diff --git a/daemons/based/based_messages.c b/daemons/based/based_messages.c index df30752d96..2bd3060242 100644 --- a/daemons/based/based_messages.c +++ b/daemons/based/based_messages.c @@ -1,540 +1,542 @@ /* - * Copyright 2004-2020 the Pacemaker project contributors + * Copyright 2004-2022 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 #include #include #include #include #include /* Maximum number of diffs to ignore while waiting for a resync */ #define MAX_DIFF_RETRY 5 gboolean cib_is_master = FALSE; xmlNode *the_cib = NULL; int revision_check(xmlNode * cib_update, xmlNode * cib_copy, int flags); int get_revision(xmlNode * xml_obj, int cur_revision); int updateList(xmlNode * local_cib, xmlNode * update_command, xmlNode * failed, int operation, const char *section); gboolean update_results(xmlNode * failed, xmlNode * target, const char *operation, int return_code); int cib_update_counter(xmlNode * xml_obj, const char *field, gboolean reset); int sync_our_cib(xmlNode * request, gboolean all); int cib_process_shutdown_req(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { const char *host = crm_element_value(req, F_ORIG); *answer = NULL; if (crm_element_value(req, F_CIB_ISREPLY) == NULL) { crm_info("Peer %s is requesting to shut down", host); return pcmk_ok; } if (cib_shutdown_flag == FALSE) { crm_err("Peer %s mistakenly thinks we wanted to shut down", host); return -EINVAL; } crm_info("Peer %s has acknowledged our shutdown request", host); terminate_cib(__func__, 0); return pcmk_ok; } int cib_process_default(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { int result = pcmk_ok; crm_trace("Processing \"%s\" event", op); *answer = NULL; if (op == NULL) { result = -EINVAL; crm_err("No operation specified"); } else if (strcasecmp(CRM_OP_NOOP, op) == 0) { ; } else { result = -EPROTONOSUPPORT; crm_err("Action [%s] is not supported by the CIB manager", op); } return result; } int cib_process_readwrite(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { int result = pcmk_ok; crm_trace("Processing \"%s\" event", op); if (pcmk__str_eq(op, CIB_OP_ISMASTER, pcmk__str_casei)) { if (cib_is_master == TRUE) { result = pcmk_ok; } else { result = -EPERM; } return result; } if (pcmk__str_eq(op, CIB_OP_MASTER, pcmk__str_casei)) { if (cib_is_master == FALSE) { crm_info("We are now in R/W mode"); cib_is_master = TRUE; } else { crm_debug("We are still in R/W mode"); } } else if (cib_is_master) { crm_info("We are now in R/O mode"); cib_is_master = FALSE; } return result; } /* Set to 1 when a sync is requested, incremented when a diff is ignored, * reset to 0 when a sync is received */ static int sync_in_progress = 0; void send_sync_request(const char *host) { xmlNode *sync_me = create_xml_node(NULL, "sync-me"); crm_info("Requesting re-sync from %s", (host? host : "all peers")); sync_in_progress = 1; crm_xml_add(sync_me, F_TYPE, "cib"); crm_xml_add(sync_me, F_CIB_OPERATION, CIB_OP_SYNC_ONE); crm_xml_add(sync_me, F_CIB_DELEGATED, cib_our_uname); send_cluster_message(host ? crm_get_peer(0, host) : NULL, crm_msg_cib, sync_me, FALSE); free_xml(sync_me); } int cib_process_ping(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { const char *host = crm_element_value(req, F_ORIG); const char *seq = crm_element_value(req, F_CIB_PING_ID); char *digest = calculate_xml_versioned_digest(the_cib, FALSE, TRUE, CRM_FEATURE_SET); static struct qb_log_callsite *cs = NULL; crm_trace("Processing \"%s\" event %s from %s", op, seq, host); *answer = create_xml_node(NULL, XML_CRM_TAG_PING); crm_xml_add(*answer, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET); crm_xml_add(*answer, XML_ATTR_DIGEST, digest); crm_xml_add(*answer, F_CIB_PING_ID, seq); if (cs == NULL) { cs = qb_log_callsite_get(__func__, __FILE__, __func__, LOG_TRACE, __LINE__, crm_trace_nonlog); } if (cs && cs->targets) { /* Append additional detail so the reciever can log the differences */ add_message_xml(*answer, F_CIB_CALLDATA, the_cib); } else { /* Always include at least the version details */ const char *tag = TYPE(the_cib); xmlNode *shallow = create_xml_node(NULL, tag); copy_in_properties(shallow, the_cib); add_message_xml(*answer, F_CIB_CALLDATA, shallow); free_xml(shallow); } crm_info("Reporting our current digest to %s: %s for %s.%s.%s (%p %d)", host, digest, crm_element_value(existing_cib, XML_ATTR_GENERATION_ADMIN), crm_element_value(existing_cib, XML_ATTR_GENERATION), crm_element_value(existing_cib, XML_ATTR_NUMUPDATES), existing_cib, cs && cs->targets); free(digest); return pcmk_ok; } int cib_process_sync(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { return sync_our_cib(req, TRUE); } int cib_process_upgrade_server(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { int rc = pcmk_ok; *answer = NULL; if(crm_element_value(req, F_CIB_SCHEMA_MAX)) { /* The originator of an upgrade request sends it to the DC, without * F_CIB_SCHEMA_MAX. If an upgrade is needed, the DC re-broadcasts the * request with F_CIB_SCHEMA_MAX, and each node performs the upgrade * (and notifies its local clients) here. */ return cib_process_upgrade( op, options, section, req, input, existing_cib, result_cib, answer); } else { int new_version = 0; int current_version = 0; xmlNode *scratch = copy_xml(existing_cib); const char *host = crm_element_value(req, F_ORIG); const char *value = crm_element_value(existing_cib, XML_ATTR_VALIDATION); const char *client_id = crm_element_value(req, F_CIB_CLIENTID); const char *call_opts = crm_element_value(req, F_CIB_CALLOPTS); const char *call_id = crm_element_value(req, F_CIB_CALLID); crm_trace("Processing \"%s\" event", op); if (value != NULL) { current_version = get_schema_version(value); } rc = update_validation(&scratch, &new_version, 0, TRUE, TRUE); if (new_version > current_version) { xmlNode *up = create_xml_node(NULL, __func__); rc = pcmk_ok; crm_notice("Upgrade request from %s verified", host); crm_xml_add(up, F_TYPE, "cib"); crm_xml_add(up, F_CIB_OPERATION, CIB_OP_UPGRADE); crm_xml_add(up, F_CIB_SCHEMA_MAX, get_schema_name(new_version)); crm_xml_add(up, F_CIB_DELEGATED, host); crm_xml_add(up, F_CIB_CLIENTID, client_id); crm_xml_add(up, F_CIB_CALLOPTS, call_opts); crm_xml_add(up, F_CIB_CALLID, call_id); if (cib_legacy_mode() && cib_is_master) { rc = cib_process_upgrade( op, options, section, up, input, existing_cib, result_cib, answer); } else { send_cluster_message(NULL, crm_msg_cib, up, FALSE); } free_xml(up); } else if(rc == pcmk_ok) { rc = -pcmk_err_schema_unchanged; } if (rc != pcmk_ok) { // Notify originating peer so it can notify its local clients crm_node_t *origin = pcmk__search_cluster_node_cache(0, host); crm_info("Rejecting upgrade request from %s: %s " CRM_XS " rc=%d peer=%s", host, pcmk_strerror(rc), rc, (origin? origin->uname : "lost")); if (origin) { xmlNode *up = create_xml_node(NULL, __func__); crm_xml_add(up, F_TYPE, "cib"); crm_xml_add(up, F_CIB_OPERATION, CIB_OP_UPGRADE); crm_xml_add(up, F_CIB_DELEGATED, host); crm_xml_add(up, F_CIB_ISREPLY, host); crm_xml_add(up, F_CIB_CLIENTID, client_id); crm_xml_add(up, F_CIB_CALLOPTS, call_opts); crm_xml_add(up, F_CIB_CALLID, call_id); crm_xml_add_int(up, F_CIB_UPGRADE_RC, rc); if (send_cluster_message(origin, crm_msg_cib, up, TRUE) == FALSE) { crm_warn("Could not send CIB upgrade result to %s", host); } free_xml(up); } } free_xml(scratch); } return rc; } int cib_process_sync_one(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { return sync_our_cib(req, FALSE); } int cib_server_process_diff(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { int rc = pcmk_ok; if (sync_in_progress > MAX_DIFF_RETRY) { /* Don't ignore diffs forever; the last request may have been lost. * If the diff fails, we'll ask for another full resync. */ sync_in_progress = 0; } /* The master should never ignore a diff */ if (sync_in_progress && !cib_is_master) { int diff_add_updates = 0; int diff_add_epoch = 0; int diff_add_admin_epoch = 0; int diff_del_updates = 0; int diff_del_epoch = 0; int diff_del_admin_epoch = 0; cib_diff_version_details(input, &diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates, &diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates); sync_in_progress++; crm_notice("Not applying diff %d.%d.%d -> %d.%d.%d (sync in progress)", diff_del_admin_epoch, diff_del_epoch, diff_del_updates, diff_add_admin_epoch, diff_add_epoch, diff_add_updates); return -pcmk_err_diff_resync; } rc = cib_process_diff(op, options, section, req, input, existing_cib, result_cib, answer); crm_trace("result: %s (%d), %s", pcmk_strerror(rc), rc, cib_is_master?"master":"slave"); if (rc == -pcmk_err_diff_resync && cib_is_master == FALSE) { free_xml(*result_cib); *result_cib = NULL; send_sync_request(NULL); } else if (rc == -pcmk_err_diff_resync) { rc = -pcmk_err_diff_failed; if (options & cib_force_diff) { crm_warn("Not requesting full refresh in R/W mode"); } } else if ((rc != pcmk_ok) && !cib_is_master && cib_legacy_mode()) { crm_warn("Requesting full CIB refresh because update failed: %s" CRM_XS " rc=%d", pcmk_strerror(rc), rc); xml_log_patchset(LOG_INFO, __func__, input); free_xml(*result_cib); *result_cib = NULL; send_sync_request(NULL); } return rc; } int cib_process_replace_svr(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { const char *tag = crm_element_name(input); int rc = cib_process_replace(op, options, section, req, input, existing_cib, result_cib, answer); if (rc == pcmk_ok && pcmk__str_eq(tag, XML_TAG_CIB, pcmk__str_casei)) { sync_in_progress = 0; } return rc; } static int delete_cib_object(xmlNode * parent, xmlNode * delete_spec) { const char *object_name = NULL; const char *object_id = NULL; xmlNode *equiv_node = NULL; int result = pcmk_ok; if (delete_spec != NULL) { object_name = crm_element_name(delete_spec); } object_id = crm_element_value(delete_spec, XML_ATTR_ID); crm_trace("Processing: <%s id=%s>", crm_str(object_name), crm_str(object_id)); if (delete_spec == NULL) { result = -EINVAL; } else if (parent == NULL) { result = -EINVAL; } else if (object_id == NULL) { /* placeholder object */ equiv_node = find_xml_node(parent, object_name, FALSE); } else { equiv_node = pcmk__xe_match(parent, object_name, XML_ATTR_ID, object_id); } if (result != pcmk_ok) { ; /* nothing */ } else if (equiv_node == NULL) { result = pcmk_ok; } else if (xml_has_children(delete_spec) == FALSE) { /* only leaves are deleted */ crm_debug("Removing leaf: <%s id=%s>", crm_str(object_name), crm_str(object_id)); free_xml(equiv_node); equiv_node = NULL; } else { xmlNode *child = NULL; for (child = pcmk__xml_first_child(delete_spec); child != NULL; child = pcmk__xml_next(child)) { int tmp_result = delete_cib_object(equiv_node, child); /* only the first error is likely to be interesting */ if (tmp_result != pcmk_ok && result == pcmk_ok) { result = tmp_result; } } } return result; } int cib_process_delete_absolute(const char *op, int options, const char *section, xmlNode * req, xmlNode * input, xmlNode * existing_cib, xmlNode ** result_cib, xmlNode ** answer) { xmlNode *failed = NULL; int result = pcmk_ok; xmlNode *update_section = NULL; crm_trace("Processing \"%s\" event for section=%s", op, crm_str(section)); if (pcmk__str_eq(XML_CIB_TAG_SECTION_ALL, section, pcmk__str_casei)) { section = NULL; } else if (pcmk__str_eq(XML_TAG_CIB, section, pcmk__str_casei)) { section = NULL; } else if (pcmk__str_eq(crm_element_name(input), XML_TAG_CIB, pcmk__str_casei)) { section = NULL; } CRM_CHECK(strcasecmp(CIB_OP_DELETE, op) == 0, return -EINVAL); if (input == NULL) { crm_err("Cannot perform modification with no data"); return -EINVAL; } failed = create_xml_node(NULL, XML_TAG_FAILED); update_section = pcmk_find_cib_element(*result_cib, section); result = delete_cib_object(update_section, input); update_results(failed, input, op, result); if ((result == pcmk_ok) && xml_has_children(failed)) { result = -EINVAL; } if (result != pcmk_ok) { crm_log_xml_err(failed, "CIB Update failures"); *answer = failed; } else { free_xml(failed); } return result; } int sync_our_cib(xmlNode * request, gboolean all) { int result = pcmk_ok; char *digest = NULL; const char *host = crm_element_value(request, F_ORIG); const char *op = crm_element_value(request, F_CIB_OPERATION); - xmlNode *replace_request = cib_msg_copy(request, FALSE); + xmlNode *replace_request = NULL; - CRM_CHECK(the_cib != NULL,;); - CRM_CHECK(replace_request != NULL,;); + CRM_CHECK(the_cib != NULL, return -EINVAL); + + replace_request = cib_msg_copy(request, FALSE); + CRM_CHECK(replace_request != NULL, return -EINVAL); crm_debug("Syncing CIB to %s", all ? "all peers" : host); if (all == FALSE && host == NULL) { crm_log_xml_err(request, "bad sync"); } /* remove the "all == FALSE" condition * * sync_from was failing, the local client wasn't being notified * because it didn't know it was a reply * setting this does not prevent the other nodes from applying it * if all == TRUE */ if (host != NULL) { crm_xml_add(replace_request, F_CIB_ISREPLY, host); } if (all) { xml_remove_prop(replace_request, F_CIB_HOST); } crm_xml_add(replace_request, F_CIB_OPERATION, CIB_OP_REPLACE); crm_xml_add(replace_request, "original_" F_CIB_OPERATION, op); pcmk__xe_set_bool_attr(replace_request, F_CIB_GLOBAL_UPDATE, true); crm_xml_add(replace_request, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET); digest = calculate_xml_versioned_digest(the_cib, FALSE, TRUE, CRM_FEATURE_SET); crm_xml_add(replace_request, XML_ATTR_DIGEST, digest); add_message_xml(replace_request, F_CIB_CALLDATA, the_cib); if (send_cluster_message (all ? NULL : crm_get_peer(0, host), crm_msg_cib, replace_request, FALSE) == FALSE) { result = -ENOTCONN; } free_xml(replace_request); free(digest); return result; } diff --git a/daemons/controld/controld_callbacks.c b/daemons/controld/controld_callbacks.c index 8d1590dc4e..42006e47b4 100644 --- a/daemons/controld/controld_callbacks.c +++ b/daemons/controld/controld_callbacks.c @@ -1,361 +1,361 @@ /* - * Copyright 2004-2021 the Pacemaker project contributors + * Copyright 2004-2022 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 /* 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, F_CRM_SYS_FROM); if (pcmk__str_eq(sys_from, CRM_SYSTEM_DC, pcmk__str_casei)) { const char *from = crm_element_value(msg, F_ORIG); if (!pcmk__str_eq(from, fsa_our_uname, pcmk__str_casei)) { int level = LOG_INFO; const char *op = crm_element_value(msg, F_CRM_TASK); /* make sure the election happens NOW */ if (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, F_CRM_SYS_TO); if (pcmk__str_eq(sys_to, CRM_SYSTEM_DC, pcmk__str_casei)) { return; } } /* crm_log_xml_trace("HA[inbound]", msg); */ route_message(C_HA_MESSAGE, msg); done: 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") bool controld_dc_left = false; 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); /* 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_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()); crm_info("Node %s is %s a peer " CRM_XS " DC=%s old=%#07x new=%#07x", node->uname, (appeared? "now" : "no longer"), (AM_I_DC? "true" : (fsa_our_dc? fsa_our_dc : "")), 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) { controld_remove_voter(node->uname); } if (!pcmk_is_set(fsa_input_register, R_CIB_CONNECTED)) { crm_trace("Ignoring peer status change because not connected to CIB"); return; } else if (fsa_state == S_STOPPING) { crm_trace("Ignoring peer status change because stopping"); return; } if (pcmk__str_eq(node->uname, fsa_our_uname, pcmk__str_casei) && !appeared) { /* 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, fsa_our_dc, pcmk__str_casei) && crm_is_peer_active(node) == FALSE) { /* Did the DC leave us? */ crm_notice("Our peer on the DC (%s) is dead", fsa_our_dc); 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(fsa_our_dc_version, "3.0.9") > 0) { controld_delete_node_state(node->uname, controld_section_attrs, cib_scope_local); } } else if (AM_I_DC || controld_dc_left || (fsa_our_dc == 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); crm_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, XML_LRM_ATTR_TASK); if (pcmk__str_eq(task, CRM_OP_FENCE, 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, CRM_OP_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(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(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) { 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(fsa_state, __func__); } abort_transition(INFINITY, tg_restart, "Node failure", NULL); fail_incompletable_actions(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, tg_restart, "Pacemaker Remote node integrated", NULL); } } /* 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(XML_CIB_TAG_STATUS, update); } free_xml(update); } trigger_fsa(); } void crmd_cib_connection_destroy(gpointer user_data) { - CRM_CHECK(user_data == fsa_cib_conn,;); + CRM_LOG_ASSERT(user_data == fsa_cib_conn); crm_trace("Invoked"); trigger_fsa(); fsa_cib_conn->state = cib_disconnected; if (!pcmk_is_set(fsa_input_register, R_CIB_CONNECTED)) { crm_info("Connection to the CIB manager terminated"); return; } // @TODO This should trigger a reconnect, not a shutdown crm_crit("Lost connection to the CIB manager, shutting down"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_CIB_CONNECTED); return; } gboolean crm_fsa_trigger(gpointer user_data) { crm_trace("Invoked (queue len: %d)", g_list_length(fsa_message_queue)); s_crmd_fsa(C_FSA_INTERNAL); crm_trace("Exited (queue len: %d)", g_list_length(fsa_message_queue)); return TRUE; } diff --git a/daemons/controld/controld_te_actions.c b/daemons/controld/controld_te_actions.c index 5599c74ac0..63b7c72359 100644 --- a/daemons/controld/controld_te_actions.c +++ b/daemons/controld/controld_te_actions.c @@ -1,641 +1,638 @@ /* - * Copyright 2004-2021 the Pacemaker project contributors + * Copyright 2004-2022 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 // lrmd_event_data_t, lrmd_free_event() #include #include #include #include #include char *te_uuid = NULL; GHashTable *te_targets = NULL; void send_rsc_command(crm_action_t * action); static void te_update_job_count(crm_action_t * action, int offset); static void te_start_action_timer(crm_graph_t * graph, crm_action_t * action) { action->timer = calloc(1, sizeof(crm_action_timer_t)); action->timer->timeout = action->timeout; action->timer->action = action; action->timer->source_id = g_timeout_add(action->timer->timeout + graph->network_delay, action_timer_callback, (void *)action->timer); CRM_ASSERT(action->timer->source_id != 0); } static gboolean te_pseudo_action(crm_graph_t * graph, crm_action_t * pseudo) { const char *task = crm_element_value(pseudo->xml, XML_LRM_ATTR_TASK); /* send to peers as well? */ if (pcmk__str_eq(task, CRM_OP_MAINTENANCE_NODES, pcmk__str_casei)) { GHashTableIter iter; crm_node_t *node = NULL; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *cmd = NULL; if (pcmk__str_eq(fsa_our_uname, node->uname, pcmk__str_casei)) { continue; } cmd = create_request(task, pseudo->xml, node->uname, CRM_SYSTEM_CRMD, CRM_SYSTEM_TENGINE, NULL); send_cluster_message(node, crm_msg_crmd, cmd, FALSE); free_xml(cmd); } remote_ra_process_maintenance_nodes(pseudo->xml); } else { /* Check action for Pacemaker Remote node side effects */ remote_ra_process_pseudo(pseudo->xml); } crm_debug("Pseudo-action %d (%s) fired and confirmed", pseudo->id, crm_element_value(pseudo->xml, XML_LRM_ATTR_TASK_KEY)); te_action_confirmed(pseudo, graph); return TRUE; } static int get_target_rc(crm_action_t * action) { int exit_status; pcmk__scan_min_int(crm_meta_value(action->params, XML_ATTR_TE_TARGET_RC), &exit_status, 0); return exit_status; } static gboolean te_crm_command(crm_graph_t * graph, crm_action_t * action) { char *counter = NULL; xmlNode *cmd = NULL; gboolean is_local = FALSE; const char *id = NULL; const char *task = NULL; const char *value = NULL; const char *on_node = NULL; const char *router_node = NULL; gboolean rc = TRUE; gboolean no_wait = FALSE; id = ID(action->xml); task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); router_node = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); if (!router_node) { router_node = on_node; if (pcmk__str_eq(task, CRM_OP_LRM_DELETE, pcmk__str_casei)) { const char *mode = crm_element_value(action->xml, PCMK__XA_MODE); if (pcmk__str_eq(mode, XML_TAG_CIB, pcmk__str_casei)) { router_node = fsa_our_uname; } } } CRM_CHECK(on_node != NULL && strlen(on_node) != 0, crm_err("Corrupted command (id=%s) %s: no node", crm_str(id), crm_str(task)); return FALSE); if (pcmk__str_eq(router_node, fsa_our_uname, pcmk__str_casei)) { is_local = TRUE; } value = crm_meta_value(action->params, XML_ATTR_TE_NOWAIT); if (crm_is_true(value)) { no_wait = TRUE; } crm_info("Executing crm-event (%s)%s%s: %s on %s", crm_str(id), (is_local? " locally" : ""), (no_wait? " without waiting" : ""), crm_str(task), on_node); if (is_local && pcmk__str_eq(task, CRM_OP_SHUTDOWN, pcmk__str_casei)) { /* defer until everything else completes */ crm_info("crm-event (%s) is a local shutdown", crm_str(id)); graph->completion_action = tg_shutdown; graph->abort_reason = "local shutdown"; te_action_confirmed(action, graph); return TRUE; } else if (pcmk__str_eq(task, CRM_OP_SHUTDOWN, pcmk__str_casei)) { crm_node_t *peer = crm_get_peer(0, router_node); pcmk__update_peer_expected(__func__, peer, CRMD_JOINSTATE_DOWN); } cmd = create_request(task, action->xml, router_node, CRM_SYSTEM_CRMD, CRM_SYSTEM_TENGINE, NULL); counter = pcmk__transition_key(transition_graph->id, action->id, get_target_rc(action), te_uuid); crm_xml_add(cmd, XML_ATTR_TRANSITION_KEY, counter); rc = send_cluster_message(crm_get_peer(0, router_node), crm_msg_crmd, cmd, TRUE); free(counter); free_xml(cmd); if (rc == FALSE) { crm_err("Action %d failed: send", action->id); return FALSE; } else if (no_wait) { te_action_confirmed(action, graph); } else { if (action->timeout <= 0) { crm_err("Action %d: %s on %s had an invalid timeout (%dms). Using %ums instead", action->id, task, on_node, action->timeout, graph->network_delay); action->timeout = (int) graph->network_delay; } te_start_action_timer(graph, action); } return TRUE; } void controld_record_action_timeout(crm_action_t *action) { lrmd_event_data_t *op = NULL; xmlNode *state = NULL; xmlNode *rsc = NULL; xmlNode *xml_op = NULL; xmlNode *action_rsc = NULL; int rc = pcmk_ok; const char *rsc_id = NULL; const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); const char *target_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID); int call_options = cib_quorum_override | cib_scope_local; int target_rc = get_target_rc(action); crm_warn("%s %d: %s on %s timed out", crm_element_name(action->xml), action->id, task_uuid, target); action_rsc = find_xml_node(action->xml, XML_CIB_TAG_RESOURCE, TRUE); if (action_rsc == NULL) { return; } rsc_id = ID(action_rsc); CRM_CHECK(rsc_id != NULL, crm_log_xml_err(action->xml, "Bad:action"); return); /* update the CIB */ state = create_xml_node(NULL, XML_CIB_TAG_STATE); crm_xml_add(state, XML_ATTR_UUID, target_uuid); crm_xml_add(state, XML_ATTR_UNAME, target); rsc = create_xml_node(state, XML_CIB_TAG_LRM); crm_xml_add(rsc, XML_ATTR_ID, target_uuid); rsc = create_xml_node(rsc, XML_LRM_TAG_RESOURCES); rsc = create_xml_node(rsc, XML_LRM_TAG_RESOURCE); crm_xml_add(rsc, XML_ATTR_ID, rsc_id); crm_copy_xml_element(action_rsc, rsc, XML_ATTR_TYPE); crm_copy_xml_element(action_rsc, rsc, XML_AGENT_ATTR_CLASS); crm_copy_xml_element(action_rsc, rsc, XML_AGENT_ATTR_PROVIDER); /* If the executor gets a timeout while waiting for the action to complete, * that will be reported via the usual callback. This timeout means that we * didn't hear from the executor or the controller that relayed the action * to the executor. */ op = pcmk__event_from_graph_action(NULL, action, PCMK_EXEC_TIMEOUT, PCMK_OCF_UNKNOWN_ERROR, "Cluster communication timeout " "(no response from executor)"); op->call_id = -1; op->user_data = pcmk__transition_key(transition_graph->id, action->id, target_rc, te_uuid); xml_op = pcmk__create_history_xml(rsc, op, CRM_FEATURE_SET, target_rc, target, __func__, LOG_INFO); lrmd_free_event(op); crm_log_xml_trace(xml_op, "Action timeout"); rc = fsa_cib_conn->cmds->update(fsa_cib_conn, XML_CIB_TAG_STATUS, state, call_options); fsa_register_cib_callback(rc, FALSE, NULL, cib_action_updated); free_xml(state); crm_trace("Sent CIB update (call ID %d) for timeout of action %d (%s on %s)", rc, action->id, task_uuid, target); crm__set_graph_action_flags(action, pcmk__graph_action_sent_update); } static gboolean te_rsc_command(crm_graph_t * graph, crm_action_t * action) { /* never overwrite stop actions in the CIB with * anything other than completed results * * Writing pending stops makes it look like the * resource is running again */ xmlNode *cmd = NULL; xmlNode *rsc_op = NULL; gboolean rc = TRUE; gboolean no_wait = FALSE; gboolean is_local = FALSE; char *counter = NULL; const char *task = NULL; const char *value = NULL; const char *on_node = NULL; const char *router_node = NULL; const char *task_uuid = NULL; CRM_ASSERT(action != NULL); CRM_ASSERT(action->xml != NULL); crm__clear_graph_action_flags(action, pcmk__graph_action_executed); on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); CRM_CHECK(on_node != NULL && strlen(on_node) != 0, crm_err("Corrupted command(id=%s) %s: no node", ID(action->xml), crm_str(task)); return FALSE); rsc_op = action->xml; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); task_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); router_node = crm_element_value(rsc_op, XML_LRM_ATTR_ROUTER_NODE); if (!router_node) { router_node = on_node; } counter = pcmk__transition_key(transition_graph->id, action->id, get_target_rc(action), te_uuid); crm_xml_add(rsc_op, XML_ATTR_TRANSITION_KEY, counter); if (pcmk__str_eq(router_node, fsa_our_uname, pcmk__str_casei)) { is_local = TRUE; } value = crm_meta_value(action->params, XML_ATTR_TE_NOWAIT); if (crm_is_true(value)) { no_wait = TRUE; } crm_notice("Initiating %s operation %s%s on %s%s "CRM_XS" action %d", task, task_uuid, (is_local? " locally" : ""), on_node, (no_wait? " without waiting" : ""), action->id); cmd = create_request(CRM_OP_INVOKE_LRM, rsc_op, router_node, CRM_SYSTEM_LRMD, CRM_SYSTEM_TENGINE, NULL); if (is_local) { /* shortcut local resource commands */ ha_msg_input_t data = { .msg = cmd, .xml = rsc_op, }; fsa_data_t msg = { .id = 0, .data = &data, .data_type = fsa_dt_ha_msg, .fsa_input = I_NULL, .fsa_cause = C_FSA_INTERNAL, .actions = A_LRM_INVOKE, .origin = __func__, }; do_lrm_invoke(A_LRM_INVOKE, C_FSA_INTERNAL, fsa_state, I_NULL, &msg); } else { rc = send_cluster_message(crm_get_peer(0, router_node), crm_msg_lrmd, cmd, TRUE); } free(counter); free_xml(cmd); crm__set_graph_action_flags(action, pcmk__graph_action_executed); if (rc == FALSE) { crm_err("Action %d failed: send", action->id); return FALSE; } else if (no_wait) { crm_info("Action %d confirmed - no wait", action->id); crm__set_graph_action_flags(action, pcmk__graph_action_confirmed); /* Just mark confirmed. * Don't bump the job count only to immediately decrement it */ pcmk__update_graph(transition_graph, action); trigger_graph(); } else if (pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { crm_debug("Action %d: %s %s on %s(timeout %dms) was already confirmed.", action->id, task, task_uuid, on_node, action->timeout); } else { if (action->timeout <= 0) { crm_err("Action %d: %s %s on %s had an invalid timeout (%dms). Using %ums instead", action->id, task, task_uuid, on_node, action->timeout, graph->network_delay); action->timeout = (int) graph->network_delay; } te_update_job_count(action, 1); te_start_action_timer(graph, action); } return TRUE; } struct te_peer_s { char *name; int jobs; int migrate_jobs; }; static void te_peer_free(gpointer p) { struct te_peer_s *peer = p; free(peer->name); free(peer); } void te_reset_job_counts(void) { GHashTableIter iter; struct te_peer_s *peer = NULL; if(te_targets == NULL) { te_targets = pcmk__strkey_table(NULL, te_peer_free); } g_hash_table_iter_init(&iter, te_targets); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & peer)) { peer->jobs = 0; peer->migrate_jobs = 0; } } static void te_update_job_count_on(const char *target, int offset, bool migrate) { struct te_peer_s *r = NULL; if(target == NULL || te_targets == NULL) { return; } r = g_hash_table_lookup(te_targets, target); if(r == NULL) { r = calloc(1, sizeof(struct te_peer_s)); r->name = strdup(target); g_hash_table_insert(te_targets, r->name, r); } r->jobs += offset; if(migrate) { r->migrate_jobs += offset; } crm_trace("jobs[%s] = %d", target, r->jobs); } static void te_update_job_count(crm_action_t * action, int offset) { const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); if (action->type != action_type_rsc || target == NULL) { /* No limit on these */ return; } /* if we have a router node, this means the action is performing * on a remote node. For now, we count all actions occurring on a * remote node against the job list on the cluster node hosting * the connection resources */ target = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); if ((target == NULL) && pcmk__strcase_any_of(task, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, NULL)) { const char *t1 = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_SOURCE); const char *t2 = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_TARGET); te_update_job_count_on(t1, offset, TRUE); te_update_job_count_on(t2, offset, TRUE); return; } else if (target == NULL) { target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); } te_update_job_count_on(target, offset, FALSE); } static gboolean te_should_perform_action_on(crm_graph_t * graph, crm_action_t * action, const char *target) { int limit = 0; struct te_peer_s *r = NULL; const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); const char *id = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); if(target == NULL) { /* No limit on these */ return TRUE; } else if(te_targets == NULL) { return FALSE; } r = g_hash_table_lookup(te_targets, target); limit = throttle_get_job_limit(target); if(r == NULL) { r = calloc(1, sizeof(struct te_peer_s)); r->name = strdup(target); g_hash_table_insert(te_targets, r->name, r); } if(limit <= r->jobs) { crm_trace("Peer %s is over their job limit of %d (%d): deferring %s", target, limit, r->jobs, id); return FALSE; } else if(graph->migration_limit > 0 && r->migrate_jobs >= graph->migration_limit) { if (pcmk__strcase_any_of(task, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, NULL)) { crm_trace("Peer %s is over their migration job limit of %d (%d): deferring %s", target, graph->migration_limit, r->migrate_jobs, id); return FALSE; } } crm_trace("Peer %s has not hit their limit yet. current jobs = %d limit= %d limit", target, r->jobs, limit); return TRUE; } static gboolean te_should_perform_action(crm_graph_t * graph, crm_action_t * action) { const char *target = NULL; const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); if (action->type != action_type_rsc) { /* No limit on these */ return TRUE; } /* if we have a router node, this means the action is performing * on a remote node. For now, we count all actions occurring on a * remote node against the job list on the cluster node hosting * the connection resources */ target = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); if ((target == NULL) && pcmk__strcase_any_of(task, CRMD_ACTION_MIGRATE, CRMD_ACTION_MIGRATED, NULL)) { target = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_SOURCE); if(te_should_perform_action_on(graph, action, target) == FALSE) { return FALSE; } target = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_TARGET); } else if (target == NULL) { target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); } return te_should_perform_action_on(graph, action, target); } /*! * \brief Confirm a graph action (and optionally update graph) * * \param[in] action Action to confirm * \param[in] graph Update and trigger this graph (if non-NULL) */ void te_action_confirmed(crm_action_t *action, crm_graph_t *graph) { if (!pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { if ((action->type == action_type_rsc) && (crm_element_value(action->xml, XML_LRM_ATTR_TARGET) != NULL)) { te_update_job_count(action, -1); } crm__set_graph_action_flags(action, pcmk__graph_action_confirmed); } if (graph) { pcmk__update_graph(graph, action); trigger_graph(); } } crm_graph_functions_t te_graph_fns = { te_pseudo_action, te_rsc_command, te_crm_command, te_fence_node, te_should_perform_action, }; void notify_crmd(crm_graph_t * graph) { const char *type = "unknown"; enum crmd_fsa_input event = I_NULL; crm_debug("Processing transition completion in state %s", fsa_state2string(fsa_state)); - if (graph->complete == FALSE) { - CRM_CHECK(graph->complete,); - graph->complete = TRUE; - } + CRM_CHECK(graph->complete, graph->complete = TRUE); switch (graph->completion_action) { case tg_stop: type = "stop"; if (fsa_state == S_TRANSITION_ENGINE) { event = I_TE_SUCCESS; } break; case tg_done: type = "done"; if (fsa_state == S_TRANSITION_ENGINE) { event = I_TE_SUCCESS; } break; case tg_restart: type = "restart"; if (fsa_state == S_TRANSITION_ENGINE) { if (transition_timer->period_ms > 0) { controld_stop_timer(transition_timer); controld_start_timer(transition_timer); } else { event = I_PE_CALC; } } else if (fsa_state == S_POLICY_ENGINE) { controld_set_fsa_action_flags(A_PE_INVOKE); trigger_fsa(); } break; case tg_shutdown: type = "shutdown"; if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)) { event = I_STOP; } else { crm_err("We didn't ask to be shut down, yet the scheduler is telling us to"); event = I_TERMINATE; } } crm_debug("Transition %d status: %s - %s", graph->id, type, crm_str(graph->abort_reason)); graph->abort_reason = NULL; graph->completion_action = tg_done; controld_clear_fsa_input_flags(R_IN_TRANSITION); if (event != I_NULL) { register_fsa_input(C_FSA_INTERNAL, event, NULL); } else if (fsa_source) { mainloop_set_trigger(fsa_source); } } diff --git a/daemons/controld/controld_transition.c b/daemons/controld/controld_transition.c index 6a49377589..9225a28aae 100644 --- a/daemons/controld/controld_transition.c +++ b/daemons/controld/controld_transition.c @@ -1,216 +1,213 @@ /* - * Copyright 2004-2021 the Pacemaker project contributors + * Copyright 2004-2022 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 extern crm_graph_functions_t te_graph_fns; static void global_cib_callback(const xmlNode * msg, int callid, int rc, xmlNode * output) { } static crm_graph_t * create_blank_graph(void) { crm_graph_t *a_graph = pcmk__unpack_graph(NULL, NULL); a_graph->complete = TRUE; a_graph->abort_reason = "DC Takeover"; a_graph->completion_action = tg_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) { gboolean init_ok = TRUE; if (action & A_TE_STOP) { if (transition_graph) { pcmk__free_graph(transition_graph); transition_graph = NULL; } if (fsa_cib_conn) { fsa_cib_conn->cmds->del_notify_callback(fsa_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(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 (te_uuid == NULL) { te_uuid = crm_generate_uuid(); crm_info("Registering TE UUID: %s", te_uuid); } if (fsa_cib_conn == NULL) { crm_err("Could not set CIB callbacks"); init_ok = FALSE; } else { if (fsa_cib_conn->cmds->add_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, te_update_diff) != pcmk_ok) { crm_err("Could not set CIB notification callback"); init_ok = FALSE; } if (fsa_cib_conn->cmds->set_op_callback(fsa_cib_conn, global_cib_callback) != pcmk_ok) { crm_err("Could not set CIB global callback"); init_ok = FALSE; } } if (init_ok) { pcmk__set_graph_functions(&te_graph_fns); if (transition_graph) { pcmk__free_graph(transition_graph); } /* create a blank one */ crm_debug("Transitioner is now active"); 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 == FALSE || (fsa_state != S_TRANSITION_ENGINE && (action & A_TE_INVOKE))) { crm_notice("No need to invoke the TE (%s) in state %s", fsa_action2string(action), fsa_state2string(fsa_state)); return; } if (action & A_TE_CANCEL) { crm_debug("Cancelling the transition: %s", transition_graph->complete ? "inactive" : "active"); abort_transition(INFINITY, tg_restart, "Peer Cancelled", NULL); if (transition_graph->complete == FALSE) { crmd_fsa_stall(FALSE); } } else if (action & A_TE_HALT) { abort_transition(INFINITY, tg_stop, "Peer Halt", NULL); if (transition_graph->complete == FALSE) { crmd_fsa_stall(FALSE); } } else if (action & A_TE_INVOKE) { const char *value = NULL; xmlNode *graph_data = NULL; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *ref = crm_element_value(input->msg, XML_ATTR_REFERENCE); const char *graph_file = crm_element_value(input->msg, F_CRM_TGRAPH); const char *graph_input = crm_element_value(input->msg, F_CRM_TGRAPH_INPUT); if (graph_file == NULL && input->xml == NULL) { crm_log_xml_err(input->msg, "Bad command"); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); return; } if (transition_graph->complete == FALSE) { crm_info("Another transition is already active"); abort_transition(INFINITY, tg_restart, "Transition