, all); } if (printed_header) { pcmk__output_xml_pop_parent(out); } g_list_free(all); return rc; } PCMK__OUTPUT_ARGS("clone", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__clone_default(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); GHashTable *stopped = NULL; GString *list_text = NULL; GList *promoted_list = NULL; GList *started_list = NULL; GList *gIter = rsc->children; const char *desc = NULL; clone_variant_data_t *clone_data = NULL; int active_instances = 0; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; desc = pe__resource_description(rsc, show_opts); get_clone_variant_data(clone_data, rsc); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) || (strstr(rsc->id, ":") != NULL && pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches)); for (; gIter != NULL; gIter = gIter->next) { gboolean print_full = FALSE; pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; gboolean partially_active = child_rsc->fns->active(child_rsc, FALSE); if (pcmk__rsc_filtered_by_node(child_rsc, only_node)) { continue; } if (child_rsc->fns->is_filtered(child_rsc, only_rsc, print_everything)) { continue; } if (pcmk_is_set(show_opts, pcmk_show_clone_detail)) { print_full = TRUE; } if (pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { // Print individual instance when unique (except stopped orphans) if (partially_active || !pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { print_full = TRUE; } // Everything else in this block is for anonymous clones } else if (pcmk_is_set(show_opts, pcmk_show_pending) && (child_rsc->pending_task != NULL) && strcmp(child_rsc->pending_task, "probe")) { // Print individual instance when non-probe action is pending print_full = TRUE; } else if (partially_active == FALSE) { // List stopped instances when requested (except orphans) if (!pcmk_is_set(child_rsc->flags, pcmk_rsc_removed) && !pcmk_is_set(show_opts, pcmk_show_clone_detail) && pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { if (stopped == NULL) { stopped = pcmk__strkey_table(free, free); } pcmk__insert_dup(stopped, child_rsc->id, "Stopped"); } } else if (is_set_recursive(child_rsc, pcmk_rsc_removed, TRUE) || !is_set_recursive(child_rsc, pcmk_rsc_managed, FALSE) || is_set_recursive(child_rsc, pcmk_rsc_failed, TRUE)) { // Print individual instance when active orphaned/unmanaged/failed print_full = TRUE; } else if (child_rsc->fns->active(child_rsc, TRUE)) { // Instance of fully active anonymous clone pcmk_node_t *location = NULL; location = child_rsc->fns->location(child_rsc, NULL, TRUE); if (location) { // Instance is active on a single node enum rsc_role_e a_role = child_rsc->fns->state(child_rsc, TRUE); if (location->details->online == FALSE && location->details->unclean) { print_full = TRUE; } else if (a_role > pcmk_role_unpromoted) { promoted_list = g_list_append(promoted_list, location); } else { started_list = g_list_append(started_list, location); } } else { /* uncolocated group - bleh */ print_full = TRUE; } } else { // Instance of partially active anonymous clone print_full = TRUE; } if (print_full) { GList *all = NULL; clone_header(out, &rc, rsc, clone_data, desc); /* Print every resource that's a child of this clone. */ all = g_list_prepend(all, (gpointer) "*"); out->message(out, crm_map_element_name(child_rsc->xml), show_opts, child_rsc, only_node, all); g_list_free(all); } } if (pcmk_is_set(show_opts, pcmk_show_clone_detail)) { PCMK__OUTPUT_LIST_FOOTER(out, rc); return pcmk_rc_ok; } /* Promoted */ promoted_list = g_list_sort(promoted_list, pe__cmp_node_name); for (gIter = promoted_list; gIter; gIter = gIter->next) { pcmk_node_t *host = gIter->data; if (!pcmk__str_in_list(host->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { continue; } pcmk__add_word(&list_text, 1024, host->details->uname); active_instances++; } g_list_free(promoted_list); if ((list_text != NULL) && (list_text->len > 0)) { clone_header(out, &rc, rsc, clone_data, desc); out->list_item(out, NULL, PROMOTED_INSTANCES ": [ %s ]", (const char *) list_text->str); g_string_truncate(list_text, 0); } /* Started/Unpromoted */ started_list = g_list_sort(started_list, pe__cmp_node_name); for (gIter = started_list; gIter; gIter = gIter->next) { pcmk_node_t *host = gIter->data; if (!pcmk__str_in_list(host->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { continue; } pcmk__add_word(&list_text, 1024, host->details->uname); active_instances++; } g_list_free(started_list); if ((list_text != NULL) && (list_text->len > 0)) { clone_header(out, &rc, rsc, clone_data, desc); if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { enum rsc_role_e role = configured_role(rsc); if (role == pcmk_role_unpromoted) { out->list_item(out, NULL, UNPROMOTED_INSTANCES " (" PCMK_META_TARGET_ROLE "): [ %s ]", (const char *) list_text->str); } else { out->list_item(out, NULL, UNPROMOTED_INSTANCES ": [ %s ]", (const char *) list_text->str); } } else { out->list_item(out, NULL, "Started: [ %s ]", (const char *) list_text->str); } } if (list_text != NULL) { g_string_free(list_text, TRUE); } if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { if (!pcmk_is_set(rsc->flags, pcmk_rsc_unique) && (clone_data->clone_max > active_instances)) { GList *nIter; GList *list = g_hash_table_get_values(rsc->allowed_nodes); /* Custom stopped table for non-unique clones */ if (stopped != NULL) { g_hash_table_destroy(stopped); stopped = NULL; } if (list == NULL) { /* Clusters with PCMK_OPT_SYMMETRIC_CLUSTER=false haven't * calculated allowed_nodes yet. If we've not probed for them * yet, the Stopped list will be empty. */ list = g_hash_table_get_values(rsc->known_on); } list = g_list_sort(list, pe__cmp_node_name); for (nIter = list; nIter != NULL; nIter = nIter->next) { pcmk_node_t *node = (pcmk_node_t *) nIter->data; if (pe_find_node(rsc->running_on, node->details->uname) == NULL && pcmk__str_in_list(node->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { xmlNode *probe_op = pe__failed_probe_for_rsc(rsc, node->details->uname); const char *state = "Stopped"; if (configured_role(rsc) == pcmk_role_stopped) { state = "Stopped (disabled)"; } if (stopped == NULL) { stopped = pcmk__strkey_table(free, free); } if (probe_op != NULL) { int rc; pcmk__scan_min_int(crm_element_value(probe_op, PCMK__XA_RC_CODE), &rc, 0); g_hash_table_insert(stopped, strdup(node->details->uname), crm_strdup_printf("Stopped (%s)", services_ocf_exitcode_str(rc))); } else { pcmk__insert_dup(stopped, node->details->uname, state); } } } g_list_free(list); } if (stopped != NULL) { GList *list = sorted_hash_table_values(stopped); clone_header(out, &rc, rsc, clone_data, desc); for (GList *status_iter = list; status_iter != NULL; status_iter = status_iter->next) { const char *status = status_iter->data; GList *nodes = nodes_with_status(stopped, status); GString *nodes_str = node_list_to_str(nodes); if (nodes_str != NULL) { if (nodes_str->len > 0) { out->list_item(out, NULL, "%s: [ %s ]", status, (const char *) nodes_str->str); } g_string_free(nodes_str, TRUE); } g_list_free(nodes); } g_list_free(list); g_hash_table_destroy(stopped); /* If there are no instances of this clone (perhaps because there are no * nodes configured), simply output the clone header by itself. This can * come up in PCS testing. */ } else if (active_instances == 0) { clone_header(out, &rc, rsc, clone_data, desc); PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } void clone_free(pcmk_resource_t * rsc) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); pcmk__rsc_trace(rsc, "Freeing %s", rsc->id); for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; CRM_ASSERT(child_rsc); pcmk__rsc_trace(child_rsc, "Freeing child %s", child_rsc->id); free_xml(child_rsc->xml); child_rsc->xml = NULL; /* There could be a saved unexpanded xml */ free_xml(child_rsc->orig_xml); child_rsc->orig_xml = NULL; child_rsc->fns->free(child_rsc); } g_list_free(rsc->children); if (clone_data) { CRM_ASSERT(clone_data->demote_notify == NULL); CRM_ASSERT(clone_data->stop_notify == NULL); CRM_ASSERT(clone_data->start_notify == NULL); CRM_ASSERT(clone_data->promote_notify == NULL); } common_free(rsc); } enum rsc_role_e clone_resource_state(const pcmk_resource_t * rsc, gboolean current) { enum rsc_role_e clone_role = pcmk_role_unknown; GList *gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; enum rsc_role_e a_role = child_rsc->fns->state(child_rsc, current); if (a_role > clone_role) { clone_role = a_role; } } pcmk__rsc_trace(rsc, "%s role: %s", rsc->id, pcmk_role_text(clone_role)); return clone_role; } /*! * \internal * \brief Check whether a clone has an instance for every node * * \param[in] rsc Clone to check * \param[in] scheduler Scheduler data */ bool pe__is_universal_clone(const pcmk_resource_t *rsc, const pcmk_scheduler_t *scheduler) { if (pcmk__is_clone(rsc)) { clone_variant_data_t *clone_data = rsc->variant_opaque; if (clone_data->clone_max == g_list_length(scheduler->nodes)) { return TRUE; } } return FALSE; } gboolean pe__clone_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent) { gboolean passes = FALSE; clone_variant_data_t *clone_data = NULL; if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches)) { passes = TRUE; } else { get_clone_variant_data(clone_data, rsc); passes = pcmk__str_in_list(pcmk__xe_id(clone_data->xml_obj_child), only_rsc, pcmk__str_star_matches); if (!passes) { for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *child_rsc = NULL; child_rsc = (const pcmk_resource_t *) iter->data; if (!child_rsc->fns->is_filtered(child_rsc, only_rsc, FALSE)) { passes = TRUE; break; } } } } return !passes; } const char * pe__clone_child_id(const pcmk_resource_t *rsc) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); return pcmk__xe_id(clone_data->xml_obj_child); } /*! * \internal * \brief Check whether a clone is ordered * * \param[in] clone Clone resource to check * * \return true if clone is ordered, otherwise false */ bool pe__clone_is_ordered(const pcmk_resource_t *clone) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); return pcmk_is_set(clone_data->flags, pcmk__clone_ordered); } /*! * \internal * \brief Set a clone flag * * \param[in,out] clone Clone resource to set flag for * \param[in] flag Clone flag to set * * \return Standard Pacemaker return code (either pcmk_rc_ok if flag was not * already set or pcmk_rc_already if it was) */ int pe__set_clone_flag(pcmk_resource_t *clone, enum pcmk__clone_flags flag) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); if (pcmk_is_set(clone_data->flags, flag)) { return pcmk_rc_already; } clone_data->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Clone", clone->id, clone_data->flags, flag, "flag"); return pcmk_rc_ok; } /*! * \internal * \brief Check whether a clone flag is set * * \param[in] group Clone resource to check * \param[in] flags Flag or flags to check * * \return \c true if all \p flags are set for \p clone, otherwise \c false */ bool pe__clone_flag_is_set(const pcmk_resource_t *clone, uint32_t flags) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); CRM_ASSERT(clone_data != NULL); return pcmk_all_flags_set(clone_data->flags, flags); } /*! * \internal * \brief Create pseudo-actions needed for promotable clones * * \param[in,out] clone Promotable clone to create actions for * \param[in] any_promoting Whether any instances will be promoted * \param[in] any_demoting Whether any instance will be demoted */ void pe__create_promotable_pseudo_ops(pcmk_resource_t *clone, bool any_promoting, bool any_demoting) { pcmk_action_t *action = NULL; pcmk_action_t *action_complete = NULL; clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); // Create a "promote" action for the clone itself action = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_PROMOTE, !any_promoting, true); // Create a "promoted" action for when all promotions are done action_complete = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_PROMOTED, !any_promoting, true); - action_complete->priority = INFINITY; + action_complete->priority = PCMK_SCORE_INFINITY; // Create notification pseudo-actions for promotion if (clone_data->promote_notify == NULL) { clone_data->promote_notify = pe__action_notif_pseudo_ops(clone, PCMK_ACTION_PROMOTE, action, action_complete); } // Create a "demote" action for the clone itself action = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_DEMOTE, !any_demoting, true); // Create a "demoted" action for when all demotions are done action_complete = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_DEMOTED, !any_demoting, true); - action_complete->priority = INFINITY; + action_complete->priority = PCMK_SCORE_INFINITY; // Create notification pseudo-actions for demotion if (clone_data->demote_notify == NULL) { clone_data->demote_notify = pe__action_notif_pseudo_ops(clone, PCMK_ACTION_DEMOTE, action, action_complete); if (clone_data->promote_notify != NULL) { order_actions(clone_data->stop_notify->post_done, clone_data->promote_notify->pre, pcmk__ar_ordered); order_actions(clone_data->start_notify->post_done, clone_data->promote_notify->pre, pcmk__ar_ordered); order_actions(clone_data->demote_notify->post_done, clone_data->promote_notify->pre, pcmk__ar_ordered); order_actions(clone_data->demote_notify->post_done, clone_data->start_notify->pre, pcmk__ar_ordered); order_actions(clone_data->demote_notify->post_done, clone_data->stop_notify->pre, pcmk__ar_ordered); } } } /*! * \internal * \brief Create all notification data and actions for a clone * * \param[in,out] clone Clone to create notifications for */ void pe__create_clone_notifications(pcmk_resource_t *clone) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); pe__create_action_notifications(clone, clone_data->start_notify); pe__create_action_notifications(clone, clone_data->stop_notify); pe__create_action_notifications(clone, clone_data->promote_notify); pe__create_action_notifications(clone, clone_data->demote_notify); } /*! * \internal * \brief Free all notification data for a clone * * \param[in,out] clone Clone to free notification data for */ void pe__free_clone_notification_data(pcmk_resource_t *clone) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); pe__free_action_notification_data(clone_data->demote_notify); clone_data->demote_notify = NULL; pe__free_action_notification_data(clone_data->stop_notify); clone_data->stop_notify = NULL; pe__free_action_notification_data(clone_data->start_notify); clone_data->start_notify = NULL; pe__free_action_notification_data(clone_data->promote_notify); clone_data->promote_notify = NULL; } /*! * \internal * \brief Create pseudo-actions for clone start/stop notifications * * \param[in,out] clone Clone to create pseudo-actions for * \param[in,out] start Start action for \p clone * \param[in,out] stop Stop action for \p clone * \param[in,out] started Started action for \p clone * \param[in,out] stopped Stopped action for \p clone */ void pe__create_clone_notif_pseudo_ops(pcmk_resource_t *clone, pcmk_action_t *start, pcmk_action_t *started, pcmk_action_t *stop, pcmk_action_t *stopped) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); if (clone_data->start_notify == NULL) { clone_data->start_notify = pe__action_notif_pseudo_ops(clone, PCMK_ACTION_START, start, started); } if (clone_data->stop_notify == NULL) { clone_data->stop_notify = pe__action_notif_pseudo_ops(clone, PCMK_ACTION_STOP, stop, stopped); if ((clone_data->start_notify != NULL) && (clone_data->stop_notify != NULL)) { order_actions(clone_data->stop_notify->post_done, clone_data->start_notify->pre, pcmk__ar_ordered); } } } /*! * \internal * \brief Get maximum clone resource instances per node * * \param[in] rsc Clone resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int pe__clone_max_per_node(const pcmk_resource_t *rsc) { const clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); return clone_data->clone_node_max; } diff --git a/lib/pengine/complex.c b/lib/pengine/complex.c index 57401bd619..7094e94550 100644 --- a/lib/pengine/complex.c +++ b/lib/pengine/complex.c @@ -1,1223 +1,1223 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include "pe_status_private.h" void populate_hash(xmlNode * nvpair_list, GHashTable * hash, const char **attrs, int attrs_length); static pcmk_node_t *active_node(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean); pcmk_rsc_methods_t resource_class_functions[] = { { native_unpack, native_find_rsc, native_parameter, native_print, native_active, native_resource_state, native_location, native_free, pe__count_common, pe__native_is_filtered, active_node, pe__primitive_max_per_node, }, { group_unpack, native_find_rsc, native_parameter, group_print, group_active, group_resource_state, native_location, group_free, pe__count_common, pe__group_is_filtered, active_node, pe__group_max_per_node, }, { clone_unpack, native_find_rsc, native_parameter, clone_print, clone_active, clone_resource_state, native_location, clone_free, pe__count_common, pe__clone_is_filtered, active_node, pe__clone_max_per_node, }, { pe__unpack_bundle, native_find_rsc, native_parameter, pe__print_bundle, pe__bundle_active, pe__bundle_resource_state, native_location, pe__free_bundle, pe__count_bundle, pe__bundle_is_filtered, pe__bundle_active_node, pe__bundle_max_per_node, } }; static enum pe_obj_types get_resource_type(const char *name) { if (pcmk__str_eq(name, PCMK_XE_PRIMITIVE, pcmk__str_casei)) { return pcmk_rsc_variant_primitive; } else if (pcmk__str_eq(name, PCMK_XE_GROUP, pcmk__str_casei)) { return pcmk_rsc_variant_group; } else if (pcmk__str_eq(name, PCMK_XE_CLONE, pcmk__str_casei)) { return pcmk_rsc_variant_clone; } else if (pcmk__str_eq(name, PCMK__XE_PROMOTABLE_LEGACY, pcmk__str_casei)) { // @COMPAT deprecated since 2.0.0 return pcmk_rsc_variant_clone; } else if (pcmk__str_eq(name, PCMK_XE_BUNDLE, pcmk__str_casei)) { return pcmk_rsc_variant_bundle; } return pcmk_rsc_variant_unknown; } /*! * \internal * \brief Insert a meta-attribute if not already present * * \param[in] key Meta-attribute name * \param[in] value Meta-attribute value to add if not already present * \param[in,out] table Meta-attribute hash table to insert into * * \note This is like pcmk__insert_meta() except it won't overwrite existing * values. */ static void dup_attr(gpointer key, gpointer value, gpointer user_data) { GHashTable *table = user_data; CRM_CHECK(table != NULL, return); if ((key != NULL) && !pcmk__str_eq((const char *) value, "#default", pcmk__str_casei|pcmk__str_null_matches) && (g_hash_table_lookup(table, key) == NULL)) { pcmk__insert_dup(table, (const char *) key, (const char *) value); } } static void expand_parents_fixed_nvpairs(pcmk_resource_t *rsc, pe_rule_eval_data_t *rule_data, GHashTable *meta_hash, pcmk_scheduler_t *scheduler) { GHashTable *parent_orig_meta = pcmk__strkey_table(free, free); pcmk_resource_t *p = rsc->parent; if (p == NULL) { return ; } /* Search all parent resources, get the fixed value of * PCMK_XE_META_ATTRIBUTES set only in the original xml, and stack it in the * hash table. The fixed value of the lower parent resource takes precedence * and is not overwritten. */ while(p != NULL) { /* A hash table for comparison is generated, including the id-ref. */ pe__unpack_dataset_nvpairs(p->xml, PCMK_XE_META_ATTRIBUTES, rule_data, parent_orig_meta, NULL, FALSE, scheduler); p = p->parent; } if (parent_orig_meta != NULL) { // This will not overwrite any values already existing for child g_hash_table_foreach(parent_orig_meta, dup_attr, meta_hash); } if (parent_orig_meta != NULL) { g_hash_table_destroy(parent_orig_meta); } return ; } void get_meta_attributes(GHashTable * meta_hash, pcmk_resource_t * rsc, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pe_rsc_eval_data_t rsc_rule_data = { .standard = crm_element_value(rsc->xml, PCMK_XA_CLASS), .provider = crm_element_value(rsc->xml, PCMK_XA_PROVIDER), .agent = crm_element_value(rsc->xml, PCMK_XA_TYPE) }; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->now, .match_data = NULL, .rsc_data = &rsc_rule_data, .op_data = NULL }; if (node) { rule_data.node_hash = node->details->attrs; } for (xmlAttrPtr a = pcmk__xe_first_attr(rsc->xml); a != NULL; a = a->next) { if (a->children != NULL) { dup_attr((gpointer) a->name, (gpointer) a->children->content, meta_hash); } } pe__unpack_dataset_nvpairs(rsc->xml, PCMK_XE_META_ATTRIBUTES, &rule_data, meta_hash, NULL, FALSE, scheduler); /* Set the PCMK_XE_META_ATTRIBUTES explicitly set in the parent resource to * the hash table of the child resource. If it is already explicitly set as * a child, it will not be overwritten. */ if (rsc->parent != NULL) { expand_parents_fixed_nvpairs(rsc, &rule_data, meta_hash, scheduler); } /* check the defaults */ pe__unpack_dataset_nvpairs(scheduler->rsc_defaults, PCMK_XE_META_ATTRIBUTES, &rule_data, meta_hash, NULL, FALSE, scheduler); /* If there is PCMK_XE_META_ATTRIBUTES that the parent resource has not * explicitly set, set a value that is not set from PCMK_XE_RSC_DEFAULTS * either. The values already set up to this point will not be overwritten. */ if (rsc->parent) { g_hash_table_foreach(rsc->parent->meta, dup_attr, meta_hash); } } void get_rsc_attributes(GHashTable *meta_hash, const pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; if (node) { rule_data.node_hash = node->details->attrs; } pe__unpack_dataset_nvpairs(rsc->xml, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, meta_hash, NULL, FALSE, scheduler); /* set anything else based on the parent */ if (rsc->parent != NULL) { get_rsc_attributes(meta_hash, rsc->parent, node, scheduler); } else { /* and finally check the defaults */ pe__unpack_dataset_nvpairs(scheduler->rsc_defaults, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, meta_hash, NULL, FALSE, scheduler); } } static char * template_op_key(xmlNode * op) { const char *name = crm_element_value(op, PCMK_XA_NAME); const char *role = crm_element_value(op, PCMK_XA_ROLE); char *key = NULL; if ((role == NULL) || pcmk__strcase_any_of(role, PCMK__ROLE_STARTED, PCMK__ROLE_UNPROMOTED, PCMK__ROLE_UNPROMOTED_LEGACY, NULL)) { role = PCMK__ROLE_UNKNOWN; } key = crm_strdup_printf("%s-%s", name, role); return key; } static gboolean unpack_template(xmlNode *xml_obj, xmlNode **expanded_xml, pcmk_scheduler_t *scheduler) { xmlNode *cib_resources = NULL; xmlNode *template = NULL; xmlNode *new_xml = NULL; xmlNode *child_xml = NULL; xmlNode *rsc_ops = NULL; xmlNode *template_ops = NULL; const char *template_ref = NULL; const char *id = NULL; if (xml_obj == NULL) { pcmk__config_err("No resource object for template unpacking"); return FALSE; } template_ref = crm_element_value(xml_obj, PCMK_XA_TEMPLATE); if (template_ref == NULL) { return TRUE; } id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("'%s' object must have a id", xml_obj->name); return FALSE; } if (pcmk__str_eq(template_ref, id, pcmk__str_none)) { pcmk__config_err("The resource object '%s' should not reference itself", id); return FALSE; } cib_resources = get_xpath_object("//" PCMK_XE_RESOURCES, scheduler->input, LOG_TRACE); if (cib_resources == NULL) { pcmk__config_err("No resources configured"); return FALSE; } template = pcmk__xe_match(cib_resources, PCMK_XE_TEMPLATE, PCMK_XA_ID, template_ref); if (template == NULL) { pcmk__config_err("No template named '%s'", template_ref); return FALSE; } new_xml = copy_xml(template); xmlNodeSetName(new_xml, xml_obj->name); crm_xml_add(new_xml, PCMK_XA_ID, id); crm_xml_add(new_xml, PCMK__META_CLONE, crm_element_value(xml_obj, PCMK__META_CLONE)); template_ops = find_xml_node(new_xml, PCMK_XE_OPERATIONS, FALSE); for (child_xml = pcmk__xe_first_child(xml_obj); child_xml != NULL; child_xml = pcmk__xe_next(child_xml)) { xmlNode *new_child = NULL; new_child = add_node_copy(new_xml, child_xml); if (pcmk__xe_is(new_child, PCMK_XE_OPERATIONS)) { rsc_ops = new_child; } } if (template_ops && rsc_ops) { xmlNode *op = NULL; GHashTable *rsc_ops_hash = pcmk__strkey_table(free, NULL); for (op = pcmk__xe_first_child(rsc_ops); op != NULL; op = pcmk__xe_next(op)) { char *key = template_op_key(op); g_hash_table_insert(rsc_ops_hash, key, op); } for (op = pcmk__xe_first_child(template_ops); op != NULL; op = pcmk__xe_next(op)) { char *key = template_op_key(op); if (g_hash_table_lookup(rsc_ops_hash, key) == NULL) { add_node_copy(rsc_ops, op); } free(key); } if (rsc_ops_hash) { g_hash_table_destroy(rsc_ops_hash); } free_xml(template_ops); } /*free_xml(*expanded_xml); */ *expanded_xml = new_xml; #if 0 /* Disable multi-level templates for now */ if (!unpack_template(new_xml, expanded_xml, scheduler)) { free_xml(*expanded_xml); *expanded_xml = NULL; return FALSE; } #endif return TRUE; } static gboolean add_template_rsc(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { const char *template_ref = NULL; const char *id = NULL; if (xml_obj == NULL) { pcmk__config_err("No resource object for processing resource list " "of template"); return FALSE; } template_ref = crm_element_value(xml_obj, PCMK_XA_TEMPLATE); if (template_ref == NULL) { return TRUE; } id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("'%s' object must have a id", xml_obj->name); return FALSE; } if (pcmk__str_eq(template_ref, id, pcmk__str_none)) { pcmk__config_err("The resource object '%s' should not reference itself", id); return FALSE; } if (add_tag_ref(scheduler->template_rsc_sets, template_ref, id) == FALSE) { return FALSE; } return TRUE; } static bool detect_promotable(pcmk_resource_t *rsc) { const char *promotable = g_hash_table_lookup(rsc->meta, PCMK_META_PROMOTABLE); if (crm_is_true(promotable)) { return TRUE; } // @COMPAT deprecated since 2.0.0 if (pcmk__xe_is(rsc->xml, PCMK__XE_PROMOTABLE_LEGACY)) { /* @TODO in some future version, pcmk__warn_once() here, * then drop support in even later version */ pcmk__insert_dup(rsc->meta, PCMK_META_PROMOTABLE, PCMK_VALUE_TRUE); return TRUE; } return FALSE; } static void free_params_table(gpointer data) { g_hash_table_destroy((GHashTable *) data); } /*! * \brief Get a table of resource parameters * * \param[in,out] rsc Resource to query * \param[in] node Node for evaluating rules (NULL for defaults) * \param[in,out] scheduler Scheduler data * * \return Hash table containing resource parameter names and values * (or NULL if \p rsc or \p scheduler is NULL) * \note The returned table will be destroyed when the resource is freed, so * callers should not destroy it. */ GHashTable * pe_rsc_params(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { GHashTable *params_on_node = NULL; /* A NULL node is used to request the resource's default parameters * (not evaluated for node), but we always want something non-NULL * as a hash table key. */ const char *node_name = ""; // Sanity check if ((rsc == NULL) || (scheduler == NULL)) { return NULL; } if ((node != NULL) && (node->details->uname != NULL)) { node_name = node->details->uname; } // Find the parameter table for given node if (rsc->parameter_cache == NULL) { rsc->parameter_cache = pcmk__strikey_table(free, free_params_table); } else { params_on_node = g_hash_table_lookup(rsc->parameter_cache, node_name); } // If none exists yet, create one with parameters evaluated for node if (params_on_node == NULL) { params_on_node = pcmk__strkey_table(free, free); get_rsc_attributes(params_on_node, rsc, node, scheduler); g_hash_table_insert(rsc->parameter_cache, strdup(node_name), params_on_node); } return params_on_node; } /*! * \internal * \brief Unpack a resource's \c PCMK_META_REQUIRES meta-attribute * * \param[in,out] rsc Resource being unpacked * \param[in] value Value of \c PCMK_META_REQUIRES meta-attribute * \param[in] is_default Whether \p value was selected by default */ static void unpack_requires(pcmk_resource_t *rsc, const char *value, bool is_default) { if (pcmk__str_eq(value, PCMK_VALUE_NOTHING, pcmk__str_casei)) { } else if (pcmk__str_eq(value, PCMK_VALUE_QUORUM, pcmk__str_casei)) { pcmk__set_rsc_flags(rsc, pcmk_rsc_needs_quorum); } else if (pcmk__str_eq(value, PCMK_VALUE_FENCING, pcmk__str_casei)) { pcmk__set_rsc_flags(rsc, pcmk_rsc_needs_fencing); if (!pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("%s requires fencing but fencing is disabled", rsc->id); } } else if (pcmk__str_eq(value, PCMK_VALUE_UNFENCING, pcmk__str_casei)) { if (pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) { pcmk__config_warn("Resetting \"" PCMK_META_REQUIRES "\" for %s " "to \"" PCMK_VALUE_QUORUM "\" because fencing " "devices cannot require unfencing", rsc->id); unpack_requires(rsc, PCMK_VALUE_QUORUM, true); return; } else if (!pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("Resetting \"" PCMK_META_REQUIRES "\" for %s " "to \"" PCMK_VALUE_QUORUM "\" because fencing is " "disabled", rsc->id); unpack_requires(rsc, PCMK_VALUE_QUORUM, true); return; } else { pcmk__set_rsc_flags(rsc, pcmk_rsc_needs_fencing |pcmk_rsc_needs_unfencing); } } else { const char *orig_value = value; if (pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) { value = PCMK_VALUE_QUORUM; } else if ((rsc->variant == pcmk_rsc_variant_primitive) && xml_contains_remote_node(rsc->xml)) { value = PCMK_VALUE_QUORUM; } else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_enable_unfencing)) { value = PCMK_VALUE_UNFENCING; } else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { value = PCMK_VALUE_FENCING; } else if (rsc->cluster->no_quorum_policy == pcmk_no_quorum_ignore) { value = PCMK_VALUE_NOTHING; } else { value = PCMK_VALUE_QUORUM; } if (orig_value != NULL) { pcmk__config_err("Resetting '" PCMK_META_REQUIRES "' for %s " "to '%s' because '%s' is not valid", rsc->id, value, orig_value); } unpack_requires(rsc, value, true); return; } pcmk__rsc_trace(rsc, "\tRequired to start: %s%s", value, (is_default? " (default)" : "")); } #ifndef PCMK__COMPAT_2_0 static void warn_about_deprecated_classes(pcmk_resource_t *rsc) { const char *std = crm_element_value(rsc->xml, PCMK_XA_CLASS); if (pcmk__str_eq(std, PCMK_RESOURCE_CLASS_UPSTART, pcmk__str_none)) { pcmk__warn_once(pcmk__wo_upstart, "Support for Upstart resources (such as %s) is " "deprecated and will be removed in a future release", rsc->id); } else if (pcmk__str_eq(std, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_none)) { pcmk__warn_once(pcmk__wo_nagios, "Support for Nagios resources (such as %s) is " "deprecated and will be removed in a future release", rsc->id); } } #endif /*! * \internal * \brief Unpack configuration XML for a given resource * * Unpack the XML object containing a resource's configuration into a new * \c pcmk_resource_t object. * * \param[in] xml_obj XML node containing the resource's configuration * \param[out] rsc Where to store the unpacked resource information * \param[in] parent Resource's parent, if any * \param[in,out] scheduler Scheduler data * * \return Standard Pacemaker return code * \note If pcmk_rc_ok is returned, \p *rsc is guaranteed to be non-NULL, and * the caller is responsible for freeing it using its variant-specific * free() method. Otherwise, \p *rsc is guaranteed to be NULL. */ int pe__unpack_resource(xmlNode *xml_obj, pcmk_resource_t **rsc, pcmk_resource_t *parent, pcmk_scheduler_t *scheduler) { xmlNode *expanded_xml = NULL; xmlNode *ops = NULL; const char *value = NULL; const char *id = NULL; bool guest_node = false; bool remote_node = false; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = NULL, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; CRM_CHECK(rsc != NULL, return EINVAL); CRM_CHECK((xml_obj != NULL) && (scheduler != NULL), *rsc = NULL; return EINVAL); rule_data.now = scheduler->now; crm_log_xml_trace(xml_obj, "[raw XML]"); id = crm_element_value(xml_obj, PCMK_XA_ID); if (id == NULL) { pcmk__config_err("Ignoring <%s> configuration without " PCMK_XA_ID, xml_obj->name); return pcmk_rc_unpack_error; } if (unpack_template(xml_obj, &expanded_xml, scheduler) == FALSE) { return pcmk_rc_unpack_error; } *rsc = calloc(1, sizeof(pcmk_resource_t)); if (*rsc == NULL) { pcmk__sched_err("Unable to allocate memory for resource '%s'", id); return ENOMEM; } (*rsc)->cluster = scheduler; if (expanded_xml) { crm_log_xml_trace(expanded_xml, "[expanded XML]"); (*rsc)->xml = expanded_xml; (*rsc)->orig_xml = xml_obj; } else { (*rsc)->xml = xml_obj; (*rsc)->orig_xml = NULL; } /* Do not use xml_obj from here on, use (*rsc)->xml in case templates are involved */ (*rsc)->parent = parent; ops = find_xml_node((*rsc)->xml, PCMK_XE_OPERATIONS, FALSE); (*rsc)->ops_xml = expand_idref(ops, scheduler->input); (*rsc)->variant = get_resource_type((const char *) (*rsc)->xml->name); if ((*rsc)->variant == pcmk_rsc_variant_unknown) { pcmk__config_err("Ignoring resource '%s' of unknown type '%s'", id, (*rsc)->xml->name); common_free(*rsc); *rsc = NULL; return pcmk_rc_unpack_error; } #ifndef PCMK__COMPAT_2_0 warn_about_deprecated_classes(*rsc); #endif (*rsc)->meta = pcmk__strkey_table(free, free); (*rsc)->allowed_nodes = pcmk__strkey_table(NULL, free); (*rsc)->known_on = pcmk__strkey_table(NULL, free); value = crm_element_value((*rsc)->xml, PCMK__META_CLONE); if (value) { (*rsc)->id = crm_strdup_printf("%s:%s", id, value); pcmk__insert_meta(*rsc, PCMK__META_CLONE, value); } else { (*rsc)->id = strdup(id); } (*rsc)->fns = &resource_class_functions[(*rsc)->variant]; get_meta_attributes((*rsc)->meta, *rsc, NULL, scheduler); (*rsc)->parameters = pe_rsc_params(*rsc, NULL, scheduler); // \deprecated (*rsc)->flags = 0; pcmk__set_rsc_flags(*rsc, pcmk_rsc_runnable|pcmk_rsc_unassigned); if (!pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) { pcmk__set_rsc_flags(*rsc, pcmk_rsc_managed); } (*rsc)->rsc_cons = NULL; (*rsc)->rsc_tickets = NULL; (*rsc)->actions = NULL; (*rsc)->role = pcmk_role_stopped; (*rsc)->next_role = pcmk_role_unknown; (*rsc)->recovery_type = pcmk_multiply_active_restart; (*rsc)->stickiness = 0; - (*rsc)->migration_threshold = INFINITY; + (*rsc)->migration_threshold = PCMK_SCORE_INFINITY; (*rsc)->failure_timeout = 0; value = g_hash_table_lookup((*rsc)->meta, PCMK_META_PRIORITY); (*rsc)->priority = char2score(value); value = g_hash_table_lookup((*rsc)->meta, PCMK_META_CRITICAL); if ((value == NULL) || crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk_rsc_critical); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_NOTIFY); if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk_rsc_notify); } if (xml_contains_remote_node((*rsc)->xml)) { (*rsc)->is_remote_node = TRUE; if (g_hash_table_lookup((*rsc)->meta, PCMK__META_CONTAINER)) { guest_node = true; } else { remote_node = true; } } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_ALLOW_MIGRATE); if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk_rsc_migratable); } else if ((value == NULL) && remote_node) { /* By default, we want remote nodes to be able * to float around the cluster without having to stop all the * resources within the remote-node before moving. Allowing * migration support enables this feature. If this ever causes * problems, migration support can be explicitly turned off with * PCMK_META_ALLOW_MIGRATE=false. */ pcmk__set_rsc_flags(*rsc, pcmk_rsc_migratable); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_IS_MANAGED); if ((value != NULL) && !pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) { if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk_rsc_managed); } else { pcmk__clear_rsc_flags(*rsc, pcmk_rsc_managed); } } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_MAINTENANCE); if (crm_is_true(value)) { pcmk__clear_rsc_flags(*rsc, pcmk_rsc_managed); pcmk__set_rsc_flags(*rsc, pcmk_rsc_maintenance); } if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) { pcmk__clear_rsc_flags(*rsc, pcmk_rsc_managed); pcmk__set_rsc_flags(*rsc, pcmk_rsc_maintenance); } if (pcmk__is_clone(pe__const_top_resource(*rsc, false))) { value = g_hash_table_lookup((*rsc)->meta, PCMK_META_GLOBALLY_UNIQUE); if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk_rsc_unique); } if (detect_promotable(*rsc)) { pcmk__set_rsc_flags(*rsc, pcmk_rsc_promotable); } } else { pcmk__set_rsc_flags(*rsc, pcmk_rsc_unique); } // @COMPAT Deprecated meta-attribute value = g_hash_table_lookup((*rsc)->meta, PCMK__META_RESTART_TYPE); if (pcmk__str_eq(value, PCMK_VALUE_RESTART, pcmk__str_casei)) { (*rsc)->restart_type = pe_restart_restart; pcmk__rsc_trace(*rsc, "%s dependency restart handling: restart", (*rsc)->id); pcmk__warn_once(pcmk__wo_restart_type, "Support for " PCMK__META_RESTART_TYPE " is deprecated " "and will be removed in a future release"); } else { (*rsc)->restart_type = pe_restart_ignore; pcmk__rsc_trace(*rsc, "%s dependency restart handling: ignore", (*rsc)->id); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_MULTIPLE_ACTIVE); if (pcmk__str_eq(value, "stop_only", pcmk__str_casei)) { (*rsc)->recovery_type = pcmk_multiply_active_stop; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: stop only", (*rsc)->id); } else if (pcmk__str_eq(value, PCMK_VALUE_BLOCK, pcmk__str_casei)) { (*rsc)->recovery_type = pcmk_multiply_active_block; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: block", (*rsc)->id); } else if (pcmk__str_eq(value, "stop_unexpected", pcmk__str_casei)) { (*rsc)->recovery_type = pcmk_multiply_active_unexpected; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: " "stop unexpected instances", (*rsc)->id); } else { // "stop_start" if (!pcmk__str_eq(value, "stop_start", pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_warn("%s is not a valid value for " PCMK_META_MULTIPLE_ACTIVE ", using default of \"stop_start\"", value); } (*rsc)->recovery_type = pcmk_multiply_active_restart; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: stop/start", (*rsc)->id); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_RESOURCE_STICKINESS); if ((value != NULL) && !pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) { (*rsc)->stickiness = char2score(value); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_MIGRATION_THRESHOLD); if ((value != NULL) && !pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) { (*rsc)->migration_threshold = char2score(value); if ((*rsc)->migration_threshold < 0) { /* @TODO We use 1 here to preserve previous behavior, but this * should probably use the default (INFINITY) or 0 (to disable) * instead. */ pcmk__warn_once(pcmk__wo_neg_threshold, PCMK_META_MIGRATION_THRESHOLD " must be non-negative, using 1 instead"); (*rsc)->migration_threshold = 1; } } if (pcmk__str_eq(crm_element_value((*rsc)->xml, PCMK_XA_CLASS), PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_fencing); pcmk__set_rsc_flags(*rsc, pcmk_rsc_fence_device); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_REQUIRES); unpack_requires(*rsc, value, false); value = g_hash_table_lookup((*rsc)->meta, PCMK_META_FAILURE_TIMEOUT); if (value != NULL) { guint interval_ms = 0U; // Stored as seconds pcmk_parse_interval_spec(value, &interval_ms); (*rsc)->failure_timeout = (int) (interval_ms / 1000); } if (remote_node) { GHashTable *params = pe_rsc_params(*rsc, NULL, scheduler); /* Grabbing the value now means that any rules based on node attributes * will evaluate to false, so such rules should not be used with * PCMK_REMOTE_RA_RECONNECT_INTERVAL. * * @TODO Evaluate per node before using */ value = g_hash_table_lookup(params, PCMK_REMOTE_RA_RECONNECT_INTERVAL); if (value) { /* reconnect delay works by setting failure_timeout and preventing the * connection from starting until the failure is cleared. */ pcmk_parse_interval_spec(value, &((*rsc)->remote_reconnect_ms)); /* We want to override any default failure_timeout in use when remote * PCMK_REMOTE_RA_RECONNECT_INTERVAL is in use. */ (*rsc)->failure_timeout = (*rsc)->remote_reconnect_ms / 1000; } } get_target_role(*rsc, &((*rsc)->next_role)); pcmk__rsc_trace(*rsc, "%s desired next state: %s", (*rsc)->id, ((*rsc)->next_role == pcmk_role_unknown)? "default" : pcmk_role_text((*rsc)->next_role)); if ((*rsc)->fns->unpack(*rsc, scheduler) == FALSE) { (*rsc)->fns->free(*rsc); *rsc = NULL; return pcmk_rc_unpack_error; } if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) { // This tag must stay exactly the same because it is tested elsewhere resource_location(*rsc, NULL, 0, "symmetric_default", scheduler); } else if (guest_node) { /* remote resources tied to a container resource must always be allowed * to opt-in to the cluster. Whether the connection resource is actually * allowed to be placed on a node is dependent on the container resource */ resource_location(*rsc, NULL, 0, "remote_connection_default", scheduler); } pcmk__rsc_trace(*rsc, "%s action notification: %s", (*rsc)->id, pcmk_is_set((*rsc)->flags, pcmk_rsc_notify)? "required" : "not required"); (*rsc)->utilization = pcmk__strkey_table(free, free); pe__unpack_dataset_nvpairs((*rsc)->xml, PCMK_XE_UTILIZATION, &rule_data, (*rsc)->utilization, NULL, FALSE, scheduler); if (expanded_xml) { if (add_template_rsc(xml_obj, scheduler) == FALSE) { (*rsc)->fns->free(*rsc); *rsc = NULL; return pcmk_rc_unpack_error; } } return pcmk_rc_ok; } gboolean is_parent(pcmk_resource_t *child, pcmk_resource_t *rsc) { pcmk_resource_t *parent = child; if (parent == NULL || rsc == NULL) { return FALSE; } while (parent->parent != NULL) { if (parent->parent == rsc) { return TRUE; } parent = parent->parent; } return FALSE; } pcmk_resource_t * uber_parent(pcmk_resource_t *rsc) { pcmk_resource_t *parent = rsc; if (parent == NULL) { return NULL; } while ((parent->parent != NULL) && (parent->parent->variant != pcmk_rsc_variant_bundle)) { parent = parent->parent; } return parent; } /*! * \internal * \brief Get the topmost parent of a resource as a const pointer * * \param[in] rsc Resource to check * \param[in] include_bundle If true, go all the way to bundle * * \return \p NULL if \p rsc is NULL, \p rsc if \p rsc has no parent, * the bundle if \p rsc is bundled and \p include_bundle is true, * otherwise the topmost parent of \p rsc up to a clone */ const pcmk_resource_t * pe__const_top_resource(const pcmk_resource_t *rsc, bool include_bundle) { const pcmk_resource_t *parent = rsc; if (parent == NULL) { return NULL; } while (parent->parent != NULL) { if (!include_bundle && (parent->parent->variant == pcmk_rsc_variant_bundle)) { break; } parent = parent->parent; } return parent; } void common_free(pcmk_resource_t * rsc) { if (rsc == NULL) { return; } pcmk__rsc_trace(rsc, "Freeing %s %d", rsc->id, rsc->variant); g_list_free(rsc->rsc_cons); g_list_free(rsc->rsc_cons_lhs); g_list_free(rsc->rsc_tickets); g_list_free(rsc->dangling_migrations); if (rsc->parameter_cache != NULL) { g_hash_table_destroy(rsc->parameter_cache); } if (rsc->meta != NULL) { g_hash_table_destroy(rsc->meta); } if (rsc->utilization != NULL) { g_hash_table_destroy(rsc->utilization); } if ((rsc->parent == NULL) && pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { free_xml(rsc->xml); rsc->xml = NULL; free_xml(rsc->orig_xml); rsc->orig_xml = NULL; /* if rsc->orig_xml, then rsc->xml is an expanded xml from a template */ } else if (rsc->orig_xml) { free_xml(rsc->xml); rsc->xml = NULL; } if (rsc->running_on) { g_list_free(rsc->running_on); rsc->running_on = NULL; } if (rsc->known_on) { g_hash_table_destroy(rsc->known_on); rsc->known_on = NULL; } if (rsc->actions) { g_list_free(rsc->actions); rsc->actions = NULL; } if (rsc->allowed_nodes) { g_hash_table_destroy(rsc->allowed_nodes); rsc->allowed_nodes = NULL; } g_list_free(rsc->fillers); g_list_free(rsc->rsc_location); pcmk__rsc_trace(rsc, "Resource freed"); free(rsc->id); free(rsc->clone_name); free(rsc->allocated_to); free(rsc->variant_opaque); free(rsc->pending_task); free(rsc); } /*! * \internal * \brief Count a node and update most preferred to it as appropriate * * \param[in] rsc An active resource * \param[in] node A node that \p rsc is active on * \param[in,out] active This will be set to \p node if \p node is more * preferred than the current value * \param[in,out] count_all If not NULL, this will be incremented * \param[in,out] count_clean If not NULL, this will be incremented if \p node * is online and clean * * \return true if the count should continue, or false if sufficiently known */ bool pe__count_active_node(const pcmk_resource_t *rsc, pcmk_node_t *node, pcmk_node_t **active, unsigned int *count_all, unsigned int *count_clean) { bool keep_looking = false; bool is_happy = false; CRM_CHECK((rsc != NULL) && (node != NULL) && (active != NULL), return false); is_happy = node->details->online && !node->details->unclean; if (count_all != NULL) { ++*count_all; } if ((count_clean != NULL) && is_happy) { ++*count_clean; } if ((count_all != NULL) || (count_clean != NULL)) { keep_looking = true; // We're counting, so go through entire list } if (rsc->partial_migration_source != NULL) { if (pcmk__same_node(node, rsc->partial_migration_source)) { *active = node; // This is the migration source } else { keep_looking = true; } } else if (!pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) { if (is_happy && ((*active == NULL) || !(*active)->details->online || (*active)->details->unclean)) { *active = node; // This is the first clean node } else { keep_looking = true; } } if (*active == NULL) { *active = node; // This is the first node checked } return keep_looking; } // Shared implementation of pcmk_rsc_methods_t:active_node() static pcmk_node_t * active_node(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean) { pcmk_node_t *active = NULL; if (count_all != NULL) { *count_all = 0; } if (count_clean != NULL) { *count_clean = 0; } if (rsc == NULL) { return NULL; } for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { if (!pe__count_active_node(rsc, (pcmk_node_t *) iter->data, &active, count_all, count_clean)) { break; // Don't waste time iterating if we don't have to } } return active; } /*! * \brief * \internal Find and count active nodes according to \c PCMK_META_REQUIRES * * \param[in] rsc Resource to check * \param[out] count If not NULL, will be set to count of active nodes * * \return An active node (or NULL if resource is not active anywhere) * * \note This is a convenience wrapper for active_node() where the count of all * active nodes or only clean active nodes is desired according to the * \c PCMK_META_REQUIRES meta-attribute. */ pcmk_node_t * pe__find_active_requires(const pcmk_resource_t *rsc, unsigned int *count) { if (rsc == NULL) { if (count != NULL) { *count = 0; } return NULL; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) { return rsc->fns->active_node(rsc, count, NULL); } else { return rsc->fns->active_node(rsc, NULL, count); } } void pe__count_common(pcmk_resource_t *rsc) { if (rsc->children != NULL) { for (GList *item = rsc->children; item != NULL; item = item->next) { ((pcmk_resource_t *) item->data)->fns->count(item->data); } } else if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed) || (rsc->role > pcmk_role_stopped)) { rsc->cluster->ninstances++; if (pe__resource_is_disabled(rsc)) { rsc->cluster->disabled_resources++; } if (pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) { rsc->cluster->blocked_resources++; } } } /*! * \internal * \brief Update a resource's next role * * \param[in,out] rsc Resource to be updated * \param[in] role Resource's new next role * \param[in] why Human-friendly reason why role is changing (for logs) */ void pe__set_next_role(pcmk_resource_t *rsc, enum rsc_role_e role, const char *why) { CRM_ASSERT((rsc != NULL) && (why != NULL)); if (rsc->next_role != role) { pcmk__rsc_trace(rsc, "Resetting next role for %s from %s to %s (%s)", rsc->id, pcmk_role_text(rsc->next_role), pcmk_role_text(role), why); rsc->next_role = role; } } diff --git a/lib/pengine/native.c b/lib/pengine/native.c index 3828a4cbaf..14c7339736 100644 --- a/lib/pengine/native.c +++ b/lib/pengine/native.c @@ -1,1477 +1,1477 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #ifdef PCMK__COMPAT_2_0 #define PROVIDER_SEP "::" #else #define PROVIDER_SEP ":" #endif /*! * \internal * \brief Check whether a resource is active on multiple nodes */ static bool is_multiply_active(const pcmk_resource_t *rsc) { unsigned int count = 0; if (rsc->variant == pcmk_rsc_variant_primitive) { pe__find_active_requires(rsc, &count); } return count > 1; } static void native_priority_to_node(pcmk_resource_t *rsc, pcmk_node_t *node, gboolean failed) { int priority = 0; if ((rsc->priority == 0) || (failed == TRUE)) { return; } if (rsc->role == pcmk_role_promoted) { // Promoted instance takes base priority + 1 priority = rsc->priority + 1; } else { priority = rsc->priority; } node->details->priority += priority; pcmk__rsc_trace(rsc, "%s now has priority %d with %s'%s' (priority: %d%s)", pcmk__node_name(node), node->details->priority, (rsc->role == pcmk_role_promoted)? "promoted " : "", rsc->id, rsc->priority, (rsc->role == pcmk_role_promoted)? " + 1" : ""); /* Priority of a resource running on a guest node is added to the cluster * node as well. */ if (node->details->remote_rsc && node->details->remote_rsc->container) { GList *gIter = node->details->remote_rsc->container->running_on; for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *a_node = gIter->data; a_node->details->priority += priority; pcmk__rsc_trace(rsc, "%s now has priority %d with %s'%s' " "(priority: %d%s) from guest node %s", pcmk__node_name(a_node), a_node->details->priority, (rsc->role == pcmk_role_promoted)? "promoted " : "", rsc->id, rsc->priority, (rsc->role == pcmk_role_promoted)? " + 1" : "", pcmk__node_name(node)); } } } void native_add_running(pcmk_resource_t *rsc, pcmk_node_t *node, pcmk_scheduler_t *scheduler, gboolean failed) { GList *gIter = rsc->running_on; CRM_CHECK(node != NULL, return); for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *a_node = (pcmk_node_t *) gIter->data; CRM_CHECK(a_node != NULL, return); if (pcmk__str_eq(a_node->details->id, node->details->id, pcmk__str_casei)) { return; } } pcmk__rsc_trace(rsc, "Adding %s to %s %s", rsc->id, pcmk__node_name(node), pcmk_is_set(rsc->flags, pcmk_rsc_managed)? "" : "(unmanaged)"); rsc->running_on = g_list_append(rsc->running_on, node); if (rsc->variant == pcmk_rsc_variant_primitive) { node->details->running_rsc = g_list_append(node->details->running_rsc, rsc); native_priority_to_node(rsc, node, failed); } if ((rsc->variant == pcmk_rsc_variant_primitive) && node->details->maintenance) { pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed); pcmk__set_rsc_flags(rsc, pcmk_rsc_maintenance); } if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk_resource_t *p = rsc->parent; pcmk__rsc_info(rsc, "resource %s isn't managed", rsc->id); - resource_location(rsc, node, INFINITY, "not_managed_default", - scheduler); + resource_location(rsc, node, PCMK_SCORE_INFINITY, + "not_managed_default", scheduler); while(p && node->details->online) { /* add without the additional location constraint */ p->running_on = g_list_append(p->running_on, node); p = p->parent; } return; } if (is_multiply_active(rsc)) { switch (rsc->recovery_type) { case pcmk_multiply_active_stop: { GHashTableIter gIter; pcmk_node_t *local_node = NULL; /* make sure it doesn't come up again */ if (rsc->allowed_nodes != NULL) { g_hash_table_destroy(rsc->allowed_nodes); } rsc->allowed_nodes = pe__node_list2table(scheduler->nodes); g_hash_table_iter_init(&gIter, rsc->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&local_node)) { - local_node->weight = -INFINITY; + local_node->weight = -PCMK_SCORE_INFINITY; } } break; case pcmk_multiply_active_block: pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed); pcmk__set_rsc_flags(rsc, pcmk_rsc_blocked); /* If the resource belongs to a group or bundle configured with * PCMK_META_MULTIPLE_ACTIVE=PCMK_VALUE_BLOCK, block the entire * entity. */ if (rsc->parent && ((rsc->parent->variant == pcmk_rsc_variant_group) || (rsc->parent->variant == pcmk_rsc_variant_bundle)) && (rsc->parent->recovery_type == pcmk_multiply_active_block)) { GList *gIter = rsc->parent->children; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = gIter->data; pcmk__clear_rsc_flags(child, pcmk_rsc_managed); pcmk__set_rsc_flags(child, pcmk_rsc_blocked); } } break; // pcmk_multiply_active_restart, pcmk_multiply_active_unexpected default: /* The scheduler will do the right thing because the relevant * variables and flags are set when unpacking the history. */ break; } crm_debug("%s is active on multiple nodes including %s: %s", rsc->id, pcmk__node_name(node), pcmk_multiply_active_text(rsc->recovery_type)); } else { pcmk__rsc_trace(rsc, "Resource %s is active on %s", rsc->id, pcmk__node_name(node)); } if (rsc->parent != NULL) { native_add_running(rsc->parent, node, scheduler, FALSE); } } static void recursive_clear_unique(pcmk_resource_t *rsc, gpointer user_data) { pcmk__clear_rsc_flags(rsc, pcmk_rsc_unique); pcmk__insert_meta(rsc, PCMK_META_GLOBALLY_UNIQUE, PCMK_VALUE_FALSE); g_list_foreach(rsc->children, (GFunc) recursive_clear_unique, NULL); } gboolean native_unpack(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { pcmk_resource_t *parent = uber_parent(rsc); const char *standard = crm_element_value(rsc->xml, PCMK_XA_CLASS); uint32_t ra_caps = pcmk_get_ra_caps(standard); pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id); // Only some agent standards support unique and promotable clones if (!pcmk_is_set(ra_caps, pcmk_ra_cap_unique) && pcmk_is_set(rsc->flags, pcmk_rsc_unique) && pcmk__is_clone(parent)) { /* @COMPAT We should probably reject this situation as an error (as we * do for promotable below) rather than warn and convert, but that would * be a backward-incompatible change that we should probably do with a * transform at a schema major version bump. */ pe__force_anon(standard, parent, rsc->id, scheduler); /* Clear PCMK_META_GLOBALLY_UNIQUE on the parent and all its descendants * unpacked so far (clearing the parent should make any future children * unpacking correct). We have to clear this resource explicitly because * it isn't hooked into the parent's children yet. */ recursive_clear_unique(parent, NULL); recursive_clear_unique(rsc, NULL); } if (!pcmk_is_set(ra_caps, pcmk_ra_cap_promotable) && pcmk_is_set(parent->flags, pcmk_rsc_promotable)) { pcmk__config_err("Resource %s is of type %s and therefore " "cannot be used as a promotable clone resource", rsc->id, standard); return FALSE; } return TRUE; } static bool rsc_is_on_node(pcmk_resource_t *rsc, const pcmk_node_t *node, int flags) { pcmk__rsc_trace(rsc, "Checking whether %s is on %s", rsc->id, pcmk__node_name(node)); if (pcmk_is_set(flags, pcmk_rsc_match_current_node) && (rsc->running_on != NULL)) { for (GList *iter = rsc->running_on; iter; iter = iter->next) { if (pcmk__same_node((pcmk_node_t *) iter->data, node)) { return true; } } } else if (pcmk_is_set(flags, pe_find_inactive) // @COMPAT deprecated && (rsc->running_on == NULL)) { return true; } else if (!pcmk_is_set(flags, pcmk_rsc_match_current_node) && (rsc->allocated_to != NULL) && pcmk__same_node(rsc->allocated_to, node)) { return true; } return false; } pcmk_resource_t * native_find_rsc(pcmk_resource_t *rsc, const char *id, const pcmk_node_t *on_node, int flags) { bool match = false; pcmk_resource_t *result = NULL; CRM_CHECK(id && rsc && rsc->id, return NULL); if (pcmk_is_set(flags, pcmk_rsc_match_clone_only)) { const char *rid = pcmk__xe_id(rsc->xml); if (!pcmk__is_clone(pe__const_top_resource(rsc, false))) { match = false; } else if (!strcmp(id, rsc->id) || pcmk__str_eq(id, rid, pcmk__str_none)) { match = true; } } else if (!strcmp(id, rsc->id)) { match = true; } else if (pcmk_is_set(flags, pcmk_rsc_match_history) && rsc->clone_name && strcmp(rsc->clone_name, id) == 0) { match = true; } else if (pcmk_is_set(flags, pcmk_rsc_match_basename) || (pcmk_is_set(flags, pcmk_rsc_match_anon_basename) && !pcmk_is_set(rsc->flags, pcmk_rsc_unique))) { match = pe_base_name_eq(rsc, id); } if (match && on_node) { if (!rsc_is_on_node(rsc, on_node, flags)) { match = false; } } if (match) { return rsc; } for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = (pcmk_resource_t *) gIter->data; result = rsc->fns->find_rsc(child, id, on_node, flags); if (result) { return result; } } return NULL; } // create is ignored char * native_parameter(pcmk_resource_t *rsc, pcmk_node_t *node, gboolean create, const char *name, pcmk_scheduler_t *scheduler) { char *value_copy = NULL; const char *value = NULL; GHashTable *params = NULL; CRM_CHECK(rsc != NULL, return NULL); CRM_CHECK(name != NULL && strlen(name) != 0, return NULL); pcmk__rsc_trace(rsc, "Looking up %s in %s", name, rsc->id); params = pe_rsc_params(rsc, node, scheduler); value = g_hash_table_lookup(params, name); if (value == NULL) { /* try meta attributes instead */ value = g_hash_table_lookup(rsc->meta, name); } pcmk__str_update(&value_copy, value); return value_copy; } gboolean native_active(pcmk_resource_t * rsc, gboolean all) { for (GList *gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) { pcmk_node_t *a_node = (pcmk_node_t *) gIter->data; if (a_node->details->unclean) { pcmk__rsc_trace(rsc, "Resource %s: %s is unclean", rsc->id, pcmk__node_name(a_node)); return TRUE; } else if (!a_node->details->online && pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk__rsc_trace(rsc, "Resource %s: %s is offline", rsc->id, pcmk__node_name(a_node)); } else { pcmk__rsc_trace(rsc, "Resource %s active on %s", rsc->id, pcmk__node_name(a_node)); return TRUE; } } return FALSE; } struct print_data_s { long options; void *print_data; }; static const char * native_pending_state(const pcmk_resource_t *rsc) { const char *pending_state = NULL; if (pcmk__str_eq(rsc->pending_task, PCMK_ACTION_START, pcmk__str_casei)) { pending_state = "Starting"; } else if (pcmk__str_eq(rsc->pending_task, PCMK_ACTION_STOP, pcmk__str_casei)) { pending_state = "Stopping"; } else if (pcmk__str_eq(rsc->pending_task, PCMK_ACTION_MIGRATE_TO, pcmk__str_casei)) { pending_state = "Migrating"; } else if (pcmk__str_eq(rsc->pending_task, PCMK_ACTION_MIGRATE_FROM, pcmk__str_casei)) { /* Work might be done in here. */ pending_state = "Migrating"; } else if (pcmk__str_eq(rsc->pending_task, PCMK_ACTION_PROMOTE, pcmk__str_casei)) { pending_state = "Promoting"; } else if (pcmk__str_eq(rsc->pending_task, PCMK_ACTION_DEMOTE, pcmk__str_casei)) { pending_state = "Demoting"; } return pending_state; } static const char * native_pending_task(const pcmk_resource_t *rsc) { const char *pending_task = NULL; if (pcmk__str_eq(rsc->pending_task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { pending_task = "Monitoring"; /* Pending probes are not printed, even if pending * operations are requested. If someone ever requests that * behavior, uncomment this and the corresponding part of * unpack.c:unpack_rsc_op(). */ /* } else if (pcmk__str_eq(rsc->pending_task, "probe", pcmk__str_casei)) { pending_task = "Checking"; */ } return pending_task; } static enum rsc_role_e native_displayable_role(const pcmk_resource_t *rsc) { enum rsc_role_e role = rsc->role; if ((role == pcmk_role_started) && pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk_rsc_promotable)) { role = pcmk_role_unpromoted; } return role; } static const char * native_displayable_state(const pcmk_resource_t *rsc, bool print_pending) { const char *rsc_state = NULL; if (print_pending) { rsc_state = native_pending_state(rsc); } if (rsc_state == NULL) { rsc_state = pcmk_role_text(native_displayable_role(rsc)); } return rsc_state; } /*! * \internal * \deprecated This function will be removed in a future release */ static void native_print_xml(pcmk_resource_t *rsc, const char *pre_text, long options, void *print_data) { const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS); const char *prov = crm_element_value(rsc->xml, PCMK_XA_PROVIDER); const char *rsc_state = native_displayable_state(rsc, pcmk_is_set(options, pe_print_pending)); const char *target_role = NULL; /* resource information. */ status_print("%sxml, PCMK_XA_TYPE)); status_print("role=\"%s\" ", rsc_state); if (rsc->meta) { target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); } if (target_role) { status_print("target_role=\"%s\" ", target_role); } status_print("active=\"%s\" ", pcmk__btoa(rsc->fns->active(rsc, TRUE))); status_print("orphaned=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_removed)); status_print("blocked=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_blocked)); status_print("managed=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_managed)); status_print("failed=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_failed)); status_print("failure_ignored=\"%s\" ", pcmk__flag_text(rsc->flags, pcmk_rsc_ignore_failure)); status_print("nodes_running_on=\"%d\" ", g_list_length(rsc->running_on)); if (options & pe_print_pending) { const char *pending_task = native_pending_task(rsc); if (pending_task) { status_print("pending=\"%s\" ", pending_task); } } /* print out the nodes this resource is running on */ if (options & pe_print_rsconly) { status_print("/>\n"); /* do nothing */ } else if (rsc->running_on != NULL) { GList *gIter = rsc->running_on; status_print(">\n"); for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = (pcmk_node_t *) gIter->data; status_print("%s \n", pre_text, pcmk__s(node->details->uname, ""), node->details->id, pcmk__btoa(!node->details->online)); } status_print("%s\n", pre_text); } else { status_print("/>\n"); } } // Append a flag to resource description string's flags list static bool add_output_flag(GString *s, const char *flag_desc, bool have_flags) { g_string_append(s, (have_flags? ", " : " (")); g_string_append(s, flag_desc); return true; } // Append a node name to resource description string's node list static bool add_output_node(GString *s, const char *node, bool have_nodes) { g_string_append(s, (have_nodes? " " : " [ ")); g_string_append(s, node); return true; } /*! * \internal * \brief Create a string description of a resource * * \param[in] rsc Resource to describe * \param[in] name Desired identifier for the resource * \param[in] node If not NULL, node that resource is "on" * \param[in] show_opts Bitmask of pcmk_show_opt_e. * \param[in] target_role Resource's target role * \param[in] show_nodes Whether to display nodes when multiply active * * \return Newly allocated string description of resource * \note Caller must free the result with g_free(). */ gchar * pcmk__native_output_string(const pcmk_resource_t *rsc, const char *name, const pcmk_node_t *node, uint32_t show_opts, const char *target_role, bool show_nodes) { const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS); const char *provider = NULL; const char *kind = crm_element_value(rsc->xml, PCMK_XA_TYPE); GString *outstr = NULL; bool have_flags = false; if (rsc->variant != pcmk_rsc_variant_primitive) { return NULL; } CRM_CHECK(name != NULL, name = "unknown"); CRM_CHECK(kind != NULL, kind = "unknown"); CRM_CHECK(class != NULL, class = "unknown"); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) { provider = crm_element_value(rsc->xml, PCMK_XA_PROVIDER); } if ((node == NULL) && (rsc->lock_node != NULL)) { node = rsc->lock_node; } if (pcmk_any_flags_set(show_opts, pcmk_show_rsc_only) || pcmk__list_of_multiple(rsc->running_on)) { node = NULL; } outstr = g_string_sized_new(128); // Resource name and agent pcmk__g_strcat(outstr, name, "\t(", class, ((provider == NULL)? "" : PROVIDER_SEP), pcmk__s(provider, ""), ":", kind, "):\t", NULL); // State on node if (pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { g_string_append(outstr, " ORPHANED"); } if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { enum rsc_role_e role = native_displayable_role(rsc); g_string_append(outstr, " FAILED"); if (role > pcmk_role_unpromoted) { pcmk__add_word(&outstr, 0, pcmk_role_text(role)); } } else { bool show_pending = pcmk_is_set(show_opts, pcmk_show_pending); pcmk__add_word(&outstr, 0, native_displayable_state(rsc, show_pending)); } if (node) { pcmk__add_word(&outstr, 0, pcmk__node_name(node)); } // Failed probe operation if (native_displayable_role(rsc) == pcmk_role_stopped) { xmlNode *probe_op = pe__failed_probe_for_rsc(rsc, node ? node->details->uname : NULL); if (probe_op != NULL) { int rc; pcmk__scan_min_int(crm_element_value(probe_op, PCMK__XA_RC_CODE), &rc, 0); pcmk__g_strcat(outstr, " (", services_ocf_exitcode_str(rc), ") ", NULL); } } // Flags, as: ( [...]) if (node && !(node->details->online) && node->details->unclean) { have_flags = add_output_flag(outstr, "UNCLEAN", have_flags); } if (node && (node == rsc->lock_node)) { have_flags = add_output_flag(outstr, "LOCKED", have_flags); } if (pcmk_is_set(show_opts, pcmk_show_pending)) { const char *pending_task = native_pending_task(rsc); if (pending_task) { have_flags = add_output_flag(outstr, pending_task, have_flags); } } if (target_role != NULL) { switch (pcmk_parse_role(target_role)) { case pcmk_role_unknown: pcmk__config_err("Invalid " PCMK_META_TARGET_ROLE " %s for resource %s", target_role, rsc->id); break; case pcmk_role_stopped: have_flags = add_output_flag(outstr, "disabled", have_flags); break; case pcmk_role_unpromoted: if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk_rsc_promotable)) { have_flags = add_output_flag(outstr, PCMK_META_TARGET_ROLE ":", have_flags); g_string_append(outstr, target_role); } break; default: /* Only show target role if it limits our abilities (i.e. ignore * Started, as it is the default anyways, and doesn't prevent * the resource from becoming promoted). */ break; } } // Blocked or maintenance implies unmanaged if (pcmk_any_flags_set(rsc->flags, pcmk_rsc_blocked|pcmk_rsc_maintenance)) { if (pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) { have_flags = add_output_flag(outstr, "blocked", have_flags); } else if (pcmk_is_set(rsc->flags, pcmk_rsc_maintenance)) { have_flags = add_output_flag(outstr, "maintenance", have_flags); } } else if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { have_flags = add_output_flag(outstr, "unmanaged", have_flags); } if (pcmk_is_set(rsc->flags, pcmk_rsc_ignore_failure)) { have_flags = add_output_flag(outstr, "failure ignored", have_flags); } if (have_flags) { g_string_append_c(outstr, ')'); } // User-supplied description if (pcmk_any_flags_set(show_opts, pcmk_show_rsc_only|pcmk_show_description) || pcmk__list_of_multiple(rsc->running_on)) { const char *desc = crm_element_value(rsc->xml, PCMK_XA_DESCRIPTION); if (desc) { g_string_append(outstr, " ("); g_string_append(outstr, desc); g_string_append(outstr, ")"); } } if (show_nodes && !pcmk_is_set(show_opts, pcmk_show_rsc_only) && pcmk__list_of_multiple(rsc->running_on)) { bool have_nodes = false; for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *n = (pcmk_node_t *) iter->data; have_nodes = add_output_node(outstr, n->details->uname, have_nodes); } if (have_nodes) { g_string_append(outstr, " ]"); } } return g_string_free(outstr, FALSE); } int pe__common_output_html(pcmk__output_t *out, const pcmk_resource_t *rsc, const char *name, const pcmk_node_t *node, uint32_t show_opts) { const char *kind = crm_element_value(rsc->xml, PCMK_XA_TYPE); const char *target_role = NULL; xmlNodePtr list_node = NULL; const char *cl = NULL; CRM_ASSERT(rsc->variant == pcmk_rsc_variant_primitive); CRM_ASSERT(kind != NULL); if (rsc->meta) { const char *is_internal = g_hash_table_lookup(rsc->meta, PCMK__META_INTERNAL_RSC); if (crm_is_true(is_internal) && !pcmk_is_set(show_opts, pcmk_show_implicit_rscs)) { crm_trace("skipping print of internal resource %s", rsc->id); return pcmk_rc_no_output; } target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); } if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { cl = PCMK__VALUE_RSC_MANAGED; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { cl = PCMK__VALUE_RSC_FAILED; } else if ((rsc->variant == pcmk_rsc_variant_primitive) && (rsc->running_on == NULL)) { cl = PCMK__VALUE_RSC_FAILED; } else if (pcmk__list_of_multiple(rsc->running_on)) { cl = PCMK__VALUE_RSC_MULTIPLE; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_ignore_failure)) { cl = PCMK__VALUE_RSC_FAILURE_IGNORED; } else { cl = PCMK__VALUE_RSC_OK; } { gchar *s = pcmk__native_output_string(rsc, name, node, show_opts, target_role, true); list_node = pcmk__output_create_html_node(out, "li", NULL, NULL, NULL); pcmk_create_html_node(list_node, PCMK__XE_SPAN, NULL, cl, s); g_free(s); } return pcmk_rc_ok; } int pe__common_output_text(pcmk__output_t *out, const pcmk_resource_t *rsc, const char *name, const pcmk_node_t *node, uint32_t show_opts) { const char *target_role = NULL; CRM_ASSERT(rsc->variant == pcmk_rsc_variant_primitive); if (rsc->meta) { const char *is_internal = g_hash_table_lookup(rsc->meta, PCMK__META_INTERNAL_RSC); if (crm_is_true(is_internal) && !pcmk_is_set(show_opts, pcmk_show_implicit_rscs)) { crm_trace("skipping print of internal resource %s", rsc->id); return pcmk_rc_no_output; } target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); } { gchar *s = pcmk__native_output_string(rsc, name, node, show_opts, target_role, true); out->list_item(out, NULL, "%s", s); g_free(s); } return pcmk_rc_ok; } /*! * \internal * \deprecated This function will be removed in a future release */ void common_print(pcmk_resource_t *rsc, const char *pre_text, const char *name, const pcmk_node_t *node, long options, void *print_data) { const char *target_role = NULL; CRM_ASSERT(rsc->variant == pcmk_rsc_variant_primitive); if (rsc->meta) { const char *is_internal = g_hash_table_lookup(rsc->meta, PCMK__META_INTERNAL_RSC); if (crm_is_true(is_internal) && !pcmk_is_set(options, pe_print_implicit)) { crm_trace("skipping print of internal resource %s", rsc->id); return; } target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); } if (options & pe_print_xml) { native_print_xml(rsc, pre_text, options, print_data); return; } if ((pre_text == NULL) && (options & pe_print_printf)) { pre_text = " "; } if (options & pe_print_html) { if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { status_print(""); } else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { status_print(""); } else if (rsc->running_on == NULL) { status_print(""); } else if (pcmk__list_of_multiple(rsc->running_on)) { status_print(""); } else if (pcmk_is_set(rsc->flags, pcmk_rsc_ignore_failure)) { status_print(""); } else { status_print(""); } } { gchar *resource_s = pcmk__native_output_string(rsc, name, node, options, target_role, false); status_print("%s%s", (pre_text? pre_text : ""), resource_s); g_free(resource_s); } if (pcmk_is_set(options, pe_print_html)) { status_print(" "); } if (!pcmk_is_set(options, pe_print_rsconly) && pcmk__list_of_multiple(rsc->running_on)) { GList *gIter = rsc->running_on; int counter = 0; if (options & pe_print_html) { status_print("
      \n"); } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print("["); } for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *n = (pcmk_node_t *) gIter->data; counter++; if (options & pe_print_html) { status_print("
    • \n%s", pcmk__node_name(n)); } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print(" %s", pcmk__node_name(n)); } else if ((options & pe_print_log)) { status_print("\t%d : %s", counter, pcmk__node_name(n)); } else { status_print("%s", pcmk__node_name(n)); } if (options & pe_print_html) { status_print("
      • \n"); } } if (options & pe_print_html) { status_print("
    \n"); } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print(" ]"); } } if (options & pe_print_html) { status_print("