Active", NULL); return; } if (fsa_pe_ref == NULL || !pcmk__str_eq(fsa_pe_ref, ref, pcmk__str_casei)) { crm_info("Transition is redundant: %s vs. %s", crm_str(fsa_pe_ref), crm_str(ref)); abort_transition(INFINITY, tg_restart, "Transition Redundant", NULL); } graph_data = input->xml; if (graph_data == NULL && graph_file != NULL) { graph_data = filename2xml(graph_file); } if (is_timer_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(transition_graph); transition_graph = pcmk__unpack_graph(graph_data, graph_input); - if (transition_graph == NULL) { - CRM_CHECK(transition_graph != NULL,); - transition_graph = create_blank_graph(); - return; - } + CRM_CHECK(transition_graph != NULL, + transition_graph = create_blank_graph(); return); crm_info("Processing graph %d (ref=%s) derived from %s", transition_graph->id, ref, graph_input); te_reset_job_counts(); value = crm_element_value(graph_data, "failed-stop-offset"); if (value) { free(failed_stop_offset); failed_stop_offset = strdup(value); } value = crm_element_value(graph_data, "failed-start-offset"); if (value) { free(failed_start_offset); failed_start_offset = strdup(value); } if ((crm_element_value_epoch(graph_data, "recheck-by", &recheck_by) != pcmk_ok) || (recheck_by < 0)) { recheck_by = 0; } trigger_graph(); pcmk__log_graph(LOG_TRACE, transition_graph); if (graph_data != input->xml) { free_xml(graph_data); } } } diff --git a/daemons/fenced/pacemaker-fenced.c b/daemons/fenced/pacemaker-fenced.c index fbe8c6790f..1b954be5a4 100644 --- a/daemons/fenced/pacemaker-fenced.c +++ b/daemons/fenced/pacemaker-fenced.c @@ -1,1710 +1,1710 @@ /* - * Copyright 2009-2021 the Pacemaker project contributors + * Copyright 2009-2022 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 // PRIu32, PRIx32 #include #include #include #include #include #include #include #include #include #include #include #include #include #include char *stonith_our_uname = NULL; long stonith_watchdog_timeout_ms = 0; GList *stonith_watchdog_targets = NULL; static GMainLoop *mainloop = NULL; gboolean stand_alone = FALSE; static gboolean no_cib_connect = FALSE; static gboolean stonith_shutdown_flag = FALSE; static qb_ipcs_service_t *ipcs = NULL; static xmlNode *local_cib = NULL; static pe_working_set_t *fenced_data_set = NULL; static cib_t *cib_api = NULL; static pcmk__output_t *out = NULL; pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_LOG, PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, { NULL, NULL, NULL } }; static void stonith_shutdown(int nsig); static void stonith_cleanup(void); static int32_t st_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid) { if (stonith_shutdown_flag) { crm_info("Ignoring new client [%d] during shutdown", pcmk__client_pid(c)); return -EPERM; } if (pcmk__new_client(c, uid, gid) == NULL) { return -EIO; } return 0; } /* Exit code means? */ static int32_t st_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; int call_options = 0; xmlNode *request = NULL; pcmk__client_t *c = pcmk__find_client(qbc); const char *op = NULL; if (c == NULL) { crm_info("Invalid client: %p", qbc); return 0; } request = pcmk__client_data2xml(c, data, &id, &flags); if (request == NULL) { pcmk__ipc_send_ack(c, id, flags, "nack", CRM_EX_PROTOCOL); return 0; } op = crm_element_value(request, F_CRM_TASK); if(pcmk__str_eq(op, CRM_OP_RM_NODE_CACHE, pcmk__str_casei)) { crm_xml_add(request, F_TYPE, T_STONITH_NG); crm_xml_add(request, F_STONITH_OPERATION, op); crm_xml_add(request, F_STONITH_CLIENTID, c->id); crm_xml_add(request, F_STONITH_CLIENTNAME, pcmk__client_name(c)); crm_xml_add(request, F_STONITH_CLIENTNODE, stonith_our_uname); send_cluster_message(NULL, crm_msg_stonith_ng, request, FALSE); free_xml(request); return 0; } if (c->name == NULL) { const char *value = crm_element_value(request, F_STONITH_CLIENTNAME); if (value == NULL) { value = "unknown"; } c->name = crm_strdup_printf("%s.%u", value, c->pid); } crm_element_value_int(request, F_STONITH_CALLOPTS, &call_options); crm_trace("Flags %#08" PRIx32 "/%#08x for command %" PRIu32 " from client %s", flags, call_options, id, pcmk__client_name(c)); if (pcmk_is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(flags & crm_ipc_client_response); CRM_LOG_ASSERT(c->request_id == 0); /* This means the client has two synchronous events in-flight */ c->request_id = id; /* Reply only to the last one */ } crm_xml_add(request, F_STONITH_CLIENTID, c->id); crm_xml_add(request, F_STONITH_CLIENTNAME, pcmk__client_name(c)); crm_xml_add(request, F_STONITH_CLIENTNODE, stonith_our_uname); stonith_command(c, id, flags, request, NULL); free_xml(request); return 0; } /* Error code means? */ static int32_t st_ipc_closed(qb_ipcs_connection_t * c) { pcmk__client_t *client = pcmk__find_client(c); if (client == NULL) { return 0; } crm_trace("Connection %p closed", c); pcmk__free_client(client); /* 0 means: yes, go ahead and destroy the connection */ return 0; } static void st_ipc_destroy(qb_ipcs_connection_t * c) { crm_trace("Connection %p destroyed", c); st_ipc_closed(c); } static void stonith_peer_callback(xmlNode * msg, void *private_data) { const char *remote_peer = crm_element_value(msg, F_ORIG); const char *op = crm_element_value(msg, F_STONITH_OPERATION); if (pcmk__str_eq(op, "poke", pcmk__str_none)) { return; } crm_log_xml_trace(msg, "Peer[inbound]"); stonith_command(NULL, 0, 0, msg, remote_peer); } #if SUPPORT_COROSYNC static void stonith_peer_ais_callback(cpg_handle_t handle, const struct cpg_name *groupName, uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len) { uint32_t kind = 0; xmlNode *xml = NULL; const char *from = NULL; char *data = pcmk_message_common_cs(handle, nodeid, pid, msg, &kind, &from); if(data == NULL) { return; } if (kind == crm_class_cluster) { xml = string2xml(data); if (xml == NULL) { crm_err("Invalid XML: '%.120s'", data); free(data); return; } crm_xml_add(xml, F_ORIG, from); /* crm_xml_add_int(xml, F_SEQ, wrapper->id); */ stonith_peer_callback(xml, NULL); } free_xml(xml); free(data); return; } static void stonith_peer_cs_destroy(gpointer user_data) { crm_crit("Lost connection to cluster layer, shutting down"); stonith_shutdown(0); } #endif void do_local_reply(xmlNode * notify_src, const char *client_id, gboolean sync_reply, gboolean from_peer) { /* send callback to originating child */ pcmk__client_t *client_obj = NULL; int local_rc = pcmk_rc_ok; crm_trace("Sending response"); client_obj = pcmk__find_client_by_id(client_id); crm_trace("Sending callback to request originator"); if (client_obj == NULL) { local_rc = EPROTO; crm_trace("No client to sent the response to. F_STONITH_CLIENTID not set."); } else { int rid = 0; if (sync_reply) { CRM_LOG_ASSERT(client_obj->request_id); rid = client_obj->request_id; client_obj->request_id = 0; crm_trace("Sending response %d to client %s%s", rid, pcmk__client_name(client_obj), (from_peer? " (originator of delegated request)" : "")); } else { crm_trace("Sending an event to client %s%s", pcmk__client_name(client_obj), (from_peer? " (originator of delegated request)" : "")); } local_rc = pcmk__ipc_send_xml(client_obj, rid, notify_src, (sync_reply? crm_ipc_flags_none : crm_ipc_server_event)); } if ((local_rc != pcmk_rc_ok) && (client_obj != NULL)) { crm_warn("%s reply to client %s failed: %s", (sync_reply? "Synchronous" : "Asynchronous"), pcmk__client_name(client_obj), pcmk_rc_str(local_rc)); } } uint64_t get_stonith_flag(const char *name) { if (pcmk__str_eq(name, T_STONITH_NOTIFY_FENCE, pcmk__str_casei)) { return st_callback_notify_fence; } else if (pcmk__str_eq(name, STONITH_OP_DEVICE_ADD, pcmk__str_casei)) { return st_callback_device_add; } else if (pcmk__str_eq(name, STONITH_OP_DEVICE_DEL, pcmk__str_casei)) { return st_callback_device_del; } else if (pcmk__str_eq(name, T_STONITH_NOTIFY_HISTORY, pcmk__str_casei)) { return st_callback_notify_history; } else if (pcmk__str_eq(name, T_STONITH_NOTIFY_HISTORY_SYNCED, pcmk__str_casei)) { return st_callback_notify_history_synced; } return st_callback_unknown; } static void stonith_notify_client(gpointer key, gpointer value, gpointer user_data) { xmlNode *update_msg = user_data; pcmk__client_t *client = value; const char *type = NULL; CRM_CHECK(client != NULL, return); CRM_CHECK(update_msg != NULL, return); type = crm_element_value(update_msg, F_SUBTYPE); CRM_CHECK(type != NULL, crm_log_xml_err(update_msg, "notify"); return); if (client->ipcs == NULL) { crm_trace("Skipping client with NULL channel"); return; } if (pcmk_is_set(client->flags, get_stonith_flag(type))) { int rc = pcmk__ipc_send_xml(client, 0, update_msg, crm_ipc_server_event|crm_ipc_server_error); if (rc != pcmk_rc_ok) { crm_warn("%s notification of client %s failed: %s " CRM_XS " id=%.8s rc=%d", type, pcmk__client_name(client), pcmk_rc_str(rc), client->id, rc); } else { crm_trace("Sent %s notification to client %s", type, pcmk__client_name(client)); } } } void do_stonith_async_timeout_update(const char *client_id, const char *call_id, int timeout) { pcmk__client_t *client = NULL; xmlNode *notify_data = NULL; if (!timeout || !call_id || !client_id) { return; } client = pcmk__find_client_by_id(client_id); if (!client) { return; } notify_data = create_xml_node(NULL, T_STONITH_TIMEOUT_VALUE); crm_xml_add(notify_data, F_TYPE, T_STONITH_TIMEOUT_VALUE); crm_xml_add(notify_data, F_STONITH_CALLID, call_id); crm_xml_add_int(notify_data, F_STONITH_TIMEOUT, timeout); crm_trace("timeout update is %d for client %s and call id %s", timeout, client_id, call_id); if (client) { pcmk__ipc_send_xml(client, 0, notify_data, crm_ipc_server_event); } free_xml(notify_data); } /*! * \internal * \brief Notify relevant IPC clients of a fencing operation result * * \param[in] type Notification type * \param[in] result Result of fencing operation (assume success if NULL) * \param[in] data If not NULL, add to notification as call data */ void fenced_send_notification(const char *type, const pcmk__action_result_t *result, xmlNode *data) { /* TODO: Standardize the contents of data */ xmlNode *update_msg = create_xml_node(NULL, "notify"); - CRM_CHECK(type != NULL,;); + CRM_LOG_ASSERT(type != NULL); crm_xml_add(update_msg, F_TYPE, T_STONITH_NOTIFY); crm_xml_add(update_msg, F_SUBTYPE, type); crm_xml_add(update_msg, F_STONITH_OPERATION, type); stonith__xe_set_result(update_msg, result); if (data != NULL) { add_message_xml(update_msg, F_STONITH_CALLDATA, data); } crm_trace("Notifying clients"); pcmk__foreach_ipc_client(stonith_notify_client, update_msg); free_xml(update_msg); crm_trace("Notify complete"); } /*! * \internal * \brief Send notifications for a configuration change to subscribed clients * * \param[in] op Notification type (STONITH_OP_DEVICE_ADD, * STONITH_OP_DEVICE_DEL, STONITH_OP_LEVEL_ADD, or * STONITH_OP_LEVEL_DEL) * \param[in] result Operation result * \param[in] desc Description of what changed * \param[in] active Current number of devices or topologies in use */ static void send_config_notification(const char *op, const pcmk__action_result_t *result, const char *desc, int active) { xmlNode *notify_data = create_xml_node(NULL, op); CRM_CHECK(notify_data != NULL, return); crm_xml_add(notify_data, F_STONITH_DEVICE, desc); crm_xml_add_int(notify_data, F_STONITH_ACTIVE, active); fenced_send_notification(op, result, notify_data); free_xml(notify_data); } /*! * \internal * \brief Send notifications for a device change to subscribed clients * * \param[in] op Notification type (STONITH_OP_DEVICE_ADD or * STONITH_OP_DEVICE_DEL) * \param[in] result Operation result * \param[in] desc ID of device that changed */ void fenced_send_device_notification(const char *op, const pcmk__action_result_t *result, const char *desc) { send_config_notification(op, result, desc, g_hash_table_size(device_list)); } /*! * \internal * \brief Send notifications for a topology level change to subscribed clients * * \param[in] op Notification type (STONITH_OP_LEVEL_ADD or * STONITH_OP_LEVEL_DEL) * \param[in] result Operation result * \param[in] desc String representation of level ([]) */ void fenced_send_level_notification(const char *op, const pcmk__action_result_t *result, const char *desc) { send_config_notification(op, result, desc, g_hash_table_size(topology)); } static void topology_remove_helper(const char *node, int level) { char *desc = NULL; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; xmlNode *data = create_xml_node(NULL, XML_TAG_FENCING_LEVEL); crm_xml_add(data, F_STONITH_ORIGIN, __func__); crm_xml_add_int(data, XML_ATTR_STONITH_INDEX, level); crm_xml_add(data, XML_ATTR_STONITH_TARGET, node); fenced_unregister_level(data, &desc, &result); fenced_send_level_notification(STONITH_OP_LEVEL_DEL, &result, desc); pcmk__reset_result(&result); free_xml(data); free(desc); } static void remove_cib_device(xmlXPathObjectPtr xpathObj) { int max = numXpathResults(xpathObj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { const char *rsc_id = NULL; const char *standard = NULL; xmlNode *match = getXpathResult(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if(match != NULL) { standard = crm_element_value(match, XML_AGENT_ATTR_CLASS); } if (!pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { continue; } rsc_id = crm_element_value(match, XML_ATTR_ID); stonith_device_remove(rsc_id, true); } } static void handle_topology_change(xmlNode *match, bool remove) { char *desc = NULL; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; CRM_CHECK(match != NULL, return); crm_trace("Updating %s", ID(match)); if(remove) { int index = 0; char *key = stonith_level_key(match, -1); crm_element_value_int(match, XML_ATTR_STONITH_INDEX, &index); topology_remove_helper(key, index); free(key); } fenced_register_level(match, &desc, &result); fenced_send_level_notification(STONITH_OP_LEVEL_ADD, &result, desc); pcmk__reset_result(&result); free(desc); } static void remove_fencing_topology(xmlXPathObjectPtr xpathObj) { int max = numXpathResults(xpathObj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if (match && crm_element_value(match, XML_DIFF_MARKER)) { /* Deletion */ int index = 0; char *target = stonith_level_key(match, -1); crm_element_value_int(match, XML_ATTR_STONITH_INDEX, &index); if (target == NULL) { crm_err("Invalid fencing target in element %s", ID(match)); } else if (index <= 0) { crm_err("Invalid level for %s in element %s", target, ID(match)); } else { topology_remove_helper(target, index); } /* } else { Deal with modifications during the 'addition' stage */ } } } static void register_fencing_topology(xmlXPathObjectPtr xpathObj) { int max = numXpathResults(xpathObj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpathObj, lpc); handle_topology_change(match, TRUE); } } /* Fencing */ static void fencing_topology_init(void) { xmlXPathObjectPtr xpathObj = NULL; const char *xpath = "//" XML_TAG_FENCING_LEVEL; crm_trace("Full topology refresh"); free_topology_list(); init_topology_list(); /* Grab everything */ xpathObj = xpath_search(local_cib, xpath); register_fencing_topology(xpathObj); freeXpathObject(xpathObj); } #define rsc_name(x) x->clone_name?x->clone_name:x->id /*! * \internal * \brief Check whether our uname is in a resource's allowed node list * * \param[in] rsc Resource to check * * \return Pointer to node object if found, NULL otherwise */ static pe_node_t * our_node_allowed_for(pe_resource_t *rsc) { GHashTableIter iter; pe_node_t *node = NULL; if (rsc && stonith_our_uname) { g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if (node && strcmp(node->details->uname, stonith_our_uname) == 0) { break; } node = NULL; } } return node; } static void watchdog_device_update(void) { if (stonith_watchdog_timeout_ms > 0) { if (!g_hash_table_lookup(device_list, STONITH_WATCHDOG_ID) && !stonith_watchdog_targets) { /* getting here watchdog-fencing enabled, no device there yet and reason isn't stonith_watchdog_targets preventing that */ int rc; xmlNode *xml; xml = create_device_registration_xml( STONITH_WATCHDOG_ID, st_namespace_internal, STONITH_WATCHDOG_AGENT, NULL, /* stonith_device_register will add our own name as PCMK_STONITH_HOST_LIST param so we can skip that here */ NULL); rc = stonith_device_register(xml, NULL, TRUE); free_xml(xml); if (rc != pcmk_ok) { crm_crit("Cannot register watchdog pseudo fence agent"); crm_exit(CRM_EX_FATAL); } } } else { /* be silent if no device - todo parameter to stonith_device_remove */ if (g_hash_table_lookup(device_list, STONITH_WATCHDOG_ID)) { stonith_device_remove(STONITH_WATCHDOG_ID, true); } } } static void update_stonith_watchdog_timeout_ms(xmlNode *cib) { xmlNode *stonith_enabled_xml = NULL; bool stonith_enabled = false; int rc = pcmk_rc_ok; long timeout_ms = 0; stonith_enabled_xml = get_xpath_object("//nvpair[@name='stonith-enabled']", cib, LOG_NEVER); rc = pcmk__xe_get_bool_attr(stonith_enabled_xml, XML_NVPAIR_ATTR_VALUE, &stonith_enabled); if (rc != pcmk_rc_ok || stonith_enabled) { xmlNode *stonith_watchdog_xml = NULL; const char *value = NULL; stonith_watchdog_xml = get_xpath_object("//nvpair[@name='stonith-watchdog-timeout']", cib, LOG_NEVER); if (stonith_watchdog_xml) { value = crm_element_value(stonith_watchdog_xml, XML_NVPAIR_ATTR_VALUE); } if (value) { timeout_ms = crm_get_msec(value); } if (timeout_ms < 0) { timeout_ms = pcmk__auto_watchdog_timeout(); } } stonith_watchdog_timeout_ms = timeout_ms; } /*! * \internal * \brief If a resource or any of its children are STONITH devices, update their * definitions given a cluster working set. * * \param[in] rsc Resource to check * \param[in] data_set Cluster working set with device information */ static void cib_device_update(pe_resource_t *rsc, pe_working_set_t *data_set) { pe_node_t *node = NULL; const char *value = NULL; const char *rclass = NULL; pe_node_t *parent = NULL; /* If this is a complex resource, check children rather than this resource itself. */ if(rsc->children) { GList *gIter = NULL; for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) { cib_device_update(gIter->data, data_set); if(pe_rsc_is_clone(rsc)) { crm_trace("Only processing one copy of the clone %s", rsc->id); break; } } return; } /* We only care about STONITH resources. */ rclass = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); if (!pcmk__str_eq(rclass, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { return; } /* If this STONITH resource is disabled, remove it. */ if (pe__resource_is_disabled(rsc)) { crm_info("Device %s has been disabled", rsc->id); return; } /* if watchdog-fencing is disabled handle any watchdog-fence resource as if it was disabled */ if ((stonith_watchdog_timeout_ms <= 0) && pcmk__str_eq(rsc->id, STONITH_WATCHDOG_ID, pcmk__str_none)) { crm_info("Watchdog-fencing disabled thus handling " "device %s as disabled", rsc->id); return; } /* Check whether our node is allowed for this resource (and its parent if in a group) */ node = our_node_allowed_for(rsc); if (rsc->parent && (rsc->parent->variant == pe_group)) { parent = our_node_allowed_for(rsc->parent); } if(node == NULL) { /* Our node is disallowed, so remove the device */ GHashTableIter iter; crm_info("Device %s has been disabled on %s: unknown", rsc->id, stonith_our_uname); g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { crm_trace("Available: %s = %d", node->details->uname, node->weight); } return; } else if(node->weight < 0 || (parent && parent->weight < 0)) { /* Our node (or its group) is disallowed by score, so remove the device */ char *score = score2char((node->weight < 0) ? node->weight : parent->weight); crm_info("Device %s has been disabled on %s: score=%s", rsc->id, stonith_our_uname, score); free(score); return; } else { /* Our node is allowed, so update the device information */ int rc; xmlNode *data; GHashTable *rsc_params = NULL; GHashTableIter gIter; stonith_key_value_t *params = NULL; const char *name = NULL; const char *agent = crm_element_value(rsc->xml, XML_EXPR_ATTR_TYPE); const char *rsc_provides = NULL; crm_debug("Device %s is allowed on %s: score=%d", rsc->id, stonith_our_uname, node->weight); rsc_params = pe_rsc_params(rsc, node, data_set); get_meta_attributes(rsc->meta, rsc, node, data_set); rsc_provides = g_hash_table_lookup(rsc->meta, PCMK_STONITH_PROVIDES); g_hash_table_iter_init(&gIter, rsc_params); while (g_hash_table_iter_next(&gIter, (gpointer *) & name, (gpointer *) & value)) { if (!name || !value) { continue; } params = stonith_key_value_add(params, name, value); crm_trace(" %s=%s", name, value); } data = create_device_registration_xml(rsc_name(rsc), st_namespace_any, agent, params, rsc_provides); stonith_key_value_freeall(params, 1, 1); rc = stonith_device_register(data, NULL, TRUE); CRM_ASSERT(rc == pcmk_ok); free_xml(data); } } /*! * \internal * \brief Update all STONITH device definitions based on current CIB */ static void cib_devices_update(void) { GHashTableIter iter; stonith_device_t *device = NULL; crm_info("Updating devices to version %s.