    \n"); } else if (options & pe_print_suppres_nl) { /* nothing */ } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print("\n"); } } /*! * \internal * \deprecated This function will be removed in a future release */ void native_print(pcmk_resource_t *rsc, const char *pre_text, long options, void *print_data) { const pcmk_node_t *node = NULL; CRM_ASSERT(rsc->variant == pcmk_rsc_variant_primitive); if (options & pe_print_xml) { native_print_xml(rsc, pre_text, options, print_data); return; } node = pcmk__current_node(rsc); if (node == NULL) { // This is set only if a non-probe action is pending on this node node = rsc->pending_node; } common_print(rsc, pre_text, rsc_printable_id(rsc), node, options, print_data); } PCMK__OUTPUT_ARGS("primitive", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__resource_xml(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); int rc = pcmk_rc_no_output; bool print_pending = pcmk_is_set(show_opts, pcmk_show_pending); const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS); const char *prov = crm_element_value(rsc->xml, PCMK_XA_PROVIDER); char ra_name[LINE_MAX]; const char *rsc_state = native_displayable_state(rsc, print_pending); const char *target_role = NULL; const char *active = pcmk__btoa(rsc->fns->active(rsc, TRUE)); const char *orphaned = pcmk__flag_text(rsc->flags, pcmk_rsc_removed); const char *blocked = pcmk__flag_text(rsc->flags, pcmk_rsc_blocked); const char *maintenance = pcmk__flag_text(rsc->flags, pcmk_rsc_maintenance); const char *managed = pcmk__flag_text(rsc->flags, pcmk_rsc_managed); const char *failed = pcmk__flag_text(rsc->flags, pcmk_rsc_failed); const char *ignored = pcmk__flag_text(rsc->flags, pcmk_rsc_ignore_failure); char *nodes_running_on = NULL; const char *pending = print_pending? native_pending_task(rsc) : NULL; const char *locked_to = NULL; const char *desc = pe__resource_description(rsc, show_opts); CRM_ASSERT(rsc->variant == pcmk_rsc_variant_primitive); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } // Resource information snprintf(ra_name, LINE_MAX, "%s%s%s:%s", class, ((prov == NULL)? "" : PROVIDER_SEP), ((prov == NULL)? "" : prov), crm_element_value(rsc->xml, PCMK_XA_TYPE)); if (rsc->meta != NULL) { target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); } nodes_running_on = pcmk__itoa(g_list_length(rsc->running_on)); if (rsc->lock_node != NULL) { locked_to = rsc->lock_node->details->uname; } rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_RESOURCE, 15, PCMK_XA_ID, rsc_printable_id(rsc), PCMK_XA_RESOURCE_AGENT, ra_name, PCMK_XA_ROLE, rsc_state, PCMK_XA_TARGET_ROLE, target_role, PCMK_XA_ACTIVE, active, PCMK_XA_ORPHANED, orphaned, PCMK_XA_BLOCKED, blocked, PCMK_XA_MAINTENANCE, maintenance, PCMK_XA_MANAGED, managed, PCMK_XA_FAILED, failed, PCMK_XA_FAILURE_IGNORED, ignored, PCMK_XA_NODES_RUNNING_ON, nodes_running_on, PCMK_XA_PENDING, pending, PCMK_XA_LOCKED_TO, locked_to, PCMK_XA_DESCRIPTION, desc); free(nodes_running_on); CRM_ASSERT(rc == pcmk_rc_ok); if (rsc->running_on != NULL) { GList *gIter = rsc->running_on; for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = (pcmk_node_t *) gIter->data; const char *cached = pcmk__btoa(node->details->online); rc = pe__name_and_nvpairs_xml(out, false, PCMK_XE_NODE, 3, PCMK_XA_NAME, node->details->uname, PCMK_XA_ID, node->details->id, PCMK_XA_CACHED, cached); CRM_ASSERT(rc == pcmk_rc_ok); } } pcmk__output_xml_pop_parent(out); return rc; } PCMK__OUTPUT_ARGS("primitive", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__resource_html(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const pcmk_node_t *node = pcmk__current_node(rsc); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } CRM_ASSERT(rsc->variant == pcmk_rsc_variant_primitive); if (node == NULL) { // This is set only if a non-probe action is pending on this node node = rsc->pending_node; } return pe__common_output_html(out, rsc, rsc_printable_id(rsc), node, show_opts); } PCMK__OUTPUT_ARGS("primitive", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__resource_text(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const pcmk_node_t *node = pcmk__current_node(rsc); CRM_ASSERT(rsc->variant == pcmk_rsc_variant_primitive); if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } if (node == NULL) { // This is set only if a non-probe action is pending on this node node = rsc->pending_node; } return pe__common_output_text(out, rsc, rsc_printable_id(rsc), node, show_opts); } void native_free(pcmk_resource_t * rsc) { pcmk__rsc_trace(rsc, "Freeing resource action list (not the data)"); common_free(rsc); } enum rsc_role_e native_resource_state(const pcmk_resource_t * rsc, gboolean current) { enum rsc_role_e role = rsc->next_role; if (current) { role = rsc->role; } pcmk__rsc_trace(rsc, "%s state: %s", rsc->id, pcmk_role_text(role)); return role; } /*! * \internal * \brief List nodes where a resource (or any of its children) is * * \param[in] rsc Resource to check * \param[out] list List to add result to * \param[in] current 0 = where allocated, 1 = where running, * 2 = where running or pending * * \return If list contains only one node, that node, or NULL otherwise */ pcmk_node_t * native_location(const pcmk_resource_t *rsc, GList **list, int current) { // @COMPAT: Accept a pcmk__rsc_node argument instead of int current pcmk_node_t *one = NULL; GList *result = NULL; if (rsc->children) { GList *gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = (pcmk_resource_t *) gIter->data; child->fns->location(child, &result, current); } } else if (current) { if (rsc->running_on) { result = g_list_copy(rsc->running_on); } if ((current == 2) && rsc->pending_node && !pe_find_node_id(result, rsc->pending_node->details->id)) { result = g_list_append(result, rsc->pending_node); } } else if (current == FALSE && rsc->allocated_to) { result = g_list_append(NULL, rsc->allocated_to); } if (result && (result->next == NULL)) { one = result->data; } if (list) { GList *gIter = result; for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = (pcmk_node_t *) gIter->data; if (*list == NULL || pe_find_node_id(*list, node->details->id) == NULL) { *list = g_list_append(*list, node); } } } g_list_free(result); return one; } static void get_rscs_brief(GList *rsc_list, GHashTable * rsc_table, GHashTable * active_table) { GList *gIter = rsc_list; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS); const char *kind = crm_element_value(rsc->xml, PCMK_XA_TYPE); int offset = 0; char buffer[LINE_MAX]; int *rsc_counter = NULL; int *active_counter = NULL; if (rsc->variant != pcmk_rsc_variant_primitive) { continue; } offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", class); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) { const char *prov = crm_element_value(rsc->xml, PCMK_XA_PROVIDER); if (prov != NULL) { offset += snprintf(buffer + offset, LINE_MAX - offset, PROVIDER_SEP "%s", prov); } } offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s", kind); CRM_LOG_ASSERT(offset > 0); if (rsc_table) { rsc_counter = g_hash_table_lookup(rsc_table, buffer); if (rsc_counter == NULL) { rsc_counter = calloc(1, sizeof(int)); *rsc_counter = 0; g_hash_table_insert(rsc_table, strdup(buffer), rsc_counter); } (*rsc_counter)++; } if (active_table) { GList *gIter2 = rsc->running_on; for (; gIter2 != NULL; gIter2 = gIter2->next) { pcmk_node_t *node = (pcmk_node_t *) gIter2->data; GHashTable *node_table = NULL; if (node->details->unclean == FALSE && node->details->online == FALSE && pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { continue; } node_table = g_hash_table_lookup(active_table, node->details->uname); if (node_table == NULL) { node_table = pcmk__strkey_table(free, free); g_hash_table_insert(active_table, strdup(node->details->uname), node_table); } active_counter = g_hash_table_lookup(node_table, buffer); if (active_counter == NULL) { active_counter = calloc(1, sizeof(int)); *active_counter = 0; g_hash_table_insert(node_table, strdup(buffer), active_counter); } (*active_counter)++; } } } } static void destroy_node_table(gpointer data) { GHashTable *node_table = data; if (node_table) { g_hash_table_destroy(node_table); } } /*! * \internal * \deprecated This function will be removed in a future release */ void print_rscs_brief(GList *rsc_list, const char *pre_text, long options, void *print_data, gboolean print_all) { GHashTable *rsc_table = pcmk__strkey_table(free, free); GHashTable *active_table = pcmk__strkey_table(free, destroy_node_table); GHashTableIter hash_iter; char *type = NULL; int *rsc_counter = NULL; get_rscs_brief(rsc_list, rsc_table, active_table); g_hash_table_iter_init(&hash_iter, rsc_table); while (g_hash_table_iter_next(&hash_iter, (gpointer *)&type, (gpointer *)&rsc_counter)) { GHashTableIter hash_iter2; char *node_name = NULL; GHashTable *node_table = NULL; int active_counter_all = 0; g_hash_table_iter_init(&hash_iter2, active_table); while (g_hash_table_iter_next(&hash_iter2, (gpointer *)&node_name, (gpointer *)&node_table)) { int *active_counter = g_hash_table_lookup(node_table, type); if (active_counter == NULL || *active_counter == 0) { continue; } else { active_counter_all += *active_counter; } if (options & pe_print_rsconly) { node_name = NULL; } if (options & pe_print_html) { status_print("
  • \n"); } if (print_all) { status_print("%s%d/%d\t(%s):\tActive %s\n", pre_text ? pre_text : "", active_counter ? *active_counter : 0, rsc_counter ? *rsc_counter : 0, type, active_counter && (*active_counter > 0) && node_name ? node_name : ""); } else { status_print("%s%d\t(%s):\tActive %s\n", pre_text ? pre_text : "", active_counter ? *active_counter : 0, type, active_counter && (*active_counter > 0) && node_name ? node_name : ""); } if (options & pe_print_html) { status_print("
    • \n"); } } if (print_all && active_counter_all == 0) { if (options & pe_print_html) { status_print("
  • \n"); } status_print("%s%d/%d\t(%s):\tActive\n", pre_text ? pre_text : "", active_counter_all, rsc_counter ? *rsc_counter : 0, type); if (options & pe_print_html) { status_print("
    • \n"); } } } if (rsc_table) { g_hash_table_destroy(rsc_table); rsc_table = NULL; } if (active_table) { g_hash_table_destroy(active_table); active_table = NULL; } } int pe__rscs_brief_output(pcmk__output_t *out, GList *rsc_list, uint32_t show_opts) { GHashTable *rsc_table = pcmk__strkey_table(free, free); GHashTable *active_table = pcmk__strkey_table(free, destroy_node_table); GList *sorted_rscs; int rc = pcmk_rc_no_output; get_rscs_brief(rsc_list, rsc_table, active_table); /* Make a list of the rsc_table keys so that it can be sorted. This is to make sure * output order stays consistent between systems. */ sorted_rscs = g_hash_table_get_keys(rsc_table); sorted_rscs = g_list_sort(sorted_rscs, (GCompareFunc) strcmp); for (GList *gIter = sorted_rscs; gIter; gIter = gIter->next) { char *type = (char *) gIter->data; int *rsc_counter = g_hash_table_lookup(rsc_table, type); GList *sorted_nodes = NULL; int active_counter_all = 0; /* Also make a list of the active_table keys so it can be sorted. If there's * more than one instance of a type of resource running, we need the nodes to * be sorted to make sure output order stays consistent between systems. */ sorted_nodes = g_hash_table_get_keys(active_table); sorted_nodes = g_list_sort(sorted_nodes, (GCompareFunc) pcmk__numeric_strcasecmp); for (GList *gIter2 = sorted_nodes; gIter2; gIter2 = gIter2->next) { char *node_name = (char *) gIter2->data; GHashTable *node_table = g_hash_table_lookup(active_table, node_name); int *active_counter = NULL; if (node_table == NULL) { continue; } active_counter = g_hash_table_lookup(node_table, type); if (active_counter == NULL || *active_counter == 0) { continue; } else { active_counter_all += *active_counter; } if (pcmk_is_set(show_opts, pcmk_show_rsc_only)) { node_name = NULL; } if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { out->list_item(out, NULL, "%d/%d\t(%s):\tActive %s", *active_counter, rsc_counter ? *rsc_counter : 0, type, (*active_counter > 0) && node_name ? node_name : ""); } else { out->list_item(out, NULL, "%d\t(%s):\tActive %s", *active_counter, type, (*active_counter > 0) && node_name ? node_name : ""); } rc = pcmk_rc_ok; } if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs) && active_counter_all == 0) { out->list_item(out, NULL, "%d/%d\t(%s):\tActive", active_counter_all, rsc_counter ? *rsc_counter : 0, type); rc = pcmk_rc_ok; } if (sorted_nodes) { g_list_free(sorted_nodes); } } if (rsc_table) { g_hash_table_destroy(rsc_table); rsc_table = NULL; } if (active_table) { g_hash_table_destroy(active_table); active_table = NULL; } if (sorted_rscs) { g_list_free(sorted_rscs); } return rc; } gboolean pe__native_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent) { if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) || pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches)) { return FALSE; } else if (check_parent && rsc->parent) { const pcmk_resource_t *up = pe__const_top_resource(rsc, true); return up->fns->is_filtered(up, only_rsc, FALSE); } return TRUE; } /*! * \internal * \brief Get maximum primitive resource instances per node * * \param[in] rsc Primitive resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int pe__primitive_max_per_node(const pcmk_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); return 1U; } diff --git a/lib/pengine/pe_health.c b/lib/pengine/pe_health.c index be1f48fb24..4f7eb10b5e 100644 --- a/lib/pengine/pe_health.c +++ b/lib/pengine/pe_health.c @@ -1,158 +1,158 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include "pe_status_private.h" /*! * \internal * \brief Set the node health values to use for \c PCMK_VALUE_RED, * \c PCMK_VALUE_YELLOW, and \c PCMK_VALUE_GREEN * * \param[in,out] scheduler Scheduler data */ void pe__unpack_node_health_scores(pcmk_scheduler_t *scheduler) { switch (pe__health_strategy(scheduler)) { case pcmk__health_strategy_none: pcmk__score_red = 0; pcmk__score_yellow = 0; pcmk__score_green = 0; break; case pcmk__health_strategy_no_red: - pcmk__score_red = -INFINITY; + pcmk__score_red = -PCMK_SCORE_INFINITY; pcmk__score_yellow = 0; pcmk__score_green = 0; break; case pcmk__health_strategy_only_green: - pcmk__score_red = -INFINITY; - pcmk__score_yellow = -INFINITY; + pcmk__score_red = -PCMK_SCORE_INFINITY; + pcmk__score_yellow = -PCMK_SCORE_INFINITY; pcmk__score_green = 0; break; default: // progressive or custom pcmk__score_red = pe__health_score(PCMK_OPT_NODE_HEALTH_RED, scheduler); pcmk__score_green = pe__health_score(PCMK_OPT_NODE_HEALTH_GREEN, scheduler); pcmk__score_yellow = pe__health_score(PCMK_OPT_NODE_HEALTH_YELLOW, scheduler); break; } if ((pcmk__score_red != 0) || (pcmk__score_yellow != 0) || (pcmk__score_green != 0)) { crm_debug("Values of node health scores: " PCMK_VALUE_RED "=%d " PCMK_VALUE_YELLOW "=%d " PCMK_VALUE_GREEN "=%d", pcmk__score_red, pcmk__score_yellow, pcmk__score_green); } } /*! * \internal * \brief Add node attribute value to an integer, if it is a health attribute * * \param[in] key Name of node attribute * \param[in] value String value of node attribute * \param[in,out] user_data Address of integer to which \p value should be * added if \p key is a node health attribute */ static void add_node_health_value(gpointer key, gpointer value, gpointer user_data) { if (pcmk__starts_with((const char *) key, "#health")) { int score = char2score((const char *) value); int *health = (int *) user_data; *health = pcmk__add_scores(score, *health); crm_trace("Combined '%s' into node health score (now %s)", (const char *) value, pcmk_readable_score(*health)); } } /*! * \internal * \brief Sum a node's health attribute scores * * \param[in] node Node whose health attributes should be added * \param[in] base_health Add this number to the total * * \return Sum of all health attribute scores of \p node plus \p base_health */ int pe__sum_node_health_scores(const pcmk_node_t *node, int base_health) { CRM_ASSERT(node != NULL); g_hash_table_foreach(node->details->attrs, add_node_health_value, &base_health); return base_health; } /*! * \internal * \brief Check the general health status for a node * * \param[in,out] node Node to check * * \return A negative value if any health attribute for \p node is red, * otherwise 0 if any attribute is yellow, otherwise a positive value. */ int pe__node_health(pcmk_node_t *node) { GHashTableIter iter; const char *name = NULL; const char *value = NULL; enum pcmk__health_strategy strategy; int score = 0; int rc = 1; CRM_ASSERT(node != NULL); strategy = pe__health_strategy(node->details->data_set); if (strategy == pcmk__health_strategy_none) { return rc; } g_hash_table_iter_init(&iter, node->details->attrs); while (g_hash_table_iter_next(&iter, (gpointer *) &name, (gpointer *) &value)) { if (pcmk__starts_with(name, "#health")) { /* It's possible that pcmk__score_red equals pcmk__score_yellow, * or pcmk__score_yellow equals pcmk__score_green, so check the * textual value first to be able to distinguish those. */ if (pcmk__str_eq(value, PCMK_VALUE_RED, pcmk__str_casei)) { return -1; } else if (pcmk__str_eq(value, PCMK_VALUE_YELLOW, pcmk__str_casei)) { rc = 0; continue; } // The value is an integer, so compare numerically score = char2score(value); if (score <= pcmk__score_red) { return -1; } else if ((score <= pcmk__score_yellow) && (pcmk__score_yellow != pcmk__score_green)) { rc = 0; } } } return rc; } diff --git a/lib/pengine/pe_notif.c b/lib/pengine/pe_notif.c index 8e5ac34bcb..549d49f8ce 100644 --- a/lib/pengine/pe_notif.c +++ b/lib/pengine/pe_notif.c @@ -1,1007 +1,1007 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "pe_status_private.h" typedef struct notify_entry_s { const pcmk_resource_t *rsc; const pcmk_node_t *node; } notify_entry_t; /*! * \internal * \brief Compare two notification entries * * Compare two notification entries, where the one with the alphabetically first * resource name (or if equal, node name) sorts as first, with NULL sorting as * less than non-NULL. * * \param[in] a First notification entry to compare * \param[in] b Second notification entry to compare * * \return -1 if \p a sorts before \p b, 0 if they are equal, otherwise 1 */ static gint compare_notify_entries(gconstpointer a, gconstpointer b) { int tmp; const notify_entry_t *entry_a = a; const notify_entry_t *entry_b = b; // NULL a or b is not actually possible if ((entry_a == NULL) && (entry_b == NULL)) { return 0; } if (entry_a == NULL) { return 1; } if (entry_b == NULL) { return -1; } // NULL resources sort first if ((entry_a->rsc == NULL) && (entry_b->rsc == NULL)) { return 0; } if (entry_a->rsc == NULL) { return 1; } if (entry_b->rsc == NULL) { return -1; } // Compare resource names tmp = strcmp(entry_a->rsc->id, entry_b->rsc->id); if (tmp != 0) { return tmp; } // Otherwise NULL nodes sort first if ((entry_a->node == NULL) && (entry_b->node == NULL)) { return 0; } if (entry_a->node == NULL) { return 1; } if (entry_b->node == NULL) { return -1; } // Finally, compare node names return strcmp(entry_a->node->details->id, entry_b->node->details->id); } /*! * \internal * \brief Duplicate a notification entry * * \param[in] entry Entry to duplicate * * \return Newly allocated duplicate of \p entry * \note It is the caller's responsibility to free the return value. */ static notify_entry_t * dup_notify_entry(const notify_entry_t *entry) { notify_entry_t *dup = calloc(1, sizeof(notify_entry_t)); CRM_ASSERT(dup != NULL); dup->rsc = entry->rsc; dup->node = entry->node; return dup; } /*! * \internal * \brief Given a list of nodes, create strings with node names * * \param[in] list List of nodes (as pcmk_node_t *) * \param[out] all_node_names If not NULL, will be set to space-separated list * of the names of all nodes in \p list * \param[out] host_node_names Same as \p all_node_names, except active * guest nodes will list the name of their host * * \note The caller is responsible for freeing the output argument values using * \p g_string_free(). */ static void get_node_names(const GList *list, GString **all_node_names, GString **host_node_names) { if (all_node_names != NULL) { *all_node_names = NULL; } if (host_node_names != NULL) { *host_node_names = NULL; } for (const GList *iter = list; iter != NULL; iter = iter->next) { const pcmk_node_t *node = (const pcmk_node_t *) iter->data; if (node->details->uname == NULL) { continue; } // Always add to list of all node names if (all_node_names != NULL) { pcmk__add_word(all_node_names, 1024, node->details->uname); } // Add to host node name list if appropriate if (host_node_names != NULL) { if (pcmk__is_guest_or_bundle_node(node) && (node->details->remote_rsc->container->running_on != NULL)) { node = pcmk__current_node(node->details->remote_rsc->container); if (node->details->uname == NULL) { continue; } } pcmk__add_word(host_node_names, 1024, node->details->uname); } } if ((all_node_names != NULL) && (*all_node_names == NULL)) { *all_node_names = g_string_new(" "); } if ((host_node_names != NULL) && (*host_node_names == NULL)) { *host_node_names = g_string_new(" "); } } /*! * \internal * \brief Create strings of instance and node names from notification entries * * \param[in,out] list List of notification entries (will be sorted here) * \param[out] rsc_names If not NULL, will be set to space-separated list * of clone instances from \p list * \param[out] node_names If not NULL, will be set to space-separated list * of node names from \p list * * \return (Possibly new) head of sorted \p list * \note The caller is responsible for freeing the output argument values using * \p g_list_free_full() and \p g_string_free(). */ static GList * notify_entries_to_strings(GList *list, GString **rsc_names, GString **node_names) { const char *last_rsc_id = NULL; // Initialize output lists to NULL if (rsc_names != NULL) { *rsc_names = NULL; } if (node_names != NULL) { *node_names = NULL; } // Sort input list for user-friendliness (and ease of filtering duplicates) list = g_list_sort(list, compare_notify_entries); for (GList *gIter = list; gIter != NULL; gIter = gIter->next) { notify_entry_t *entry = (notify_entry_t *) gIter->data; // Entry must have a resource (with ID) CRM_LOG_ASSERT((entry != NULL) && (entry->rsc != NULL) && (entry->rsc->id != NULL)); if ((entry == NULL) || (entry->rsc == NULL) || (entry->rsc->id == NULL)) { continue; } // Entry must have a node unless listing inactive resources CRM_LOG_ASSERT((node_names == NULL) || (entry->node != NULL)); if ((node_names != NULL) && (entry->node == NULL)) { continue; } // Don't add duplicates of a particular clone instance if (pcmk__str_eq(entry->rsc->id, last_rsc_id, pcmk__str_none)) { continue; } last_rsc_id = entry->rsc->id; if (rsc_names != NULL) { pcmk__add_word(rsc_names, 1024, entry->rsc->id); } if ((node_names != NULL) && (entry->node->details->uname != NULL)) { pcmk__add_word(node_names, 1024, entry->node->details->uname); } } // If there are no entries, return "empty" lists if ((rsc_names != NULL) && (*rsc_names == NULL)) { *rsc_names = g_string_new(" "); } if ((node_names != NULL) && (*node_names == NULL)) { *node_names = g_string_new(" "); } return list; } /*! * \internal * \brief Copy a meta-attribute into a notify action * * \param[in] key Name of meta-attribute to copy * \param[in] value Value of meta-attribute to copy * \param[in,out] user_data Notify action to copy into */ static void copy_meta_to_notify(gpointer key, gpointer value, gpointer user_data) { pcmk_action_t *notify = (pcmk_action_t *) user_data; /* Any existing meta-attributes (for example, the action timeout) are for * the notify action itself, so don't override those. */ if (g_hash_table_lookup(notify->meta, (const char *) key) != NULL) { return; } pcmk__insert_dup(notify->meta, (const char *) key, (const char *) value); } static void add_notify_data_to_action_meta(const notify_data_t *n_data, pcmk_action_t *action) { for (const GSList *item = n_data->keys; item; item = item->next) { const pcmk_nvpair_t *nvpair = (const pcmk_nvpair_t *) item->data; pcmk__insert_meta(action, nvpair->name, nvpair->value); } } /*! * \internal * \brief Create a new notify pseudo-action for a clone resource * * \param[in,out] rsc Clone resource that notification is for * \param[in] action Action to use in notify action key * \param[in] notif_action PCMK_ACTION_NOTIFY or PCMK_ACTION_NOTIFIED * \param[in] notif_type "pre", "post", "confirmed-pre", "confirmed-post" * * \return Newly created notify pseudo-action */ static pcmk_action_t * new_notify_pseudo_action(pcmk_resource_t *rsc, const pcmk_action_t *action, const char *notif_action, const char *notif_type) { pcmk_action_t *notify = NULL; notify = custom_action(rsc, pcmk__notify_key(rsc->id, notif_type, action->task), notif_action, NULL, pcmk_is_set(action->flags, pcmk_action_optional), rsc->cluster); pcmk__set_action_flags(notify, pcmk_action_pseudo); pcmk__insert_meta(notify, "notify_key_type", notif_type); pcmk__insert_meta(notify, "notify_key_operation", action->task); return notify; } /*! * \internal * \brief Create a new notify action for a clone instance * * \param[in,out] rsc Clone instance that notification is for * \param[in] node Node that notification is for * \param[in,out] op Action that notification is for * \param[in,out] notify_done Parent pseudo-action for notifications complete * \param[in] n_data Notification values to add to action meta-data * * \return Newly created notify action */ static pcmk_action_t * new_notify_action(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_action_t *op, pcmk_action_t *notify_done, const notify_data_t *n_data) { char *key = NULL; pcmk_action_t *notify_action = NULL; const char *value = NULL; const char *task = NULL; const char *skip_reason = NULL; CRM_CHECK((rsc != NULL) && (node != NULL), return NULL); // Ensure we have all the info we need if (op == NULL) { skip_reason = "no action"; } else if (notify_done == NULL) { skip_reason = "no parent notification"; } else if (!node->details->online) { skip_reason = "node offline"; } else if (!pcmk_is_set(op->flags, pcmk_action_runnable)) { skip_reason = "original action not runnable"; } if (skip_reason != NULL) { pcmk__rsc_trace(rsc, "Skipping notify action for %s on %s: %s", rsc->id, pcmk__node_name(node), skip_reason); return NULL; } value = g_hash_table_lookup(op->meta, "notify_type"); // "pre" or "post" task = g_hash_table_lookup(op->meta, "notify_operation"); // original action pcmk__rsc_trace(rsc, "Creating notify action for %s on %s (%s-%s)", rsc->id, pcmk__node_name(node), value, task); // Create the notify action key = pcmk__notify_key(rsc->id, value, task); notify_action = custom_action(rsc, key, op->task, node, pcmk_is_set(op->flags, pcmk_action_optional), rsc->cluster); // Add meta-data to notify action g_hash_table_foreach(op->meta, copy_meta_to_notify, notify_action); add_notify_data_to_action_meta(n_data, notify_action); // Order notify after original action and before parent notification order_actions(op, notify_action, pcmk__ar_ordered); order_actions(notify_action, notify_done, pcmk__ar_ordered); return notify_action; } /*! * \internal * \brief Create a new "post-" notify action for a clone instance * * \param[in,out] rsc Clone instance that notification is for * \param[in] node Node that notification is for * \param[in,out] n_data Notification values to add to action meta-data */ static void new_post_notify_action(pcmk_resource_t *rsc, const pcmk_node_t *node, notify_data_t *n_data) { pcmk_action_t *notify = NULL; CRM_ASSERT(n_data != NULL); // Create the "post-" notify action for specified instance notify = new_notify_action(rsc, node, n_data->post, n_data->post_done, n_data); if (notify != NULL) { - notify->priority = INFINITY; + notify->priority = PCMK_SCORE_INFINITY; } // Order recurring monitors after all "post-" notifications complete if (n_data->post_done == NULL) { return; } for (GList *iter = rsc->actions; iter != NULL; iter = iter->next) { pcmk_action_t *mon = (pcmk_action_t *) iter->data; const char *interval_ms_s = NULL; interval_ms_s = g_hash_table_lookup(mon->meta, PCMK_META_INTERVAL); if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches) || pcmk__str_eq(mon->task, PCMK_ACTION_CANCEL, pcmk__str_none)) { continue; // Not a recurring monitor } order_actions(n_data->post_done, mon, pcmk__ar_ordered); } } /*! * \internal * \brief Create and order notification pseudo-actions for a clone action * * In addition to the actual notify actions needed for each clone instance, * clone notifications also require pseudo-actions to provide ordering points * in the notification process. This creates the notification data, along with * appropriate pseudo-actions and their orderings. * * For example, the ordering sequence for starting a clone is: * * "pre-" notify pseudo-action for clone * -> "pre-" notify actions for each clone instance * -> "pre-" notifications complete pseudo-action for clone * -> start actions for each clone instance * -> "started" pseudo-action for clone * -> "post-" notify pseudo-action for clone * -> "post-" notify actions for each clone instance * -> "post-" notifications complete pseudo-action for clone * * \param[in,out] rsc Clone that notifications are for * \param[in] task Name of action that notifications are for * \param[in,out] action If not NULL, create a "pre-" pseudo-action ordered * before a "pre-" complete pseudo-action, ordered * before this action * \param[in,out] complete If not NULL, create a "post-" pseudo-action ordered * after this action, and a "post-" complete * pseudo-action ordered after that * * \return Newly created notification data */ notify_data_t * pe__action_notif_pseudo_ops(pcmk_resource_t *rsc, const char *task, pcmk_action_t *action, pcmk_action_t *complete) { notify_data_t *n_data = NULL; if (!pcmk_is_set(rsc->flags, pcmk_rsc_notify)) { return NULL; } n_data = calloc(1, sizeof(notify_data_t)); CRM_ASSERT(n_data != NULL); n_data->action = task; if (action != NULL) { // Need "pre-" pseudo-actions // Create "pre-" notify pseudo-action for clone n_data->pre = new_notify_pseudo_action(rsc, action, PCMK_ACTION_NOTIFY, "pre"); pcmk__set_action_flags(n_data->pre, pcmk_action_runnable); pcmk__insert_meta(n_data->pre, "notify_type", "pre"); pcmk__insert_meta(n_data->pre, "notify_operation", n_data->action); // Create "pre-" notifications complete pseudo-action for clone n_data->pre_done = new_notify_pseudo_action(rsc, action, PCMK_ACTION_NOTIFIED, "confirmed-pre"); pcmk__set_action_flags(n_data->pre_done, pcmk_action_runnable); pcmk__insert_meta(n_data->pre_done, "notify_type", "pre"); pcmk__insert_meta(n_data->pre_done, "notify_operation", n_data->action); // Order "pre-" -> "pre-" complete -> original action order_actions(n_data->pre, n_data->pre_done, pcmk__ar_ordered); order_actions(n_data->pre_done, action, pcmk__ar_ordered); } if (complete != NULL) { // Need "post-" pseudo-actions // Create "post-" notify pseudo-action for clone n_data->post = new_notify_pseudo_action(rsc, complete, PCMK_ACTION_NOTIFY, "post"); - n_data->post->priority = INFINITY; + n_data->post->priority = PCMK_SCORE_INFINITY; if (pcmk_is_set(complete->flags, pcmk_action_runnable)) { pcmk__set_action_flags(n_data->post, pcmk_action_runnable); } else { pcmk__clear_action_flags(n_data->post, pcmk_action_runnable); } pcmk__insert_meta(n_data->post, "notify_type", "post"); pcmk__insert_meta(n_data->post, "notify_operation", n_data->action); // Create "post-" notifications complete pseudo-action for clone n_data->post_done = new_notify_pseudo_action(rsc, complete, PCMK_ACTION_NOTIFIED, "confirmed-post"); - n_data->post_done->priority = INFINITY; + n_data->post_done->priority = PCMK_SCORE_INFINITY; if (pcmk_is_set(complete->flags, pcmk_action_runnable)) { pcmk__set_action_flags(n_data->post_done, pcmk_action_runnable); } else { pcmk__clear_action_flags(n_data->post_done, pcmk_action_runnable); } pcmk__insert_meta(n_data->post_done, "notify_type", "post"); pcmk__insert_meta(n_data->post_done, "notify_operation", n_data->action); // Order original action complete -> "post-" -> "post-" complete order_actions(complete, n_data->post, pcmk__ar_first_implies_then); order_actions(n_data->post, n_data->post_done, pcmk__ar_first_implies_then); } // If we created both, order "pre-" complete -> "post-" if ((action != NULL) && (complete != NULL)) { order_actions(n_data->pre_done, n_data->post, pcmk__ar_ordered); } return n_data; } /*! * \internal * \brief Create a new notification entry * * \param[in] rsc Resource for notification * \param[in] node Node for notification * * \return Newly allocated notification entry * \note The caller is responsible for freeing the return value. */ static notify_entry_t * new_notify_entry(const pcmk_resource_t *rsc, const pcmk_node_t *node) { notify_entry_t *entry = calloc(1, sizeof(notify_entry_t)); CRM_ASSERT(entry != NULL); entry->rsc = rsc; entry->node = node; return entry; } /*! * \internal * \brief Add notification data for resource state and optionally actions * * \param[in] rsc Clone or clone instance being notified * \param[in] activity Whether to add notification entries for actions * \param[in,out] n_data Notification data for clone */ static void collect_resource_data(const pcmk_resource_t *rsc, bool activity, notify_data_t *n_data) { const GList *iter = NULL; notify_entry_t *entry = NULL; const pcmk_node_t *node = NULL; if (n_data == NULL) { return; } if (n_data->allowed_nodes == NULL) { n_data->allowed_nodes = rsc->allowed_nodes; } // If this is a clone, call recursively for each instance if (rsc->children != NULL) { for (iter = rsc->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *child = (const pcmk_resource_t *) iter->data; collect_resource_data(child, activity, n_data); } return; } // This is a notification for a single clone instance if (rsc->running_on != NULL) { node = rsc->running_on->data; // First is sufficient } entry = new_notify_entry(rsc, node); // Add notification indicating the resource state switch (rsc->role) { case pcmk_role_stopped: n_data->inactive = g_list_prepend(n_data->inactive, entry); break; case pcmk_role_started: n_data->active = g_list_prepend(n_data->active, entry); break; case pcmk_role_unpromoted: n_data->unpromoted = g_list_prepend(n_data->unpromoted, entry); n_data->active = g_list_prepend(n_data->active, dup_notify_entry(entry)); break; case pcmk_role_promoted: n_data->promoted = g_list_prepend(n_data->promoted, entry); n_data->active = g_list_prepend(n_data->active, dup_notify_entry(entry)); break; default: pcmk__sched_err("Resource %s role on %s (%s) is not supported for " "notifications (bug?)", rsc->id, pcmk__node_name(node), pcmk_role_text(rsc->role)); free(entry); break; } if (!activity) { return; } // Add notification entries for each of the resource's actions for (iter = rsc->actions; iter != NULL; iter = iter->next) { const pcmk_action_t *op = (const pcmk_action_t *) iter->data; if (!pcmk_is_set(op->flags, pcmk_action_optional) && (op->node != NULL)) { enum action_tasks task = pcmk_parse_action(op->task); if ((task == pcmk_action_stop) && op->node->details->unclean) { // Create anyway (additional noise if node can't be fenced) } else if (!pcmk_is_set(op->flags, pcmk_action_runnable)) { continue; } entry = new_notify_entry(rsc, op->node); switch (task) { case pcmk_action_start: n_data->start = g_list_prepend(n_data->start, entry); break; case pcmk_action_stop: n_data->stop = g_list_prepend(n_data->stop, entry); break; case pcmk_action_promote: n_data->promote = g_list_prepend(n_data->promote, entry); break; case pcmk_action_demote: n_data->demote = g_list_prepend(n_data->demote, entry); break; default: free(entry); break; } } } } // For (char *) value #define add_notify_env(n_data, key, value) do { \ n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, value); \ } while (0) // For (GString *) value #define add_notify_env_gs(n_data, key, value) do { \ n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, \ (const char *) value->str); \ } while (0) // For (GString *) value #define add_notify_env_free_gs(n_data, key, value) do { \ n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, \ (const char *) value->str); \ g_string_free(value, TRUE); value = NULL; \ } while (0) /*! * \internal * \brief Create notification name/value pairs from structured data * * \param[in] rsc Resource that notification is for * \param[in,out] n_data Notification data */ static void add_notif_keys(const pcmk_resource_t *rsc, notify_data_t *n_data) { bool required = false; // Whether to make notify actions required GString *rsc_list = NULL; GString *node_list = NULL; GString *metal_list = NULL; const char *source = NULL; GList *nodes = NULL; n_data->stop = notify_entries_to_strings(n_data->stop, &rsc_list, &node_list); if ((strcmp(" ", (const char *) rsc_list->str) != 0) && pcmk__str_eq(n_data->action, PCMK_ACTION_STOP, pcmk__str_none)) { required = true; } add_notify_env_free_gs(n_data, "notify_stop_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_stop_uname", node_list); if ((n_data->start != NULL) && pcmk__str_eq(n_data->action, PCMK_ACTION_START, pcmk__str_none)) { required = true; } n_data->start = notify_entries_to_strings(n_data->start, &rsc_list, &node_list); add_notify_env_free_gs(n_data, "notify_start_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_start_uname", node_list); if ((n_data->demote != NULL) && pcmk__str_eq(n_data->action, PCMK_ACTION_DEMOTE, pcmk__str_none)) { required = true; } n_data->demote = notify_entries_to_strings(n_data->demote, &rsc_list, &node_list); add_notify_env_free_gs(n_data, "notify_demote_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_demote_uname", node_list); if ((n_data->promote != NULL) && pcmk__str_eq(n_data->action, PCMK_ACTION_PROMOTE, pcmk__str_none)) { required = true; } n_data->promote = notify_entries_to_strings(n_data->promote, &rsc_list, &node_list); add_notify_env_free_gs(n_data, "notify_promote_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_promote_uname", node_list); n_data->active = notify_entries_to_strings(n_data->active, &rsc_list, &node_list); add_notify_env_free_gs(n_data, "notify_active_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_active_uname", node_list); n_data->unpromoted = notify_entries_to_strings(n_data->unpromoted, &rsc_list, &node_list); add_notify_env_gs(n_data, "notify_unpromoted_resource", rsc_list); add_notify_env_gs(n_data, "notify_unpromoted_uname", node_list); // Deprecated: kept for backward compatibility with older resource agents add_notify_env_free_gs(n_data, "notify_slave_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_slave_uname", node_list); n_data->promoted = notify_entries_to_strings(n_data->promoted, &rsc_list, &node_list); add_notify_env_gs(n_data, "notify_promoted_resource", rsc_list); add_notify_env_gs(n_data, "notify_promoted_uname", node_list); // Deprecated: kept for backward compatibility with older resource agents add_notify_env_free_gs(n_data, "notify_master_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_master_uname", node_list); n_data->inactive = notify_entries_to_strings(n_data->inactive, &rsc_list, NULL); add_notify_env_free_gs(n_data, "notify_inactive_resource", rsc_list); nodes = g_hash_table_get_values(n_data->allowed_nodes); if (!pcmk__is_daemon) { /* For display purposes, sort the node list, for consistent * regression test output (while avoiding the performance hit * for the live cluster). */ nodes = g_list_sort(nodes, pe__cmp_node_name); } get_node_names(nodes, &node_list, NULL); add_notify_env_free_gs(n_data, "notify_available_uname", node_list); g_list_free(nodes); source = g_hash_table_lookup(rsc->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); if (pcmk__str_eq(PCMK_VALUE_HOST, source, pcmk__str_none)) { get_node_names(rsc->cluster->nodes, &node_list, &metal_list); add_notify_env_free_gs(n_data, "notify_all_hosts", metal_list); } else { get_node_names(rsc->cluster->nodes, &node_list, NULL); } add_notify_env_free_gs(n_data, "notify_all_uname", node_list); if (required && (n_data->pre != NULL)) { pcmk__clear_action_flags(n_data->pre, pcmk_action_optional); pcmk__clear_action_flags(n_data->pre_done, pcmk_action_optional); } if (required && (n_data->post != NULL)) { pcmk__clear_action_flags(n_data->post, pcmk_action_optional); pcmk__clear_action_flags(n_data->post_done, pcmk_action_optional); } } /* * \internal * \brief Find any remote connection start relevant to an action * * \param[in] action Action to check * * \return If action is behind a remote connection, connection's start */ static pcmk_action_t * find_remote_start(pcmk_action_t *action) { if ((action != NULL) && (action->node != NULL)) { pcmk_resource_t *remote_rsc = action->node->details->remote_rsc; if (remote_rsc != NULL) { return find_first_action(remote_rsc->actions, NULL, PCMK_ACTION_START, NULL); } } return NULL; } /*! * \internal * \brief Create notify actions, and add notify data to original actions * * \param[in,out] rsc Clone or clone instance that notification is for * \param[in,out] n_data Clone notification data for some action */ static void create_notify_actions(pcmk_resource_t *rsc, notify_data_t *n_data) { GList *iter = NULL; pcmk_action_t *stop = NULL; pcmk_action_t *start = NULL; enum action_tasks task = pcmk_parse_action(n_data->action); // If this is a clone, call recursively for each instance if (rsc->children != NULL) { g_list_foreach(rsc->children, (GFunc) create_notify_actions, n_data); return; } // Add notification meta-attributes to original actions for (iter = rsc->actions; iter != NULL; iter = iter->next) { pcmk_action_t *op = (pcmk_action_t *) iter->data; if (!pcmk_is_set(op->flags, pcmk_action_optional) && (op->node != NULL)) { switch (pcmk_parse_action(op->task)) { case pcmk_action_start: case pcmk_action_stop: case pcmk_action_promote: case pcmk_action_demote: add_notify_data_to_action_meta(n_data, op); break; default: break; } } } // Skip notify action itself if original action was not needed switch (task) { case pcmk_action_start: if (n_data->start == NULL) { pcmk__rsc_trace(rsc, "No notify action needed for %s %s", rsc->id, n_data->action); return; } break; case pcmk_action_promote: if (n_data->promote == NULL) { pcmk__rsc_trace(rsc, "No notify action needed for %s %s", rsc->id, n_data->action); return; } break; case pcmk_action_demote: if (n_data->demote == NULL) { pcmk__rsc_trace(rsc, "No notify action needed for %s %s", rsc->id, n_data->action); return; } break; default: // We cannot do same for stop because it might be implied by fencing break; } pcmk__rsc_trace(rsc, "Creating notify actions for %s %s", rsc->id, n_data->action); // Create notify actions for stop or demote if ((rsc->role != pcmk_role_stopped) && ((task == pcmk_action_stop) || (task == pcmk_action_demote))) { stop = find_first_action(rsc->actions, NULL, PCMK_ACTION_STOP, NULL); for (iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *current_node = (pcmk_node_t *) iter->data; /* If a stop is a pseudo-action implied by fencing, don't try to * notify the node getting fenced. */ if ((stop != NULL) && pcmk_is_set(stop->flags, pcmk_action_pseudo) && (current_node->details->unclean || current_node->details->remote_requires_reset)) { continue; } new_notify_action(rsc, current_node, n_data->pre, n_data->pre_done, n_data); if ((task == pcmk_action_demote) || (stop == NULL) || pcmk_is_set(stop->flags, pcmk_action_optional)) { new_post_notify_action(rsc, current_node, n_data); } } } // Create notify actions for start or promote if ((rsc->next_role != pcmk_role_stopped) && ((task == pcmk_action_start) || (task == pcmk_action_promote))) { start = find_first_action(rsc->actions, NULL, PCMK_ACTION_START, NULL); if (start != NULL) { pcmk_action_t *remote_start = find_remote_start(start); if ((remote_start != NULL) && !pcmk_is_set(remote_start->flags, pcmk_action_runnable)) { /* Start and promote actions for a clone instance behind * a Pacemaker Remote connection happen after the * connection starts. If the connection start is blocked, do * not schedule notifications for these actions. */ return; } } if (rsc->allocated_to == NULL) { pcmk__sched_err("Next role '%s' but %s is not allocated", pcmk_role_text(rsc->next_role), rsc->id); return; } if ((task != pcmk_action_start) || (start == NULL) || pcmk_is_set(start->flags, pcmk_action_optional)) { new_notify_action(rsc, rsc->allocated_to, n_data->pre, n_data->pre_done, n_data); } new_post_notify_action(rsc, rsc->allocated_to, n_data); } } /*! * \internal * \brief Create notification data and actions for one clone action * * \param[in,out] rsc Clone resource that notification is for * \param[in,out] n_data Clone notification data for some action */ void pe__create_action_notifications(pcmk_resource_t *rsc, notify_data_t *n_data) { if ((rsc == NULL) || (n_data == NULL)) { return; } collect_resource_data(rsc, true, n_data); add_notif_keys(rsc, n_data); create_notify_actions(rsc, n_data); } /*! * \internal * \brief Free notification data for one action * * \param[in,out] n_data Notification data to free */ void pe__free_action_notification_data(notify_data_t *n_data) { if (n_data == NULL) { return; } g_list_free_full(n_data->stop, free); g_list_free_full(n_data->start, free); g_list_free_full(n_data->demote, free); g_list_free_full(n_data->promote, free); g_list_free_full(n_data->promoted, free); g_list_free_full(n_data->unpromoted, free); g_list_free_full(n_data->active, free); g_list_free_full(n_data->inactive, free); pcmk_free_nvpairs(n_data->keys); free(n_data); } /*! * \internal * \brief Order clone "notifications complete" pseudo-action after fencing * * If a stop action is implied by fencing, the usual notification pseudo-actions * will not be sufficient to order things properly, or even create all needed * notifications if the clone is also stopping on another node, and another * clone is ordered after it. This function creates new notification * pseudo-actions relative to the fencing to ensure everything works properly. * * \param[in] stop Stop action implied by fencing * \param[in,out] rsc Clone resource that notification is for * \param[in,out] stonith_op Fencing action that implies \p stop */ void pe__order_notifs_after_fencing(const pcmk_action_t *stop, pcmk_resource_t *rsc, pcmk_action_t *stonith_op) { notify_data_t *n_data; crm_info("Ordering notifications for implied %s after fencing", stop->uuid); n_data = pe__action_notif_pseudo_ops(rsc, PCMK_ACTION_STOP, NULL, stonith_op); if (n_data != NULL) { collect_resource_data(rsc, false, n_data); add_notify_env(n_data, "notify_stop_resource", rsc->id); add_notify_env(n_data, "notify_stop_uname", stop->node->details->uname); create_notify_actions(uber_parent(rsc), n_data); pe__free_action_notification_data(n_data); } } diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c index 8bc1519d77..7fceea8916 100644 --- a/lib/pengine/unpack.c +++ b/lib/pengine/unpack.c @@ -1,5158 +1,5158 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include CRM_TRACE_INIT_DATA(pe_status); // A (parsed) resource action history entry struct action_history { pcmk_resource_t *rsc; // Resource that history is for pcmk_node_t *node; // Node that history is for xmlNode *xml; // History entry XML // Parsed from entry XML const char *id; // XML ID of history entry const char *key; // Operation key of action const char *task; // Action name const char *exit_reason; // Exit reason given for result guint interval_ms; // Action interval int call_id; // Call ID of action int expected_exit_status; // Expected exit status of action int exit_status; // Actual exit status of action int execution_status; // Execution status of action }; /* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than * use pcmk__set_scheduler_flags()/pcmk__clear_scheduler_flags() so that the * flag is stringified more readably in log messages. */ #define set_config_flag(scheduler, option, flag) do { \ GHashTable *config_hash = (scheduler)->config_hash; \ const char *scf_value = pcmk__cluster_option(config_hash, (option)); \ \ if (scf_value != NULL) { \ if (crm_is_true(scf_value)) { \ (scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", \ crm_system_name, (scheduler)->flags, \ (flag), #flag); \ } else { \ (scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", \ crm_system_name, (scheduler)->flags, \ (flag), #flag); \ } \ } \ } while(0) static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *failed); static void determine_remote_online_status(pcmk_scheduler_t *scheduler, pcmk_node_t *this_node); static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite, pcmk_scheduler_t *scheduler); static void determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node, pcmk_scheduler_t *scheduler); static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml, pcmk_scheduler_t *scheduler); static gboolean is_dangling_guest_node(pcmk_node_t *node) { /* we are looking for a remote-node that was supposed to be mapped to a * container resource, but all traces of that container have disappeared * from both the config and the status section. */ if (pcmk__is_pacemaker_remote_node(node) && (node->details->remote_rsc != NULL) && (node->details->remote_rsc->container == NULL) && pcmk_is_set(node->details->remote_rsc->flags, pcmk_rsc_removed_filler)) { return TRUE; } return FALSE; } /*! * \brief Schedule a fence action for a node * * \param[in,out] scheduler Scheduler data * \param[in,out] node Node to fence * \param[in] reason Text description of why fencing is needed * \param[in] priority_delay Whether to consider * \c PCMK_OPT_PRIORITY_FENCING_DELAY */ void pe_fence_node(pcmk_scheduler_t *scheduler, pcmk_node_t *node, const char *reason, bool priority_delay) { CRM_CHECK(node, return); /* A guest node is fenced by marking its container as failed */ if (pcmk__is_guest_or_bundle_node(node)) { pcmk_resource_t *rsc = node->details->remote_rsc->container; if (!pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { crm_notice("Not fencing guest node %s " "(otherwise would because %s): " "its guest resource %s is unmanaged", pcmk__node_name(node), reason, rsc->id); } else { pcmk__sched_warn("Guest node %s will be fenced " "(by recovering its guest resource %s): %s", pcmk__node_name(node), rsc->id, reason); /* We don't mark the node as unclean because that would prevent the * node from running resources. We want to allow it to run resources * in this transition if the recovery succeeds. */ node->details->remote_requires_reset = TRUE; pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); } } } else if (is_dangling_guest_node(node)) { crm_info("Cleaning up dangling connection for guest node %s: " "fencing was already done because %s, " "and guest resource no longer exists", pcmk__node_name(node), reason); pcmk__set_rsc_flags(node->details->remote_rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); } else if (pcmk__is_remote_node(node)) { pcmk_resource_t *rsc = node->details->remote_rsc; if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { crm_notice("Not fencing remote node %s " "(otherwise would because %s): connection is unmanaged", pcmk__node_name(node), reason); } else if(node->details->remote_requires_reset == FALSE) { node->details->remote_requires_reset = TRUE; pcmk__sched_warn("Remote node %s %s: %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "will be fenced" : "is unclean", reason); } node->details->unclean = TRUE; // No need to apply PCMK_OPT_PRIORITY_FENCING_DELAY for remote nodes pe_fence_op(node, NULL, TRUE, reason, FALSE, scheduler); } else if (node->details->unclean) { crm_trace("Cluster node %s %s because %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean", reason); } else { pcmk__sched_warn("Cluster node %s %s: %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "will be fenced" : "is unclean", reason); node->details->unclean = TRUE; pe_fence_op(node, NULL, TRUE, reason, priority_delay, scheduler); } } // @TODO xpaths can't handle templates, rules, or id-refs // nvpair with provides or requires set to unfencing #define XPATH_UNFENCING_NVPAIR PCMK_XE_NVPAIR \ "[(@" PCMK_XA_NAME "='" PCMK_STONITH_PROVIDES "'" \ "or @" PCMK_XA_NAME "='" PCMK_META_REQUIRES "') " \ "and @" PCMK_XA_VALUE "='" PCMK_VALUE_UNFENCING "']" // unfencing in rsc_defaults or any resource #define XPATH_ENABLE_UNFENCING \ "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \ "//" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR \ "|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RSC_DEFAULTS \ "/" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR static void set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler) { xmlXPathObjectPtr result = NULL; if (!pcmk_is_set(scheduler->flags, flag)) { result = xpath_search(scheduler->input, xpath); if (result && (numXpathResults(result) > 0)) { pcmk__set_scheduler_flags(scheduler, flag); } freeXpathObject(result); } } gboolean unpack_config(xmlNode *config, pcmk_scheduler_t *scheduler) { const char *value = NULL; guint interval_ms = 0U; GHashTable *config_hash = pcmk__strkey_table(free, free); pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; scheduler->config_hash = config_hash; pe__unpack_dataset_nvpairs(config, PCMK_XE_CLUSTER_PROPERTY_SET, &rule_data, config_hash, PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, FALSE, scheduler); pcmk__validate_cluster_options(config_hash); set_config_flag(scheduler, PCMK_OPT_ENABLE_STARTUP_PROBES, pcmk_sched_probe_resources); if (!pcmk_is_set(scheduler->flags, pcmk_sched_probe_resources)) { crm_info("Startup probes: disabled (dangerous)"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_HAVE_WATCHDOG); if (value && crm_is_true(value)) { crm_info("Watchdog-based self-fencing will be performed via SBD if " "fencing is required and " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " is nonzero"); pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_fencing); } /* Set certain flags via xpath here, so they can be used before the relevant * configuration sections are unpacked. */ set_if_xpath(pcmk_sched_enable_unfencing, XPATH_ENABLE_UNFENCING, scheduler); value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT); pcmk_parse_interval_spec(value, &interval_ms); if (interval_ms >= INT_MAX) { scheduler->stonith_timeout = INT_MAX; } else { scheduler->stonith_timeout = (int) interval_ms; } crm_debug("STONITH timeout: %d", scheduler->stonith_timeout); set_config_flag(scheduler, PCMK_OPT_STONITH_ENABLED, pcmk_sched_fencing_enabled); if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { crm_debug("STONITH of failed nodes is enabled"); } else { crm_debug("STONITH of failed nodes is disabled"); } scheduler->stonith_action = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_ACTION); if (!strcmp(scheduler->stonith_action, PCMK__ACTION_POWEROFF)) { pcmk__warn_once(pcmk__wo_poweroff, "Support for " PCMK_OPT_STONITH_ACTION " of " "'" PCMK__ACTION_POWEROFF "' is deprecated and will be " "removed in a future release " "(use '" PCMK_ACTION_OFF "' instead)"); scheduler->stonith_action = PCMK_ACTION_OFF; } crm_trace("STONITH will %s nodes", scheduler->stonith_action); set_config_flag(scheduler, PCMK_OPT_CONCURRENT_FENCING, pcmk_sched_concurrent_fencing); if (pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing)) { crm_debug("Concurrent fencing is enabled"); } else { crm_debug("Concurrent fencing is disabled"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_PRIORITY_FENCING_DELAY); if (value) { pcmk_parse_interval_spec(value, &interval_ms); scheduler->priority_fencing_delay = (int) (interval_ms / 1000); crm_trace("Priority fencing delay is %ds", scheduler->priority_fencing_delay); } set_config_flag(scheduler, PCMK_OPT_STOP_ALL_RESOURCES, pcmk_sched_stop_all); crm_debug("Stop all active resources: %s", pcmk__flag_text(scheduler->flags, pcmk_sched_stop_all)); set_config_flag(scheduler, PCMK_OPT_SYMMETRIC_CLUSTER, pcmk_sched_symmetric_cluster); if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) { crm_debug("Cluster is symmetric" " - resources can run anywhere by default"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_NO_QUORUM_POLICY); if (pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_ignore; } else if (pcmk__str_eq(value, PCMK_VALUE_FREEZE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_freeze; } else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_demote; } else if (pcmk__str_eq(value, PCMK_VALUE_FENCE_LEGACY, pcmk__str_casei)) { if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { int do_panic = 0; crm_element_value_int(scheduler->input, PCMK_XA_NO_QUORUM_PANIC, &do_panic); if (do_panic || pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) { scheduler->no_quorum_policy = pcmk_no_quorum_fence; } else { crm_notice("Resetting " PCMK_OPT_NO_QUORUM_POLICY " to 'stop': cluster has never had quorum"); scheduler->no_quorum_policy = pcmk_no_quorum_stop; } } else { pcmk__config_err("Resetting " PCMK_OPT_NO_QUORUM_POLICY " to 'stop' because fencing is disabled"); scheduler->no_quorum_policy = pcmk_no_quorum_stop; } } else { scheduler->no_quorum_policy = pcmk_no_quorum_stop; } switch (scheduler->no_quorum_policy) { case pcmk_no_quorum_freeze: crm_debug("On loss of quorum: Freeze resources"); break; case pcmk_no_quorum_stop: crm_debug("On loss of quorum: Stop ALL resources"); break; case pcmk_no_quorum_demote: crm_debug("On loss of quorum: " "Demote promotable resources and stop other resources"); break; case pcmk_no_quorum_fence: crm_notice("On loss of quorum: Fence all remaining nodes"); break; case pcmk_no_quorum_ignore: crm_notice("On loss of quorum: Ignore"); break; } set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_RESOURCES, pcmk_sched_stop_removed_resources); if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { crm_trace("Orphan resources are stopped"); } else { crm_trace("Orphan resources are ignored"); } set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_ACTIONS, pcmk_sched_cancel_removed_actions); if (pcmk_is_set(scheduler->flags, pcmk_sched_cancel_removed_actions)) { crm_trace("Orphan resource actions are stopped"); } else { crm_trace("Orphan resource actions are ignored"); } value = pcmk__cluster_option(config_hash, PCMK__OPT_REMOVE_AFTER_STOP); if (value != NULL) { if (crm_is_true(value)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_remove_after_stop); #ifndef PCMK__COMPAT_2_0 pcmk__warn_once(pcmk__wo_remove_after, "Support for the " PCMK__OPT_REMOVE_AFTER_STOP " cluster property is deprecated and will be " "removed in a future release"); #endif } else { pcmk__clear_scheduler_flags(scheduler, pcmk_sched_remove_after_stop); } } set_config_flag(scheduler, PCMK_OPT_MAINTENANCE_MODE, pcmk_sched_in_maintenance); crm_trace("Maintenance mode: %s", pcmk__flag_text(scheduler->flags, pcmk_sched_in_maintenance)); set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL, pcmk_sched_start_failure_fatal); if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) { crm_trace("Start failures are always fatal"); } else { crm_trace("Start failures are handled by failcount"); } if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING, pcmk_sched_startup_fencing); } if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) { crm_trace("Unseen nodes will be fenced"); } else { pcmk__warn_once(pcmk__wo_blind, "Blind faith: not fencing unseen nodes"); } pe__unpack_node_health_scores(scheduler); scheduler->placement_strategy = pcmk__cluster_option(config_hash, PCMK_OPT_PLACEMENT_STRATEGY); crm_trace("Placement strategy: %s", scheduler->placement_strategy); set_config_flag(scheduler, PCMK_OPT_SHUTDOWN_LOCK, pcmk_sched_shutdown_lock); if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { value = pcmk__cluster_option(config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT); pcmk_parse_interval_spec(value, &(scheduler->shutdown_lock)); scheduler->shutdown_lock /= 1000; crm_trace("Resources will be locked to nodes that were cleanly " "shut down (locks expire after %s)", pcmk__readable_interval(scheduler->shutdown_lock)); } else { crm_trace("Resources will not be locked to nodes that were cleanly " "shut down"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT); pcmk_parse_interval_spec(value, &(scheduler->node_pending_timeout)); scheduler->node_pending_timeout /= 1000; if (scheduler->node_pending_timeout == 0) { crm_trace("Do not fence pending nodes"); } else { crm_trace("Fence pending nodes after %s", pcmk__readable_interval(scheduler->node_pending_timeout * 1000)); } return TRUE; } pcmk_node_t * pe_create_node(const char *id, const char *uname, const char *type, const char *score, pcmk_scheduler_t *scheduler) { pcmk_node_t *new_node = NULL; if (pe_find_node(scheduler->nodes, uname) != NULL) { pcmk__config_warn("More than one node entry has name '%s'", uname); } new_node = calloc(1, sizeof(pcmk_node_t)); if (new_node == NULL) { pcmk__sched_err("Could not allocate memory for node %s", uname); return NULL; } new_node->weight = char2score(score); new_node->details = calloc(1, sizeof(struct pe_node_shared_s)); if (new_node->details == NULL) { free(new_node); pcmk__sched_err("Could not allocate memory for node %s", uname); return NULL; } crm_trace("Creating node for entry %s/%s", uname, id); new_node->details->id = id; new_node->details->uname = uname; new_node->details->online = FALSE; new_node->details->shutdown = FALSE; new_node->details->rsc_discovery_enabled = TRUE; new_node->details->running_rsc = NULL; new_node->details->data_set = scheduler; if (pcmk__str_eq(type, PCMK_VALUE_MEMBER, pcmk__str_null_matches|pcmk__str_casei)) { new_node->details->type = pcmk_node_variant_cluster; } else if (pcmk__str_eq(type, PCMK_VALUE_REMOTE, pcmk__str_casei)) { new_node->details->type = pcmk_node_variant_remote; pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_remote_nodes); } else { /* @COMPAT 'ping' is the default for backward compatibility, but it * should be changed to 'member' at a compatibility break */ if (!pcmk__str_eq(type, PCMK__VALUE_PING, pcmk__str_casei)) { pcmk__config_warn("Node %s has unrecognized type '%s', " "assuming '" PCMK__VALUE_PING "'", pcmk__s(uname, "without name"), type); } pcmk__warn_once(pcmk__wo_ping_node, "Support for nodes of type '" PCMK__VALUE_PING "' " "(such as %s) is deprecated and will be removed in a " "future release", pcmk__s(uname, "unnamed node")); new_node->details->type = node_ping; } new_node->details->attrs = pcmk__strkey_table(free, free); if (pcmk__is_pacemaker_remote_node(new_node)) { pcmk__insert_dup(new_node->details->attrs, CRM_ATTR_KIND, "remote"); } else { pcmk__insert_dup(new_node->details->attrs, CRM_ATTR_KIND, "cluster"); } new_node->details->utilization = pcmk__strkey_table(free, free); new_node->details->digest_cache = pcmk__strkey_table(free, pe__free_digests); scheduler->nodes = g_list_insert_sorted(scheduler->nodes, new_node, pe__cmp_node_name); return new_node; } static const char * expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pcmk_scheduler_t *data) { xmlNode *attr_set = NULL; xmlNode *attr = NULL; const char *container_id = pcmk__xe_id(xml_obj); const char *remote_name = NULL; const char *remote_server = NULL; const char *remote_port = NULL; const char *connect_timeout = "60s"; const char *remote_allow_migrate=NULL; const char *is_managed = NULL; for (attr_set = pcmk__xe_first_child(xml_obj); attr_set != NULL; attr_set = pcmk__xe_next(attr_set)) { if (!pcmk__xe_is(attr_set, PCMK_XE_META_ATTRIBUTES)) { continue; } for (attr = pcmk__xe_first_child(attr_set); attr != NULL; attr = pcmk__xe_next(attr)) { const char *value = crm_element_value(attr, PCMK_XA_VALUE); const char *name = crm_element_value(attr, PCMK_XA_NAME); if (name == NULL) { // Sanity continue; } if (strcmp(name, PCMK_META_REMOTE_NODE) == 0) { remote_name = value; } else if (strcmp(name, PCMK_META_REMOTE_ADDR) == 0) { remote_server = value; } else if (strcmp(name, PCMK_META_REMOTE_PORT) == 0) { remote_port = value; } else if (strcmp(name, PCMK_META_REMOTE_CONNECT_TIMEOUT) == 0) { connect_timeout = value; } else if (strcmp(name, PCMK_META_REMOTE_ALLOW_MIGRATE) == 0) { remote_allow_migrate = value; } else if (strcmp(name, PCMK_META_IS_MANAGED) == 0) { is_managed = value; } } } if (remote_name == NULL) { return NULL; } if (pe_find_resource(data->resources, remote_name) != NULL) { return NULL; } pe_create_remote_xml(parent, remote_name, container_id, remote_allow_migrate, is_managed, connect_timeout, remote_server, remote_port); return remote_name; } static void handle_startup_fencing(pcmk_scheduler_t *scheduler, pcmk_node_t *new_node) { if ((new_node->details->type == pcmk_node_variant_remote) && (new_node->details->remote_rsc == NULL)) { /* Ignore fencing for remote nodes that don't have a connection resource * associated with them. This happens when remote node entries get left * in the nodes section after the connection resource is removed. */ return; } if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) { // All nodes are unclean until we've seen their status entry new_node->details->unclean = TRUE; } else { // Blind faith ... new_node->details->unclean = FALSE; } /* We need to be able to determine if a node's status section * exists or not separate from whether the node is unclean. */ new_node->details->unseen = TRUE; } gboolean unpack_nodes(xmlNode *xml_nodes, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; pcmk_node_t *new_node = NULL; const char *id = NULL; const char *uname = NULL; const char *type = NULL; const char *score = NULL; for (xml_obj = pcmk__xe_first_child(xml_nodes); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (pcmk__xe_is(xml_obj, PCMK_XE_NODE)) { new_node = NULL; id = crm_element_value(xml_obj, PCMK_XA_ID); uname = crm_element_value(xml_obj, PCMK_XA_UNAME); type = crm_element_value(xml_obj, PCMK_XA_TYPE); score = crm_element_value(xml_obj, PCMK_XA_SCORE); crm_trace("Processing node %s/%s", uname, id); if (id == NULL) { pcmk__config_err("Ignoring <" PCMK_XE_NODE "> entry in configuration without id"); continue; } new_node = pe_create_node(id, uname, type, score, scheduler); if (new_node == NULL) { return FALSE; } handle_startup_fencing(scheduler, new_node); add_node_attrs(xml_obj, new_node, FALSE, scheduler); crm_trace("Done with node %s", crm_element_value(xml_obj, PCMK_XA_UNAME)); } } if (scheduler->localhost && (pe_find_node(scheduler->nodes, scheduler->localhost) == NULL)) { crm_info("Creating a fake local node"); pe_create_node(scheduler->localhost, scheduler->localhost, NULL, 0, scheduler); } return TRUE; } static void setup_container(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { const char *container_id = NULL; if (rsc->children) { g_list_foreach(rsc->children, (GFunc) setup_container, scheduler); return; } container_id = g_hash_table_lookup(rsc->meta, PCMK__META_CONTAINER); if (container_id && !pcmk__str_eq(container_id, rsc->id, pcmk__str_casei)) { pcmk_resource_t *container = pe_find_resource(scheduler->resources, container_id); if (container) { rsc->container = container; pcmk__set_rsc_flags(container, pcmk_rsc_has_filler); container->fillers = g_list_append(container->fillers, rsc); pcmk__rsc_trace(rsc, "Resource %s's container is %s", rsc->id, container_id); } else { pcmk__config_err("Resource %s: Unknown resource container (%s)", rsc->id, container_id); } } } gboolean unpack_remote_nodes(xmlNode *xml_resources, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; /* Create remote nodes and guest nodes from the resource configuration * before unpacking resources. */ for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { const char *new_node_id = NULL; /* Check for remote nodes, which are defined by ocf:pacemaker:remote * primitives. */ if (xml_contains_remote_node(xml_obj)) { new_node_id = pcmk__xe_id(xml_obj); /* The "pe_find_node" check is here to make sure we don't iterate over * an expanded node that has already been added to the node list. */ if (new_node_id && (pe_find_node(scheduler->nodes, new_node_id) == NULL)) { crm_trace("Found remote node %s defined by resource %s", new_node_id, pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } continue; } /* Check for guest nodes, which are defined by special meta-attributes * of a primitive of any type (for example, VirtualDomain or Xen). */ if (pcmk__xe_is(xml_obj, PCMK_XE_PRIMITIVE)) { /* This will add an ocf:pacemaker:remote primitive to the * configuration for the guest node's connection, to be unpacked * later. */ new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources, scheduler); if (new_node_id && (pe_find_node(scheduler->nodes, new_node_id) == NULL)) { crm_trace("Found guest node %s in resource %s", new_node_id, pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } continue; } /* Check for guest nodes inside a group. Clones are currently not * supported as guest nodes. */ if (pcmk__xe_is(xml_obj, PCMK_XE_GROUP)) { xmlNode *xml_obj2 = NULL; for (xml_obj2 = pcmk__xe_first_child(xml_obj); xml_obj2 != NULL; xml_obj2 = pcmk__xe_next(xml_obj2)) { new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources, scheduler); if (new_node_id && (pe_find_node(scheduler->nodes, new_node_id) == NULL)) { crm_trace("Found guest node %s in resource %s inside group %s", new_node_id, pcmk__xe_id(xml_obj2), pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } } } } return TRUE; } /* Call this after all the nodes and resources have been * unpacked, but before the status section is read. * * A remote node's online status is reflected by the state * of the remote node's connection resource. We need to link * the remote node to this connection resource so we can have * easy access to the connection resource during the scheduler calculations. */ static void link_rsc2remotenode(pcmk_scheduler_t *scheduler, pcmk_resource_t *new_rsc) { pcmk_node_t *remote_node = NULL; if (new_rsc->is_remote_node == FALSE) { return; } if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { /* remote_nodes and remote_resources are not linked in quick location calculations */ return; } remote_node = pe_find_node(scheduler->nodes, new_rsc->id); CRM_CHECK(remote_node != NULL, return); pcmk__rsc_trace(new_rsc, "Linking remote connection resource %s to %s", new_rsc->id, pcmk__node_name(remote_node)); remote_node->details->remote_rsc = new_rsc; if (new_rsc->container == NULL) { /* Handle start-up fencing for remote nodes (as opposed to guest nodes) * the same as is done for cluster nodes. */ handle_startup_fencing(scheduler, remote_node); } else { /* pe_create_node() marks the new node as "remote" or "cluster"; now * that we know the node is a guest node, update it correctly. */ pcmk__insert_dup(remote_node->details->attrs, CRM_ATTR_KIND, "container"); } } static void destroy_tag(gpointer data) { pcmk_tag_t *tag = data; if (tag) { free(tag->id); g_list_free_full(tag->refs, free); free(tag); } } /*! * \internal * \brief Parse configuration XML for resource information * * \param[in] xml_resources Top of resource configuration XML * \param[in,out] scheduler Scheduler data * * \return TRUE * * \note unpack_remote_nodes() MUST be called before this, so that the nodes can * be used when pe__unpack_resource() calls resource_location() */ gboolean unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; GList *gIter = NULL; scheduler->template_rsc_sets = pcmk__strkey_table(free, destroy_tag); for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { pcmk_resource_t *new_rsc = NULL; const char *id = pcmk__xe_id(xml_obj); if (pcmk__str_empty(id)) { pcmk__config_err("Ignoring <%s> resource without ID", xml_obj->name); continue; } if (pcmk__xe_is(xml_obj, PCMK_XE_TEMPLATE)) { if (g_hash_table_lookup_extended(scheduler->template_rsc_sets, id, NULL, NULL) == FALSE) { /* Record the template's ID for the knowledge of its existence anyway. */ pcmk__insert_dup(scheduler->template_rsc_sets, id, NULL); } continue; } crm_trace("Unpacking <%s " PCMK_XA_ID "='%s'>", xml_obj->name, id); if (pe__unpack_resource(xml_obj, &new_rsc, NULL, scheduler) == pcmk_rc_ok) { scheduler->resources = g_list_append(scheduler->resources, new_rsc); pcmk__rsc_trace(new_rsc, "Added resource %s", new_rsc->id); } else { pcmk__config_err("Ignoring <%s> resource '%s' " "because configuration is invalid", xml_obj->name, id); } } for (gIter = scheduler->resources; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; setup_container(rsc, scheduler); link_rsc2remotenode(scheduler, rsc); } scheduler->resources = g_list_sort(scheduler->resources, pe__cmp_rsc_priority); if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { /* Ignore */ } else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled) && !pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) { pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined"); pcmk__config_err("Either configure some or disable STONITH with the " PCMK_OPT_STONITH_ENABLED " option"); pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity"); } return TRUE; } gboolean unpack_tags(xmlNode *xml_tags, pcmk_scheduler_t *scheduler) { xmlNode *xml_tag = NULL; scheduler->tags = pcmk__strkey_table(free, destroy_tag); for (xml_tag = pcmk__xe_first_child(xml_tags); xml_tag != NULL; xml_tag = pcmk__xe_next(xml_tag)) { xmlNode *xml_obj_ref = NULL; const char *tag_id = pcmk__xe_id(xml_tag); if (!pcmk__xe_is(xml_tag, PCMK_XE_TAG)) { continue; } if (tag_id == NULL) { pcmk__config_err("Ignoring <%s> without " PCMK_XA_ID, (const char *) xml_tag->name); continue; } for (xml_obj_ref = pcmk__xe_first_child(xml_tag); xml_obj_ref != NULL; xml_obj_ref = pcmk__xe_next(xml_obj_ref)) { const char *obj_ref = pcmk__xe_id(xml_obj_ref); if (!pcmk__xe_is(xml_obj_ref, PCMK_XE_OBJ_REF)) { continue; } if (obj_ref == NULL) { pcmk__config_err("Ignoring <%s> for tag '%s' without " PCMK_XA_ID, xml_obj_ref->name, tag_id); continue; } if (add_tag_ref(scheduler->tags, tag_id, obj_ref) == FALSE) { return FALSE; } } } return TRUE; } /* The ticket state section: * "/cib/status/tickets/ticket_state" */ static gboolean unpack_ticket_state(xmlNode *xml_ticket, pcmk_scheduler_t *scheduler) { const char *ticket_id = NULL; const char *granted = NULL; const char *last_granted = NULL; const char *standby = NULL; xmlAttrPtr xIter = NULL; pcmk_ticket_t *ticket = NULL; ticket_id = pcmk__xe_id(xml_ticket); if (pcmk__str_empty(ticket_id)) { return FALSE; } crm_trace("Processing ticket state for %s", ticket_id); ticket = g_hash_table_lookup(scheduler->tickets, ticket_id); if (ticket == NULL) { ticket = ticket_new(ticket_id, scheduler); if (ticket == NULL) { return FALSE; } } for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = pcmk__xml_attr_value(xIter); if (pcmk__str_eq(prop_name, PCMK_XA_ID, pcmk__str_none)) { continue; } pcmk__insert_dup(ticket->state, prop_name, prop_value); } granted = g_hash_table_lookup(ticket->state, PCMK__XA_GRANTED); if (granted && crm_is_true(granted)) { ticket->granted = TRUE; crm_info("We have ticket '%s'", ticket->id); } else { ticket->granted = FALSE; crm_info("We do not have ticket '%s'", ticket->id); } last_granted = g_hash_table_lookup(ticket->state, PCMK_XA_LAST_GRANTED); if (last_granted) { long long last_granted_ll; pcmk__scan_ll(last_granted, &last_granted_ll, 0LL); ticket->last_granted = (time_t) last_granted_ll; } standby = g_hash_table_lookup(ticket->state, PCMK_XA_STANDBY); if (standby && crm_is_true(standby)) { ticket->standby = TRUE; if (ticket->granted) { crm_info("Granted ticket '%s' is in standby-mode", ticket->id); } } else { ticket->standby = FALSE; } crm_trace("Done with ticket state for %s", ticket_id); return TRUE; } static gboolean unpack_tickets_state(xmlNode *xml_tickets, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; for (xml_obj = pcmk__xe_first_child(xml_tickets); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (!pcmk__xe_is(xml_obj, PCMK__XE_TICKET_STATE)) { continue; } unpack_ticket_state(xml_obj, scheduler); } return TRUE; } static void unpack_handle_remote_attrs(pcmk_node_t *this_node, const xmlNode *state, pcmk_scheduler_t *scheduler) { const char *discovery = NULL; const xmlNode *attrs = NULL; pcmk_resource_t *rsc = NULL; if (!pcmk__xe_is(state, PCMK__XE_NODE_STATE)) { return; } if ((this_node == NULL) || !pcmk__is_pacemaker_remote_node(this_node)) { return; } crm_trace("Processing Pacemaker Remote node %s", pcmk__node_name(this_node)); pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_IN_MAINTENANCE), &(this_node->details->remote_maintenance), 0); rsc = this_node->details->remote_rsc; if (this_node->details->remote_requires_reset == FALSE) { this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; } attrs = find_xml_node(state, PCMK__XE_TRANSIENT_ATTRIBUTES, FALSE); add_node_attrs(attrs, this_node, TRUE, scheduler); if (pe__shutdown_requested(this_node)) { crm_info("%s is shutting down", pcmk__node_name(this_node)); this_node->details->shutdown = TRUE; } if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_STANDBY, NULL, pcmk__rsc_node_current))) { crm_info("%s is in standby mode", pcmk__node_name(this_node)); this_node->details->standby = TRUE; } if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_MAINTENANCE, NULL, pcmk__rsc_node_current)) || ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_managed))) { crm_info("%s is in maintenance mode", pcmk__node_name(this_node)); this_node->details->maintenance = TRUE; } discovery = pcmk__node_attr(this_node, PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED, NULL, pcmk__rsc_node_current); if ((discovery != NULL) && !crm_is_true(discovery)) { pcmk__warn_once(pcmk__wo_rdisc_enabled, "Support for the " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " node attribute is deprecated and will be removed" " (and behave as 'true') in a future release."); if (pcmk__is_remote_node(this_node) && !pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("Ignoring " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " attribute on Pacemaker Remote node %s" " because fencing is disabled", pcmk__node_name(this_node)); } else { /* This is either a remote node with fencing enabled, or a guest * node. We don't care whether fencing is enabled when fencing guest * nodes, because they are "fenced" by recovering their containing * resource. */ crm_info("%s has resource discovery disabled", pcmk__node_name(this_node)); this_node->details->rsc_discovery_enabled = FALSE; } } } /*! * \internal * \brief Unpack a cluster node's transient attributes * * \param[in] state CIB node state XML * \param[in,out] node Cluster node whose attributes are being unpacked * \param[in,out] scheduler Scheduler data */ static void unpack_transient_attributes(const xmlNode *state, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { const char *discovery = NULL; const xmlNode *attrs = find_xml_node(state, PCMK__XE_TRANSIENT_ATTRIBUTES, FALSE); add_node_attrs(attrs, node, TRUE, scheduler); if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_STANDBY, NULL, pcmk__rsc_node_current))) { crm_info("%s is in standby mode", pcmk__node_name(node)); node->details->standby = TRUE; } if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_MAINTENANCE, NULL, pcmk__rsc_node_current))) { crm_info("%s is in maintenance mode", pcmk__node_name(node)); node->details->maintenance = TRUE; } discovery = pcmk__node_attr(node, PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED, NULL, pcmk__rsc_node_current); if ((discovery != NULL) && !crm_is_true(discovery)) { pcmk__config_warn("Ignoring " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " attribute for %s because disabling resource" " discovery is not allowed for cluster nodes", pcmk__node_name(node)); } } /*! * \internal * \brief Unpack a node state entry (first pass) * * Unpack one node state entry from status. This unpacks information from the * \C PCMK__XE_NODE_STATE element itself and node attributes inside it, but not * the resource history inside it. Multiple passes through the status are needed * to fully unpack everything. * * \param[in] state CIB node state XML * \param[in,out] scheduler Scheduler data */ static void unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *uname = NULL; pcmk_node_t *this_node = NULL; id = crm_element_value(state, PCMK_XA_ID); if (id == NULL) { pcmk__config_err("Ignoring invalid " PCMK__XE_NODE_STATE " entry without " PCMK_XA_ID); crm_log_xml_info(state, "missing-id"); return; } uname = crm_element_value(state, PCMK_XA_UNAME); if (uname == NULL) { /* If a joining peer makes the cluster acquire the quorum from corosync * meanwhile it has not joined CPG membership of pacemaker-controld yet, * it's possible that the created PCMK__XE_NODE_STATE entry doesn't have * a PCMK_XA_UNAME yet. We should recognize the node as `pending` and * wait for it to join CPG. */ crm_trace("Handling " PCMK__XE_NODE_STATE " entry with id=\"%s\" " "without " PCMK_XA_UNAME, id); } this_node = pe_find_node_any(scheduler->nodes, id, uname); if (this_node == NULL) { pcmk__config_warn("Ignoring recorded node state for id=\"%s\" (%s) " "because it is no longer in the configuration", id, pcmk__s(uname, "uname unknown")); return; } if (pcmk__is_pacemaker_remote_node(this_node)) { /* We can't determine the online status of Pacemaker Remote nodes until * after all resource history has been unpacked. In this first pass, we * do need to mark whether the node has been fenced, as this plays a * role during unpacking cluster node resource state. */ pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_FENCED), &(this_node->details->remote_was_fenced), 0); return; } unpack_transient_attributes(state, this_node, scheduler); /* Provisionally mark this cluster node as clean. We have at least seen it * in the current cluster's lifetime. */ this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; crm_trace("Determining online status of cluster node %s (id %s)", pcmk__node_name(this_node), id); determine_online_status(state, this_node, scheduler); if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate) && this_node->details->online && (scheduler->no_quorum_policy == pcmk_no_quorum_fence)) { /* Everything else should flow from this automatically * (at least until the scheduler becomes able to migrate off * healthy resources) */ pe_fence_node(scheduler, this_node, "cluster does not have quorum", FALSE); } } /*! * \internal * \brief Unpack nodes' resource history as much as possible * * Unpack as many nodes' resource history as possible in one pass through the * status. We need to process Pacemaker Remote nodes' connections/containers * before unpacking their history; the connection/container history will be * in another node's history, so it might take multiple passes to unpack * everything. * * \param[in] status CIB XML status section * \param[in] fence If true, treat any not-yet-unpacked nodes as unseen * \param[in,out] scheduler Scheduler data * * \return Standard Pacemaker return code (specifically pcmk_rc_ok if done, * or EAGAIN if more unpacking remains to be done) */ static int unpack_node_history(const xmlNode *status, bool fence, pcmk_scheduler_t *scheduler) { int rc = pcmk_rc_ok; // Loop through all PCMK__XE_NODE_STATE entries in CIB status for (const xmlNode *state = first_named_child(status, PCMK__XE_NODE_STATE); state != NULL; state = crm_next_same_xml(state)) { const char *id = pcmk__xe_id(state); const char *uname = crm_element_value(state, PCMK_XA_UNAME); pcmk_node_t *this_node = NULL; if ((id == NULL) || (uname == NULL)) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history from malformed " PCMK__XE_NODE_STATE " without id and/or uname"); continue; } this_node = pe_find_node_any(scheduler->nodes, id, uname); if (this_node == NULL) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history for node %s because " "no longer in configuration", id); continue; } if (this_node->details->unpacked) { crm_trace("Not unpacking resource history for node %s because " "already unpacked", id); continue; } if (fence) { // We're processing all remaining nodes } else if (pcmk__is_guest_or_bundle_node(this_node)) { /* We can unpack a guest node's history only after we've unpacked * other resource history to the point that we know that the node's * connection and containing resource are both up. */ pcmk_resource_t *rsc = this_node->details->remote_rsc; if ((rsc == NULL) || (rsc->role != pcmk_role_started) || (rsc->container->role != pcmk_role_started)) { crm_trace("Not unpacking resource history for guest node %s " "because container and connection are not known to " "be up", id); continue; } } else if (pcmk__is_remote_node(this_node)) { /* We can unpack a remote node's history only after we've unpacked * other resource history to the point that we know that the node's * connection is up, with the exception of when shutdown locks are * in use. */ pcmk_resource_t *rsc = this_node->details->remote_rsc; if ((rsc == NULL) || (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock) && (rsc->role != pcmk_role_started))) { crm_trace("Not unpacking resource history for remote node %s " "because connection is not known to be up", id); continue; } /* If fencing and shutdown locks are disabled and we're not processing * unseen nodes, then we don't want to unpack offline nodes until online * nodes have been unpacked. This allows us to number active clone * instances first. */ } else if (!pcmk_any_flags_set(scheduler->flags, pcmk_sched_fencing_enabled |pcmk_sched_shutdown_lock) && !this_node->details->online) { crm_trace("Not unpacking resource history for offline " "cluster node %s", id); continue; } if (pcmk__is_pacemaker_remote_node(this_node)) { determine_remote_online_status(scheduler, this_node); unpack_handle_remote_attrs(this_node, state, scheduler); } crm_trace("Unpacking resource history for %snode %s", (fence? "unseen " : ""), id); this_node->details->unpacked = TRUE; unpack_node_lrm(this_node, state, scheduler); rc = EAGAIN; // Other node histories might depend on this one } return rc; } /* remove nodes that are down, stopping */ /* create positive rsc_to_node constraints between resources and the nodes they are running on */ /* anything else? */ gboolean unpack_status(xmlNode *status, pcmk_scheduler_t *scheduler) { xmlNode *state = NULL; crm_trace("Beginning unpack"); if (scheduler->tickets == NULL) { scheduler->tickets = pcmk__strkey_table(free, destroy_ticket); } for (state = pcmk__xe_first_child(status); state != NULL; state = pcmk__xe_next(state)) { if (pcmk__xe_is(state, PCMK_XE_TICKETS)) { unpack_tickets_state((xmlNode *) state, scheduler); } else if (pcmk__xe_is(state, PCMK__XE_NODE_STATE)) { unpack_node_state(state, scheduler); } } while (unpack_node_history(status, FALSE, scheduler) == EAGAIN) { crm_trace("Another pass through node resource histories is needed"); } // Now catch any nodes we didn't see unpack_node_history(status, pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled), scheduler); /* Now that we know where resources are, we can schedule stops of containers * with failed bundle connections */ if (scheduler->stop_needed != NULL) { for (GList *item = scheduler->stop_needed; item; item = item->next) { pcmk_resource_t *container = item->data; pcmk_node_t *node = pcmk__current_node(container); if (node) { stop_action(container, node, FALSE); } } g_list_free(scheduler->stop_needed); scheduler->stop_needed = NULL; } /* Now that we know status of all Pacemaker Remote connections and nodes, * we can stop connections for node shutdowns, and check the online status * of remote/guest nodes that didn't have any node history to unpack. */ for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *this_node = gIter->data; if (!pcmk__is_pacemaker_remote_node(this_node)) { continue; } if (this_node->details->shutdown && (this_node->details->remote_rsc != NULL)) { pe__set_next_role(this_node->details->remote_rsc, pcmk_role_stopped, "remote shutdown"); } if (!this_node->details->unpacked) { determine_remote_online_status(scheduler, this_node); } } return TRUE; } /*! * \internal * \brief Unpack node's time when it became a member at the cluster layer * * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry * \param[in,out] scheduler Scheduler data * * \return Epoch time when node became a cluster member * (or scheduler effective time for legacy entries) if a member, * 0 if not a member, or -1 if no valid information available */ static long long unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler) { const char *member_time = crm_element_value(node_state, PCMK__XA_IN_CCM); int member = 0; if (member_time == NULL) { return -1LL; } else if (crm_str_to_boolean(member_time, &member) == 1) { /* If in_ccm=0, we'll return 0 here. If in_ccm=1, either the entry was * recorded as a boolean for a DC < 2.1.7, or the node is pending * shutdown and has left the CPG, in which case it was set to 1 to avoid * fencing for PCMK_OPT_NODE_PENDING_TIMEOUT. * * We return the effective time for in_ccm=1 because what's important to * avoid fencing is that effective time minus this value is less than * the pending node timeout. */ return member? (long long) get_effective_time(scheduler) : 0LL; } else { long long when_member = 0LL; if ((pcmk__scan_ll(member_time, &when_member, 0LL) != pcmk_rc_ok) || (when_member < 0LL)) { crm_warn("Unrecognized value '%s' for " PCMK__XA_IN_CCM " in " PCMK__XE_NODE_STATE " entry", member_time); return -1LL; } return when_member; } } /*! * \internal * \brief Unpack node's time when it became online in process group * * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry * * \return Epoch time when node became online in process group (or 0 if not * online, or 1 for legacy online entries) */ static long long unpack_node_online(const xmlNode *node_state) { const char *peer_time = crm_element_value(node_state, PCMK_XA_CRMD); // @COMPAT Entries recorded for DCs < 2.1.7 have "online" or "offline" if (pcmk__str_eq(peer_time, PCMK_VALUE_OFFLINE, pcmk__str_casei|pcmk__str_null_matches)) { return 0LL; } else if (pcmk__str_eq(peer_time, PCMK_VALUE_ONLINE, pcmk__str_casei)) { return 1LL; } else { long long when_online = 0LL; if ((pcmk__scan_ll(peer_time, &when_online, 0LL) != pcmk_rc_ok) || (when_online < 0)) { crm_warn("Unrecognized value '%s' for " PCMK_XA_CRMD " in " PCMK__XE_NODE_STATE " entry, assuming offline", peer_time); return 0LL; } return when_online; } } /*! * \internal * \brief Unpack node attribute for user-requested fencing * * \param[in] node Node to check * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry in CIB status * * \return \c true if fencing has been requested for \p node, otherwise \c false */ static bool unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state) { long long value = 0LL; int value_i = 0; const char *value_s = pcmk__node_attr(node, PCMK_NODE_ATTR_TERMINATE, NULL, pcmk__rsc_node_current); // Value may be boolean or an epoch time if (crm_str_to_boolean(value_s, &value_i) == 1) { return (value_i != 0); } if (pcmk__scan_ll(value_s, &value, 0LL) == pcmk_rc_ok) { return (value > 0); } crm_warn("Ignoring unrecognized value '%s' for " PCMK_NODE_ATTR_TERMINATE "node attribute for %s", value_s, pcmk__node_name(node)); return false; } static gboolean determine_online_status_no_fencing(pcmk_scheduler_t *scheduler, const xmlNode *node_state, pcmk_node_t *this_node) { gboolean online = FALSE; const char *join = crm_element_value(node_state, PCMK__XA_JOIN); const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); long long when_member = unpack_node_member(node_state, scheduler); long long when_online = unpack_node_online(node_state); if (when_member <= 0) { crm_trace("Node %s is %sdown", pcmk__node_name(this_node), ((when_member < 0)? "presumed " : "")); } else if (when_online > 0) { if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { online = TRUE; } else { crm_debug("Node %s is not ready to run resources: %s", pcmk__node_name(this_node), join); } } else if (this_node->details->expected_up == FALSE) { crm_trace("Node %s controller is down: " "member@%lld online@%lld join=%s expected=%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, "")); } else { /* mark it unclean */ pe_fence_node(scheduler, this_node, "peer is unexpectedly down", FALSE); crm_info("Node %s member@%lld online@%lld join=%s expected=%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, "")); } return online; } /*! * \internal * \brief Check whether a node has taken too long to join controller group * * \param[in,out] scheduler Scheduler data * \param[in] node Node to check * \param[in] when_member Epoch time when node became a cluster member * \param[in] when_online Epoch time when node joined controller group * * \return true if node has been pending (on the way up) longer than * \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false * \note This will also update the cluster's recheck time if appropriate. */ static inline bool pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, long long when_member, long long when_online) { if ((scheduler->node_pending_timeout > 0) && (when_member > 0) && (when_online <= 0)) { // There is a timeout on pending nodes, and node is pending time_t timeout = when_member + scheduler->node_pending_timeout; if (get_effective_time(node->details->data_set) >= timeout) { return true; // Node has timed out } // Node is pending, but still has time pe__update_recheck_time(timeout, scheduler, "pending node timeout"); } return false; } static bool determine_online_status_fencing(pcmk_scheduler_t *scheduler, const xmlNode *node_state, pcmk_node_t *this_node) { bool termination_requested = unpack_node_terminate(this_node, node_state); const char *join = crm_element_value(node_state, PCMK__XA_JOIN); const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); long long when_member = unpack_node_member(node_state, scheduler); long long when_online = unpack_node_online(node_state); /* - PCMK__XA_JOIN ::= member|down|pending|banned - PCMK_XA_EXPECTED ::= member|down @COMPAT with entries recorded for DCs < 2.1.7 - PCMK__XA_IN_CCM ::= true|false - PCMK_XA_CRMD ::= online|offline Since crm_feature_set 3.18.0 (pacemaker-2.1.7): - PCMK__XA_IN_CCM ::= |0 Since when node has been a cluster member. A value 0 of means the node is not a cluster member. - PCMK_XA_CRMD ::= |0 Since when peer has been online in CPG. A value 0 means the peer is offline in CPG. */ crm_trace("Node %s member@%lld online@%lld join=%s expected=%s%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, ""), (termination_requested? " (termination requested)" : "")); if (this_node->details->shutdown) { crm_debug("%s is shutting down", pcmk__node_name(this_node)); /* Slightly different criteria since we can't shut down a dead peer */ return (when_online > 0); } if (when_member < 0) { pe_fence_node(scheduler, this_node, "peer has not been seen by the cluster", FALSE); return false; } if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_none)) { pe_fence_node(scheduler, this_node, "peer failed Pacemaker membership criteria", FALSE); } else if (termination_requested) { if ((when_member <= 0) && (when_online <= 0) && pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_none)) { crm_info("%s was fenced as requested", pcmk__node_name(this_node)); return false; } pe_fence_node(scheduler, this_node, "fencing was requested", false); } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN, pcmk__str_null_matches)) { if (pending_too_long(scheduler, this_node, when_member, when_online)) { pe_fence_node(scheduler, this_node, "peer pending timed out on joining the process group", FALSE); } else if ((when_member > 0) || (when_online > 0)) { crm_info("- %s is not ready to run resources", pcmk__node_name(this_node)); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { crm_trace("%s is down or still coming up", pcmk__node_name(this_node)); } } else if (when_member <= 0) { // Consider PCMK_OPT_PRIORITY_FENCING_DELAY for lost nodes pe_fence_node(scheduler, this_node, "peer is no longer part of the cluster", TRUE); } else if (when_online <= 0) { pe_fence_node(scheduler, this_node, "peer process is no longer available", FALSE); /* Everything is running at this point, now check join state */ } else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_none)) { crm_info("%s is active", pcmk__node_name(this_node)); } else if (pcmk__str_any_of(join, CRMD_JOINSTATE_PENDING, CRMD_JOINSTATE_DOWN, NULL)) { crm_info("%s is not ready to run resources", pcmk__node_name(this_node)); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { pe_fence_node(scheduler, this_node, "peer was in an unknown state", FALSE); } return (when_member > 0); } static void determine_remote_online_status(pcmk_scheduler_t *scheduler, pcmk_node_t *this_node) { pcmk_resource_t *rsc = this_node->details->remote_rsc; pcmk_resource_t *container = NULL; pcmk_node_t *host = NULL; /* If there is a node state entry for a (former) Pacemaker Remote node * but no resource creating that node, the node's connection resource will * be NULL. Consider it an offline remote node in that case. */ if (rsc == NULL) { this_node->details->online = FALSE; goto remote_online_done; } container = rsc->container; if (container && pcmk__list_of_1(rsc->running_on)) { host = rsc->running_on->data; } /* If the resource is currently started, mark it online. */ if (rsc->role == pcmk_role_started) { crm_trace("%s node %s presumed ONLINE because connection resource is started", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = TRUE; } /* consider this node shutting down if transitioning start->stop */ if ((rsc->role == pcmk_role_started) && (rsc->next_role == pcmk_role_stopped)) { crm_trace("%s node %s shutting down because connection resource is stopping", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->shutdown = TRUE; } /* Now check all the failure conditions. */ if(container && pcmk_is_set(container->flags, pcmk_rsc_failed)) { crm_trace("Guest node %s UNCLEAN because guest resource failed", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { crm_trace("%s node %s OFFLINE because connection resource failed", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; } else if ((rsc->role == pcmk_role_stopped) || ((container != NULL) && (container->role == pcmk_role_stopped))) { crm_trace("%s node %s OFFLINE because its resource is stopped", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = FALSE; } else if (host && (host->details->online == FALSE) && host->details->unclean) { crm_trace("Guest node %s UNCLEAN because host is unclean", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } remote_online_done: crm_trace("Remote node %s online=%s", this_node->details->id, this_node->details->online ? "TRUE" : "FALSE"); } static void determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node, pcmk_scheduler_t *scheduler) { gboolean online = FALSE; const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); CRM_CHECK(this_node != NULL, return); this_node->details->shutdown = FALSE; this_node->details->expected_up = FALSE; if (pe__shutdown_requested(this_node)) { this_node->details->shutdown = TRUE; } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { this_node->details->expected_up = TRUE; } if (this_node->details->type == node_ping) { this_node->details->unclean = FALSE; online = FALSE; /* As far as resource management is concerned, * the node is safely offline. * Anyone caught abusing this logic will be shot */ } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { online = determine_online_status_no_fencing(scheduler, node_state, this_node); } else { online = determine_online_status_fencing(scheduler, node_state, this_node); } if (online) { this_node->details->online = TRUE; } else { /* remove node from contention */ this_node->fixed = TRUE; // @COMPAT deprecated and unused - this_node->weight = -INFINITY; + this_node->weight = -PCMK_SCORE_INFINITY; } if (online && this_node->details->shutdown) { /* don't run resources here */ this_node->fixed = TRUE; // @COMPAT deprecated and unused - this_node->weight = -INFINITY; + this_node->weight = -PCMK_SCORE_INFINITY; } if (this_node->details->type == node_ping) { crm_info("%s is not a Pacemaker node", pcmk__node_name(this_node)); } else if (this_node->details->unclean) { pcmk__sched_warn("%s is unclean", pcmk__node_name(this_node)); } else if (this_node->details->online) { crm_info("%s is %s", pcmk__node_name(this_node), this_node->details->shutdown ? "shutting down" : this_node->details->pending ? "pending" : this_node->details->standby ? "standby" : this_node->details->maintenance ? "maintenance" : "online"); } else { crm_trace("%s is offline", pcmk__node_name(this_node)); } } /*! * \internal * \brief Find the end of a resource's name, excluding any clone suffix * * \param[in] id Resource ID to check * * \return Pointer to last character of resource's base name */ const char * pe_base_name_end(const char *id) { if (!pcmk__str_empty(id)) { const char *end = id + strlen(id) - 1; for (const char *s = end; s > id; --s) { switch (*s) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': break; case ':': return (s == end)? s : (s - 1); default: return end; } } return end; } return NULL; } /*! * \internal * \brief Get a resource name excluding any clone suffix * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_strip(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); char *basename = NULL; CRM_ASSERT(end); basename = strndup(last_rsc_id, end - last_rsc_id + 1); CRM_ASSERT(basename); return basename; } /*! * \internal * \brief Get the name of the first instance of a cloned resource * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name plus :0 * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_zero(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); size_t base_name_len = end - last_rsc_id + 1; char *zero = NULL; CRM_ASSERT(end); zero = calloc(base_name_len + 3, sizeof(char)); CRM_ASSERT(zero); memcpy(zero, last_rsc_id, base_name_len); zero[base_name_len] = ':'; zero[base_name_len + 1] = '0'; return zero; } static pcmk_resource_t * create_fake_resource(const char *rsc_id, const xmlNode *rsc_entry, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = NULL; xmlNode *xml_rsc = create_xml_node(NULL, PCMK_XE_PRIMITIVE); copy_in_properties(xml_rsc, rsc_entry); crm_xml_add(xml_rsc, PCMK_XA_ID, rsc_id); crm_log_xml_debug(xml_rsc, "Orphan resource"); if (pe__unpack_resource(xml_rsc, &rsc, NULL, scheduler) != pcmk_rc_ok) { return NULL; } if (xml_contains_remote_node(xml_rsc)) { pcmk_node_t *node; crm_debug("Detected orphaned remote node %s", rsc_id); node = pe_find_node(scheduler->nodes, rsc_id); if (node == NULL) { node = pe_create_node(rsc_id, rsc_id, PCMK_VALUE_REMOTE, NULL, scheduler); } link_rsc2remotenode(scheduler, rsc); if (node) { crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id); node->details->shutdown = TRUE; } } if (crm_element_value(rsc_entry, PCMK__META_CONTAINER)) { /* This orphaned rsc needs to be mapped to a container. */ crm_trace("Detected orphaned container filler %s", rsc_id); pcmk__set_rsc_flags(rsc, pcmk_rsc_removed_filler); } pcmk__set_rsc_flags(rsc, pcmk_rsc_removed); scheduler->resources = g_list_append(scheduler->resources, rsc); return rsc; } /*! * \internal * \brief Create orphan instance for anonymous clone resource history * * \param[in,out] parent Clone resource that orphan will be added to * \param[in] rsc_id Orphan's resource ID * \param[in] node Where orphan is active (for logging only) * \param[in,out] scheduler Scheduler data * * \return Newly added orphaned instance of \p parent */ static pcmk_resource_t * create_anonymous_orphan(pcmk_resource_t *parent, const char *rsc_id, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk_resource_t *top = pe__create_clone_child(parent, scheduler); // find_rsc() because we might be a cloned group pcmk_resource_t *orphan = top->fns->find_rsc(top, rsc_id, NULL, pcmk_rsc_match_clone_only); pcmk__rsc_debug(parent, "Created orphan %s for %s: %s on %s", top->id, parent->id, rsc_id, pcmk__node_name(node)); return orphan; } /*! * \internal * \brief Check a node for an instance of an anonymous clone * * Return a child instance of the specified anonymous clone, in order of * preference: (1) the instance running on the specified node, if any; * (2) an inactive instance (i.e. within the total of \c PCMK_META_CLONE_MAX * instances); (3) a newly created orphan (that is, \c PCMK_META_CLONE_MAX * instances are already active). * * \param[in,out] scheduler Scheduler data * \param[in] node Node on which to check for instance * \param[in,out] parent Clone to check * \param[in] rsc_id Name of cloned resource in history (no instance) */ static pcmk_resource_t * find_anonymous_clone(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, pcmk_resource_t *parent, const char *rsc_id) { GList *rIter = NULL; pcmk_resource_t *rsc = NULL; pcmk_resource_t *inactive_instance = NULL; gboolean skip_inactive = FALSE; CRM_ASSERT(parent != NULL); CRM_ASSERT(pcmk__is_clone(parent)); CRM_ASSERT(!pcmk_is_set(parent->flags, pcmk_rsc_unique)); // Check for active (or partially active, for cloned groups) instance pcmk__rsc_trace(parent, "Looking for %s on %s in %s", rsc_id, pcmk__node_name(node), parent->id); for (rIter = parent->children; rsc == NULL && rIter; rIter = rIter->next) { GList *locations = NULL; pcmk_resource_t *child = rIter->data; /* Check whether this instance is already known to be active or pending * anywhere, at this stage of unpacking. Because this function is called * for a resource before the resource's individual operation history * entries are unpacked, locations will generally not contain the * desired node. * * However, there are three exceptions: * (1) when child is a cloned group and we have already unpacked the * history of another member of the group on the same node; * (2) when we've already unpacked the history of another numbered * instance on the same node (which can happen if * PCMK_META_GLOBALLY_UNIQUE was flipped from true to false); and * (3) when we re-run calculations on the same scheduler data as part of * a simulation. */ child->fns->location(child, &locations, 2); if (locations) { /* We should never associate the same numbered anonymous clone * instance with multiple nodes, and clone instances can't migrate, * so there must be only one location, regardless of history. */ CRM_LOG_ASSERT(locations->next == NULL); if (pcmk__same_node((pcmk_node_t *) locations->data, node)) { /* This child instance is active on the requested node, so check * for a corresponding configured resource. We use find_rsc() * instead of child because child may be a cloned group, and we * need the particular member corresponding to rsc_id. * * If the history entry is orphaned, rsc will be NULL. */ rsc = parent->fns->find_rsc(child, rsc_id, NULL, pcmk_rsc_match_clone_only); if (rsc) { /* If there are multiple instance history entries for an * anonymous clone in a single node's history (which can * happen if PCMK_META_GLOBALLY_UNIQUE is switched from true * to false), we want to consider the instances beyond the * first as orphans, even if there are inactive instance * numbers available. */ if (rsc->running_on) { crm_notice("Active (now-)anonymous clone %s has " "multiple (orphan) instance histories on %s", parent->id, pcmk__node_name(node)); skip_inactive = TRUE; rsc = NULL; } else { pcmk__rsc_trace(parent, "Resource %s, active", rsc->id); } } } g_list_free(locations); } else { pcmk__rsc_trace(parent, "Resource %s, skip inactive", child->id); if (!skip_inactive && !inactive_instance && !pcmk_is_set(child->flags, pcmk_rsc_blocked)) { // Remember one inactive instance in case we don't find active inactive_instance = parent->fns->find_rsc(child, rsc_id, NULL, pcmk_rsc_match_clone_only); /* ... but don't use it if it was already associated with a * pending action on another node */ if ((inactive_instance != NULL) && (inactive_instance->pending_node != NULL) && !pcmk__same_node(inactive_instance->pending_node, node)) { inactive_instance = NULL; } } } } if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) { pcmk__rsc_trace(parent, "Resource %s, empty slot", inactive_instance->id); rsc = inactive_instance; } /* If the resource has PCMK_META_REQUIRES set to PCMK_VALUE_QUORUM or * PCMK_VALUE_NOTHING, and we don't have a clone instance for every node, we * don't want to consume a valid instance number for unclean nodes. Such * instances may appear to be active according to the history, but should be * considered inactive, so we can start an instance elsewhere. Treat such * instances as orphans. * * An exception is instances running on guest nodes -- since guest node * "fencing" is actually just a resource stop, requires shouldn't apply. * * @TODO Ideally, we'd use an inactive instance number if it is not needed * for any clean instances. However, we don't know that at this point. */ if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing) && (!node->details->online || node->details->unclean) && !pcmk__is_guest_or_bundle_node(node) && !pe__is_universal_clone(parent, scheduler)) { rsc = NULL; } if (rsc == NULL) { rsc = create_anonymous_orphan(parent, rsc_id, node, scheduler); pcmk__rsc_trace(parent, "Resource %s, orphan", rsc->id); } return rsc; } static pcmk_resource_t * unpack_find_resource(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, const char *rsc_id) { pcmk_resource_t *rsc = NULL; pcmk_resource_t *parent = NULL; crm_trace("looking for %s", rsc_id); rsc = pe_find_resource(scheduler->resources, rsc_id); if (rsc == NULL) { /* If we didn't find the resource by its name in the operation history, * check it again as a clone instance. Even when PCMK_META_CLONE_MAX=0, * we create a single :0 orphan to match against here. */ char *clone0_id = clone_zero(rsc_id); pcmk_resource_t *clone0 = pe_find_resource(scheduler->resources, clone0_id); if (clone0 && !pcmk_is_set(clone0->flags, pcmk_rsc_unique)) { rsc = clone0; parent = uber_parent(clone0); crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id); } else { crm_trace("%s is not known as %s either (orphan)", rsc_id, clone0_id); } free(clone0_id); } else if (rsc->variant > pcmk_rsc_variant_primitive) { crm_trace("Resource history for %s is orphaned because it is no longer primitive", rsc_id); return NULL; } else { parent = uber_parent(rsc); } if (pcmk__is_anonymous_clone(parent)) { if (pcmk__is_bundled(parent)) { rsc = pe__find_bundle_replica(parent->parent, node); } else { char *base = clone_strip(rsc_id); rsc = find_anonymous_clone(scheduler, node, parent, base); free(base); CRM_ASSERT(rsc != NULL); } } if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_casei) && !pcmk__str_eq(rsc_id, rsc->clone_name, pcmk__str_casei)) { pcmk__str_update(&rsc->clone_name, rsc_id); pcmk__rsc_debug(rsc, "Internally renamed %s on %s to %s%s", rsc_id, pcmk__node_name(node), rsc->id, pcmk_is_set(rsc->flags, pcmk_rsc_removed)? " (ORPHAN)" : ""); } return rsc; } static pcmk_resource_t * process_orphan_resource(const xmlNode *rsc_entry, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = NULL; const char *rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID); crm_debug("Detected orphan resource %s on %s", rsc_id, pcmk__node_name(node)); rsc = create_fake_resource(rsc_id, rsc_entry, scheduler); if (rsc == NULL) { return NULL; } if (!pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed); } else { CRM_CHECK(rsc != NULL, return NULL); pcmk__rsc_trace(rsc, "Added orphan %s", rsc->id); - resource_location(rsc, NULL, -INFINITY, "__orphan_do_not_run__", - scheduler); + resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, + "__orphan_do_not_run__", scheduler); } return rsc; } static void process_rsc_state(pcmk_resource_t *rsc, pcmk_node_t *node, enum action_fail_response on_fail) { pcmk_node_t *tmpnode = NULL; char *reason = NULL; enum action_fail_response save_on_fail = pcmk_on_fail_ignore; CRM_ASSERT(rsc); pcmk__rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s", rsc->id, pcmk_role_text(rsc->role), pcmk__node_name(node), pcmk_on_fail_text(on_fail)); /* process current state */ if (rsc->role != pcmk_role_unknown) { pcmk_resource_t *iter = rsc; while (iter) { if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) { pcmk_node_t *n = pe__copy_node(node); pcmk__rsc_trace(rsc, "%s%s%s known on %s", rsc->id, ((rsc->clone_name == NULL)? "" : " also known as "), ((rsc->clone_name == NULL)? "" : rsc->clone_name), pcmk__node_name(n)); g_hash_table_insert(iter->known_on, (gpointer) n->details->id, n); } if (pcmk_is_set(iter->flags, pcmk_rsc_unique)) { break; } iter = iter->parent; } } /* If a managed resource is believed to be running, but node is down ... */ if ((rsc->role > pcmk_role_stopped) && node->details->online == FALSE && node->details->maintenance == FALSE && pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { gboolean should_fence = FALSE; /* If this is a guest node, fence it (regardless of whether fencing is * enabled, because guest node fencing is done by recovery of the * container resource rather than by the fencer). Mark the resource * we're processing as failed. When the guest comes back up, its * operation history in the CIB will be cleared, freeing the affected * resource to run again once we are sure we know its state. */ if (pcmk__is_guest_or_bundle_node(node)) { pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); should_fence = TRUE; } else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { if (pcmk__is_remote_node(node) && (node->details->remote_rsc != NULL) && !pcmk_is_set(node->details->remote_rsc->flags, pcmk_rsc_failed)) { /* Setting unseen means that fencing of the remote node will * occur only if the connection resource is not going to start * somewhere. This allows connection resources on a failed * cluster node to move to another node without requiring the * remote nodes to be fenced as well. */ node->details->unseen = TRUE; reason = crm_strdup_printf("%s is active there (fencing will be" " revoked if remote connection can " "be re-established elsewhere)", rsc->id); } should_fence = TRUE; } if (should_fence) { if (reason == NULL) { reason = crm_strdup_printf("%s is thought to be active there", rsc->id); } pe_fence_node(rsc->cluster, node, reason, FALSE); } free(reason); } /* In order to calculate priority_fencing_delay correctly, save the failure information and pass it to native_add_running(). */ save_on_fail = on_fail; if (node->details->unclean) { /* No extra processing needed * Also allows resources to be started again after a node is shot */ on_fail = pcmk_on_fail_ignore; } switch (on_fail) { case pcmk_on_fail_ignore: /* nothing to do */ break; case pcmk_on_fail_demote: pcmk__set_rsc_flags(rsc, pcmk_rsc_failed); demote_action(rsc, node, FALSE); break; case pcmk_on_fail_fence_node: /* treat it as if it is still running * but also mark the node as unclean */ reason = crm_strdup_printf("%s failed there", rsc->id); pe_fence_node(rsc->cluster, node, reason, FALSE); free(reason); break; case pcmk_on_fail_standby_node: node->details->standby = TRUE; node->details->standby_onfail = TRUE; break; case pcmk_on_fail_block: /* is_managed == FALSE will prevent any * actions being sent for the resource */ pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed); pcmk__set_rsc_flags(rsc, pcmk_rsc_blocked); break; case pcmk_on_fail_ban: /* make sure it comes up somewhere else * or not at all */ - resource_location(rsc, node, -INFINITY, "__action_migration_auto__", - rsc->cluster); + resource_location(rsc, node, -PCMK_SCORE_INFINITY, + "__action_migration_auto__", rsc->cluster); break; case pcmk_on_fail_stop: pe__set_next_role(rsc, pcmk_role_stopped, PCMK_META_ON_FAIL "=" PCMK_VALUE_STOP); break; case pcmk_on_fail_restart: if ((rsc->role != pcmk_role_stopped) && (rsc->role != pcmk_role_unknown)) { pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); stop_action(rsc, node, FALSE); } break; case pcmk_on_fail_restart_container: pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); if ((rsc->container != NULL) && pcmk__is_bundled(rsc)) { /* A bundle's remote connection can run on a different node than * the bundle's container. We don't necessarily know where the * container is running yet, so remember it and add a stop * action for it later. */ rsc->cluster->stop_needed = g_list_prepend(rsc->cluster->stop_needed, rsc->container); } else if (rsc->container) { stop_action(rsc->container, node, FALSE); } else if ((rsc->role != pcmk_role_stopped) && (rsc->role != pcmk_role_unknown)) { stop_action(rsc, node, FALSE); } break; case pcmk_on_fail_reset_remote: pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { tmpnode = NULL; if (rsc->is_remote_node) { tmpnode = pe_find_node(rsc->cluster->nodes, rsc->id); } if (pcmk__is_remote_node(tmpnode) && !(tmpnode->details->remote_was_fenced)) { /* The remote connection resource failed in a way that * should result in fencing the remote node. */ pe_fence_node(rsc->cluster, tmpnode, "remote connection is unrecoverable", FALSE); } } /* require the stop action regardless if fencing is occurring or not. */ if (rsc->role > pcmk_role_stopped) { stop_action(rsc, node, FALSE); } /* if reconnect delay is in use, prevent the connection from exiting the * "STOPPED" role until the failure is cleared by the delay timeout. */ if (rsc->remote_reconnect_ms) { pe__set_next_role(rsc, pcmk_role_stopped, "remote reset"); } break; } /* ensure a remote-node connection failure forces an unclean remote-node * to be fenced. By setting unseen = FALSE, the remote-node failure will * result in a fencing operation regardless if we're going to attempt to * reconnect to the remote-node in this transition or not. */ if (pcmk_is_set(rsc->flags, pcmk_rsc_failed) && rsc->is_remote_node) { tmpnode = pe_find_node(rsc->cluster->nodes, rsc->id); if (tmpnode && tmpnode->details->unclean) { tmpnode->details->unseen = FALSE; } } if ((rsc->role != pcmk_role_stopped) && (rsc->role != pcmk_role_unknown)) { if (pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { if (pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk__config_warn("Detected active orphan %s running on %s", rsc->id, pcmk__node_name(node)); } else { pcmk__config_warn("Resource '%s' must be stopped manually on " "%s because cluster is configured not to " "stop active orphans", rsc->id, pcmk__node_name(node)); } } native_add_running(rsc, node, rsc->cluster, (save_on_fail != pcmk_on_fail_ignore)); switch (on_fail) { case pcmk_on_fail_ignore: break; case pcmk_on_fail_demote: case pcmk_on_fail_block: pcmk__set_rsc_flags(rsc, pcmk_rsc_failed); break; default: pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); break; } } else if (rsc->clone_name && strchr(rsc->clone_name, ':') != NULL) { /* Only do this for older status sections that included instance numbers * Otherwise stopped instances will appear as orphans */ pcmk__rsc_trace(rsc, "Resetting clone_name %s for %s (stopped)", rsc->clone_name, rsc->id); free(rsc->clone_name); rsc->clone_name = NULL; } else { GList *possible_matches = pe__resource_actions(rsc, node, PCMK_ACTION_STOP, FALSE); GList *gIter = possible_matches; for (; gIter != NULL; gIter = gIter->next) { pcmk_action_t *stop = (pcmk_action_t *) gIter->data; pcmk__set_action_flags(stop, pcmk_action_optional); } g_list_free(possible_matches); } /* A successful stop after migrate_to on the migration source doesn't make * the partially migrated resource stopped on the migration target. */ if ((rsc->role == pcmk_role_stopped) && rsc->partial_migration_source && rsc->partial_migration_source->details == node->details && rsc->partial_migration_target && rsc->running_on) { rsc->role = pcmk_role_started; } } /* create active recurring operations as optional */ static void process_recurring(pcmk_node_t *node, pcmk_resource_t *rsc, int start_index, int stop_index, GList *sorted_op_list, pcmk_scheduler_t *scheduler) { int counter = -1; const char *task = NULL; const char *status = NULL; GList *gIter = sorted_op_list; CRM_ASSERT(rsc); pcmk__rsc_trace(rsc, "%s: Start index %d, stop index = %d", rsc->id, start_index, stop_index); for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; guint interval_ms = 0; char *key = NULL; const char *id = pcmk__xe_id(rsc_op); counter++; if (node->details->online == FALSE) { pcmk__rsc_trace(rsc, "Skipping %s on %s: node is offline", rsc->id, pcmk__node_name(node)); break; /* Need to check if there's a monitor for role="Stopped" */ } else if (start_index < stop_index && counter <= stop_index) { pcmk__rsc_trace(rsc, "Skipping %s on %s: resource is not active", id, pcmk__node_name(node)); continue; } else if (counter < start_index) { pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d", id, pcmk__node_name(node), counter); continue; } crm_element_value_ms(rsc_op, PCMK_META_INTERVAL, &interval_ms); if (interval_ms == 0) { pcmk__rsc_trace(rsc, "Skipping %s on %s: non-recurring", id, pcmk__node_name(node)); continue; } status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS); if (pcmk__str_eq(status, "-1", pcmk__str_casei)) { pcmk__rsc_trace(rsc, "Skipping %s on %s: status", id, pcmk__node_name(node)); continue; } task = crm_element_value(rsc_op, PCMK_XA_OPERATION); /* create the action */ key = pcmk__op_key(rsc->id, task, interval_ms); pcmk__rsc_trace(rsc, "Creating %s on %s", key, pcmk__node_name(node)); custom_action(rsc, key, task, node, TRUE, scheduler); } } void calculate_active_ops(const GList *sorted_op_list, int *start_index, int *stop_index) { int counter = -1; int implied_monitor_start = -1; int implied_clone_start = -1; const char *task = NULL; const char *status = NULL; *stop_index = -1; *start_index = -1; for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) { const xmlNode *rsc_op = (const xmlNode *) iter->data; counter++; task = crm_element_value(rsc_op, PCMK_XA_OPERATION); status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS); if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_casei) && pcmk__str_eq(status, "0", pcmk__str_casei)) { *stop_index = counter; } else if (pcmk__strcase_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { *start_index = counter; } else if ((implied_monitor_start <= *stop_index) && pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { const char *rc = crm_element_value(rsc_op, PCMK__XA_RC_CODE); if (pcmk__strcase_any_of(rc, "0", "8", NULL)) { implied_monitor_start = counter; } } else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, NULL)) { implied_clone_start = counter; } } if (*start_index == -1) { if (implied_clone_start != -1) { *start_index = implied_clone_start; } else if (implied_monitor_start != -1) { *start_index = implied_monitor_start; } } } // If resource history entry has shutdown lock, remember lock node and time static void unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { time_t lock_time = 0; // When lock started (i.e. node shutdown time) if ((crm_element_value_epoch(rsc_entry, PCMK_OPT_SHUTDOWN_LOCK, &lock_time) == pcmk_ok) && (lock_time != 0)) { if ((scheduler->shutdown_lock > 0) && (get_effective_time(scheduler) > (lock_time + scheduler->shutdown_lock))) { pcmk__rsc_info(rsc, "Shutdown lock for %s on %s expired", rsc->id, pcmk__node_name(node)); pe__clear_resource_history(rsc, node); } else { /* @COMPAT I don't like breaking const signatures, but * rsc->lock_node should really be const -- we just can't change it * until the next API compatibility break. */ rsc->lock_node = (pcmk_node_t *) node; rsc->lock_time = lock_time; } } } /*! * \internal * \brief Unpack one \c PCMK__XE_LRM_RESOURCE entry from a node's CIB status * * \param[in,out] node Node whose status is being unpacked * \param[in] rsc_entry \c PCMK__XE_LRM_RESOURCE XML being unpacked * \param[in,out] scheduler Scheduler data * * \return Resource corresponding to the entry, or NULL if no operation history */ static pcmk_resource_t * unpack_lrm_resource(pcmk_node_t *node, const xmlNode *lrm_resource, pcmk_scheduler_t *scheduler) { GList *gIter = NULL; int stop_index = -1; int start_index = -1; enum rsc_role_e req_role = pcmk_role_unknown; const char *rsc_id = pcmk__xe_id(lrm_resource); pcmk_resource_t *rsc = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; xmlNode *rsc_op = NULL; xmlNode *last_failure = NULL; enum action_fail_response on_fail = pcmk_on_fail_ignore; enum rsc_role_e saved_role = pcmk_role_unknown; if (rsc_id == NULL) { pcmk__config_err("Ignoring invalid " PCMK__XE_LRM_RESOURCE " entry: No " PCMK_XA_ID); crm_log_xml_info(lrm_resource, "missing-id"); return NULL; } crm_trace("Unpacking " PCMK__XE_LRM_RESOURCE " for %s on %s", rsc_id, pcmk__node_name(node)); /* Build a list of individual PCMK__XE_LRM_RSC_OP entries, so we can sort * them */ for (rsc_op = first_named_child(lrm_resource, PCMK__XE_LRM_RSC_OP); rsc_op != NULL; rsc_op = crm_next_same_xml(rsc_op)) { op_list = g_list_prepend(op_list, rsc_op); } if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } } /* find the resource */ rsc = unpack_find_resource(scheduler, node, rsc_id); if (rsc == NULL) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } else { rsc = process_orphan_resource(lrm_resource, node, scheduler); } } CRM_ASSERT(rsc != NULL); // Check whether the resource is "shutdown-locked" to this node if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { unpack_shutdown_lock(lrm_resource, rsc, node, scheduler); } /* process operations */ saved_role = rsc->role; rsc->role = pcmk_role_unknown; sorted_op_list = g_list_sort(op_list, sort_op_by_callid); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail); } /* create active recurring operations as optional */ calculate_active_ops(sorted_op_list, &start_index, &stop_index); process_recurring(node, rsc, start_index, stop_index, sorted_op_list, scheduler); /* no need to free the contents */ g_list_free(sorted_op_list); process_rsc_state(rsc, node, on_fail); if (get_target_role(rsc, &req_role)) { if ((rsc->next_role == pcmk_role_unknown) || (req_role < rsc->next_role)) { pe__set_next_role(rsc, req_role, PCMK_META_TARGET_ROLE); } else if (req_role > rsc->next_role) { pcmk__rsc_info(rsc, "%s: Not overwriting calculated next role %s" " with requested next role %s", rsc->id, pcmk_role_text(rsc->next_role), pcmk_role_text(req_role)); } } if (saved_role > rsc->role) { rsc->role = saved_role; } return rsc; } static void handle_orphaned_container_fillers(const xmlNode *lrm_rsc_list, pcmk_scheduler_t *scheduler) { for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) { pcmk_resource_t *rsc; pcmk_resource_t *container; const char *rsc_id; const char *container_id; if (!pcmk__xe_is(rsc_entry, PCMK__XE_LRM_RESOURCE)) { continue; } container_id = crm_element_value(rsc_entry, PCMK__META_CONTAINER); rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID); if (container_id == NULL || rsc_id == NULL) { continue; } container = pe_find_resource(scheduler->resources, container_id); if (container == NULL) { continue; } rsc = pe_find_resource(scheduler->resources, rsc_id); if ((rsc == NULL) || (rsc->container != NULL) || !pcmk_is_set(rsc->flags, pcmk_rsc_removed_filler)) { continue; } pcmk__rsc_trace(rsc, "Mapped container of orphaned resource %s to %s", rsc->id, container_id); rsc->container = container; container->fillers = g_list_append(container->fillers, rsc); } } /*! * \internal * \brief Unpack one node's lrm status section * * \param[in,out] node Node whose status is being unpacked * \param[in] xml CIB node state XML * \param[in,out] scheduler Scheduler data */ static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml, pcmk_scheduler_t *scheduler) { bool found_orphaned_container_filler = false; // Drill down to PCMK__XE_LRM_RESOURCES section xml = find_xml_node(xml, PCMK__XE_LRM, FALSE); if (xml == NULL) { return; } xml = find_xml_node(xml, PCMK__XE_LRM_RESOURCES, FALSE); if (xml == NULL) { return; } // Unpack each PCMK__XE_LRM_RESOURCE entry for (const xmlNode *rsc_entry = first_named_child(xml, PCMK__XE_LRM_RESOURCE); rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) { pcmk_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, scheduler); if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk_rsc_removed_filler)) { found_orphaned_container_filler = true; } } /* Now that all resource state has been unpacked for this node, map any * orphaned container fillers to their container resource. */ if (found_orphaned_container_filler) { handle_orphaned_container_fillers(xml, scheduler); } } static void set_active(pcmk_resource_t *rsc) { const pcmk_resource_t *top = pe__const_top_resource(rsc, false); if (top && pcmk_is_set(top->flags, pcmk_rsc_promotable)) { rsc->role = pcmk_role_unpromoted; } else { rsc->role = pcmk_role_started; } } static void set_node_score(gpointer key, gpointer value, gpointer user_data) { pcmk_node_t *node = value; int *score = user_data; node->weight = *score; } #define XPATH_NODE_STATE "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE #define SUB_XPATH_LRM_RESOURCE "/" PCMK__XE_LRM \ "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE #define SUB_XPATH_LRM_RSC_OP "/" PCMK__XE_LRM_RSC_OP static xmlNode * find_lrm_op(const char *resource, const char *op, const char *node, const char *source, int target_rc, pcmk_scheduler_t *scheduler) { GString *xpath = NULL; xmlNode *xml = NULL; CRM_CHECK((resource != NULL) && (op != NULL) && (node != NULL), return NULL); xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node, "']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", resource, "']" SUB_XPATH_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='", op, "'", NULL); /* Need to check against transition_magic too? */ if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_TO) == 0)) { pcmk__g_strcat(xpath, " and @" PCMK__META_MIGRATE_TARGET "='", source, "']", NULL); } else if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) { pcmk__g_strcat(xpath, " and @" PCMK__META_MIGRATE_SOURCE "='", source, "']", NULL); } else { g_string_append_c(xpath, ']'); } xml = get_xpath_object((const char *) xpath->str, scheduler->input, LOG_DEBUG); g_string_free(xpath, TRUE); if (xml && target_rc >= 0) { int rc = PCMK_OCF_UNKNOWN_ERROR; int status = PCMK_EXEC_ERROR; crm_element_value_int(xml, PCMK__XA_RC_CODE, &rc); crm_element_value_int(xml, PCMK__XA_OP_STATUS, &status); if ((rc != target_rc) || (status != PCMK_EXEC_DONE)) { return NULL; } } return xml; } static xmlNode * find_lrm_resource(const char *rsc_id, const char *node_name, pcmk_scheduler_t *scheduler) { GString *xpath = NULL; xmlNode *xml = NULL; CRM_CHECK((rsc_id != NULL) && (node_name != NULL), return NULL); xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node_name, "']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", rsc_id, "']", NULL); xml = get_xpath_object((const char *) xpath->str, scheduler->input, LOG_DEBUG); g_string_free(xpath, TRUE); return xml; } /*! * \internal * \brief Check whether a resource has no completed action history on a node * * \param[in,out] rsc Resource to check * \param[in] node_name Node to check * * \return true if \p rsc_id is unknown on \p node_name, otherwise false */ static bool unknown_on_node(pcmk_resource_t *rsc, const char *node_name) { bool result = false; xmlXPathObjectPtr search; char *xpath = NULL; xpath = crm_strdup_printf(XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" SUB_XPATH_LRM_RSC_OP "[@" PCMK__XA_RC_CODE "!='%d']", node_name, rsc->id, PCMK_OCF_UNKNOWN); search = xpath_search(rsc->cluster->input, xpath); result = (numXpathResults(search) == 0); freeXpathObject(search); free(xpath); return result; } /*! * \brief Check whether a probe/monitor indicating the resource was not running * on a node happened after some event * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] xml_op Event that monitor is being compared to * \param[in] same_node Whether the operations are on the same node * \param[in,out] scheduler Scheduler data * * \return true if such a monitor happened after event, false otherwise */ static bool monitor_not_running_after(const char *rsc_id, const char *node_name, const xmlNode *xml_op, bool same_node, pcmk_scheduler_t *scheduler) { /* Any probe/monitor operation on the node indicating it was not running * there */ xmlNode *monitor = find_lrm_op(rsc_id, PCMK_ACTION_MONITOR, node_name, NULL, PCMK_OCF_NOT_RUNNING, scheduler); return (monitor && pe__is_newer_op(monitor, xml_op, same_node) > 0); } /*! * \brief Check whether any non-monitor operation on a node happened after some * event * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] xml_op Event that non-monitor is being compared to * \param[in] same_node Whether the operations are on the same node * \param[in,out] scheduler Scheduler data * * \return true if such a operation happened after event, false otherwise */ static bool non_monitor_after(const char *rsc_id, const char *node_name, const xmlNode *xml_op, bool same_node, pcmk_scheduler_t *scheduler) { xmlNode *lrm_resource = NULL; lrm_resource = find_lrm_resource(rsc_id, node_name, scheduler); if (lrm_resource == NULL) { return false; } for (xmlNode *op = first_named_child(lrm_resource, PCMK__XE_LRM_RSC_OP); op != NULL; op = crm_next_same_xml(op)) { const char * task = NULL; if (op == xml_op) { continue; } task = crm_element_value(op, PCMK_XA_OPERATION); if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL) && pe__is_newer_op(op, xml_op, same_node) > 0) { return true; } } return false; } /*! * \brief Check whether the resource has newer state on a node after a migration * attempt * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] migrate_to Any migrate_to event that is being compared to * \param[in] migrate_from Any migrate_from event that is being compared to * \param[in,out] scheduler Scheduler data * * \return true if such a operation happened after event, false otherwise */ static bool newer_state_after_migrate(const char *rsc_id, const char *node_name, const xmlNode *migrate_to, const xmlNode *migrate_from, pcmk_scheduler_t *scheduler) { const xmlNode *xml_op = migrate_to; const char *source = NULL; const char *target = NULL; bool same_node = false; if (migrate_from) { xml_op = migrate_from; } source = crm_element_value(xml_op, PCMK__META_MIGRATE_SOURCE); target = crm_element_value(xml_op, PCMK__META_MIGRATE_TARGET); /* It's preferred to compare to the migrate event on the same node if * existing, since call ids are more reliable. */ if (pcmk__str_eq(node_name, target, pcmk__str_casei)) { if (migrate_from) { xml_op = migrate_from; same_node = true; } else { xml_op = migrate_to; } } else if (pcmk__str_eq(node_name, source, pcmk__str_casei)) { if (migrate_to) { xml_op = migrate_to; same_node = true; } else { xml_op = migrate_from; } } /* If there's any newer non-monitor operation on the node, or any newer * probe/monitor operation on the node indicating it was not running there, * the migration events potentially no longer matter for the node. */ return non_monitor_after(rsc_id, node_name, xml_op, same_node, scheduler) || monitor_not_running_after(rsc_id, node_name, xml_op, same_node, scheduler); } /*! * \internal * \brief Parse migration source and target node names from history entry * * \param[in] entry Resource history entry for a migration action * \param[in] source_node If not NULL, source must match this node * \param[in] target_node If not NULL, target must match this node * \param[out] source_name Where to store migration source node name * \param[out] target_name Where to store migration target node name * * \return Standard Pacemaker return code */ static int get_migration_node_names(const xmlNode *entry, const pcmk_node_t *source_node, const pcmk_node_t *target_node, const char **source_name, const char **target_name) { *source_name = crm_element_value(entry, PCMK__META_MIGRATE_SOURCE); *target_name = crm_element_value(entry, PCMK__META_MIGRATE_TARGET); if ((*source_name == NULL) || (*target_name == NULL)) { pcmk__config_err("Ignoring resource history entry %s without " PCMK__META_MIGRATE_SOURCE " and " PCMK__META_MIGRATE_TARGET, pcmk__xe_id(entry)); return pcmk_rc_unpack_error; } if ((source_node != NULL) && !pcmk__str_eq(*source_name, source_node->details->uname, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_err("Ignoring resource history entry %s because " PCMK__META_MIGRATE_SOURCE "='%s' does not match %s", pcmk__xe_id(entry), *source_name, pcmk__node_name(source_node)); return pcmk_rc_unpack_error; } if ((target_node != NULL) && !pcmk__str_eq(*target_name, target_node->details->uname, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_err("Ignoring resource history entry %s because " PCMK__META_MIGRATE_TARGET "='%s' does not match %s", pcmk__xe_id(entry), *target_name, pcmk__node_name(target_node)); return pcmk_rc_unpack_error; } return pcmk_rc_ok; } /* * \internal * \brief Add a migration source to a resource's list of dangling migrations * * If the migrate_to and migrate_from actions in a live migration both * succeeded, but there is no stop on the source, the migration is considered * "dangling." Add the source to the resource's dangling migration list, which * will be used to schedule a stop on the source without affecting the target. * * \param[in,out] rsc Resource involved in migration * \param[in] node Migration source */ static void add_dangling_migration(pcmk_resource_t *rsc, const pcmk_node_t *node) { pcmk__rsc_trace(rsc, "Dangling migration of %s requires stop on %s", rsc->id, pcmk__node_name(node)); rsc->role = pcmk_role_stopped; rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, (gpointer) node); } /*! * \internal * \brief Update resource role etc. after a successful migrate_to action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_to_success(struct action_history *history) { /* A complete migration sequence is: * 1. migrate_to on source node (which succeeded if we get to this function) * 2. migrate_from on target node * 3. stop on source node * * If no migrate_from has happened, the migration is considered to be * "partial". If the migrate_from succeeded but no stop has happened, the * migration is considered to be "dangling". * * If a successful migrate_to and stop have happened on the source node, we * still need to check for a partial migration, due to scenarios (easier to * produce with batch-limit=1) like: * * - A resource is migrating from node1 to node2, and a migrate_to is * initiated for it on node1. * * - node2 goes into standby mode while the migrate_to is pending, which * aborts the transition. * * - Upon completion of the migrate_to, a new transition schedules a stop * on both nodes and a start on node1. * * - If the new transition is aborted for any reason while the resource is * stopping on node1, the transition after that stop completes will see * the migrate_to and stop on the source, but it's still a partial * migration, and the resource must be stopped on node2 because it is * potentially active there due to the migrate_to. * * We also need to take into account that either node's history may be * cleared at any point in the migration process. */ int from_rc = PCMK_OCF_OK; int from_status = PCMK_EXEC_PENDING; pcmk_node_t *target_node = NULL; xmlNode *migrate_from = NULL; const char *source = NULL; const char *target = NULL; bool source_newer_op = false; bool target_newer_state = false; bool active_on_target = false; // Get source and target node names from XML if (get_migration_node_names(history->xml, history->node, NULL, &source, &target) != pcmk_rc_ok) { return; } // Check for newer state on the source source_newer_op = non_monitor_after(history->rsc->id, source, history->xml, true, history->rsc->cluster); // Check for a migrate_from action from this source on the target migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM, target, source, -1, history->rsc->cluster); if (migrate_from != NULL) { if (source_newer_op) { /* There's a newer non-monitor operation on the source and a * migrate_from on the target, so this migrate_to is irrelevant to * the resource's state. */ return; } crm_element_value_int(migrate_from, PCMK__XA_RC_CODE, &from_rc); crm_element_value_int(migrate_from, PCMK__XA_OP_STATUS, &from_status); } /* If the resource has newer state on both the source and target after the * migration events, this migrate_to is irrelevant to the resource's state. */ target_newer_state = newer_state_after_migrate(history->rsc->id, target, history->xml, migrate_from, history->rsc->cluster); if (source_newer_op && target_newer_state) { return; } /* Check for dangling migration (migrate_from succeeded but stop not done). * We know there's no stop because we already returned if the target has a * migrate_from and the source has any newer non-monitor operation. */ if ((from_rc == PCMK_OCF_OK) && (from_status == PCMK_EXEC_DONE)) { add_dangling_migration(history->rsc, history->node); return; } /* Without newer state, this migrate_to implies the resource is active. * (Clones are not allowed to migrate, so role can't be promoted.) */ history->rsc->role = pcmk_role_started; target_node = pe_find_node(history->rsc->cluster->nodes, target); active_on_target = !target_newer_state && (target_node != NULL) && target_node->details->online; if (from_status != PCMK_EXEC_PENDING) { // migrate_from failed on target if (active_on_target) { native_add_running(history->rsc, target_node, history->rsc->cluster, TRUE); } else { // Mark resource as failed, require recovery, and prevent migration pcmk__set_rsc_flags(history->rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); pcmk__clear_rsc_flags(history->rsc, pcmk_rsc_migratable); } return; } // The migrate_from is pending, complete but erased, or to be scheduled /* If there is no history at all for the resource on an online target, then * it was likely cleaned. Just return, and we'll schedule a probe. Once we * have the probe result, it will be reflected in target_newer_state. */ if ((target_node != NULL) && target_node->details->online && unknown_on_node(history->rsc, target)) { return; } if (active_on_target) { pcmk_node_t *source_node = pe_find_node(history->rsc->cluster->nodes, source); native_add_running(history->rsc, target_node, history->rsc->cluster, FALSE); if ((source_node != NULL) && source_node->details->online) { /* This is a partial migration: the migrate_to completed * successfully on the source, but the migrate_from has not * completed. Remember the source and target; if the newly * chosen target remains the same when we schedule actions * later, we may continue with the migration. */ history->rsc->partial_migration_target = target_node; history->rsc->partial_migration_source = source_node; } } else if (!source_newer_op) { // Mark resource as failed, require recovery, and prevent migration pcmk__set_rsc_flags(history->rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); pcmk__clear_rsc_flags(history->rsc, pcmk_rsc_migratable); } } /*! * \internal * \brief Update resource role etc. after a failed migrate_to action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_to_failure(struct action_history *history) { xmlNode *target_migrate_from = NULL; const char *source = NULL; const char *target = NULL; // Get source and target node names from XML if (get_migration_node_names(history->xml, history->node, NULL, &source, &target) != pcmk_rc_ok) { return; } /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ history->rsc->role = pcmk_role_started; // Check for migrate_from on the target target_migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM, target, source, PCMK_OCF_OK, history->rsc->cluster); if (/* If the resource state is unknown on the target, it will likely be * probed there. * Don't just consider it running there. We will get back here anyway in * case the probe detects it's running there. */ !unknown_on_node(history->rsc, target) /* If the resource has newer state on the target after the migration * events, this migrate_to no longer matters for the target. */ && !newer_state_after_migrate(history->rsc->id, target, history->xml, target_migrate_from, history->rsc->cluster)) { /* The resource has no newer state on the target, so assume it's still * active there. * (if it is up). */ pcmk_node_t *target_node = pe_find_node(history->rsc->cluster->nodes, target); if (target_node && target_node->details->online) { native_add_running(history->rsc, target_node, history->rsc->cluster, FALSE); } } else if (!non_monitor_after(history->rsc->id, source, history->xml, true, history->rsc->cluster)) { /* We know the resource has newer state on the target, but this * migrate_to still matters for the source as long as there's no newer * non-monitor operation there. */ // Mark node as having dangling migration so we can force a stop later history->rsc->dangling_migrations = g_list_prepend(history->rsc->dangling_migrations, (gpointer) history->node); } } /*! * \internal * \brief Update resource role etc. after a failed migrate_from action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_from_failure(struct action_history *history) { xmlNode *source_migrate_to = NULL; const char *source = NULL; const char *target = NULL; // Get source and target node names from XML if (get_migration_node_names(history->xml, NULL, history->node, &source, &target) != pcmk_rc_ok) { return; } /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ history->rsc->role = pcmk_role_started; // Check for a migrate_to on the source source_migrate_to = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_TO, source, target, PCMK_OCF_OK, history->rsc->cluster); if (/* If the resource state is unknown on the source, it will likely be * probed there. * Don't just consider it running there. We will get back here anyway in * case the probe detects it's running there. */ !unknown_on_node(history->rsc, source) /* If the resource has newer state on the source after the migration * events, this migrate_from no longer matters for the source. */ && !newer_state_after_migrate(history->rsc->id, source, source_migrate_to, history->xml, history->rsc->cluster)) { /* The resource has no newer state on the source, so assume it's still * active there (if it is up). */ pcmk_node_t *source_node = pe_find_node(history->rsc->cluster->nodes, source); if (source_node && source_node->details->online) { native_add_running(history->rsc, source_node, history->rsc->cluster, TRUE); } } } /*! * \internal * \brief Add an action to cluster's list of failed actions * * \param[in,out] history Parsed action result history */ static void record_failed_op(struct action_history *history) { if (!