%s.%s", crm_element_value(local_cib, XML_ATTR_GENERATION_ADMIN), crm_element_value(local_cib, XML_ATTR_GENERATION), crm_element_value(local_cib, XML_ATTR_NUMUPDATES)); CRM_ASSERT(fenced_data_set != NULL); fenced_data_set->input = local_cib; fenced_data_set->now = crm_time_new(NULL); fenced_data_set->localhost = stonith_our_uname; pe__set_working_set_flags(fenced_data_set, pe_flag_quick_location); cluster_status(fenced_data_set); pcmk__schedule_actions(fenced_data_set, NULL, NULL); g_hash_table_iter_init(&iter, device_list); while (g_hash_table_iter_next(&iter, NULL, (void **)&device)) { if (device->cib_registered) { device->dirty = TRUE; } } /* have list repopulated if cib has a watchdog-fencing-resource TODO: keep a cached list for queries happening while we are refreshing */ g_list_free_full(stonith_watchdog_targets, free); stonith_watchdog_targets = NULL; g_list_foreach(fenced_data_set->resources, (GFunc) cib_device_update, fenced_data_set); g_hash_table_iter_init(&iter, device_list); while (g_hash_table_iter_next(&iter, NULL, (void **)&device)) { if (device->dirty) { g_hash_table_iter_remove(&iter); } } fenced_data_set->input = NULL; // Wasn't a copy, so don't let API free it pe_reset_working_set(fenced_data_set); } static void update_cib_stonith_devices_v2(const char *event, xmlNode * msg) { xmlNode *change = NULL; char *reason = NULL; bool needs_update = FALSE; xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); for (change = pcmk__xml_first_child(patchset); change != NULL; change = pcmk__xml_next(change)) { const char *op = crm_element_value(change, XML_DIFF_OP); const char *xpath = crm_element_value(change, XML_DIFF_PATH); const char *shortpath = NULL; if ((op == NULL) || (strcmp(op, "move") == 0) || strstr(xpath, "/"XML_CIB_TAG_STATUS)) { continue; } else if (pcmk__str_eq(op, "delete", pcmk__str_casei) && strstr(xpath, "/"XML_CIB_TAG_RESOURCE)) { const char *rsc_id = NULL; char *search = NULL; char *mutable = NULL; if (strstr(xpath, XML_TAG_ATTR_SETS) || strstr(xpath, XML_TAG_META_SETS)) { needs_update = TRUE; reason = strdup("(meta) attribute deleted from resource"); break; } mutable = strdup(xpath); rsc_id = strstr(mutable, "primitive[@id=\'"); if (rsc_id != NULL) { rsc_id += strlen("primitive[@id=\'"); search = strchr(rsc_id, '\''); } if (search != NULL) { *search = 0; stonith_device_remove(rsc_id, true); /* watchdog_device_update called afterwards to fall back to implicit definition if needed */ } else { crm_warn("Ignoring malformed CIB update (resource deletion)"); } free(mutable); } else if (strstr(xpath, "/"XML_CIB_TAG_RESOURCES) || strstr(xpath, "/"XML_CIB_TAG_CONSTRAINTS) || strstr(xpath, "/"XML_CIB_TAG_RSCCONFIG)) { shortpath = strrchr(xpath, '/'); CRM_ASSERT(shortpath); reason = crm_strdup_printf("%s %s", op, shortpath+1); needs_update = TRUE; break; } } if(needs_update) { crm_info("Updating device list from CIB: %s", reason); cib_devices_update(); } else { crm_trace("No updates for device list found in CIB"); } free(reason); } static void update_cib_stonith_devices_v1(const char *event, xmlNode * msg) { const char *reason = "none"; gboolean needs_update = FALSE; xmlXPathObjectPtr xpath_obj = NULL; /* process new constraints */ xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" XML_CONS_TAG_RSC_LOCATION); if (numXpathResults(xpath_obj) > 0) { int max = numXpathResults(xpath_obj), lpc = 0; /* Safest and simplest to always recompute */ needs_update = TRUE; reason = "new location constraint"; for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpath_obj, lpc); crm_log_xml_trace(match, "new constraint"); } } freeXpathObject(xpath_obj); /* process deletions */ xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_CIB_TAG_RESOURCE); if (numXpathResults(xpath_obj) > 0) { remove_cib_device(xpath_obj); } freeXpathObject(xpath_obj); /* process additions */ xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_CIB_TAG_RESOURCE); if (numXpathResults(xpath_obj) > 0) { int max = numXpathResults(xpath_obj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { const char *rsc_id = NULL; const char *standard = NULL; xmlNode *match = getXpathResult(xpath_obj, lpc); rsc_id = crm_element_value(match, XML_ATTR_ID); standard = crm_element_value(match, XML_AGENT_ATTR_CLASS); if (!pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { continue; } crm_trace("Fencing resource %s was added or modified", rsc_id); reason = "new resource"; needs_update = TRUE; } } freeXpathObject(xpath_obj); if(needs_update) { crm_info("Updating device list from CIB: %s", reason); cib_devices_update(); } } static void update_cib_stonith_devices(const char *event, xmlNode * msg) { int format = 1; xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); CRM_ASSERT(patchset); crm_element_value_int(patchset, "format", &format); switch(format) { case 1: update_cib_stonith_devices_v1(event, msg); break; case 2: update_cib_stonith_devices_v2(event, msg); break; default: crm_warn("Unknown patch format: %d", format); } } /* Needs to hold node name + attribute name + attribute value + 75 */ #define XPATH_MAX 512 /*! * \internal * \brief Check whether a node has a specific attribute name/value * * \param[in] node Name of node to check * \param[in] name Name of an attribute to look for * \param[in] value The value the named attribute needs to be set to in order to be considered a match * * \return TRUE if the locally cached CIB has the specified node attribute */ gboolean node_has_attr(const char *node, const char *name, const char *value) { char xpath[XPATH_MAX]; xmlNode *match; int n; CRM_CHECK(local_cib != NULL, return FALSE); /* Search for the node's attributes in the CIB. While the schema allows * multiple sets of instance attributes, and allows instance attributes to * use id-ref to reference values elsewhere, that is intended for resources, * so we ignore that here. */ n = snprintf(xpath, XPATH_MAX, "//" XML_CIB_TAG_NODES "/" XML_CIB_TAG_NODE "[@uname='%s']/" XML_TAG_ATTR_SETS "/" XML_CIB_TAG_NVPAIR "[@name='%s' and @value='%s']", node, name, value); match = get_xpath_object(xpath, local_cib, LOG_NEVER); CRM_CHECK(n < XPATH_MAX, return FALSE); return (match != NULL); } /*! * \internal * \brief Check whether a node does watchdog-fencing * * \param[in] node Name of node to check * * \return TRUE if node found in stonith_watchdog_targets * or stonith_watchdog_targets is empty indicating * all nodes are doing watchdog-fencing */ gboolean node_does_watchdog_fencing(const char *node) { return ((stonith_watchdog_targets == NULL) || pcmk__str_in_list(node, stonith_watchdog_targets, pcmk__str_casei)); } static void update_fencing_topology(const char *event, xmlNode * msg) { int format = 1; const char *xpath; xmlXPathObjectPtr xpathObj = NULL; xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); CRM_ASSERT(patchset); crm_element_value_int(patchset, "format", &format); if(format == 1) { /* Process deletions (only) */ xpath = "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_TAG_FENCING_LEVEL; xpathObj = xpath_search(msg, xpath); remove_fencing_topology(xpathObj); freeXpathObject(xpathObj); /* Process additions and changes */ xpath = "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_TAG_FENCING_LEVEL; xpathObj = xpath_search(msg, xpath); register_fencing_topology(xpathObj); freeXpathObject(xpathObj); } else if(format == 2) { xmlNode *change = NULL; int add[] = { 0, 0, 0 }; int del[] = { 0, 0, 0 }; xml_patch_versions(patchset, add, del); for (change = pcmk__xml_first_child(patchset); change != NULL; change = pcmk__xml_next(change)) { const char *op = crm_element_value(change, XML_DIFF_OP); const char *xpath = crm_element_value(change, XML_DIFF_PATH); if(op == NULL) { continue; } else if(strstr(xpath, "/" XML_TAG_FENCING_LEVEL) != NULL) { /* Change to a specific entry */ crm_trace("Handling %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); if(strcmp(op, "move") == 0) { continue; } else if(strcmp(op, "create") == 0) { handle_topology_change(change->children, FALSE); } else if(strcmp(op, "modify") == 0) { xmlNode *match = first_named_child(change, XML_DIFF_RESULT); if(match) { handle_topology_change(match->children, TRUE); } } else if(strcmp(op, "delete") == 0) { /* Nuclear option, all we have is the path and an id... not enough to remove a specific entry */ crm_info("Re-initializing fencing topology after %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); fencing_topology_init(); return; } } else if (strstr(xpath, "/" XML_TAG_FENCING_TOPOLOGY) != NULL) { /* Change to the topology in general */ crm_info("Re-initializing fencing topology after top-level %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); fencing_topology_init(); return; } else if (strstr(xpath, "/" XML_CIB_TAG_CONFIGURATION)) { /* Changes to the whole config section, possibly including the topology as a whild */ if(first_named_child(change, XML_TAG_FENCING_TOPOLOGY) == NULL) { crm_trace("Nothing for us in %s operation %d.%d.%d for %s.", op, add[0], add[1], add[2], xpath); } else if(strcmp(op, "delete") == 0 || strcmp(op, "create") == 0) { crm_info("Re-initializing fencing topology after top-level %s operation %d.%d.%d for %s.", op, add[0], add[1], add[2], xpath); fencing_topology_init(); return; } } else { crm_trace("Nothing for us in %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); } } } else { crm_warn("Unknown patch format: %d", format); } } static bool have_cib_devices = FALSE; static void update_cib_cache_cb(const char *event, xmlNode * msg) { int rc = pcmk_ok; xmlNode *stonith_enabled_xml = NULL; static gboolean stonith_enabled_saved = TRUE; long timeout_ms_saved = stonith_watchdog_timeout_ms; gboolean need_full_refresh = FALSE; bool value = false; if(!have_cib_devices) { crm_trace("Skipping updates until we get a full dump"); return; } else if(msg == NULL) { crm_trace("Missing %s update", event); return; } /* Maintain a local copy of the CIB so that we have full access * to device definitions, location constraints, and node attributes */ if (local_cib != NULL) { int rc = pcmk_ok; xmlNode *patchset = NULL; crm_element_value_int(msg, F_CIB_RC, &rc); if (rc != pcmk_ok) { return; } patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); xml_log_patchset(LOG_TRACE, "Config update", patchset); rc = xml_apply_patchset(local_cib, patchset, TRUE); switch (rc) { case pcmk_ok: case -pcmk_err_old_data: break; case -pcmk_err_diff_resync: case -pcmk_err_diff_failed: crm_notice("[%s] Patch aborted: %s (%d)", event, pcmk_strerror(rc), rc); free_xml(local_cib); local_cib = NULL; break; default: crm_warn("[%s] ABORTED: %s (%d)", event, pcmk_strerror(rc), rc); free_xml(local_cib); local_cib = NULL; } } if (local_cib == NULL) { crm_trace("Re-requesting full CIB"); rc = cib_api->cmds->query(cib_api, NULL, &local_cib, cib_scope_local | cib_sync_call); if(rc != pcmk_ok) { crm_err("Couldn't retrieve the CIB: %s (%d)", pcmk_strerror(rc), rc); return; } CRM_ASSERT(local_cib != NULL); stonith_enabled_saved = FALSE; /* Trigger a full refresh below */ } pcmk__refresh_node_caches_from_cib(local_cib); update_stonith_watchdog_timeout_ms(local_cib); stonith_enabled_xml = get_xpath_object("//nvpair[@name='stonith-enabled']", local_cib, LOG_NEVER); if (pcmk__xe_get_bool_attr(stonith_enabled_xml, XML_NVPAIR_ATTR_VALUE, &value) == pcmk_rc_ok && !value) { crm_trace("Ignoring CIB updates while fencing is disabled"); stonith_enabled_saved = FALSE; } else if (stonith_enabled_saved == FALSE) { crm_info("Updating fencing device and topology lists " "now that fencing is enabled"); stonith_enabled_saved = TRUE; need_full_refresh = TRUE; } else { if (timeout_ms_saved != stonith_watchdog_timeout_ms) { need_full_refresh = TRUE; } else { update_fencing_topology(event, msg); update_cib_stonith_devices(event, msg); watchdog_device_update(); } } if (need_full_refresh) { fencing_topology_init(); cib_devices_update(); watchdog_device_update(); } } static void init_cib_cache_cb(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { crm_info("Updating device list from CIB"); have_cib_devices = TRUE; local_cib = copy_xml(output); pcmk__refresh_node_caches_from_cib(local_cib); update_stonith_watchdog_timeout_ms(local_cib); fencing_topology_init(); cib_devices_update(); watchdog_device_update(); } static void stonith_shutdown(int nsig) { crm_info("Terminating with %d clients", pcmk__ipc_client_count()); stonith_shutdown_flag = TRUE; if (mainloop != NULL && g_main_loop_is_running(mainloop)) { g_main_loop_quit(mainloop); } else { stonith_cleanup(); crm_exit(CRM_EX_OK); } } static void cib_connection_destroy(gpointer user_data) { if (stonith_shutdown_flag) { crm_info("Connection to the CIB manager closed"); return; } else { crm_crit("Lost connection to the CIB manager, shutting down"); } if (cib_api) { cib_api->cmds->signoff(cib_api); } stonith_shutdown(0); } static void stonith_cleanup(void) { if (cib_api) { cib_api->cmds->del_notify_callback(cib_api, T_CIB_DIFF_NOTIFY, update_cib_cache_cb); cib_api->cmds->signoff(cib_api); } if (ipcs) { qb_ipcs_destroy(ipcs); } crm_peer_destroy(); pcmk__client_cleanup(); free_stonith_remote_op_list(); free_topology_list(); free_device_list(); free_metadata_cache(); free(stonith_our_uname); stonith_our_uname = NULL; free_xml(local_cib); local_cib = NULL; } static pcmk__cli_option_t long_options[] = { // long option, argument type, storage, short option, description, flags { "stand-alone", no_argument, 0, 's', NULL, pcmk__option_default }, { "stand-alone-w-cpg", no_argument, 0, 'c', NULL, pcmk__option_default }, { "logfile", required_argument, 0, 'l', NULL, pcmk__option_default }, { "verbose", no_argument, 0, 'V', NULL, pcmk__option_default }, { "version", no_argument, 0, '$', NULL, pcmk__option_default }, { "help", no_argument, 0, '?', NULL, pcmk__option_default }, { 0, 0, 0, 0 } }; static void setup_cib(void) { int rc, retries = 0; cib_api = cib_new(); if (cib_api == NULL) { crm_err("No connection to the CIB manager"); return; } do { sleep(retries); rc = cib_api->cmds->signon(cib_api, CRM_SYSTEM_STONITHD, cib_command); } while (rc == -ENOTCONN && ++retries < 5); if (rc != pcmk_ok) { crm_err("Could not connect to the CIB manager: %s (%d)", pcmk_strerror(rc), rc); } else if (pcmk_ok != cib_api->cmds->add_notify_callback(cib_api, T_CIB_DIFF_NOTIFY, update_cib_cache_cb)) { crm_err("Could not set CIB notification callback"); } else { rc = cib_api->cmds->query(cib_api, NULL, NULL, cib_scope_local); cib_api->cmds->register_callback(cib_api, rc, 120, FALSE, NULL, "init_cib_cache_cb", init_cib_cache_cb); cib_api->cmds->set_connection_dnotify(cib_api, cib_connection_destroy); crm_info("Watching for fencing topology changes"); } } struct qb_ipcs_service_handlers ipc_callbacks = { .connection_accept = st_ipc_accept, .connection_created = NULL, .msg_process = st_ipc_dispatch, .connection_closed = st_ipc_closed, .connection_destroyed = st_ipc_destroy }; /*! * \internal * \brief Callback for peer status changes * * \param[in] type What changed * \param[in] node What peer had the change * \param[in] data Previous value of what changed */ static void st_peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data) { if ((type != crm_status_processes) && !pcmk_is_set(node->flags, crm_remote_node)) { /* * This is a hack until we can send to a nodeid and/or we fix node name lookups * These messages are ignored in stonith_peer_callback() */ xmlNode *query = create_xml_node(NULL, "stonith_command"); crm_xml_add(query, F_XML_TAGNAME, "stonith_command"); crm_xml_add(query, F_TYPE, T_STONITH_NG); crm_xml_add(query, F_STONITH_OPERATION, "poke"); crm_debug("Broadcasting our uname because of node %u", node->id); send_cluster_message(NULL, crm_msg_stonith_ng, query, FALSE); free_xml(query); } } int main(int argc, char **argv) { int flag; int lpc = 0; int argerr = 0; int option_index = 0; crm_cluster_t *cluster = NULL; const char *actions[] = { "reboot", "off", "on", "list", "monitor", "status" }; crm_ipc_t *old_instance = NULL; int rc = pcmk_rc_ok; crm_log_preinit(NULL, argc, argv); pcmk__set_cli_options(NULL, "[options]", long_options, "daemon for executing fencing devices in a " "Pacemaker cluster"); while (1) { flag = pcmk__next_cli_option(argc, argv, &option_index, NULL); if (flag == -1) { break; } switch (flag) { case 'V': crm_bump_log_level(argc, argv); break; case 'l': { int rc = pcmk__add_logfile(optarg); if (rc != pcmk_rc_ok) { /* Logging has not yet been initialized, so stderr is * the only way to get information out */ fprintf(stderr, "Logging to %s is disabled: %s\n", optarg, pcmk_rc_str(rc)); } } break; case 's': stand_alone = TRUE; break; case 'c': stand_alone = FALSE; no_cib_connect = TRUE; break; case '$': case '?': pcmk__cli_help(flag, CRM_EX_OK); break; default: ++argerr; break; } } if (argc - optind == 1 && pcmk__str_eq("metadata", argv[optind], pcmk__str_casei)) { printf("\n"); printf("\n"); printf(" 1.0\n"); printf(" Instance attributes available for all \"stonith\"-class resources" " and used by Pacemaker's fence daemon, formerly known as stonithd\n"); printf(" Instance attributes available for all \"stonith\"-class resources\n"); printf(" \n"); #if 0 // priority is not implemented yet printf(" \n"); printf(" Devices that are not in a topology " "are tried in order of highest to lowest integer priority\n"); printf(" \n"); printf(" \n"); #endif printf(" \n", PCMK_STONITH_HOST_ARGUMENT); printf (" Advanced use only: An alternate parameter to supply instead of 'port'\n"); printf (" Some devices do not support the standard 'port' parameter or may provide additional ones.