(history->node->details->online)) { return; } for (const xmlNode *xIter = history->rsc->cluster->failed->children; xIter != NULL; xIter = xIter->next) { const char *key = pcmk__xe_history_key(xIter); const char *uname = crm_element_value(xIter, PCMK_XA_UNAME); if (pcmk__str_eq(history->key, key, pcmk__str_none) && pcmk__str_eq(uname, history->node->details->uname, pcmk__str_casei)) { crm_trace("Skipping duplicate entry %s on %s", history->key, pcmk__node_name(history->node)); return; } } crm_trace("Adding entry for %s on %s to failed action list", history->key, pcmk__node_name(history->node)); crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->details->uname); crm_xml_add(history->xml, PCMK__XA_RSC_ID, history->rsc->id); add_node_copy(history->rsc->cluster->failed, history->xml); } static char * last_change_str(const xmlNode *xml_op) { time_t when; char *result = NULL; if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &when) == pcmk_ok) { char *when_s = pcmk__epoch2str(&when, 0); const char *p = strchr(when_s, ' '); // Skip day of week to make message shorter if ((p != NULL) && (*(++p) != '\0')) { result = strdup(p); CRM_ASSERT(result != NULL); } free(when_s); } if (result == NULL) { result = strdup("unknown time"); CRM_ASSERT(result != NULL); } return result; } /*! * \internal * \brief Compare two on-fail values * * \param[in] first One on-fail value to compare * \param[in] second The other on-fail value to compare * * \return A negative number if second is more severe than first, zero if they * are equal, or a positive number if first is more severe than second. * \note This is only needed until the action_fail_response values can be * renumbered at the next API compatibility break. */ static int cmp_on_fail(enum action_fail_response first, enum action_fail_response second) { switch (first) { case pcmk_on_fail_demote: switch (second) { case pcmk_on_fail_ignore: return 1; case pcmk_on_fail_demote: return 0; default: return -1; } break; case pcmk_on_fail_reset_remote: switch (second) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: return 1; case pcmk_on_fail_reset_remote: return 0; default: return -1; } break; case pcmk_on_fail_restart_container: switch (second) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: case pcmk_on_fail_reset_remote: return 1; case pcmk_on_fail_restart_container: return 0; default: return -1; } break; default: break; } switch (second) { case pcmk_on_fail_demote: return (first == pcmk_on_fail_ignore)? -1 : 1; case pcmk_on_fail_reset_remote: switch (first) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: return -1; default: return 1; } break; case pcmk_on_fail_restart_container: switch (first) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: case pcmk_on_fail_reset_remote: return -1; default: return 1; } break; default: break; } return first - second; } /*! * \internal * \brief Ban a resource (or its clone if an anonymous instance) from all nodes * * \param[in,out] rsc Resource to ban */ static void ban_from_all_nodes(pcmk_resource_t *rsc) { - int score = -INFINITY; + int score = -PCMK_SCORE_INFINITY; pcmk_resource_t *fail_rsc = rsc; if (fail_rsc->parent != NULL) { pcmk_resource_t *parent = uber_parent(fail_rsc); if (pcmk__is_anonymous_clone(parent)) { /* For anonymous clones, if an operation with * PCMK_META_ON_FAIL=PCMK_VALUE_STOP fails for any instance, the * entire clone must stop. */ fail_rsc = parent; } } // Ban the resource from all nodes crm_notice("%s will not be started under current conditions", fail_rsc->id); if (fail_rsc->allowed_nodes != NULL) { g_hash_table_destroy(fail_rsc->allowed_nodes); } fail_rsc->allowed_nodes = pe__node_list2table(rsc->cluster->nodes); g_hash_table_foreach(fail_rsc->allowed_nodes, set_node_score, &score); } /*! * \internal * \brief Get configured failure handling and role after failure for an action * * \param[in,out] history Unpacked action history entry * \param[out] on_fail Where to set configured failure handling * \param[out] fail_role Where to set to role after failure */ static void unpack_failure_handling(struct action_history *history, enum action_fail_response *on_fail, enum rsc_role_e *fail_role) { xmlNode *config = pcmk__find_action_config(history->rsc, history->task, history->interval_ms, true); GHashTable *meta = pcmk__unpack_action_meta(history->rsc, history->node, history->task, history->interval_ms, config); const char *on_fail_str = g_hash_table_lookup(meta, PCMK_META_ON_FAIL); *on_fail = pcmk__parse_on_fail(history->rsc, history->task, history->interval_ms, on_fail_str); *fail_role = pcmk__role_after_failure(history->rsc, history->task, *on_fail, meta); g_hash_table_destroy(meta); } /*! * \internal * \brief Update resource role, failure handling, etc., after a failed action * * \param[in,out] history Parsed action result history * \param[in] config_on_fail Action failure handling from configuration * \param[in] fail_role Resource's role after failure of this action * \param[out] last_failure This will be set to the history XML * \param[in,out] on_fail Actual handling of action result */ static void unpack_rsc_op_failure(struct action_history *history, enum action_fail_response config_on_fail, enum rsc_role_e fail_role, xmlNode **last_failure, enum action_fail_response *on_fail) { bool is_probe = false; char *last_change_s = NULL; *last_failure = history->xml; is_probe = pcmk_xe_is_probe(history->xml); last_change_s = last_change_str(history->xml); if (!pcmk_is_set(history->rsc->cluster->flags, pcmk_sched_symmetric_cluster) && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) { crm_trace("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " CRM_XS " exit-status=%d id=%s", services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), (is_probe? "probe" : history->task), history->rsc->id, pcmk__node_name(history->node), last_change_s, history->exit_status, history->id); } else { pcmk__sched_warn("Unexpected result (%s%s%s) was recorded for %s of " "%s on %s at %s " CRM_XS " exit-status=%d id=%s", services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), (is_probe? "probe" : history->task), history->rsc->id, pcmk__node_name(history->node), last_change_s, history->exit_status, history->id); if (is_probe && (history->exit_status != PCMK_OCF_OK) && (history->exit_status != PCMK_OCF_NOT_RUNNING) && (history->exit_status != PCMK_OCF_RUNNING_PROMOTED)) { /* A failed (not just unexpected) probe result could mean the user * didn't know resources will be probed even where they can't run. */ crm_notice("If it is not possible for %s to run on %s, see " "the " PCMK_XA_RESOURCE_DISCOVERY " option for location " "constraints", history->rsc->id, pcmk__node_name(history->node)); } record_failed_op(history); } free(last_change_s); if (cmp_on_fail(*on_fail, config_on_fail) < 0) { pcmk__rsc_trace(history->rsc, "on-fail %s -> %s for %s", pcmk_on_fail_text(*on_fail), pcmk_on_fail_text(config_on_fail), history->key); *on_fail = config_on_fail; } if (strcmp(history->task, PCMK_ACTION_STOP) == 0) { - resource_location(history->rsc, history->node, -INFINITY, + resource_location(history->rsc, history->node, -PCMK_SCORE_INFINITY, "__stop_fail__", history->rsc->cluster); } else if (strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) { unpack_migrate_to_failure(history); } else if (strcmp(history->task, PCMK_ACTION_MIGRATE_FROM) == 0) { unpack_migrate_from_failure(history); } else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) { history->rsc->role = pcmk_role_promoted; } else if (strcmp(history->task, PCMK_ACTION_DEMOTE) == 0) { if (config_on_fail == pcmk_on_fail_block) { history->rsc->role = pcmk_role_promoted; pe__set_next_role(history->rsc, pcmk_role_stopped, "demote with on-fail=block"); } else if (history->exit_status == PCMK_OCF_NOT_RUNNING) { history->rsc->role = pcmk_role_stopped; } else { /* Staying in the promoted role would put the scheduler and * controller into a loop. Setting the role to unpromoted is not * dangerous because the resource will be stopped as part of * recovery, and any promotion will be ordered after that stop. */ history->rsc->role = pcmk_role_unpromoted; } } if (is_probe && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) { /* leave stopped */ pcmk__rsc_trace(history->rsc, "Leaving %s stopped", history->rsc->id); history->rsc->role = pcmk_role_stopped; } else if (history->rsc->role < pcmk_role_started) { pcmk__rsc_trace(history->rsc, "Setting %s active", history->rsc->id); set_active(history->rsc); } pcmk__rsc_trace(history->rsc, "Resource %s: role=%s unclean=%s on_fail=%s fail_role=%s", history->rsc->id, pcmk_role_text(history->rsc->role), pcmk__btoa(history->node->details->unclean), pcmk_on_fail_text(config_on_fail), pcmk_role_text(fail_role)); if ((fail_role != pcmk_role_started) && (history->rsc->next_role < fail_role)) { pe__set_next_role(history->rsc, fail_role, "failure"); } if (fail_role == pcmk_role_stopped) { ban_from_all_nodes(history->rsc); } } /*! * \internal * \brief Block a resource with a failed action if it cannot be recovered * * If resource action is a failed stop and fencing is not possible, mark the * resource as unmanaged and blocked, since recovery cannot be done. * * \param[in,out] history Parsed action history entry */ static void block_if_unrecoverable(struct action_history *history) { char *last_change_s = NULL; if (strcmp(history->task, PCMK_ACTION_STOP) != 0) { return; // All actions besides stop are always recoverable } if (pe_can_fence(history->node->details->data_set, history->node)) { return; // Failed stops are recoverable via fencing } last_change_s = last_change_str(history->xml); pcmk__sched_err("No further recovery can be attempted for %s " "because %s on %s failed (%s%s%s) at %s " CRM_XS " rc=%d id=%s", history->rsc->id, history->task, pcmk__node_name(history->node), services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), last_change_s, history->exit_status, history->id); free(last_change_s); pcmk__clear_rsc_flags(history->rsc, pcmk_rsc_managed); pcmk__set_rsc_flags(history->rsc, pcmk_rsc_blocked); } /*! * \internal * \brief Update action history's execution status and why * * \param[in,out] history Parsed action history entry * \param[out] why Where to store reason for update * \param[in] value New value * \param[in] reason Description of why value was changed */ static inline void remap_because(struct action_history *history, const char **why, int value, const char *reason) { if (history->execution_status != value) { history->execution_status = value; *why = reason; } } /*! * \internal * \brief Remap informational monitor results and operation status * * For the monitor results, certain OCF codes are for providing extended information * to the user about services that aren't yet failed but not entirely healthy either. * These must be treated as the "normal" result by Pacemaker. * * For operation status, the action result can be used to determine an appropriate * status for the purposes of responding to the action. The status provided by the * executor is not directly usable since the executor does not know what was expected. * * \param[in,out] history Parsed action history entry * \param[in,out] on_fail What should be done about the result * \param[in] expired Whether result is expired * * \note If the result is remapped and the node is not shutting down or failed, * the operation will be recorded in the scheduler data's list of failed * operations to highlight it for the user. * * \note This may update the resource's current and next role. */ static void remap_operation(struct action_history *history, enum action_fail_response *on_fail, bool expired) { bool is_probe = false; int orig_exit_status = history->exit_status; int orig_exec_status = history->execution_status; const char *why = NULL; const char *task = history->task; // Remap degraded results to their successful counterparts history->exit_status = pcmk__effective_rc(history->exit_status); if (history->exit_status != orig_exit_status) { why = "degraded result"; if (!expired && (!history->node->details->shutdown || history->node->details->online)) { record_failed_op(history); } } if (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml) && ((history->execution_status != PCMK_EXEC_DONE) || (history->exit_status != PCMK_OCF_NOT_RUNNING))) { history->execution_status = PCMK_EXEC_DONE; history->exit_status = PCMK_OCF_NOT_RUNNING; why = "equivalent probe result"; } /* If the executor reported an execution status of anything but done or * error, consider that final. But for done or error, we know better whether * it should be treated as a failure or not, because we know the expected * result. */ switch (history->execution_status) { case PCMK_EXEC_DONE: case PCMK_EXEC_ERROR: break; // These should be treated as node-fatal case PCMK_EXEC_NO_FENCE_DEVICE: case PCMK_EXEC_NO_SECRETS: remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "node-fatal error"); goto remap_done; default: goto remap_done; } is_probe = pcmk_xe_is_probe(history->xml); if (is_probe) { task = "probe"; } if (history->expected_exit_status < 0) { /* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with * Heartbeat 2.0.7 or earlier as the cluster layer, did not include the * expected exit status in the transition key, which (along with the * similar case of a corrupted transition key in the CIB) will be * reported to this function as -1. Pacemaker 2.0+ does not support * rolling upgrades from those versions or processing of saved CIB files * from those versions, so we do not need to care much about this case. */ remap_because(history, &why, PCMK_EXEC_ERROR, "obsolete history format"); pcmk__config_warn("Expected result not found for %s on %s " "(corrupt or obsolete CIB?)", history->key, pcmk__node_name(history->node)); } else if (history->exit_status == history->expected_exit_status) { remap_because(history, &why, PCMK_EXEC_DONE, "expected result"); } else { remap_because(history, &why, PCMK_EXEC_ERROR, "unexpected result"); pcmk__rsc_debug(history->rsc, "%s on %s: expected %d (%s), got %d (%s%s%s)", history->key, pcmk__node_name(history->node), history->expected_exit_status, services_ocf_exitcode_str(history->expected_exit_status), history->exit_status, services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, "")); } switch (history->exit_status) { case PCMK_OCF_OK: if (is_probe && (history->expected_exit_status == PCMK_OCF_NOT_RUNNING)) { char *last_change_s = last_change_str(history->xml); remap_because(history, &why, PCMK_EXEC_DONE, "probe"); pcmk__rsc_info(history->rsc, "Probe found %s active on %s at %s", history->rsc->id, pcmk__node_name(history->node), last_change_s); free(last_change_s); } break; case PCMK_OCF_NOT_RUNNING: if (is_probe || (history->expected_exit_status == history->exit_status) || !pcmk_is_set(history->rsc->flags, pcmk_rsc_managed)) { /* For probes, recurring monitors for the Stopped role, and * unmanaged resources, "not running" is not considered a * failure. */ remap_because(history, &why, PCMK_EXEC_DONE, "exit status"); history->rsc->role = pcmk_role_stopped; *on_fail = pcmk_on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "not running"); } break; case PCMK_OCF_RUNNING_PROMOTED: if (is_probe && (history->exit_status != history->expected_exit_status)) { char *last_change_s = last_change_str(history->xml); remap_because(history, &why, PCMK_EXEC_DONE, "probe"); pcmk__rsc_info(history->rsc, "Probe found %s active and promoted on %s at %s", history->rsc->id, pcmk__node_name(history->node), last_change_s); free(last_change_s); } if (!expired || (history->exit_status == history->expected_exit_status)) { history->rsc->role = pcmk_role_promoted; } break; case PCMK_OCF_FAILED_PROMOTED: if (!expired) { history->rsc->role = pcmk_role_promoted; } remap_because(history, &why, PCMK_EXEC_ERROR, "exit status"); break; case PCMK_OCF_NOT_CONFIGURED: remap_because(history, &why, PCMK_EXEC_ERROR_FATAL, "exit status"); break; case PCMK_OCF_UNIMPLEMENT_FEATURE: { guint interval_ms = 0; crm_element_value_ms(history->xml, PCMK_META_INTERVAL, &interval_ms); if (interval_ms == 0) { if (!expired) { block_if_unrecoverable(history); } remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status"); } else { remap_because(history, &why, PCMK_EXEC_NOT_SUPPORTED, "exit status"); } } break; case PCMK_OCF_NOT_INSTALLED: case PCMK_OCF_INVALID_PARAM: case PCMK_OCF_INSUFFICIENT_PRIV: if (!expired) { block_if_unrecoverable(history); } remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status"); break; default: if (history->execution_status == PCMK_EXEC_DONE) { char *last_change_s = last_change_str(history->xml); crm_info("Treating unknown exit status %d from %s of %s " "on %s at %s as failure", history->exit_status, task, history->rsc->id, pcmk__node_name(history->node), last_change_s); remap_because(history, &why, PCMK_EXEC_ERROR, "unknown exit status"); free(last_change_s); } break; } remap_done: if (why != NULL) { pcmk__rsc_trace(history->rsc, "Remapped %s result from [%s: %s] to [%s: %s] " "because of %s", history->key, pcmk_exec_status_str(orig_exec_status), crm_exit_str(orig_exit_status), pcmk_exec_status_str(history->execution_status), crm_exit_str(history->exit_status), why); } } // return TRUE if start or monitor last failure but parameters changed static bool should_clear_for_param_change(const xmlNode *xml_op, const char *task, pcmk_resource_t *rsc, pcmk_node_t *node) { if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) { if (pe__bundle_needs_remote_name(rsc)) { /* We haven't allocated resources yet, so we can't reliably * substitute addr parameters for the REMOTE_CONTAINER_HACK. * When that's needed, defer the check until later. */ pe__add_param_check(xml_op, rsc, node, pcmk__check_last_failure, rsc->cluster); } else { pcmk__op_digest_t *digest_data = NULL; digest_data = rsc_action_digest_cmp(rsc, xml_op, node, rsc->cluster); switch (digest_data->rc) { case pcmk__digest_unknown: crm_trace("Resource %s history entry %s on %s" " has no digest to compare", rsc->id, pcmk__xe_history_key(xml_op), node->details->id); break; case pcmk__digest_match: break; default: return TRUE; } } } return FALSE; } // Order action after fencing of remote node, given connection rsc static void order_after_remote_fencing(pcmk_action_t *action, pcmk_resource_t *remote_conn, pcmk_scheduler_t *scheduler) { pcmk_node_t *remote_node = pe_find_node(scheduler->nodes, remote_conn->id); if (remote_node) { pcmk_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL, FALSE, scheduler); order_actions(fence, action, pcmk__ar_first_implies_then); } } static bool should_ignore_failure_timeout(const pcmk_resource_t *rsc, const char *task, guint interval_ms, bool is_last_failure) { /* Clearing failures of recurring monitors has special concerns. The * executor reports only changes in the monitor result, so if the * monitor is still active and still getting the same failure result, * that will go undetected after the failure is cleared. * * Also, the operation history will have the time when the recurring * monitor result changed to the given code, not the time when the * result last happened. * * @TODO We probably should clear such failures only when the failure * timeout has passed since the last occurrence of the failed result. * However we don't record that information. We could maybe approximate * that by clearing only if there is a more recent successful monitor or * stop result, but we don't even have that information at this point * since we are still unpacking the resource's operation history. * * This is especially important for remote connection resources with a * reconnect interval, so in that case, we skip clearing failures * if the remote node hasn't been fenced. */ if (rsc->remote_reconnect_ms && pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled) && (interval_ms != 0) && pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { pcmk_node_t *remote_node = pe_find_node(rsc->cluster->nodes, rsc->id); if (remote_node && !remote_node->details->remote_was_fenced) { if (is_last_failure) { crm_info("Waiting to clear monitor failure for remote node %s" " until fencing has occurred", rsc->id); } return TRUE; } } return FALSE; } /*! * \internal * \brief Check operation age and schedule failure clearing when appropriate * * This function has two distinct purposes. The first is to check whether an * operation history entry is expired (i.e. the resource has a failure timeout, * the entry is older than the timeout, and the resource either has no fail * count or its fail count is entirely older than the timeout). The second is to * schedule fail count clearing when appropriate (i.e. the operation is expired * and either the resource has an expired fail count or the operation is a * last_failure for a remote connection resource with a reconnect interval, * or the operation is a last_failure for a start or monitor operation and the * resource's parameters have changed since the operation). * * \param[in,out] history Parsed action result history * * \return true if operation history entry is expired, otherwise false */ static bool check_operation_expiry(struct action_history *history) { bool expired = false; bool is_last_failure = pcmk__ends_with(history->id, "_last_failure_0"); time_t last_run = 0; int unexpired_fail_count = 0; const char *clear_reason = NULL; if (history->execution_status == PCMK_EXEC_NOT_INSTALLED) { pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not expired: " "Not Installed does not expire", history->id, pcmk__node_name(history->node)); return false; // "Not installed" must always be cleared manually } if ((history->rsc->failure_timeout > 0) && (crm_element_value_epoch(history->xml, PCMK_XA_LAST_RC_CHANGE, &last_run) == 0)) { /* Resource has a PCMK_META_FAILURE_TIMEOUT and history entry has a * timestamp */ time_t now = get_effective_time(history->rsc->cluster); time_t last_failure = 0; // Is this particular operation history older than the failure timeout? if ((now >= (last_run + history->rsc->failure_timeout)) && !should_ignore_failure_timeout(history->rsc, history->task, history->interval_ms, is_last_failure)) { expired = true; } // Does the resource as a whole have an unexpired fail count? unexpired_fail_count = pe_get_failcount(history->node, history->rsc, &last_failure, pcmk__fc_effective, history->xml); // Update scheduler recheck time according to *last* failure crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d timeout=%ds" " last-failure@%lld", history->id, (long long) last_run, (expired? "" : "not "), (long long) now, unexpired_fail_count, history->rsc->failure_timeout, (long long) last_failure); last_failure += history->rsc->failure_timeout + 1; if (unexpired_fail_count && (now < last_failure)) { pe__update_recheck_time(last_failure, history->rsc->cluster, "fail count expiration"); } } if (expired) { if (pe_get_failcount(history->node, history->rsc, NULL, pcmk__fc_default, history->xml)) { // There is a fail count ignoring timeout if (unexpired_fail_count == 0) { // There is no fail count considering timeout clear_reason = "it expired"; } else { /* This operation is old, but there is an unexpired fail count. * In a properly functioning cluster, this should only be * possible if this operation is not a failure (otherwise the * fail count should be expired too), so this is really just a * failsafe. */ pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not " "expired: Unexpired fail count", history->id, pcmk__node_name(history->node)); expired = false; } } else if (is_last_failure && (history->rsc->remote_reconnect_ms != 0)) { /* Clear any expired last failure when reconnect interval is set, * even if there is no fail count. */ clear_reason = "reconnect interval is set"; } } if (!expired && is_last_failure && should_clear_for_param_change(history->xml, history->task, history->rsc, history->node)) { clear_reason = "resource parameters have changed"; } if (clear_reason != NULL) { pcmk_action_t *clear_op = NULL; // Schedule clearing of the fail count clear_op = pe__clear_failcount(history->rsc, history->node, clear_reason, history->rsc->cluster); if (pcmk_is_set(history->rsc->cluster->flags, pcmk_sched_fencing_enabled) && (history->rsc->remote_reconnect_ms != 0)) { /* If we're clearing a remote connection due to a reconnect * interval, we want to wait until any scheduled fencing * completes. * * We could limit this to remote_node->details->unclean, but at * this point, that's always true (it won't be reliable until * after unpack_node_history() is done). */ crm_info("Clearing %s failure will wait until any scheduled " "fencing of %s completes", history->task, history->rsc->id); order_after_remote_fencing(clear_op, history->rsc, history->rsc->cluster); } } if (expired && (history->interval_ms == 0) && pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { switch (history->exit_status) { case PCMK_OCF_OK: case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_PROMOTED: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: // Don't expire probes that return these values pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not " "expired: Probe result", history->id, pcmk__node_name(history->node)); expired = false; break; } } return expired; } int pe__target_rc_from_xml(const xmlNode *xml_op) { int target_rc = 0; const char *key = crm_element_value(xml_op, PCMK__XA_TRANSITION_KEY); if (key == NULL) { return -1; } decode_transition_key(key, NULL, NULL, NULL, &target_rc); return target_rc; } /*! * \internal * \brief Update a resource's state for an action result * * \param[in,out] history Parsed action history entry * \param[in] exit_status Exit status to base new state on * \param[in] last_failure Resource's last_failure entry, if known * \param[in,out] on_fail Resource's current failure handling */ static void update_resource_state(struct action_history *history, int exit_status, const xmlNode *last_failure, enum action_fail_response *on_fail) { bool clear_past_failure = false; if ((exit_status == PCMK_OCF_NOT_INSTALLED) || (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml))) { history->rsc->role = pcmk_role_stopped; } else if (exit_status == PCMK_OCF_NOT_RUNNING) { clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { if ((last_failure != NULL) && pcmk__str_eq(history->key, pcmk__xe_history_key(last_failure), pcmk__str_none)) { clear_past_failure = true; } if (history->rsc->role < pcmk_role_started) { set_active(history->rsc); } } else if (pcmk__str_eq(history->task, PCMK_ACTION_START, pcmk__str_none)) { history->rsc->role = pcmk_role_started; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_STOP, pcmk__str_none)) { history->rsc->role = pcmk_role_stopped; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_PROMOTE, pcmk__str_none)) { history->rsc->role = pcmk_role_promoted; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_DEMOTE, pcmk__str_none)) { if (*on_fail == pcmk_on_fail_demote) { /* Demote clears an error only if * PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE */ clear_past_failure = true; } history->rsc->role = pcmk_role_unpromoted; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_FROM, pcmk__str_none)) { history->rsc->role = pcmk_role_started; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)) { unpack_migrate_to_success(history); } else if (history->rsc->role < pcmk_role_started) { pcmk__rsc_trace(history->rsc, "%s active on %s", history->rsc->id, pcmk__node_name(history->node)); set_active(history->rsc); } if (!clear_past_failure) { return; } switch (*on_fail) { case pcmk_on_fail_stop: case pcmk_on_fail_ban: case pcmk_on_fail_standby_node: case pcmk_on_fail_fence_node: pcmk__rsc_trace(history->rsc, "%s (%s) is not cleared by a completed %s", history->rsc->id, pcmk_on_fail_text(*on_fail), history->task); break; case pcmk_on_fail_block: case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: case pcmk_on_fail_restart_container: *on_fail = pcmk_on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "clear past failures"); break; case pcmk_on_fail_reset_remote: if (history->rsc->remote_reconnect_ms == 0) { /* With no reconnect interval, the connection is allowed to * start again after the remote node is fenced and * completely stopped. (With a reconnect interval, we wait * for the failure to be cleared entirely before attempting * to reconnect.) */ *on_fail = pcmk_on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "clear past failures and reset remote"); } break; } } /*! * \internal * \brief Check whether a given history entry matters for resource state * * \param[in] history Parsed action history entry * * \return true if action can affect resource state, otherwise false */ static inline bool can_affect_state(struct action_history *history) { #if 0 /* @COMPAT It might be better to parse only actions we know we're interested * in, rather than exclude a couple we don't. However that would be a * behavioral change that should be done at a major or minor series release. * Currently, unknown operations can affect whether a resource is considered * active and/or failed. */ return pcmk__str_any_of(history->task, PCMK_ACTION_MONITOR, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, "asyncmon", NULL); #else return !pcmk__str_any_of(history->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_META_DATA, NULL); #endif } /*! * \internal * \brief Unpack execution/exit status and exit reason from a history entry * * \param[in,out] history Action history entry to unpack * * \return Standard Pacemaker return code */ static int unpack_action_result(struct action_history *history) { if ((crm_element_value_int(history->xml, PCMK__XA_OP_STATUS, &(history->execution_status)) < 0) || (history->execution_status < PCMK_EXEC_PENDING) || (history->execution_status > PCMK_EXEC_MAX) || (history->execution_status == PCMK_EXEC_CANCELLED)) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "with invalid " PCMK__XA_OP_STATUS " '%s'", history->id, history->rsc->id, pcmk__node_name(history->node), pcmk__s(crm_element_value(history->xml, PCMK__XA_OP_STATUS), "")); return pcmk_rc_unpack_error; } if ((crm_element_value_int(history->xml, PCMK__XA_RC_CODE, &(history->exit_status)) < 0) || (history->exit_status < 0) || (history->exit_status > CRM_EX_MAX)) { #if 0 /* @COMPAT We should ignore malformed entries, but since that would * change behavior, it should be done at a major or minor series * release. */ pcmk__config_err("Ignoring resource history entry %s for %s on %s " "with invalid " PCMK__XA_RC_CODE " '%s'", history->id, history->rsc->id, pcmk__node_name(history->node), pcmk__s(crm_element_value(history->xml, PCMK__XA_RC_CODE), "")); return pcmk_rc_unpack_error; #else history->exit_status = CRM_EX_ERROR; #endif } history->exit_reason = crm_element_value(history->xml, PCMK_XA_EXIT_REASON); return pcmk_rc_ok; } /*! * \internal * \brief Process an action history entry whose result expired * * \param[in,out] history Parsed action history entry * \param[in] orig_exit_status Action exit status before remapping * * \return Standard Pacemaker return code (in particular, pcmk_rc_ok means the * entry needs no further processing) */ static int process_expired_result(struct action_history *history, int orig_exit_status) { if (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml) && (orig_exit_status != history->expected_exit_status)) { if (history->rsc->role <= pcmk_role_stopped) { history->rsc->role = pcmk_role_unknown; } crm_trace("Ignoring resource history entry %s for probe of %s on %s: " "Masked failure expired", history->id, history->rsc->id, pcmk__node_name(history->node)); return pcmk_rc_ok; } if (history->exit_status == history->expected_exit_status) { return pcmk_rc_undetermined; // Only failures expire } if (history->interval_ms == 0) { crm_notice("Ignoring resource history entry %s for %s of %s on %s: " "Expired failure", history->id, history->task, history->rsc->id, pcmk__node_name(history->node)); return pcmk_rc_ok; } if (history->node->details->online && !history->node->details->unclean) { /* Reschedule the recurring action. schedule_cancel() won't work at * this stage, so as a hacky workaround, forcibly change the restart * digest so pcmk__check_action_config() does what we want later. * * @TODO We should skip this if there is a newer successful monitor. * Also, this causes rescheduling only if the history entry * has a PCMK__XA_OP_DIGEST (which the expire-non-blocked-failure * scheduler regression test doesn't, but that may not be a * realistic scenario in production). */ crm_notice("Rescheduling %s-interval %s of %s on %s " "after failure expired", pcmk__readable_interval(history->interval_ms), history->task, history->rsc->id, pcmk__node_name(history->node)); crm_xml_add(history->xml, PCMK__XA_OP_RESTART_DIGEST, "calculated-failure-timeout"); return pcmk_rc_ok; } return pcmk_rc_undetermined; } /*! * \internal * \brief Process a masked probe failure * * \param[in,out] history Parsed action history entry * \param[in] orig_exit_status Action exit status before remapping * \param[in] last_failure Resource's last_failure entry, if known * \param[in,out] on_fail Resource's current failure handling */ static void mask_probe_failure(struct action_history *history, int orig_exit_status, const xmlNode *last_failure, enum action_fail_response *on_fail) { pcmk_resource_t *ban_rsc = history->rsc; if (!pcmk_is_set(history->rsc->flags, pcmk_rsc_unique)) { ban_rsc = uber_parent(history->rsc); } crm_notice("Treating probe result '%s' for %s on %s as 'not running'", services_ocf_exitcode_str(orig_exit_status), history->rsc->id, pcmk__node_name(history->node)); update_resource_state(history, history->expected_exit_status, last_failure, on_fail); crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->details->uname); record_failed_op(history); - resource_location(ban_rsc, history->node, -INFINITY, "masked-probe-failure", - history->rsc->cluster); + resource_location(ban_rsc, history->node, -PCMK_SCORE_INFINITY, + "masked-probe-failure", history->rsc->cluster); } /*! * \internal Check whether a given failure is for a given pending action * * \param[in] history Parsed history entry for pending action * \param[in] last_failure Resource's last_failure entry, if known * * \return true if \p last_failure is failure of pending action in \p history, * otherwise false * \note Both \p history and \p last_failure must come from the same * \c PCMK__XE_LRM_RESOURCE block, as node and resource are assumed to be * the same. */ static bool failure_is_newer(const struct action_history *history, const xmlNode *last_failure) { guint failure_interval_ms = 0U; long long failure_change = 0LL; long long this_change = 0LL; if (last_failure == NULL) { return false; // Resource has no last_failure entry } if (!pcmk__str_eq(history->task, crm_element_value(last_failure, PCMK_XA_OPERATION), pcmk__str_none)) { return false; // last_failure is for different action } if ((crm_element_value_ms(last_failure, PCMK_META_INTERVAL, &failure_interval_ms) != pcmk_ok) || (history->interval_ms != failure_interval_ms)) { return false; // last_failure is for action with different interval } if ((pcmk__scan_ll(crm_element_value(history->xml, PCMK_XA_LAST_RC_CHANGE), &this_change, 0LL) != pcmk_rc_ok) || (pcmk__scan_ll(crm_element_value(last_failure, PCMK_XA_LAST_RC_CHANGE), &failure_change, 0LL) != pcmk_rc_ok) || (failure_change < this_change)) { return false; // Failure is not known to be newer } return true; } /*! * \internal * \brief Update a resource's role etc. for a pending action * * \param[in,out] history Parsed history entry for pending action * \param[in] last_failure Resource's last_failure entry, if known */ static void process_pending_action(struct action_history *history, const xmlNode *last_failure) { /* For recurring monitors, a failure is recorded only in RSC_last_failure_0, * and there might be a RSC_monitor_INTERVAL entry with the last successful * or pending result. * * If last_failure contains the failure of the pending recurring monitor * we're processing here, and is newer, the action is no longer pending. * (Pending results have call ID -1, which sorts last, so the last failure * if any should be known.) */ if (failure_is_newer(history, last_failure)) { return; } if (strcmp(history->task, PCMK_ACTION_START) == 0) { pcmk__set_rsc_flags(history->rsc, pcmk_rsc_start_pending); set_active(history->rsc); } else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) { history->rsc->role = pcmk_role_promoted; } else if ((strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) && history->node->details->unclean) { /* A migrate_to action is pending on a unclean source, so force a stop * on the target. */ const char *migrate_target = NULL; pcmk_node_t *target = NULL; migrate_target = crm_element_value(history->xml, PCMK__META_MIGRATE_TARGET); target = pe_find_node(history->rsc->cluster->nodes, migrate_target); if (target != NULL) { stop_action(history->rsc, target, FALSE); } } if (history->rsc->pending_task != NULL) { /* There should never be multiple pending actions, but as a failsafe, * just remember the first one processed for display purposes. */ return; } if (pcmk_is_probe(history->task, history->interval_ms)) { /* Pending probes are currently never displayed, even if pending * operations are requested. If we ever want to change that, * enable the below and the corresponding part of * native.c:native_pending_task(). */ #if 0 history->rsc->pending_task = strdup("probe"); history->rsc->pending_node = history->node; #endif } else { history->rsc->pending_task = strdup(history->task); history->rsc->pending_node = history->node; } } static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *on_fail) { int old_rc = 0; bool expired = false; pcmk_resource_t *parent = rsc; enum rsc_role_e fail_role = pcmk_role_unknown; enum action_fail_response failure_strategy = pcmk_on_fail_restart; struct action_history history = { .rsc = rsc, .node = node, .xml = xml_op, .execution_status = PCMK_EXEC_UNKNOWN, }; CRM_CHECK(rsc && node && xml_op, return); history.id = pcmk__xe_id(xml_op); if (history.id == NULL) { pcmk__config_err("Ignoring resource history entry for %s on %s " "without ID", rsc->id, pcmk__node_name(node)); return; } // Task and interval history.task = crm_element_value(xml_op, PCMK_XA_OPERATION); if (history.task == NULL) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "without " PCMK_XA_OPERATION, history.id, rsc->id, pcmk__node_name(node)); return; } crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &(history.interval_ms)); if (!can_affect_state(&history)) { pcmk__rsc_trace(rsc, "Ignoring resource history entry %s for %s on %s " "with irrelevant action '%s'", history.id, rsc->id, pcmk__node_name(node), history.task); return; } if (unpack_action_result(&history) != pcmk_rc_ok) { return; // Error already logged } history.expected_exit_status = pe__target_rc_from_xml(xml_op); history.key = pcmk__xe_history_key(xml_op); crm_element_value_int(xml_op, PCMK__XA_CALL_ID, &(history.call_id)); pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)", history.id, history.task, history.call_id, pcmk__node_name(node), pcmk_exec_status_str(history.execution_status), crm_exit_str(history.exit_status)); if (node->details->unclean) { pcmk__rsc_trace(rsc, "%s is running on %s, which is unclean (further action " "depends on value of stop's on-fail attribute)", rsc->id, pcmk__node_name(node)); } expired = check_operation_expiry(&history); old_rc = history.exit_status; remap_operation(&history, on_fail, expired); if (expired && (process_expired_result(&history, old_rc) == pcmk_rc_ok)) { goto done; } if (!pcmk__is_bundled(rsc) && pcmk_xe_mask_probe_failure(xml_op)) { mask_probe_failure(&history, old_rc, *last_failure, on_fail); goto done; } if (!pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { parent = uber_parent(rsc); } switch (history.execution_status) { case PCMK_EXEC_PENDING: process_pending_action(&history, *last_failure); goto done; case PCMK_EXEC_DONE: update_resource_state(&history, history.exit_status, *last_failure, on_fail); goto done; case PCMK_EXEC_NOT_INSTALLED: unpack_failure_handling(&history, &failure_strategy, &fail_role); if (failure_strategy == pcmk_on_fail_ignore) { crm_warn("Cannot ignore failed %s of %s on %s: " "Resource agent doesn't exist " CRM_XS " status=%d rc=%d id=%s", history.task, rsc->id, pcmk__node_name(node), history.execution_status, history.exit_status, history.id); /* Also for printing it as "FAILED" by marking it as * pcmk_rsc_failed later */ *on_fail = pcmk_on_fail_ban; } - resource_location(parent, node, -INFINITY, "hard-error", - rsc->cluster); + resource_location(parent, node, -PCMK_SCORE_INFINITY, + "hard-error", rsc->cluster); unpack_rsc_op_failure(&history, failure_strategy, fail_role, last_failure, on_fail); goto done; case PCMK_EXEC_NOT_CONNECTED: if (pcmk__is_pacemaker_remote_node(node) && pcmk_is_set(node->details->remote_rsc->flags, pcmk_rsc_managed)) { /* We should never get into a situation where a managed remote * connection resource is considered OK but a resource action * behind the connection gets a "not connected" status. But as a * fail-safe in case a bug or unusual circumstances do lead to * that, ensure the remote connection is considered failed. */ pcmk__set_rsc_flags(node->details->remote_rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); } break; // Not done, do error handling case PCMK_EXEC_ERROR: case PCMK_EXEC_ERROR_HARD: case PCMK_EXEC_ERROR_FATAL: case PCMK_EXEC_TIMEOUT: case PCMK_EXEC_NOT_SUPPORTED: case PCMK_EXEC_INVALID: break; // Not done, do error handling default: // No other value should be possible at this point break; } unpack_failure_handling(&history, &failure_strategy, &fail_role); if ((failure_strategy == pcmk_on_fail_ignore) || ((failure_strategy == pcmk_on_fail_restart_container) && (strcmp(history.task, PCMK_ACTION_STOP) == 0))) { char *last_change_s = last_change_str(xml_op); crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s succeeded " CRM_XS " %s", history.task, services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), rsc->id, pcmk__node_name(node), last_change_s, history.id); free(last_change_s); update_resource_state(&history, history.expected_exit_status, *last_failure, on_fail); crm_xml_add(xml_op, PCMK_XA_UNAME, node->details->uname); pcmk__set_rsc_flags(rsc, pcmk_rsc_ignore_failure); record_failed_op(&history); if ((failure_strategy == pcmk_on_fail_restart_container) && cmp_on_fail(*on_fail, pcmk_on_fail_restart) <= 0) { *on_fail = failure_strategy; } } else { unpack_rsc_op_failure(&history, failure_strategy, fail_role, last_failure, on_fail); if (history.execution_status == PCMK_EXEC_ERROR_HARD) { uint8_t log_level = LOG_ERR; if (history.exit_status == PCMK_OCF_NOT_INSTALLED) { log_level = LOG_NOTICE; } do_crm_log(log_level, "Preventing %s from restarting on %s because " "of hard failure (%s%s%s) " CRM_XS " %s", parent->id, pcmk__node_name(node), services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), history.id); - resource_location(parent, node, -INFINITY, "hard-error", - rsc->cluster); + resource_location(parent, node, -PCMK_SCORE_INFINITY, + "hard-error", rsc->cluster); } else if (history.execution_status == PCMK_EXEC_ERROR_FATAL) { pcmk__sched_err("Preventing %s from restarting anywhere because " "of fatal failure (%s%s%s) " CRM_XS " %s", parent->id, services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), history.id); - resource_location(parent, NULL, -INFINITY, "fatal-error", - rsc->cluster); + resource_location(parent, NULL, -PCMK_SCORE_INFINITY, + "fatal-error", rsc->cluster); } } done: pcmk__rsc_trace(rsc, "%s role on %s after %s is %s (next %s)", rsc->id, pcmk__node_name(node), history.id, pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role)); } static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite, pcmk_scheduler_t *scheduler) { const char *cluster_name = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; pcmk__insert_dup(node->details->attrs, CRM_ATTR_UNAME, node->details->uname); pcmk__insert_dup(node->details->attrs, CRM_ATTR_ID, node->details->id); if (pcmk__str_eq(node->details->id, scheduler->dc_uuid, pcmk__str_casei)) { scheduler->dc_node = node; node->details->is_dc = TRUE; pcmk__insert_dup(node->details->attrs, CRM_ATTR_IS_DC, PCMK_VALUE_TRUE); } else { pcmk__insert_dup(node->details->attrs, CRM_ATTR_IS_DC, PCMK_VALUE_FALSE); } cluster_name = g_hash_table_lookup(scheduler->config_hash, PCMK_OPT_CLUSTER_NAME); if (cluster_name) { pcmk__insert_dup(node->details->attrs, CRM_ATTR_CLUSTER_NAME, cluster_name); } pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, node->details->attrs, NULL, overwrite, scheduler); pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_UTILIZATION, &rule_data, node->details->utilization, NULL, FALSE, scheduler); if (pcmk__node_attr(node, CRM_ATTR_SITE_NAME, NULL, pcmk__rsc_node_current) == NULL) { const char *site_name = pcmk__node_attr(node, "site-name", NULL, pcmk__rsc_node_current); if (site_name) { pcmk__insert_dup(node->details->attrs, CRM_ATTR_SITE_NAME, site_name); } else if (cluster_name) { /* Default to cluster-name if unset */ pcmk__insert_dup(node->details->attrs, CRM_ATTR_SITE_NAME, cluster_name); } } } static GList * extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter) { int counter = -1; int stop_index = -1; int start_index = -1; xmlNode *rsc_op = NULL; GList *gIter = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; /* extract operations */ op_list = NULL; sorted_op_list = NULL; for (rsc_op = pcmk__xe_first_child(rsc_entry); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) { if (pcmk__xe_is(rsc_op, PCMK__XE_LRM_RSC_OP)) { crm_xml_add(rsc_op, PCMK_XA_RESOURCE, rsc); crm_xml_add(rsc_op, PCMK_XA_UNAME, node); op_list = g_list_prepend(op_list, rsc_op); } } if (op_list == NULL) { /* if there are no operations, there is nothing to do */ return NULL; } sorted_op_list = g_list_sort(op_list, sort_op_by_callid); /* create active recurring operations as optional */ if (active_filter == FALSE) { return sorted_op_list; } op_list = NULL; calculate_active_ops(sorted_op_list, &start_index, &stop_index); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; if (start_index < stop_index) { crm_trace("Skipping %s: not active", pcmk__xe_id(rsc_entry)); break; } else if (counter < start_index) { crm_trace("Skipping %s: old", pcmk__xe_id(rsc_op)); continue; } op_list = g_list_append(op_list, rsc_op); } g_list_free(sorted_op_list); return op_list; } GList * find_operations(const char *rsc, const char *node, gboolean active_filter, pcmk_scheduler_t *scheduler) { GList *output = NULL; GList *intermediate = NULL; xmlNode *tmp = NULL; xmlNode *status = find_xml_node(scheduler->input, PCMK_XE_STATUS, TRUE); pcmk_node_t *this_node = NULL; xmlNode *node_state = NULL; for (node_state = pcmk__xe_first_child(status); node_state != NULL; node_state = pcmk__xe_next(node_state)) { if (pcmk__xe_is(node_state, PCMK__XE_NODE_STATE)) { const char *uname = crm_element_value(node_state, PCMK_XA_UNAME); if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) { continue; } this_node = pe_find_node(scheduler->nodes, uname); if(this_node == NULL) { CRM_LOG_ASSERT(this_node != NULL); continue; } else if (pcmk__is_pacemaker_remote_node(this_node)) { determine_remote_online_status(scheduler, this_node); } else { determine_online_status(node_state, this_node, scheduler); } if (this_node->details->online || pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { /* offline nodes run no resources... * unless stonith is enabled in which case we need to * make sure rsc start events happen after the stonith */ xmlNode *lrm_rsc = NULL; tmp = find_xml_node(node_state, PCMK__XE_LRM, FALSE); tmp = find_xml_node(tmp, PCMK__XE_LRM_RESOURCES, FALSE); for (lrm_rsc = pcmk__xe_first_child(tmp); lrm_rsc != NULL; lrm_rsc = pcmk__xe_next(lrm_rsc)) { if (pcmk__xe_is(lrm_rsc, PCMK__XE_LRM_RESOURCE)) { const char *rsc_id = crm_element_value(lrm_rsc, PCMK_XA_ID); if (rsc != NULL && !pcmk__str_eq(rsc_id, rsc, pcmk__str_casei)) { continue; } intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter); output = g_list_concat(output, intermediate); } } } } } return output; } diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c index ed9612a76e..fad44589ca 100644 --- a/lib/pengine/utils.c +++ b/lib/pengine/utils.c @@ -1,918 +1,918 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include "pe_status_private.h" extern bool pcmk__is_daemon; gboolean ghash_free_str_str(gpointer key, gpointer value, gpointer user_data); /*! * \internal * \brief Check whether we can fence a particular node * * \param[in] scheduler Scheduler data * \param[in] node Name of node to check * * \return true if node can be fenced, false otherwise */ bool pe_can_fence(const pcmk_scheduler_t *scheduler, const pcmk_node_t *node) { if (pcmk__is_guest_or_bundle_node(node)) { /* Guest nodes are fenced by stopping their container resource. We can * do that if the container's host is either online or fenceable. */ pcmk_resource_t *rsc = node->details->remote_rsc->container; for (GList *n = rsc->running_on; n != NULL; n = n->next) { pcmk_node_t *container_node = n->data; if (!container_node->details->online && !pe_can_fence(scheduler, container_node)) { return false; } } return true; } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { return false; /* Turned off */ } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) { return false; /* No devices */ } else if (pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) { return true; } else if (scheduler->no_quorum_policy == pcmk_no_quorum_ignore) { return true; } else if(node == NULL) { return false; } else if(node->details->online) { crm_notice("We can fence %s without quorum because they're in our membership", pcmk__node_name(node)); return true; } crm_trace("Cannot fence %s", pcmk__node_name(node)); return false; } /*! * \internal * \brief Copy a node object * * \param[in] this_node Node object to copy * * \return Newly allocated shallow copy of this_node * \note This function asserts on errors and is guaranteed to return non-NULL. */ pcmk_node_t * pe__copy_node(const pcmk_node_t *this_node) { pcmk_node_t *new_node = NULL; CRM_ASSERT(this_node != NULL); new_node = calloc(1, sizeof(pcmk_node_t)); CRM_ASSERT(new_node != NULL); new_node->rsc_discover_mode = this_node->rsc_discover_mode; new_node->weight = this_node->weight; new_node->fixed = this_node->fixed; // @COMPAT deprecated and unused new_node->count = this_node->count; new_node->details = this_node->details; return new_node; } /*! * \internal * \brief Create a node hash table from a node list * * \param[in] list Node list * * \return Hash table equivalent of node list */ GHashTable * pe__node_list2table(const GList *list) { GHashTable *result = NULL; result = pcmk__strkey_table(NULL, free); for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) { pcmk_node_t *new_node = NULL; new_node = pe__copy_node((const pcmk_node_t *) gIter->data); g_hash_table_insert(result, (gpointer) new_node->details->id, new_node); } return result; } /*! * \internal * \brief Compare two nodes by name, with numeric portions sorted numerically * * Sort two node names case-insensitively like strcasecmp(), but with any * numeric portions of the name sorted numerically. For example, "node10" will * sort higher than "node9" but lower than "remotenode9". * * \param[in] a First node to compare (can be \c NULL) * \param[in] b Second node to compare (can be \c NULL) * * \retval -1 \c a comes before \c b (or \c a is \c NULL and \c b is not) * \retval 0 \c a and \c b are equal (or both are \c NULL) * \retval 1 \c a comes after \c b (or \c b is \c NULL and \c a is not) */ gint pe__cmp_node_name(gconstpointer a, gconstpointer b) { const pcmk_node_t *node1 = (const pcmk_node_t *) a; const pcmk_node_t *node2 = (const pcmk_node_t *) b; if ((node1 == NULL) && (node2 == NULL)) { return 0; } if (node1 == NULL) { return -1; } if (node2 == NULL) { return 1; } return pcmk__numeric_strcasecmp(node1->details->uname, node2->details->uname); } /*! * \internal * \brief Output node weights to stdout * * \param[in] rsc Use allowed nodes for this resource * \param[in] comment Text description to prefix lines with * \param[in] nodes If rsc is not specified, use these nodes * \param[in,out] scheduler Scheduler data */ static void pe__output_node_weights(const pcmk_resource_t *rsc, const char *comment, GHashTable *nodes, pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv; // Sort the nodes so the output is consistent for regression tests GList *list = g_list_sort(g_hash_table_get_values(nodes), pe__cmp_node_name); for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) { const pcmk_node_t *node = (const pcmk_node_t *) gIter->data; out->message(out, "node-weight", rsc, comment, node->details->uname, pcmk_readable_score(node->weight)); } g_list_free(list); } /*! * \internal * \brief Log node weights at trace level * * \param[in] file Caller's filename * \param[in] function Caller's function name * \param[in] line Caller's line number * \param[in] rsc If not NULL, include this resource's ID in logs * \param[in] comment Text description to prefix lines with * \param[in] nodes Nodes whose scores should be logged */ static void pe__log_node_weights(const char *file, const char *function, int line, const pcmk_resource_t *rsc, const char *comment, GHashTable *nodes) { GHashTableIter iter; pcmk_node_t *node = NULL; // Don't waste time if we're not tracing at this point pcmk__if_tracing({}, return); g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (rsc) { qb_log_from_external_source(function, file, "%s: %s allocation score on %s: %s", LOG_TRACE, line, 0, comment, rsc->id, pcmk__node_name(node), pcmk_readable_score(node->weight)); } else { qb_log_from_external_source(function, file, "%s: %s = %s", LOG_TRACE, line, 0, comment, pcmk__node_name(node), pcmk_readable_score(node->weight)); } } } /*! * \internal * \brief Log or output node weights * * \param[in] file Caller's filename * \param[in] function Caller's function name * \param[in] line Caller's line number * \param[in] to_log Log if true, otherwise output * \param[in] rsc If not NULL, use this resource's ID in logs, * and show scores recursively for any children * \param[in] comment Text description to prefix lines with * \param[in] nodes Nodes whose scores should be shown * \param[in,out] scheduler Scheduler data */ void pe__show_node_scores_as(const char *file, const char *function, int line, bool to_log, const pcmk_resource_t *rsc, const char *comment, GHashTable *nodes, pcmk_scheduler_t *scheduler) { if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { // Don't show allocation scores for orphans return; } if (nodes == NULL) { // Nothing to show return; } if (to_log) { pe__log_node_weights(file, function, line, rsc, comment, nodes); } else { pe__output_node_weights(rsc, comment, nodes, scheduler); } // If this resource has children, repeat recursively for each if (rsc && rsc->children) { for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = (pcmk_resource_t *) gIter->data; pe__show_node_scores_as(file, function, line, to_log, child, comment, child->allowed_nodes, scheduler); } } } /*! * \internal * \brief Compare two resources by priority * * \param[in] a First resource to compare (can be \c NULL) * \param[in] b Second resource to compare (can be \c NULL) * * \retval -1 \c a->priority > \c b->priority (or \c b is \c NULL and \c a is * not) * \retval 0 \c a->priority == \c b->priority (or both \c a and \c b are * \c NULL) * \retval 1 \c a->priority < \c b->priority (or \c a is \c NULL and \c b is * not) */ gint pe__cmp_rsc_priority(gconstpointer a, gconstpointer b) { const pcmk_resource_t *resource1 = (const pcmk_resource_t *)a; const pcmk_resource_t *resource2 = (const pcmk_resource_t *)b; if (a == NULL && b == NULL) { return 0; } if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (resource1->priority > resource2->priority) { return -1; } if (resource1->priority < resource2->priority) { return 1; } return 0; } static void resource_node_score(pcmk_resource_t *rsc, const pcmk_node_t *node, int score, const char *tag) { pcmk_node_t *match = NULL; if ((rsc->exclusive_discover || (node->rsc_discover_mode == pcmk_probe_never)) && pcmk__str_eq(tag, "symmetric_default", pcmk__str_casei)) { /* This string comparision may be fragile, but exclusive resources and * exclusive nodes should not have the symmetric_default constraint * applied to them. */ return; } else if (rsc->children) { GList *gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; resource_node_score(child_rsc, node, score, tag); } } match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (match == NULL) { match = pe__copy_node(node); g_hash_table_insert(rsc->allowed_nodes, (gpointer) match->details->id, match); } match->weight = pcmk__add_scores(match->weight, score); pcmk__rsc_trace(rsc, "Enabling %s preference (%s) for %s on %s (now %s)", tag, pcmk_readable_score(score), rsc->id, pcmk__node_name(node), pcmk_readable_score(match->weight)); } void resource_location(pcmk_resource_t *rsc, const pcmk_node_t *node, int score, const char *tag, pcmk_scheduler_t *scheduler) { if (node != NULL) { resource_node_score(rsc, node, score, tag); } else if (scheduler != NULL) { GList *gIter = scheduler->nodes; for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node_iter = (pcmk_node_t *) gIter->data; resource_node_score(rsc, node_iter, score, tag); } } else { GHashTableIter iter; pcmk_node_t *node_iter = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node_iter)) { resource_node_score(rsc, node_iter, score, tag); } } - if (node == NULL && score == -INFINITY) { + if ((node == NULL) && (score == -PCMK_SCORE_INFINITY)) { if (rsc->allocated_to) { crm_info("Deallocating %s from %s", rsc->id, pcmk__node_name(rsc->allocated_to)); free(rsc->allocated_to); rsc->allocated_to = NULL; } } } time_t get_effective_time(pcmk_scheduler_t *scheduler) { if(scheduler) { if (scheduler->now == NULL) { crm_trace("Recording a new 'now'"); scheduler->now = crm_time_new(NULL); } return crm_time_get_seconds_since_epoch(scheduler->now); } crm_trace("Defaulting to 'now'"); return time(NULL); } gboolean get_target_role(const pcmk_resource_t *rsc, enum rsc_role_e *role) { enum rsc_role_e local_role = pcmk_role_unknown; const char *value = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); CRM_CHECK(role != NULL, return FALSE); if (pcmk__str_eq(value, "started", pcmk__str_null_matches | pcmk__str_casei) || pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) { return FALSE; } local_role = pcmk_parse_role(value); if (local_role == pcmk_role_unknown) { pcmk__config_err("Ignoring '" PCMK_META_TARGET_ROLE "' for %s " "because '%s' is not valid", rsc->id, value); return FALSE; } else if (local_role > pcmk_role_started) { if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk_rsc_promotable)) { if (local_role > pcmk_role_unpromoted) { /* This is what we'd do anyway, just leave the default to avoid messing up the placement algorithm */ return FALSE; } } else { pcmk__config_err("Ignoring '" PCMK_META_TARGET_ROLE "' for %s " "because '%s' only makes sense for promotable " "clones", rsc->id, value); return FALSE; } } *role = local_role; return TRUE; } gboolean order_actions(pcmk_action_t *lh_action, pcmk_action_t *rh_action, uint32_t flags) { GList *gIter = NULL; pcmk__related_action_t *wrapper = NULL; GList *list = NULL; if (flags == pcmk__ar_none) { return FALSE; } if (lh_action == NULL || rh_action == NULL) { return FALSE; } crm_trace("Creating action wrappers for ordering: %s then %s", lh_action->uuid, rh_action->uuid); /* Ensure we never create a dependency on ourselves... it's happened */ CRM_ASSERT(lh_action != rh_action); /* Filter dups, otherwise update_action_states() has too much work to do */ gIter = lh_action->actions_after; for (; gIter != NULL; gIter = gIter->next) { pcmk__related_action_t *after = gIter->data; if (after->action == rh_action && (after->type & flags)) { return FALSE; } } wrapper = calloc(1, sizeof(pcmk__related_action_t)); wrapper->action = rh_action; wrapper->type = flags; list = lh_action->actions_after; list = g_list_prepend(list, wrapper); lh_action->actions_after = list; wrapper = calloc(1, sizeof(pcmk__related_action_t)); wrapper->action = lh_action; wrapper->type = flags; list = rh_action->actions_before; list = g_list_prepend(list, wrapper); rh_action->actions_before = list; return TRUE; } void destroy_ticket(gpointer data) { pcmk_ticket_t *ticket = data; if (ticket->state) { g_hash_table_destroy(ticket->state); } free(ticket->id); free(ticket); } pcmk_ticket_t * ticket_new(const char *ticket_id, pcmk_scheduler_t *scheduler) { pcmk_ticket_t *ticket = NULL; if (pcmk__str_empty(ticket_id)) { return NULL; } if (scheduler->tickets == NULL) { scheduler->tickets = pcmk__strkey_table(free, destroy_ticket); } ticket = g_hash_table_lookup(scheduler->tickets, ticket_id); if (ticket == NULL) { ticket = calloc(1, sizeof(pcmk_ticket_t)); if (ticket == NULL) { pcmk__sched_err("Cannot allocate ticket '%s'", ticket_id); return NULL; } crm_trace("Creaing ticket entry for %s", ticket_id); ticket->id = strdup(ticket_id); ticket->granted = FALSE; ticket->last_granted = -1; ticket->standby = FALSE; ticket->state = pcmk__strkey_table(free, free); g_hash_table_insert(scheduler->tickets, strdup(ticket->id), ticket); } return ticket; } const char * rsc_printable_id(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { return rsc->id; } return pcmk__xe_id(rsc->xml); } void pe__clear_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags) { pcmk__clear_rsc_flags(rsc, flags); for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe__clear_resource_flags_recursive((pcmk_resource_t *) gIter->data, flags); } } void pe__clear_resource_flags_on_all(pcmk_scheduler_t *scheduler, uint64_t flag) { for (GList *lpc = scheduler->resources; lpc != NULL; lpc = lpc->next) { pcmk_resource_t *r = (pcmk_resource_t *) lpc->data; pe__clear_resource_flags_recursive(r, flag); } } void pe__set_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags) { pcmk__set_rsc_flags(rsc, flags); for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe__set_resource_flags_recursive((pcmk_resource_t *) gIter->data, flags); } } void trigger_unfencing(pcmk_resource_t *rsc, pcmk_node_t *node, const char *reason, pcmk_action_t *dependency, pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_enable_unfencing)) { /* No resources require it */ return; } else if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) { /* Wasn't a stonith device */ return; } else if(node && node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { pcmk_action_t *unfence = pe_fence_op(node, PCMK_ACTION_ON, FALSE, reason, FALSE, scheduler); if(dependency) { order_actions(unfence, dependency, pcmk__ar_ordered); } } else if(rsc) { GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if(node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { trigger_unfencing(rsc, node, reason, dependency, scheduler); } } } } gboolean add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref) { pcmk_tag_t *tag = NULL; GList *gIter = NULL; gboolean is_existing = FALSE; CRM_CHECK(tags && tag_name && obj_ref, return FALSE); tag = g_hash_table_lookup(tags, tag_name); if (tag == NULL) { tag = calloc(1, sizeof(pcmk_tag_t)); if (tag == NULL) { pcmk__sched_err("Could not allocate memory for tag %s", tag_name); return FALSE; } tag->id = strdup(tag_name); tag->refs = NULL; g_hash_table_insert(tags, strdup(tag_name), tag); } for (gIter = tag->refs; gIter != NULL; gIter = gIter->next) { const char *existing_ref = (const char *) gIter->data; if (pcmk__str_eq(existing_ref, obj_ref, pcmk__str_none)){ is_existing = TRUE; break; } } if (is_existing == FALSE) { tag->refs = g_list_append(tag->refs, strdup(obj_ref)); crm_trace("Added: tag=%s ref=%s", tag->id, obj_ref); } return TRUE; } /*! * \internal * \brief Check whether shutdown has been requested for a node * * \param[in] node Node to check * * \return TRUE if node has shutdown attribute set and nonzero, FALSE otherwise * \note This differs from simply using node->details->shutdown in that it can * be used before that has been determined (and in fact to determine it), * and it can also be used to distinguish requested shutdown from implicit * shutdown of remote nodes by virtue of their connection stopping. */ bool pe__shutdown_requested(const pcmk_node_t *node) { const char *shutdown = pcmk__node_attr(node, PCMK__NODE_ATTR_SHUTDOWN, NULL, pcmk__rsc_node_current); return !pcmk__str_eq(shutdown, "0", pcmk__str_null_matches); } /*! * \internal * \brief Update "recheck by" time in scheduler data * * \param[in] recheck Epoch time when recheck should happen * \param[in,out] scheduler Scheduler data * \param[in] reason What time is being updated for (for logs) */ void pe__update_recheck_time(time_t recheck, pcmk_scheduler_t *scheduler, const char *reason) { if ((recheck > get_effective_time(scheduler)) && ((scheduler->recheck_by == 0) || (scheduler->recheck_by > recheck))) { scheduler->recheck_by = recheck; crm_debug("Updated next scheduler recheck to %s for %s", pcmk__trim(ctime(&recheck)), reason); } } /*! * \internal * \brief Extract nvpair blocks contained by a CIB XML element into a hash table * * \param[in] xml_obj XML element containing blocks of nvpair elements * \param[in] set_name If not NULL, only use blocks of this element * \param[in] rule_data Matching parameters to use when unpacking * \param[out] hash Where to store extracted name/value pairs * \param[in] always_first If not NULL, process block with this ID first * \param[in] overwrite Whether to replace existing values with same name * \param[in,out] scheduler Scheduler data containing \p xml_obj */ void pe__unpack_dataset_nvpairs(const xmlNode *xml_obj, const char *set_name, const pe_rule_eval_data_t *rule_data, GHashTable *hash, const char *always_first, gboolean overwrite, pcmk_scheduler_t *scheduler) { crm_time_t *next_change = crm_time_new_undefined(); pe_eval_nvpairs(scheduler->input, xml_obj, set_name, rule_data, hash, always_first, overwrite, next_change); if (crm_time_is_defined(next_change)) { time_t recheck = (time_t) crm_time_get_seconds_since_epoch(next_change); pe__update_recheck_time(recheck, scheduler, "rule evaluation"); } crm_time_free(next_change); } bool pe__resource_is_disabled(const pcmk_resource_t *rsc) { const char *target_role = NULL; CRM_CHECK(rsc != NULL, return false); target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); if (target_role) { // If invalid, we've already logged an error when unpacking enum rsc_role_e target_role_e = pcmk_parse_role(target_role); if ((target_role_e == pcmk_role_stopped) || ((target_role_e == pcmk_role_unpromoted) && pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk_rsc_promotable))) { return true; } } return false; } /*! * \internal * \brief Check whether a resource is running only on given node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p rsc is running only on \p node, otherwise false */ bool pe__rsc_running_on_only(const pcmk_resource_t *rsc, const pcmk_node_t *node) { return (rsc != NULL) && pcmk__list_of_1(rsc->running_on) && pcmk__same_node((const pcmk_node_t *) rsc->running_on->data, node); } bool pe__rsc_running_on_any(pcmk_resource_t *rsc, GList *node_list) { for (GList *ele = rsc->running_on; ele; ele = ele->next) { pcmk_node_t *node = (pcmk_node_t *) ele->data; if (pcmk__str_in_list(node->details->uname, node_list, pcmk__str_star_matches|pcmk__str_casei)) { return true; } } return false; } bool pcmk__rsc_filtered_by_node(pcmk_resource_t *rsc, GList *only_node) { return (rsc->fns->active(rsc, FALSE) && !pe__rsc_running_on_any(rsc, only_node)); } GList * pe__filter_rsc_list(GList *rscs, GList *filter) { GList *retval = NULL; for (GList *gIter = rscs; gIter; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; /* I think the second condition is safe here for all callers of this * function. If not, it needs to move into pe__node_text. */ if (pcmk__str_in_list(rsc_printable_id(rsc), filter, pcmk__str_star_matches) || (rsc->parent && pcmk__str_in_list(rsc_printable_id(rsc->parent), filter, pcmk__str_star_matches))) { retval = g_list_prepend(retval, rsc); } } return retval; } GList * pe__build_node_name_list(pcmk_scheduler_t *scheduler, const char *s) { GList *nodes = NULL; if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) { /* Nothing was given so return a list of all node names. Or, '*' was * given. This would normally fall into the pe__unames_with_tag branch * where it will return an empty list. Catch it here instead. */ nodes = g_list_prepend(nodes, strdup("*")); } else { pcmk_node_t *node = pe_find_node(scheduler->nodes, s); if (node) { /* The given string was a valid uname for a node. Return a * singleton list containing just that uname. */ nodes = g_list_prepend(nodes, strdup(s)); } else { /* The given string was not a valid uname. It's either a tag or * it's a typo or something. In the first case, we'll return a * list of all the unames of the nodes with the given tag. In the * second case, we'll return a NULL pointer and nothing will * get displayed. */ nodes = pe__unames_with_tag(scheduler, s); } } return nodes; } GList * pe__build_rsc_list(pcmk_scheduler_t *scheduler, const char *s) { GList *resources = NULL; if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) { resources = g_list_prepend(resources, strdup("*")); } else { const uint32_t flags = pcmk_rsc_match_history|pcmk_rsc_match_basename; pcmk_resource_t *rsc = pe_find_resource_with_flags(scheduler->resources, s, flags); if (rsc) { /* A colon in the name we were given means we're being asked to filter * on a specific instance of a cloned resource. Put that exact string * into the filter list. Otherwise, use the printable ID of whatever * resource was found that matches what was asked for. */ if (strstr(s, ":") != NULL) { resources = g_list_prepend(resources, strdup(rsc->id)); } else { resources = g_list_prepend(resources, strdup(rsc_printable_id(rsc))); } } else { /* The given string was not a valid resource name. It's a tag or a * typo or something. See pe__build_node_name_list() for more * detail. */ resources = pe__rscs_with_tag(scheduler, s); } } return resources; } xmlNode * pe__failed_probe_for_rsc(const pcmk_resource_t *rsc, const char *name) { const pcmk_resource_t *parent = pe__const_top_resource(rsc, false); const char *rsc_id = rsc->id; if (parent->variant == pcmk_rsc_variant_clone) { rsc_id = pe__clone_child_id(parent); } for (xmlNode *xml_op = pcmk__xml_first_child(rsc->cluster->failed); xml_op != NULL; xml_op = pcmk__xml_next(xml_op)) { const char *value = NULL; char *op_id = NULL; /* This resource operation is not a failed probe. */ if (!pcmk_xe_mask_probe_failure(xml_op)) { continue; } /* This resource operation was not run on the given node. Note that if name is * NULL, this will always succeed. */ value = crm_element_value(xml_op, PCMK__META_ON_NODE); if (value == NULL || !pcmk__str_eq(value, name, pcmk__str_casei|pcmk__str_null_matches)) { continue; } if (!parse_op_key(pcmk__xe_history_key(xml_op), &op_id, NULL, NULL)) { continue; // This history entry is missing an operation key } /* This resource operation's ID does not match the rsc_id we are looking for. */ if (!pcmk__str_eq(op_id, rsc_id, pcmk__str_none)) { free(op_id); continue; } free(op_id); return xml_op; } return NULL; }