\n" "Use this to specify an alternate, device-specific, parameter that should indicate the machine to be fenced.\n" "A value of 'none' can be used to tell the cluster not to supply any additional parameters.\n" " \n"); printf(" \n"); printf(" \n"); printf(" \n", PCMK_STONITH_HOST_MAP); printf (" A mapping of host names to ports numbers for devices that do not support host names.\n"); printf (" Eg. node1:1;node2:2,3 would tell the cluster to use port 1 for node1 and ports 2 and 3 for node2\n"); printf(" \n"); printf(" \n"); printf(" \n", PCMK_STONITH_HOST_LIST); printf(" A list of machines controlled by " "this device (Optional unless %s=static-list).\n", PCMK_STONITH_HOST_CHECK); printf(" \n"); printf(" \n"); printf(" \n", PCMK_STONITH_HOST_CHECK); printf (" How to determine which machines are controlled by the device.\n"); printf(" Allowed values: dynamic-list " "(query the device via the 'list' command), static-list " "(check the " PCMK_STONITH_HOST_LIST " attribute), status " "(query the device via the 'status' command), none (assume " "every device can fence every machine)\n"); printf(" \n"); printf(" \n"); printf(" \n", PCMK_STONITH_DELAY_MAX); printf(" Enable a delay of no more than the " "time specified before executing fencing actions. Pacemaker " "derives the overall delay by taking the value of " PCMK_STONITH_DELAY_BASE " and adding a random delay value such " "that the sum is kept below this maximum.\n"); printf(" This prevents double fencing when " "using slow devices such as sbd.\nUse this to enable a random " "delay for fencing actions.\nThe overall delay is derived from " "this random delay value adding a static delay so that the sum " "is kept below the maximum delay.\n"); printf(" \n"); printf(" \n"); printf(" \n", PCMK_STONITH_DELAY_BASE); printf(" Enable a base delay for " "fencing actions and specify base delay value.\n"); printf(" This prevents double fencing when " "different delays are configured on the nodes.\nUse this to " "enable a static delay for fencing actions.\nThe overall delay " "is derived from a random delay value adding this static delay " "so that the sum is kept below the maximum delay.\nSet to eg. " "node1:1s;node2:5 to set different value per node.\n"); printf(" \n"); printf(" \n"); printf(" \n", PCMK_STONITH_ACTION_LIMIT); printf (" The maximum number of actions can be performed in parallel on this device\n"); printf (" Cluster property concurrent-fencing=true needs to be configured first.\n" "Then use this to specify the maximum number of actions can be performed in parallel on this device. -1 is unlimited.\n"); printf(" \n"); printf(" \n"); for (lpc = 0; lpc < PCMK__NELEM(actions); lpc++) { printf(" \n", actions[lpc]); printf (" Advanced use only: An alternate command to run instead of '%s'\n", actions[lpc]); printf (" Some devices do not support the standard commands or may provide additional ones.\n" "Use this to specify an alternate, device-specific, command that implements the '%s' action.\n", actions[lpc]); printf(" \n", actions[lpc]); printf(" \n"); printf(" \n", actions[lpc]); printf (" Advanced use only: Specify an alternate timeout to use for %s actions instead of stonith-timeout\n", actions[lpc]); printf (" Some devices need much more/less time to complete than normal.\n" "Use this to specify an alternate, device-specific, timeout for '%s' actions.\n", actions[lpc]); printf(" \n"); printf(" \n"); printf(" \n", actions[lpc]); printf (" Advanced use only: The maximum number of times to retry the '%s' command within the timeout period\n", actions[lpc]); printf(" Some devices do not support multiple connections." " Operations may 'fail' if the device is busy with another task so Pacemaker will automatically retry the operation, if there is time remaining." " Use this option to alter the number of times Pacemaker retries '%s' actions before giving up." "\n", actions[lpc]); printf(" \n"); printf(" \n"); } printf(" \n"); printf("\n"); return CRM_EX_OK; } if (optind != argc) { ++argerr; } if (argerr) { pcmk__cli_help('?', CRM_EX_USAGE); } crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); crm_notice("Starting Pacemaker fencer"); old_instance = crm_ipc_new("stonith-ng", 0); if (crm_ipc_connect(old_instance)) { /* IPC end-point already up */ crm_ipc_close(old_instance); crm_ipc_destroy(old_instance); crm_err("pacemaker-fenced is already active, aborting startup"); crm_exit(CRM_EX_OK); } else { /* not up or not authentic, we'll proceed either way */ crm_ipc_destroy(old_instance); old_instance = NULL; } mainloop_add_signal(SIGTERM, stonith_shutdown); crm_peer_init(); fenced_data_set = pe_new_working_set(); CRM_ASSERT(fenced_data_set != NULL); pe__set_working_set_flags(fenced_data_set, pe_flag_no_counts|pe_flag_no_compat); pe__set_working_set_flags(fenced_data_set, pe_flag_show_utilization); cluster = calloc(1, sizeof(crm_cluster_t)); CRM_ASSERT(cluster != NULL); if (stand_alone == FALSE) { if (is_corosync_cluster()) { #if SUPPORT_COROSYNC cluster->destroy = stonith_peer_cs_destroy; cluster->cpg.cpg_deliver_fn = stonith_peer_ais_callback; cluster->cpg.cpg_confchg_fn = pcmk_cpg_membership; #endif } crm_set_status_callback(&st_peer_update_callback); if (crm_cluster_connect(cluster) == FALSE) { crm_crit("Cannot sign in to the cluster... terminating"); crm_exit(CRM_EX_FATAL); } stonith_our_uname = strdup(cluster->uname); if (no_cib_connect == FALSE) { setup_cib(); } } else { stonith_our_uname = strdup("localhost"); } init_device_list(); init_topology_list(); pcmk__serve_fenced_ipc(&ipcs, &ipc_callbacks); pcmk__register_formats(NULL, formats); rc = pcmk__output_new(&out, "log", NULL, argv); if ((rc != pcmk_rc_ok) || (out == NULL)) { crm_err("Can't log resource details due to internal error: %s\n", pcmk_rc_str(rc)); crm_exit(CRM_EX_FATAL); } pe__register_messages(out); pcmk__register_lib_messages(out); pcmk__output_set_log_level(out, LOG_TRACE); fenced_data_set->priv = out; /* Create the mainloop and run it... */ mainloop = g_main_loop_new(NULL, FALSE); crm_notice("Pacemaker fencer successfully started and accepting connections"); g_main_loop_run(mainloop); stonith_cleanup(); free(cluster->uuid); free(cluster->uname); free(cluster); pe_free_working_set(fenced_data_set); out->finish(out, CRM_EX_OK, true, NULL); pcmk__output_free(out); pcmk__unregister_formats(); crm_exit(CRM_EX_OK); } diff --git a/lib/pacemaker/pcmk_output.c b/lib/pacemaker/pcmk_output.c index d9672fcc9e..1a6798ce6f 100644 --- a/lib/pacemaker/pcmk_output.c +++ b/lib/pacemaker/pcmk_output.c @@ -1,1846 +1,1846 @@ /* - * Copyright 2019-2021 the Pacemaker project contributors + * Copyright 2019-2022 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 static char * colocations_header(pe_resource_t *rsc, pcmk__colocation_t *cons, gboolean dependents) { char *score = NULL; char *retval = NULL; score = score2char(cons->score); if (cons->primary_role > RSC_ROLE_STARTED) { retval = crm_strdup_printf("%s (score=%s, %s role=%s, id=%s)", rsc->id, score, dependents ? "needs" : "with", role2text(cons->primary_role), cons->id); } else { retval = crm_strdup_printf("%s (score=%s, id=%s)", rsc->id, score, cons->id); } free(score); return retval; } static void colocations_xml_node(pcmk__output_t *out, pe_resource_t *rsc, pcmk__colocation_t *cons) { char *score = NULL; xmlNodePtr node = NULL; score = score2char(cons->score); node = pcmk__output_create_xml_node(out, XML_CONS_TAG_RSC_DEPEND, "id", cons->id, "rsc", cons->dependent->id, "with-rsc", cons->primary->id, "score", score, NULL); if (cons->node_attribute) { xmlSetProp(node, (pcmkXmlStr) "node-attribute", (pcmkXmlStr) cons->node_attribute); } if (cons->dependent_role != RSC_ROLE_UNKNOWN) { xmlSetProp(node, (pcmkXmlStr) "rsc-role", (pcmkXmlStr) role2text(cons->dependent_role)); } if (cons->primary_role != RSC_ROLE_UNKNOWN) { xmlSetProp(node, (pcmkXmlStr) "with-rsc-role", (pcmkXmlStr) role2text(cons->primary_role)); } free(score); } static int do_locations_list_xml(pcmk__output_t *out, pe_resource_t *rsc, bool add_header) { GList *lpc = NULL; GList *list = rsc->rsc_location; int rc = pcmk_rc_no_output; for (lpc = list; lpc != NULL; lpc = lpc->next) { pe__location_t *cons = lpc->data; GList *lpc2 = NULL; for (lpc2 = cons->node_list_rh; lpc2 != NULL; lpc2 = lpc2->next) { pe_node_t *node = (pe_node_t *) lpc2->data; char *score = score2char(node->weight); if (add_header) { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "locations"); } pcmk__output_create_xml_node(out, XML_CONS_TAG_RSC_LOCATION, "node", node->details->uname, "rsc", rsc->id, "id", cons->id, "score", score, NULL); free(score); } } if (add_header) { PCMK__OUTPUT_LIST_FOOTER(out, rc); } return rc; } PCMK__OUTPUT_ARGS("rsc-action-item", "const char *", "pe_resource_t *", "pe_node_t *", "pe_node_t *", "pe_action_t *", "pe_action_t *") static int rsc_action_item(pcmk__output_t *out, va_list args) { const char *change = va_arg(args, const char *); pe_resource_t *rsc = va_arg(args, pe_resource_t *); pe_node_t *origin = va_arg(args, pe_node_t *); pe_node_t *destination = va_arg(args, pe_node_t *); pe_action_t *action = va_arg(args, pe_action_t *); pe_action_t *source = va_arg(args, pe_action_t *); int len = 0; char *reason = NULL; char *details = NULL; bool same_host = FALSE; bool same_role = FALSE; bool need_role = FALSE; static int rsc_width = 5; static int detail_width = 5; CRM_ASSERT(action); CRM_ASSERT(destination != NULL || origin != NULL); if(source == NULL) { source = action; } len = strlen(rsc->id); if(len > rsc_width) { rsc_width = len + 2; } if ((rsc->role > RSC_ROLE_STARTED) || (rsc->next_role > RSC_ROLE_UNPROMOTED)) { need_role = TRUE; } if(origin != NULL && destination != NULL && origin->details == destination->details) { same_host = TRUE; } if(rsc->role == rsc->next_role) { same_role = TRUE; } if (need_role && (origin == NULL)) { /* Starting and promoting a promotable clone instance */ details = crm_strdup_printf("%s -> %s %s", role2text(rsc->role), role2text(rsc->next_role), destination->details->uname); } else if (origin == NULL) { /* Starting a resource */ details = crm_strdup_printf("%s", destination->details->uname); } else if (need_role && (destination == NULL)) { /* Stopping a promotable clone instance */ details = crm_strdup_printf("%s %s", role2text(rsc->role), origin->details->uname); } else if (destination == NULL) { /* Stopping a resource */ details = crm_strdup_printf("%s", origin->details->uname); } else if (need_role && same_role && same_host) { /* Recovering, restarting or re-promoting a promotable clone instance */ details = crm_strdup_printf("%s %s", role2text(rsc->role), origin->details->uname); } else if (same_role && same_host) { /* Recovering or Restarting a normal resource */ details = crm_strdup_printf("%s", origin->details->uname); } else if (need_role && same_role) { /* Moving a promotable clone instance */ details = crm_strdup_printf("%s -> %s %s", origin->details->uname, destination->details->uname, role2text(rsc->role)); } else if (same_role) { /* Moving a normal resource */ details = crm_strdup_printf("%s -> %s", origin->details->uname, destination->details->uname); } else if (same_host) { /* Promoting or demoting a promotable clone instance */ details = crm_strdup_printf("%s -> %s %s", role2text(rsc->role), role2text(rsc->next_role), origin->details->uname); } else { /* Moving and promoting/demoting */ details = crm_strdup_printf("%s %s -> %s %s", role2text(rsc->role), origin->details->uname, role2text(rsc->next_role), destination->details->uname); } len = strlen(details); if(len > detail_width) { detail_width = len; } if(source->reason && !pcmk_is_set(action->flags, pe_action_runnable)) { reason = crm_strdup_printf("due to %s (blocked)", source->reason); } else if(source->reason) { reason = crm_strdup_printf("due to %s", source->reason); } else if (!pcmk_is_set(action->flags, pe_action_runnable)) { reason = strdup("blocked"); } out->list_item(out, NULL, "%-8s %-*s ( %*s )%s%s", change, rsc_width, rsc->id, detail_width, details, reason ? " " : "", reason ? reason : ""); free(details); free(reason); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("rsc-action-item", "const char *", "pe_resource_t *", "pe_node_t *", "pe_node_t *", "pe_action_t *", "pe_action_t *") static int rsc_action_item_xml(pcmk__output_t *out, va_list args) { const char *change = va_arg(args, const char *); pe_resource_t *rsc = va_arg(args, pe_resource_t *); pe_node_t *origin = va_arg(args, pe_node_t *); pe_node_t *destination = va_arg(args, pe_node_t *); pe_action_t *action = va_arg(args, pe_action_t *); pe_action_t *source = va_arg(args, pe_action_t *); char *change_str = NULL; bool same_host = FALSE; bool same_role = FALSE; bool need_role = FALSE; xmlNode *xml = NULL; CRM_ASSERT(action); CRM_ASSERT(destination != NULL || origin != NULL); if (source == NULL) { source = action; } if ((rsc->role > RSC_ROLE_STARTED) || (rsc->next_role > RSC_ROLE_UNPROMOTED)) { need_role = TRUE; } if(origin != NULL && destination != NULL && origin->details == destination->details) { same_host = TRUE; } if(rsc->role == rsc->next_role) { same_role = TRUE; } change_str = g_ascii_strdown(change, -1); xml = pcmk__output_create_xml_node(out, "rsc_action", "action", change_str, "resource", rsc->id, NULL); g_free(change_str); if (need_role && (origin == NULL)) { /* Starting and promoting a promotable clone instance */ pcmk__xe_set_props(xml, "role", role2text(rsc->role), "next-role", role2text(rsc->next_role), "dest", destination->details->uname, NULL); } else if (origin == NULL) { /* Starting a resource */ crm_xml_add(xml, "node", destination->details->uname); } else if (need_role && (destination == NULL)) { /* Stopping a promotable clone instance */ pcmk__xe_set_props(xml, "role", role2text(rsc->role), "node", origin->details->uname, NULL); } else if (destination == NULL) { /* Stopping a resource */ crm_xml_add(xml, "node", origin->details->uname); } else if (need_role && same_role && same_host) { /* Recovering, restarting or re-promoting a promotable clone instance */ pcmk__xe_set_props(xml, "role", role2text(rsc->role), "source", origin->details->uname, NULL); } else if (same_role && same_host) { /* Recovering or Restarting a normal resource */ crm_xml_add(xml, "source", origin->details->uname); } else if (need_role && same_role) { /* Moving a promotable clone instance */ pcmk__xe_set_props(xml, "source", origin->details->uname, "dest", destination->details->uname, "role", role2text(rsc->role), NULL); } else if (same_role) { /* Moving a normal resource */ pcmk__xe_set_props(xml, "source", origin->details->uname, "dest", destination->details->uname, NULL); } else if (same_host) { /* Promoting or demoting a promotable clone instance */ pcmk__xe_set_props(xml, "role", role2text(rsc->role), "next-role", role2text(rsc->next_role), "source", origin->details->uname, NULL); } else { /* Moving and promoting/demoting */ pcmk__xe_set_props(xml, "role", role2text(rsc->role), "source", origin->details->uname, "next-role", role2text(rsc->next_role), "dest", destination->details->uname, NULL); } if (source->reason && !pcmk_is_set(action->flags, pe_action_runnable)) { pcmk__xe_set_props(xml, "reason", source->reason, "blocked", "true", NULL); } else if(source->reason) { crm_xml_add(xml, "reason", source->reason); } else if (!pcmk_is_set(action->flags, pe_action_runnable)) { pcmk__xe_set_bool_attr(xml, "blocked", true); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("rsc-is-colocated-with-list", "pe_resource_t *", "gboolean") static int rsc_is_colocated_with_list(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); gboolean recursive = va_arg(args, gboolean); int rc = pcmk_rc_no_output; if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { return rc; } pe__set_resource_flags(rsc, pe_rsc_allocating); for (GList *lpc = rsc->rsc_cons; lpc != NULL; lpc = lpc->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data; char *hdr = NULL; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Resources %s is colocated with", rsc->id); if (pcmk_is_set(cons->primary->flags, pe_rsc_allocating)) { out->list_item(out, NULL, "%s (id=%s - loop)", cons->primary->id, cons->id); continue; } hdr = colocations_header(cons->primary, cons, FALSE); out->list_item(out, NULL, "%s", hdr); free(hdr); /* Empty list header just for indentation of information about this resource. */ out->begin_list(out, NULL, NULL, NULL); out->message(out, "locations-list", cons->primary); if (recursive) { out->message(out, "rsc-is-colocated-with-list", cons->primary, recursive); } out->end_list(out); } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("rsc-is-colocated-with-list", "pe_resource_t *", "gboolean") static int rsc_is_colocated_with_list_xml(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); gboolean recursive = va_arg(args, gboolean); int rc = pcmk_rc_no_output; if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { return rc; } pe__set_resource_flags(rsc, pe_rsc_allocating); for (GList *lpc = rsc->rsc_cons; lpc != NULL; lpc = lpc->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data; if (pcmk_is_set(cons->primary->flags, pe_rsc_allocating)) { colocations_xml_node(out, cons->primary, cons); continue; } colocations_xml_node(out, cons->primary, cons); do_locations_list_xml(out, cons->primary, false); if (recursive) { out->message(out, "rsc-is-colocated-with-list", cons->primary, recursive); } } return rc; } PCMK__OUTPUT_ARGS("rscs-colocated-with-list", "pe_resource_t *", "gboolean") static int rscs_colocated_with_list(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); gboolean recursive = va_arg(args, gboolean); int rc = pcmk_rc_no_output; if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { return rc; } pe__set_resource_flags(rsc, pe_rsc_allocating); for (GList *lpc = rsc->rsc_cons_lhs; lpc != NULL; lpc = lpc->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data; char *hdr = NULL; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Resources colocated with %s", rsc->id); if (pcmk_is_set(cons->dependent->flags, pe_rsc_allocating)) { out->list_item(out, NULL, "%s (id=%s - loop)", cons->dependent->id, cons->id); continue; } hdr = colocations_header(cons->dependent, cons, TRUE); out->list_item(out, NULL, "%s", hdr); free(hdr); /* Empty list header just for indentation of information about this resource. */ out->begin_list(out, NULL, NULL, NULL); out->message(out, "locations-list", cons->dependent); if (recursive) { out->message(out, "rscs-colocated-with-list", cons->dependent, recursive); } out->end_list(out); } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("rscs-colocated-with-list", "pe_resource_t *", "gboolean") static int rscs_colocated_with_list_xml(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); gboolean recursive = va_arg(args, gboolean); int rc = pcmk_rc_no_output; if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { return rc; } pe__set_resource_flags(rsc, pe_rsc_allocating); for (GList *lpc = rsc->rsc_cons_lhs; lpc != NULL; lpc = lpc->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data; if (pcmk_is_set(cons->dependent->flags, pe_rsc_allocating)) { colocations_xml_node(out, cons->dependent, cons); continue; } colocations_xml_node(out, cons->dependent, cons); do_locations_list_xml(out, cons->dependent, false); if (recursive) { out->message(out, "rscs-colocated-with-list", cons->dependent, recursive); } } return rc; } PCMK__OUTPUT_ARGS("locations-list", "pe_resource_t *") static int locations_list(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); GList *lpc = NULL; GList *list = rsc->rsc_location; int rc = pcmk_rc_no_output; for (lpc = list; lpc != NULL; lpc = lpc->next) { pe__location_t *cons = lpc->data; GList *lpc2 = NULL; for (lpc2 = cons->node_list_rh; lpc2 != NULL; lpc2 = lpc2->next) { pe_node_t *node = (pe_node_t *) lpc2->data; char *score = score2char(node->weight); PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Locations"); out->list_item(out, NULL, "Node %s (score=%s, id=%s, rsc=%s)", node->details->uname, score, cons->id, rsc->id); free(score); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("locations-list", "pe_resource_t *") static int locations_list_xml(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); return do_locations_list_xml(out, rsc, true); } PCMK__OUTPUT_ARGS("stacks-constraints", "pe_resource_t *", "pe_working_set_t *", "gboolean") static int stacks_and_constraints(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); pe_working_set_t *data_set = va_arg(args, pe_working_set_t *); gboolean recursive = va_arg(args, gboolean); pcmk__unpack_constraints(data_set); // Constraints apply to group/clone, not member/instance rsc = uber_parent(rsc); out->message(out, "locations-list", rsc); pe__clear_resource_flags_on_all(data_set, pe_rsc_allocating); out->message(out, "rscs-colocated-with-list", rsc, recursive); pe__clear_resource_flags_on_all(data_set, pe_rsc_allocating); out->message(out, "rsc-is-colocated-with-list", rsc, recursive); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("stacks-constraints", "pe_resource_t *", "pe_working_set_t *", "gboolean") static int stacks_and_constraints_xml(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); pe_working_set_t *data_set = va_arg(args, pe_working_set_t *); gboolean recursive = va_arg(args, gboolean); pcmk__unpack_constraints(data_set); // Constraints apply to group/clone, not member/instance rsc = uber_parent(rsc); pcmk__output_xml_create_parent(out, "constraints", NULL); do_locations_list_xml(out, rsc, false); pe__clear_resource_flags_on_all(data_set, pe_rsc_allocating); out->message(out, "rscs-colocated-with-list", rsc, recursive); pe__clear_resource_flags_on_all(data_set, pe_rsc_allocating); out->message(out, "rsc-is-colocated-with-list", rsc, recursive); pcmk__output_xml_pop_parent(out); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("health", "const char *", "const char *", "const char *", "const char *") static int health_text(pcmk__output_t *out, va_list args) { const char *sys_from G_GNUC_UNUSED = va_arg(args, const char *); const char *host_from = va_arg(args, const char *); const char *fsa_state = va_arg(args, const char *); const char *result = va_arg(args, const char *); if (!out->is_quiet(out)) { return out->info(out, "Controller on %s in state %s: %s", crm_str(host_from), crm_str(fsa_state), crm_str(result)); } else if (fsa_state != NULL) { pcmk__formatted_printf(out, "%s\n", fsa_state); return pcmk_rc_ok; } return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("health", "const char *", "const char *", "const char *", "const char *") static int health_xml(pcmk__output_t *out, va_list args) { const char *sys_from = va_arg(args, const char *); const char *host_from = va_arg(args, const char *); const char *fsa_state = va_arg(args, const char *); const char *result = va_arg(args, const char *); pcmk__output_create_xml_node(out, crm_str(sys_from), "node_name", crm_str(host_from), "state", crm_str(fsa_state), "result", crm_str(result), NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("pacemakerd-health", "const char *", "const char *", "const char *") static int pacemakerd_health_text(pcmk__output_t *out, va_list args) { const char *sys_from = va_arg(args, const char *); const char *state = va_arg(args, const char *); const char *last_updated = va_arg(args, const char *); if (!out->is_quiet(out)) { return out->info(out, "Status of %s: '%s' %s %s", crm_str(sys_from), crm_str(state), (!pcmk__str_empty(last_updated))? "last updated":"", crm_str(last_updated)); } else { pcmk__formatted_printf(out, "%s\n", crm_str(state)); return pcmk_rc_ok; } return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("pacemakerd-health", "const char *", "const char *", "const char *") static int pacemakerd_health_xml(pcmk__output_t *out, va_list args) { const char *sys_from = va_arg(args, const char *); const char *state = va_arg(args, const char *); const char *last_updated = va_arg(args, const char *); pcmk__output_create_xml_node(out, crm_str(sys_from), "state", crm_str(state), "last_updated", crm_str(last_updated), NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("profile", "const char *", "clock_t", "clock_t") static int profile_default(pcmk__output_t *out, va_list args) { const char *xml_file = va_arg(args, const char *); clock_t start = va_arg(args, clock_t); clock_t end = va_arg(args, clock_t); out->list_item(out, NULL, "Testing %s ... %.2f secs", xml_file, (end - start) / (float) CLOCKS_PER_SEC); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("profile", "const char *", "clock_t", "clock_t") static int profile_xml(pcmk__output_t *out, va_list args) { const char *xml_file = va_arg(args, const char *); clock_t start = va_arg(args, clock_t); clock_t end = va_arg(args, clock_t); char *duration = pcmk__ftoa((end - start) / (float) CLOCKS_PER_SEC); pcmk__output_create_xml_node(out, "timing", "file", xml_file, "duration", duration, NULL); free(duration); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("dc", "const char *") static int dc_text(pcmk__output_t *out, va_list args) { const char *dc = va_arg(args, const char *); if (!out->is_quiet(out)) { return out->info(out, "Designated Controller is: %s", crm_str(dc)); } else if (dc != NULL) { pcmk__formatted_printf(out, "%s\n", dc); return pcmk_rc_ok; } return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("dc", "const char *") static int dc_xml(pcmk__output_t *out, va_list args) { const char *dc = va_arg(args, const char *); pcmk__output_create_xml_node(out, "dc", "node_name", crm_str(dc), NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("crmadmin-node", "const char *", "const char *", "const char *", "gboolean") static int crmadmin_node_text(pcmk__output_t *out, va_list args) { const char *type = va_arg(args, const char *); const char *name = va_arg(args, const char *); const char *id = va_arg(args, const char *); gboolean BASH_EXPORT = va_arg(args, gboolean); if (out->is_quiet(out)) { pcmk__formatted_printf(out, "%s\n", crm_str(name)); return pcmk_rc_ok; } else if (BASH_EXPORT) { return out->info(out, "export %s=%s", crm_str(name), crm_str(id)); } else { return out->info(out, "%s node: %s (%s)", type ? type : "cluster", crm_str(name), crm_str(id)); } } PCMK__OUTPUT_ARGS("crmadmin-node", "const char *", "const char *", "const char *", "gboolean") static int crmadmin_node_xml(pcmk__output_t *out, va_list args) { const char *type = va_arg(args, const char *); const char *name = va_arg(args, const char *); const char *id = va_arg(args, const char *); pcmk__output_create_xml_node(out, "node", "type", type ? type : "cluster", "name", crm_str(name), "id", crm_str(id), NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("digests", "pe_resource_t *", "pe_node_t *", "const char *", "guint", "op_digest_cache_t *") static int digests_text(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); pe_node_t *node = va_arg(args, pe_node_t *); const char *task = va_arg(args, const char *); guint interval_ms = va_arg(args, guint); op_digest_cache_t *digests = va_arg(args, op_digest_cache_t *); char *action_desc = NULL; const char *rsc_desc = "unknown resource"; const char *node_desc = "unknown node"; if (interval_ms != 0) { action_desc = crm_strdup_printf("%ums-interval %s action", interval_ms, ((task == NULL)? "unknown" : task)); } else if (pcmk__str_eq(task, "monitor", pcmk__str_none)) { action_desc = strdup("probe action"); } else { action_desc = crm_strdup_printf("%s action", ((task == NULL)? "unknown" : task)); } if ((rsc != NULL) && (rsc->id != NULL)) { rsc_desc = rsc->id; } if ((node != NULL) && (node->details->uname != NULL)) { node_desc = node->details->uname; } out->begin_list(out, NULL, NULL, "Digests for %s %s on %s", rsc_desc, action_desc, node_desc); free(action_desc); if (digests == NULL) { out->list_item(out, NULL, "none"); out->end_list(out); return pcmk_rc_ok; } if (digests->digest_all_calc != NULL) { out->list_item(out, NULL, "%s (all parameters)", digests->digest_all_calc); } if (digests->digest_secure_calc != NULL) { out->list_item(out, NULL, "%s (non-private parameters)", digests->digest_secure_calc); } if (digests->digest_restart_calc != NULL) { out->list_item(out, NULL, "%s (non-reloadable parameters)", digests->digest_restart_calc); } out->end_list(out); return pcmk_rc_ok; } static void add_digest_xml(xmlNode *parent, const char *type, const char *digest, xmlNode *digest_source) { if (digest != NULL) { xmlNodePtr digest_xml = create_xml_node(parent, "digest"); crm_xml_add(digest_xml, "type", ((type == NULL)? "unspecified" : type)); crm_xml_add(digest_xml, "hash", digest); if (digest_source != NULL) { add_node_copy(digest_xml, digest_source); } } } PCMK__OUTPUT_ARGS("digests", "pe_resource_t *", "pe_node_t *", "const char *", "guint", "op_digest_cache_t *") static int digests_xml(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); pe_node_t *node = va_arg(args, pe_node_t *); const char *task = va_arg(args, const char *); guint interval_ms = va_arg(args, guint); op_digest_cache_t *digests = va_arg(args, op_digest_cache_t *); char *interval_s = crm_strdup_printf("%ums", interval_ms); xmlNode *xml = NULL; xml = pcmk__output_create_xml_node(out, "digests", "resource", crm_str(rsc->id), "node", crm_str(node->details->uname), "task", crm_str(task), "interval", interval_s, NULL); free(interval_s); if (digests != NULL) { add_digest_xml(xml, "all", digests->digest_all_calc, digests->params_all); add_digest_xml(xml, "nonprivate", digests->digest_secure_calc, digests->params_secure); add_digest_xml(xml, "nonreloadable", digests->digest_restart_calc, digests->params_restart); } return pcmk_rc_ok; } #define STOP_SANITY_ASSERT(lineno) do { \ if(current && current->details->unclean) { \ /* It will be a pseudo op */ \ } else if(stop == NULL) { \ crm_err("%s:%d: No stop action exists for %s", \ __func__, lineno, rsc->id); \ CRM_ASSERT(stop != NULL); \ } else if (pcmk_is_set(stop->flags, pe_action_optional)) { \ crm_err("%s:%d: Action %s is still optional", \ __func__, lineno, stop->uuid); \ CRM_ASSERT(!pcmk_is_set(stop->flags, pe_action_optional)); \ } \ } while(0) PCMK__OUTPUT_ARGS("rsc-action", "pe_resource_t *", "pe_node_t *", "pe_node_t *", "gboolean") static int rsc_action_default(pcmk__output_t *out, va_list args) { pe_resource_t *rsc = va_arg(args, pe_resource_t *); pe_node_t *current = va_arg(args, pe_node_t *); pe_node_t *next = va_arg(args, pe_node_t *); gboolean moving = va_arg(args, gboolean); GList *possible_matches = NULL; char *key = NULL; int rc = pcmk_rc_no_output; pe_node_t *start_node = NULL; pe_action_t *start = NULL; pe_action_t *stop = NULL; pe_action_t *promote = NULL; pe_action_t *demote = NULL; if (!pcmk_is_set(rsc->flags, pe_rsc_managed) || (current == NULL && next == NULL)) { pe_rsc_info(rsc, "Leave %s\t(%s%s)", rsc->id, role2text(rsc->role), !pcmk_is_set(rsc->flags, pe_rsc_managed)? " unmanaged" : ""); return rc; } if (current != NULL && next != NULL && !pcmk__str_eq(current->details->id, next->details->id, pcmk__str_casei)) { moving = TRUE; } possible_matches = pe__resource_actions(rsc, next, RSC_START, FALSE); if (possible_matches) { start = possible_matches->data; g_list_free(possible_matches); } if ((start == NULL) || !pcmk_is_set(start->flags, pe_action_runnable)) { start_node = NULL; } else { start_node = current; } possible_matches = pe__resource_actions(rsc, start_node, RSC_STOP, FALSE); if (possible_matches) { stop = possible_matches->data; g_list_free(possible_matches); } possible_matches = pe__resource_actions(rsc, next, RSC_PROMOTE, FALSE); if (possible_matches) { promote = possible_matches->data; g_list_free(possible_matches); } possible_matches = pe__resource_actions(rsc, next, RSC_DEMOTE, FALSE); if (possible_matches) { demote = possible_matches->data; g_list_free(possible_matches); } if (rsc->role == rsc->next_role) { pe_action_t *migrate_op = NULL; + CRM_CHECK(next != NULL, return rc); + possible_matches = pe__resource_actions(rsc, next, RSC_MIGRATED, FALSE); if (possible_matches) { migrate_op = possible_matches->data; } - CRM_CHECK(next != NULL,); - if (next == NULL) { - } else if ((migrate_op != NULL) && (current != NULL) + if ((migrate_op != NULL) && (current != NULL) && pcmk_is_set(migrate_op->flags, pe_action_runnable)) { rc = out->message(out, "rsc-action-item", "Migrate", rsc, current, next, start, NULL); } else if (pcmk_is_set(rsc->flags, pe_rsc_reload)) { rc = out->message(out, "rsc-action-item", "Reload", rsc, current, next, start, NULL); } else if (start == NULL || pcmk_is_set(start->flags, pe_action_optional)) { if ((demote != NULL) && (promote != NULL) && !pcmk_is_set(demote->flags, pe_action_optional) && !pcmk_is_set(promote->flags, pe_action_optional)) { rc = out->message(out, "rsc-action-item", "Re-promote", rsc, current, next, promote, demote); } else { pe_rsc_info(rsc, "Leave %s\t(%s %s)", rsc->id, role2text(rsc->role), next->details->uname); } } else if (!pcmk_is_set(start->flags, pe_action_runnable)) { rc = out->message(out, "rsc-action-item", "Stop", rsc, current, NULL, stop, (stop && stop->reason)? stop : start); STOP_SANITY_ASSERT(__LINE__); } else if (moving && current) { rc = out->message(out, "rsc-action-item", pcmk_is_set(rsc->flags, pe_rsc_failed)? "Recover" : "Move", rsc, current, next, stop, NULL); } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { rc = out->message(out, "rsc-action-item", "Recover", rsc, current, NULL, stop, NULL); STOP_SANITY_ASSERT(__LINE__); } else { rc = out->message(out, "rsc-action-item", "Restart", rsc, current, next, start, NULL); /* STOP_SANITY_ASSERT(__LINE__); False positive for migrate-fail-7 */ } g_list_free(possible_matches); return rc; } if(stop && (rsc->next_role == RSC_ROLE_STOPPED || (start && !pcmk_is_set(start->flags, pe_action_runnable)))) { GList *gIter = NULL; key = stop_key(rsc); for (gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; pe_action_t *stop_op = NULL; possible_matches = find_actions(rsc->actions, key, node); if (possible_matches) { stop_op = possible_matches->data; g_list_free(possible_matches); } if (stop_op && (stop_op->flags & pe_action_runnable)) { STOP_SANITY_ASSERT(__LINE__); } if (out->message(out, "rsc-action-item", "Stop", rsc, node, NULL, stop_op, (stop_op && stop_op->reason)? stop_op : start) == pcmk_rc_ok) { rc = pcmk_rc_ok; } } free(key); } else if ((stop != NULL) && pcmk_all_flags_set(rsc->flags, pe_rsc_failed|pe_rsc_stop)) { /* 'stop' may be NULL if the failure was ignored */ rc = out->message(out, "rsc-action-item", "Recover", rsc, current, next, stop, start); STOP_SANITY_ASSERT(__LINE__); } else if (moving) { rc = out->message(out, "rsc-action-item", "Move", rsc, current, next, stop, NULL); STOP_SANITY_ASSERT(__LINE__); } else if (pcmk_is_set(rsc->flags, pe_rsc_reload)) { rc = out->message(out, "rsc-action-item", "Reload", rsc, current, next, start, NULL); } else if (stop != NULL && !pcmk_is_set(stop->flags, pe_action_optional)) { rc = out->message(out, "rsc-action-item", "Restart", rsc, current, next, start, NULL); STOP_SANITY_ASSERT(__LINE__); } else if (rsc->role == RSC_ROLE_PROMOTED) { CRM_LOG_ASSERT(current != NULL); rc = out->message(out, "rsc-action-item", "Demote", rsc, current, next, demote, NULL); } else if (rsc->next_role == RSC_ROLE_PROMOTED) { CRM_LOG_ASSERT(next); rc = out->message(out, "rsc-action-item", "Promote", rsc, current, next, promote, NULL); } else if (rsc->role == RSC_ROLE_STOPPED && rsc->next_role > RSC_ROLE_STOPPED) { rc = out->message(out, "rsc-action-item", "Start", rsc, current, next, start, NULL); } return rc; } PCMK__OUTPUT_ARGS("node-action", "char *", "char *", "char *") static int node_action(pcmk__output_t *out, va_list args) { char *task = va_arg(args, char *); char *node_name = va_arg(args, char *); char *reason = va_arg(args, char *); if (task == NULL) { return pcmk_rc_no_output; } else if (reason) { out->list_item(out, NULL, "%s %s '%s'", task, node_name, reason); } else { crm_notice(" * %s %s", task, node_name); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-action", "char *", "char *", "char *") static int node_action_xml(pcmk__output_t *out, va_list args) { char *task = va_arg(args, char *); char *node_name = va_arg(args, char *); char *reason = va_arg(args, char *); if (task == NULL) { return pcmk_rc_no_output; } else if (reason) { pcmk__output_create_xml_node(out, "node_action", "task", task, "node", node_name, "reason", reason, NULL); } else { crm_notice(" * %s %s", task, node_name); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-cluster-action", "const char *", "const char *", "xmlNodePtr") static int inject_cluster_action(pcmk__output_t *out, va_list args) { const char *node = va_arg(args, const char *); const char *task = va_arg(args, const char *); xmlNodePtr rsc = va_arg(args, xmlNodePtr); if (out->is_quiet(out)) { return pcmk_rc_no_output; } if(rsc) { out->list_item(out, NULL, "Cluster action: %s for %s on %s", task, ID(rsc), node); } else { out->list_item(out, NULL, "Cluster action: %s on %s", task, node); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-cluster-action", "const char *", "const char *", "xmlNodePtr") static int inject_cluster_action_xml(pcmk__output_t *out, va_list args) { const char *node = va_arg(args, const char *); const char *task = va_arg(args, const char *); xmlNodePtr rsc = va_arg(args, xmlNodePtr); xmlNodePtr xml_node = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } xml_node = pcmk__output_create_xml_node(out, "cluster_action", "task", task, "node", node, NULL); if (rsc) { crm_xml_add(xml_node, "id", ID(rsc)); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-fencing-action", "char *", "const char *") static int inject_fencing_action(pcmk__output_t *out, va_list args) { char *target = va_arg(args, char *); const char *op = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } out->list_item(out, NULL, "Fencing %s (%s)", target, op); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-fencing-action", "char *", "const char *") static int inject_fencing_action_xml(pcmk__output_t *out, va_list args) { char *target = va_arg(args, char *); const char *op = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } pcmk__output_create_xml_node(out, "fencing_action", "target", target, "op", op, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-attr", "const char *", "const char *", "xmlNodePtr") static int inject_attr(pcmk__output_t *out, va_list args) { const char *name = va_arg(args, const char *); const char *value = va_arg(args, const char *); xmlNodePtr cib_node = va_arg(args, xmlNodePtr); xmlChar *node_path = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } node_path = xmlGetNodePath(cib_node); out->list_item(out, NULL, "Injecting attribute %s=%s into %s '%s'", name, value, node_path, ID(cib_node)); free(node_path); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-attr", "const char *", "const char *", "xmlNodePtr") static int inject_attr_xml(pcmk__output_t *out, va_list args) { const char *name = va_arg(args, const char *); const char *value = va_arg(args, const char *); xmlNodePtr cib_node = va_arg(args, xmlNodePtr); xmlChar *node_path = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } node_path = xmlGetNodePath(cib_node); pcmk__output_create_xml_node(out, "inject_attr", "name", name, "value", value, "node_path", node_path, "cib_node", ID(cib_node), NULL); free(node_path); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-spec", "char *") static int inject_spec(pcmk__output_t *out, va_list args) { char *spec = va_arg(args, char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } out->list_item(out, NULL, "Injecting %s into the configuration", spec); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-spec", "char *") static int inject_spec_xml(pcmk__output_t *out, va_list args) { char *spec = va_arg(args, char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } pcmk__output_create_xml_node(out, "inject_spec", "spec", spec, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-modify-config", "const char *", "const char *") static int inject_modify_config(pcmk__output_t *out, va_list args) { const char *quorum = va_arg(args, const char *); const char *watchdog = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } out->begin_list(out, NULL, NULL, "Performing Requested Modifications"); if (quorum) { out->list_item(out, NULL, "Setting quorum: %s", quorum); } if (watchdog) { out->list_item(out, NULL, "Setting watchdog: %s", watchdog); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-modify-config", "const char *", "const char *") static int inject_modify_config_xml(pcmk__output_t *out, va_list args) { const char *quorum = va_arg(args, const char *); const char *watchdog = va_arg(args, const char *); xmlNodePtr node = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } node = pcmk__output_xml_create_parent(out, "modifications", NULL); if (quorum) { crm_xml_add(node, "quorum", quorum); } if (watchdog) { crm_xml_add(node, "watchdog", watchdog); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-modify-node", "const char *", "char *") static int inject_modify_node(pcmk__output_t *out, va_list args) { const char *action = va_arg(args, const char *); char *node = va_arg(args, char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } if (pcmk__str_eq(action, "Online", pcmk__str_none)) { out->list_item(out, NULL, "Bringing node %s online", node); return pcmk_rc_ok; } else if (pcmk__str_eq(action, "Offline", pcmk__str_none)) { out->list_item(out, NULL, "Taking node %s offline", node); return pcmk_rc_ok; } else if (pcmk__str_eq(action, "Failing", pcmk__str_none)) { out->list_item(out, NULL, "Failing node %s", node); return pcmk_rc_ok; } return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("inject-modify-node", "const char *", "char *") static int inject_modify_node_xml(pcmk__output_t *out, va_list args) { const char *action = va_arg(args, const char *); char *node = va_arg(args, char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } pcmk__output_create_xml_node(out, "modify_node", "action", action, "node", node, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-modify-ticket", "const char *", "char *") static int inject_modify_ticket(pcmk__output_t *out, va_list args) { const char *action = va_arg(args, const char *); char *ticket = va_arg(args, char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } if (pcmk__str_eq(action, "Standby", pcmk__str_none)) { out->list_item(out, NULL, "Making ticket %s standby", ticket); } else { out->list_item(out, NULL, "%s ticket %s", action, ticket); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-modify-ticket", "const char *", "char *") static int inject_modify_ticket_xml(pcmk__output_t *out, va_list args) { const char *action = va_arg(args, const char *); char *ticket = va_arg(args, char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } pcmk__output_create_xml_node(out, "modify_ticket", "action", action, "ticket", ticket, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-pseudo-action", "const char *", "const char *") static int inject_pseudo_action(pcmk__output_t *out, va_list args) { const char *node = va_arg(args, const char *); const char *task = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } out->list_item(out, NULL, "Pseudo action: %s%s%s", task, node ? " on " : "", node ? node : ""); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-pseudo-action", "const char *", "const char *") static int inject_pseudo_action_xml(pcmk__output_t *out, va_list args) { const char *node = va_arg(args, const char *); const char *task = va_arg(args, const char *); xmlNodePtr xml_node = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } xml_node = pcmk__output_create_xml_node(out, "pseudo_action", "task", task, NULL); if (node) { crm_xml_add(xml_node, "node", node); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-rsc-action", "const char *", "const char *", "char *", "guint") static int inject_rsc_action(pcmk__output_t *out, va_list args) { const char *rsc = va_arg(args, const char *); const char *operation = va_arg(args, const char *); char *node = va_arg(args, char *); guint interval_ms = va_arg(args, guint); if (out->is_quiet(out)) { return pcmk_rc_no_output; } if (interval_ms) { out->list_item(out, NULL, "Resource action: %-15s %s=%u on %s", rsc, operation, interval_ms, node); } else { out->list_item(out, NULL, "Resource action: %-15s %s on %s", rsc, operation, node); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-rsc-action", "const char *", "const char *", "char *", "guint") static int inject_rsc_action_xml(pcmk__output_t *out, va_list args) { const char *rsc = va_arg(args, const char *); const char *operation = va_arg(args, const char *); char *node = va_arg(args, char *); guint interval_ms = va_arg(args, guint); xmlNodePtr xml_node = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } xml_node = pcmk__output_create_xml_node(out, "rsc_action", "resource", rsc, "op", operation, "node", node, NULL); if (interval_ms) { char *interval_s = pcmk__itoa(interval_ms); crm_xml_add(xml_node, "interval", interval_s); free(interval_s); } return pcmk_rc_ok; } #define CHECK_RC(retcode, retval) \ if (retval == pcmk_rc_ok) { \ retcode = pcmk_rc_ok; \ } PCMK__OUTPUT_ARGS("cluster-status", "pe_working_set_t *", "crm_exit_t", "stonith_history_t *", "gboolean", "unsigned int", "unsigned int", "const char *", "GList *", "GList *") int pcmk__cluster_status_text(pcmk__output_t *out, va_list args) { pe_working_set_t *data_set = va_arg(args, pe_working_set_t *); crm_exit_t history_rc = va_arg(args, crm_exit_t); stonith_history_t *stonith_history = va_arg(args, stonith_history_t *); gboolean fence_history = va_arg(args, gboolean); unsigned int section_opts = va_arg(args, unsigned int); unsigned int show_opts = va_arg(args, unsigned int); const char *prefix = va_arg(args, const char *); GList *unames = va_arg(args, GList *); GList *resources = va_arg(args, GList *); int rc = pcmk_rc_no_output; bool already_printed_failure = false; CHECK_RC(rc, out->message(out, "cluster-summary", data_set, section_opts, show_opts)); if (pcmk_is_set(section_opts, pcmk_section_nodes) && unames) { CHECK_RC(rc, out->message(out, "node-list", data_set->nodes, unames, resources, show_opts, rc == pcmk_rc_ok)); } /* Print resources section, if needed */ if (pcmk_is_set(section_opts, pcmk_section_resources)) { CHECK_RC(rc, out->message(out, "resource-list", data_set, show_opts, TRUE, unames, resources, rc == pcmk_rc_ok)); } /* print Node Attributes section if requested */ if (pcmk_is_set(section_opts, pcmk_section_attributes)) { CHECK_RC(rc, out->message(out, "node-attribute-list", data_set, show_opts, rc == pcmk_rc_ok, unames, resources)); } /* If requested, print resource operations (which includes failcounts) * or just failcounts */ if (pcmk_any_flags_set(section_opts, pcmk_section_operations | pcmk_section_failcounts)) { CHECK_RC(rc, out->message(out, "node-summary", data_set, unames, resources, section_opts, show_opts, rc == pcmk_rc_ok)); } /* If there were any failed actions, print them */ if (pcmk_is_set(section_opts, pcmk_section_failures) && xml_has_children(data_set->failed)) { CHECK_RC(rc, out->message(out, "failed-action-list", data_set, unames, resources, show_opts, rc == pcmk_rc_ok)); } /* Print failed stonith actions */ if (pcmk_is_set(section_opts, pcmk_section_fence_failed) && fence_history) { if (history_rc == 0) { stonith_history_t *hp = stonith__first_matching_event(stonith_history, stonith__event_state_eq, GINT_TO_POINTER(st_failed)); if (hp) { CHECK_RC(rc, out->message(out, "failed-fencing-list", stonith_history, unames, section_opts, rc == pcmk_rc_ok)); } } else { PCMK__OUTPUT_SPACER_IF(out, rc == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Failed Fencing Actions"); out->list_item(out, NULL, "Failed to get fencing history: %s", crm_exit_str(history_rc)); out->end_list(out); already_printed_failure = true; } } /* Print tickets if requested */ if (pcmk_is_set(section_opts, pcmk_section_tickets)) { CHECK_RC(rc, out->message(out, "ticket-list", data_set, rc == pcmk_rc_ok)); } /* Print negative location constraints if requested */ if (pcmk_is_set(section_opts, pcmk_section_bans)) { CHECK_RC(rc, out->message(out, "ban-list", data_set, prefix, resources, show_opts, rc == pcmk_rc_ok)); } /* Print stonith history */ if (fence_history && pcmk_any_flags_set(section_opts, pcmk_section_fencing_all)) { if (history_rc != 0) { if (!already_printed_failure) { PCMK__OUTPUT_SPACER_IF(out, rc == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Failed Fencing Actions"); out->list_item(out, NULL, "Failed to get fencing history: %s", crm_exit_str(history_rc)); out->end_list(out); } } else if (pcmk_is_set(section_opts, pcmk_section_fence_worked)) { stonith_history_t *hp = stonith__first_matching_event(stonith_history, stonith__event_state_neq, GINT_TO_POINTER(st_failed)); if (hp) { CHECK_RC(rc, out->message(out, "fencing-list", hp, unames, section_opts, rc == pcmk_rc_ok)); } } else if (pcmk_is_set(section_opts, pcmk_section_fence_pending)) { stonith_history_t *hp = stonith__first_matching_event(stonith_history, stonith__event_state_pending, NULL); if (hp) { CHECK_RC(rc, out->message(out, "pending-fencing-list", hp, unames, section_opts, rc == pcmk_rc_ok)); } } } return rc; } PCMK__OUTPUT_ARGS("cluster-status", "pe_working_set_t *", "crm_exit_t", "stonith_history_t *", "gboolean", "unsigned int", "unsigned int", "const char *", "GList *", "GList *") static int cluster_status_xml(pcmk__output_t *out, va_list args) { pe_working_set_t *data_set = va_arg(args, pe_working_set_t *); crm_exit_t history_rc = va_arg(args, crm_exit_t); stonith_history_t *stonith_history = va_arg(args, stonith_history_t *); gboolean fence_history = va_arg(args, gboolean); unsigned int section_opts = va_arg(args, unsigned int); unsigned int show_opts = va_arg(args, unsigned int); const char *prefix = va_arg(args, const char *); GList *unames = va_arg(args, GList *); GList *resources = va_arg(args, GList *); out->message(out, "cluster-summary", data_set, section_opts, show_opts); /*** NODES ***/ if (pcmk_is_set(section_opts, pcmk_section_nodes)) { out->message(out, "node-list", data_set->nodes, unames, resources, show_opts, FALSE); } /* Print resources section, if needed */ if (pcmk_is_set(section_opts, pcmk_section_resources)) { /* XML output always displays full details. */ unsigned int full_show_opts = show_opts & ~pcmk_show_brief; out->message(out, "resource-list", data_set, full_show_opts, FALSE, unames, resources, FALSE); } /* print Node Attributes section if requested */ if (pcmk_is_set(section_opts, pcmk_section_attributes)) { out->message(out, "node-attribute-list", data_set, show_opts, FALSE, unames, resources); } /* If requested, print resource operations (which includes failcounts) * or just failcounts */ if (pcmk_any_flags_set(section_opts, pcmk_section_operations | pcmk_section_failcounts)) { out->message(out, "node-summary", data_set, unames, resources, section_opts, show_opts, FALSE); } /* If there were any failed actions, print them */ if (pcmk_is_set(section_opts, pcmk_section_failures) && xml_has_children(data_set->failed)) { out->message(out, "failed-action-list", data_set, unames, resources, show_opts, FALSE); } /* Print stonith history */ if (pcmk_is_set(section_opts, pcmk_section_fencing_all) && fence_history) { out->message(out, "full-fencing-list", history_rc, stonith_history, unames, section_opts, FALSE); } /* Print tickets if requested */ if (pcmk_is_set(section_opts, pcmk_section_tickets)) { out->message(out, "ticket-list", data_set, FALSE); } /* Print negative location constraints if requested */ if (pcmk_is_set(section_opts, pcmk_section_bans)) { out->message(out, "ban-list", data_set, prefix, resources, show_opts, FALSE); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-status", "pe_working_set_t *", "crm_exit_t", "stonith_history_t *", "gboolean", "unsigned int", "unsigned int", "const char *", "GList *", "GList *") static int cluster_status_html(pcmk__output_t *out, va_list args) { pe_working_set_t *data_set = va_arg(args, pe_working_set_t *); crm_exit_t history_rc = va_arg(args, crm_exit_t); stonith_history_t *stonith_history = va_arg(args, stonith_history_t *); gboolean fence_history = va_arg(args, gboolean); unsigned int section_opts = va_arg(args, unsigned int); unsigned int show_opts = va_arg(args, unsigned int); const char *prefix = va_arg(args, const char *); GList *unames = va_arg(args, GList *); GList *resources = va_arg(args, GList *); bool already_printed_failure = false; out->message(out, "cluster-summary", data_set, section_opts, show_opts); /*** NODE LIST ***/ if (pcmk_is_set(section_opts, pcmk_section_nodes) && unames) { out->message(out, "node-list", data_set->nodes, unames, resources, show_opts, FALSE); } /* Print resources section, if needed */ if (pcmk_is_set(section_opts, pcmk_section_resources)) { out->message(out, "resource-list", data_set, show_opts, TRUE, unames, resources, FALSE); } /* print Node Attributes section if requested */ if (pcmk_is_set(section_opts, pcmk_section_attributes)) { out->message(out, "node-attribute-list", data_set, show_opts, FALSE, unames, resources); } /* If requested, print resource operations (which includes failcounts) * or just failcounts */ if (pcmk_any_flags_set(section_opts, pcmk_section_operations | pcmk_section_failcounts)) { out->message(out, "node-summary", data_set, unames, resources, section_opts, show_opts, FALSE); } /* If there were any failed actions, print them */ if (pcmk_is_set(section_opts, pcmk_section_failures) && xml_has_children(data_set->failed)) { out->message(out, "failed-action-list", data_set, unames, resources, show_opts, FALSE); } /* Print failed stonith actions */ if (pcmk_is_set(section_opts, pcmk_section_fence_failed) && fence_history) { if (history_rc == 0) { stonith_history_t *hp = stonith__first_matching_event(stonith_history, stonith__event_state_eq, GINT_TO_POINTER(st_failed)); if (hp) { out->message(out, "failed-fencing-list", stonith_history, unames, section_opts, FALSE); } } else { out->begin_list(out, NULL, NULL, "Failed Fencing Actions"); out->list_item(out, NULL, "Failed to get fencing history: %s", crm_exit_str(history_rc)); out->end_list(out); } } /* Print stonith history */ if (fence_history && pcmk_any_flags_set(section_opts, pcmk_section_fencing_all)) { if (history_rc != 0) { if (!already_printed_failure) { out->begin_list(out, NULL, NULL, "Failed Fencing Actions"); out->list_item(out, NULL, "Failed to get fencing history: %s", crm_exit_str(history_rc)); out->end_list(out); } } else if (pcmk_is_set(section_opts, pcmk_section_fence_worked)) { stonith_history_t *hp = stonith__first_matching_event(stonith_history, stonith__event_state_neq, GINT_TO_POINTER(st_failed)); if (hp) { out->message(out, "fencing-list", hp, unames, section_opts, FALSE); } } else if (pcmk_is_set(section_opts, pcmk_section_fence_pending)) { stonith_history_t *hp = stonith__first_matching_event(stonith_history, stonith__event_state_pending, NULL); if (hp) { out->message(out, "pending-fencing-list", hp, unames, section_opts, FALSE); } } } /* Print tickets if requested */ if (pcmk_is_set(section_opts, pcmk_section_tickets)) { out->message(out, "ticket-list", data_set, FALSE); } /* Print negative location constraints if requested */ if (pcmk_is_set(section_opts, pcmk_section_bans)) { out->message(out, "ban-list", data_set, prefix, resources, show_opts, FALSE); } return pcmk_rc_ok; } static pcmk__message_entry_t fmt_functions[] = { { "cluster-status", "default", pcmk__cluster_status_text }, { "cluster-status", "html", cluster_status_html }, { "cluster-status", "xml", cluster_status_xml }, { "crmadmin-node", "default", crmadmin_node_text }, { "crmadmin-node", "xml", crmadmin_node_xml }, { "dc", "default", dc_text }, { "dc", "xml", dc_xml }, { "digests", "default", digests_text }, { "digests", "xml", digests_xml }, { "health", "default", health_text }, { "health", "xml", health_xml }, { "inject-attr", "default", inject_attr }, { "inject-attr", "xml", inject_attr_xml }, { "inject-cluster-action", "default", inject_cluster_action }, { "inject-cluster-action", "xml", inject_cluster_action_xml }, { "inject-fencing-action", "default", inject_fencing_action }, { "inject-fencing-action", "xml", inject_fencing_action_xml }, { "inject-modify-config", "default", inject_modify_config }, { "inject-modify-config", "xml", inject_modify_config_xml }, { "inject-modify-node", "default", inject_modify_node }, { "inject-modify-node", "xml", inject_modify_node_xml }, { "inject-modify-ticket", "default", inject_modify_ticket }, { "inject-modify-ticket", "xml", inject_modify_ticket_xml }, { "inject-pseudo-action", "default", inject_pseudo_action }, { "inject-pseudo-action", "xml", inject_pseudo_action_xml }, { "inject-rsc-action", "default", inject_rsc_action }, { "inject-rsc-action", "xml", inject_rsc_action_xml }, { "inject-spec", "default", inject_spec }, { "inject-spec", "xml", inject_spec_xml }, { "locations-list", "default", locations_list }, { "locations-list", "xml", locations_list_xml }, { "node-action", "default", node_action }, { "node-action", "xml", node_action_xml }, { "pacemakerd-health", "default", pacemakerd_health_text }, { "pacemakerd-health", "xml", pacemakerd_health_xml }, { "profile", "default", profile_default, }, { "profile", "xml", profile_xml }, { "rsc-action", "default", rsc_action_default }, { "rsc-action-item", "default", rsc_action_item }, { "rsc-action-item", "xml", rsc_action_item_xml }, { "rsc-is-colocated-with-list", "default", rsc_is_colocated_with_list }, { "rsc-is-colocated-with-list", "xml", rsc_is_colocated_with_list_xml }, { "rscs-colocated-with-list", "default", rscs_colocated_with_list }, { "rscs-colocated-with-list", "xml", rscs_colocated_with_list_xml }, { "stacks-constraints", "default", stacks_and_constraints }, { "stacks-constraints", "xml", stacks_and_constraints_xml }, { NULL, NULL, NULL } }; void pcmk__register_lib_messages(pcmk__output_t *out) { pcmk__register_messages(out, fmt_functions); } diff --git a/lib/pacemaker/pcmk_simulate.c b/lib/pacemaker/pcmk_simulate.c index ceab6afe0e..d97134d15a 100644 --- a/lib/pacemaker/pcmk_simulate.c +++ b/lib/pacemaker/pcmk_simulate.c @@ -1,560 +1,560 @@ /* - * Copyright 2021 the Pacemaker project contributors + * Copyright 2021-2022 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 "libpacemaker_private.h" static char * create_action_name(pe_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) { action_host = action->node->details->uname; } else if (!pcmk_is_set(action->flags, pe_action_pseudo)) { action_host = ""; } if (pcmk__str_eq(action->task, RSC_CANCEL, pcmk__str_casei)) { prefix = "Cancel "; task = action->cancel_task; } if (action->rsc && action->rsc->clone_name) { clone_name = action->rsc->clone_name; } if (clone_name) { char *key = NULL; guint interval_ms = 0; if (pcmk__guint_from_hash(action->meta, XML_LRM_ATTR_INTERVAL_MS, 0, &interval_ms) != pcmk_rc_ok) { interval_ms = 0; } if (pcmk__strcase_any_of(action->task, RSC_NOTIFY, RSC_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) { 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, CRM_OP_FENCE, pcmk__str_casei)) { const char *op = g_hash_table_lookup(action->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; } static void print_cluster_status(pe_working_set_t * data_set, unsigned int show_opts, unsigned int section_opts, const char *title, bool print_spacer) { pcmk__output_t *out = data_set->priv; GList *all = NULL; section_opts |= pcmk_section_nodes | pcmk_section_resources; 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", data_set, 0, NULL, FALSE, section_opts, show_opts | pcmk_show_inactive_rscs | pcmk_show_failed_detail, NULL, all, all); out->end_list(out); g_list_free(all); } static void print_transition_summary(pe_working_set_t *data_set, bool print_spacer) { pcmk__output_t *out = data_set->priv; PCMK__OUTPUT_SPACER_IF(out, print_spacer); out->begin_list(out, NULL, NULL, "Transition Summary"); pcmk__output_actions(data_set); out->end_list(out); } static void reset(pe_working_set_t *data_set, xmlNodePtr input, pcmk__output_t *out, char *use_date, unsigned int flags) { data_set->input = input; data_set->priv = out; pcmk__set_effective_date(data_set, true, use_date); if (pcmk_is_set(flags, pcmk_sim_sanitized)) { pe__set_working_set_flags(data_set, pe_flag_sanitized); } if (pcmk_is_set(flags, pcmk_sim_show_scores)) { pe__set_working_set_flags(data_set, pe_flag_show_scores); } if (pcmk_is_set(flags, pcmk_sim_show_utilization)) { pe__set_working_set_flags(data_set, pe_flag_show_utilization); } } int pcmk__write_sim_dotfile(pe_working_set_t *data_set, const char *dot_file, bool all_actions, bool verbose) { GList *gIter = NULL; FILE *dot_strm = fopen(dot_file, "w"); if (dot_strm == NULL) { return errno; } fprintf(dot_strm, " digraph \"g\" {\n"); for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->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, pe_action_pseudo)) { font = "orange"; } if (pcmk_is_set(action->flags, pe_action_dumped)) { style = "bold"; color = "green"; } else if ((action->rsc != NULL) && !pcmk_is_set(action->rsc->flags, pe_rsc_managed)) { color = "red"; font = "purple"; if (!all_actions) { goto do_not_write; } } else if (pcmk_is_set(action->flags, pe_action_optional)) { color = "blue"; if (!all_actions) { goto do_not_write; } } else { color = "red"; - CRM_CHECK(!pcmk_is_set(action->flags, pe_action_runnable), ;); + CRM_LOG_ASSERT(!pcmk_is_set(action->flags, pe_action_runnable)); } pe__set_action_flags(action, pe_action_dumped); fprintf(dot_strm, "\"%s\" [ style=%s color=\"%s\" fontcolor=\"%s\"]\n", action_name, style, color, font); do_not_write: free(action_name); } for (gIter = data_set->actions; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; GList *gIter2 = NULL; for (gIter2 = action->actions_before; gIter2 != NULL; gIter2 = gIter2->next) { pe_action_wrapper_t *before = (pe_action_wrapper_t *) gIter2->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 = "bold"; } else if (before->type == pe_order_none) { continue; } else if (pcmk_is_set(before->action->flags, pe_action_dumped) && pcmk_is_set(action->flags, pe_action_dumped) && before->type != pe_order_load) { 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; } void pcmk__profile_file(const char *xml_file, long long repeat, pe_working_set_t *data_set, char *use_date) { pcmk__output_t *out = data_set->priv; xmlNode *cib_object = NULL; clock_t start = 0; clock_t end; CRM_ASSERT(out != NULL); cib_object = filename2xml(xml_file); start = clock(); if (pcmk_find_cib_element(cib_object, XML_CIB_TAG_STATUS) == NULL) { create_xml_node(cib_object, XML_CIB_TAG_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; } for (int i = 0; i < repeat; ++i) { xmlNode *input = (repeat == 1)? cib_object : copy_xml(cib_object); data_set->input = input; pcmk__set_effective_date(data_set, false, use_date); pcmk__schedule_actions(data_set, input, NULL); pe_reset_working_set(data_set); } end = clock(); out->message(out, "profile", xml_file, start, end); } void pcmk__profile_dir(const char *dir, long long repeat, pe_working_set_t *data_set, char *use_date) { pcmk__output_t *out = data_set->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)) { pcmk__profile_file(buffer, repeat, data_set, use_date); } free(namelist[file_num]); } free(namelist); out->end_list(out); } } void pcmk__set_effective_date(pe_working_set_t *data_set, bool print_original, char *use_date) { pcmk__output_t *out = data_set->priv; time_t original_date = 0; CRM_ASSERT(out != NULL); crm_element_value_epoch(data_set->input, "execution-date", &original_date); if (use_date) { data_set->now = crm_time_new(use_date); out->info(out, "Setting effective cluster time: %s", use_date); crm_time_log(LOG_NOTICE, "Pretending 'now' is", data_set->now, crm_time_log_date | crm_time_log_timeofday); } else if (original_date) { data_set->now = crm_time_new(NULL); crm_time_set_timet(data_set->now, &original_date); if (print_original) { char *when = crm_time_as_string(data_set->now, crm_time_log_date|crm_time_log_timeofday); out->info(out, "Using the original execution date of: %s", when); free(when); } } } int pcmk__simulate(pe_working_set_t *data_set, pcmk__output_t *out, pcmk_injections_t *injections, unsigned int flags, unsigned int section_opts, char *use_date, char *input_file, char *graph_file, char *dot_file) { int printed = pcmk_rc_no_output; int rc = pcmk_rc_ok; xmlNodePtr input = NULL; cib_t *cib = NULL; bool modified = false; rc = cib__signon_query(&cib, &input); if (rc != pcmk_rc_ok) { goto simulate_done; } reset(data_set, input, out, use_date, flags); cluster_status(data_set); if (!out->is_quiet(out)) { if (pcmk_is_set(data_set->flags, pe_flag_maintenance_mode)) { printed = out->message(out, "maint-mode", data_set->flags); } if (data_set->disabled_resources || data_set->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", data_set->disabled_resources, data_set->ninstances, data_set->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(data_set, pcmk_is_set(flags, pcmk_sim_show_pending) ? pcmk_show_pending : 0, section_opts, "Current cluster status", printed == pcmk_rc_ok); printed = pcmk_rc_ok; } modified = 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 || injections->watchdog != NULL; if (modified) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); modify_configuration(data_set, 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(data_set); reset(data_set, input, out, use_date, flags); cluster_status(data_set); } 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)) { crm_time_t *local_date = NULL; pcmk__output_t *logger_out = NULL; if (pcmk_all_flags_set(data_set->flags, pe_flag_show_scores|pe_flag_show_utilization)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Allocation Scores and Utilization Information"); printed = pcmk_rc_ok; } else if (pcmk_is_set(data_set->flags, pe_flag_show_scores)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Allocation Scores"); printed = pcmk_rc_ok; } else if (pcmk_is_set(data_set->flags, pe_flag_show_utilization)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Utilization Information"); printed = pcmk_rc_ok; } else { logger_out = pcmk__new_logger(); if (logger_out == NULL) { rc = pcmk_rc_error; goto simulate_done; } data_set->priv = logger_out; } pcmk__schedule_actions(data_set, input, local_date); if (logger_out == NULL) { out->end_list(out); } else { logger_out->finish(logger_out, CRM_EX_OK, true, NULL); pcmk__output_free(logger_out); data_set->priv = out; } input = NULL; /* Don't try and free it twice */ if (graph_file != NULL) { rc = write_xml_file(data_set->graph, graph_file, FALSE); if (rc < 0) { rc = pcmk_rc_graph_error; goto simulate_done; } } if (dot_file != NULL) { rc = pcmk__write_sim_dotfile(data_set, 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(data_set, printed == pcmk_rc_ok); } } rc = pcmk_rc_ok; if (pcmk_is_set(flags, pcmk_sim_simulate)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); if (run_simulation(data_set, cib, injections->op_fail) != transition_complete) { rc = pcmk_rc_invalid_transition; } if (!out->is_quiet(out)) { pcmk__set_effective_date(data_set, true, use_date); if (pcmk_is_set(flags, pcmk_sim_show_scores)) { pe__set_working_set_flags(data_set, pe_flag_show_scores); } if (pcmk_is_set(flags, pcmk_sim_show_utilization)) { pe__set_working_set_flags(data_set, pe_flag_show_utilization); } cluster_status(data_set); print_cluster_status(data_set, 0, section_opts, "Revised Cluster Status", true); } } simulate_done: if (cib) { cib->cmds->signoff(cib); cib_delete(cib); } return rc; } int pcmk_simulate(xmlNodePtr *xml, pe_working_set_t *data_set, pcmk_injections_t *injections, unsigned int flags, unsigned int section_opts, char *use_date, char *input_file, char *graph_file, char *dot_file) { pcmk__output_t *out = NULL; int rc = pcmk_rc_ok; rc = pcmk__out_prologue(&out, xml); if (rc != pcmk_rc_ok) { return rc; } pe__register_messages(out); pcmk__register_lib_messages(out); rc = pcmk__simulate(data_set, out, injections, flags, section_opts, use_date, input_file, graph_file, dot_file); pcmk__out_epilogue(out, xml, rc); return rc; } void pcmk_free_injections(pcmk_injections_t *injections) { if (injections == NULL) { return; } g_list_free_full(injections->node_up, g_free); g_list_free_full(injections->node_down, g_free); g_list_free_full(injections->node_fail, g_free); g_list_free_full(injections->op_fail, g_free); g_list_free_full(injections->op_inject, g_free); g_list_free_full(injections->ticket_grant, g_free); g_list_free_full(injections->ticket_revoke, g_free); g_list_free_full(injections->ticket_standby, g_free); g_list_free_full(injections->ticket_activate, g_free); free(injections->quorum); free(injections->watchdog); free(injections); } diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c index b1e84110a2..46d48e0170 100644 --- a/lib/pengine/unpack.c +++ b/lib/pengine/unpack.c @@ -1,4115 +1,4115 @@ /* - * Copyright 2004-2021 the Pacemaker project contributors + * Copyright 2004-2022 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 #include CRM_TRACE_INIT_DATA(pe_status); /* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than * use pe__set_working_set_flags()/pe__clear_working_set_flags() so that the * flag is stringified more readably in log messages. */ #define set_config_flag(data_set, option, flag) do { \ const char *scf_value = pe_pref((data_set)->config_hash, (option)); \ if (scf_value != NULL) { \ if (crm_is_true(scf_value)) { \ (data_set)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Working set", \ crm_system_name, (data_set)->flags, \ (flag), #flag); \ } else { \ (data_set)->flags = pcmk__clear_flags_as(__func__, __LINE__,\ LOG_TRACE, "Working set", \ crm_system_name, (data_set)->flags, \ (flag), #flag); \ } \ } \ } while(0) static void unpack_rsc_op(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *failed, pe_working_set_t *data_set); static void determine_remote_online_status(pe_working_set_t *data_set, pe_node_t *this_node); static void add_node_attrs(xmlNode *attrs, pe_node_t *node, bool overwrite, pe_working_set_t *data_set); static void determine_online_status(xmlNode *node_state, pe_node_t *this_node, pe_working_set_t *data_set); static void unpack_node_lrm(pe_node_t *node, xmlNode *xml, pe_working_set_t *data_set); // Bitmask for warnings we only want to print once uint32_t pe_wo = 0; static gboolean is_dangling_guest_node(pe_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 (pe__is_guest_or_remote_node(node) && node->details->remote_rsc && node->details->remote_rsc->container == NULL && pcmk_is_set(node->details->remote_rsc->flags, pe_rsc_orphan_container_filler)) { return TRUE; } return FALSE; } /*! * \brief Schedule a fence action for a node * * \param[in,out] data_set Current working set of cluster * \param[in,out] node Node to fence * \param[in] reason Text description of why fencing is needed * \param[in] priority_delay Whether to consider `priority-fencing-delay` */ void pe_fence_node(pe_working_set_t * data_set, pe_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 (pe__is_guest_node(node)) { pe_resource_t *rsc = node->details->remote_rsc->container; if (!pcmk_is_set(rsc->flags, pe_rsc_failed)) { if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { crm_notice("Not fencing guest node %s " "(otherwise would because %s): " "its guest resource %s is unmanaged", node->details->uname, reason, rsc->id); } else { crm_warn("Guest node %s will be fenced " "(by recovering its guest resource %s): %s", node->details->uname, 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; pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); } } } 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", node->details->uname, reason); pe__set_resource_flags(node->details->remote_rsc, pe_rsc_failed|pe_rsc_stop); } else if (pe__is_remote_node(node)) { pe_resource_t *rsc = node->details->remote_rsc; if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_managed)) { crm_notice("Not fencing remote node %s " "(otherwise would because %s): connection is unmanaged", node->details->uname, reason); } else if(node->details->remote_requires_reset == FALSE) { node->details->remote_requires_reset = TRUE; crm_warn("Remote node %s %s: %s", node->details->uname, pe_can_fence(data_set, node)? "will be fenced" : "is unclean", reason); } node->details->unclean = TRUE; // No need to apply `priority-fencing-delay` for remote nodes pe_fence_op(node, NULL, TRUE, reason, FALSE, data_set); } else if (node->details->unclean) { crm_trace("Cluster node %s %s because %s", node->details->uname, pe_can_fence(data_set, node)? "would also be fenced" : "also is unclean", reason); } else { crm_warn("Cluster node %s %s: %s", node->details->uname, pe_can_fence(data_set, node)? "will be fenced" : "is unclean", reason); node->details->unclean = TRUE; pe_fence_op(node, NULL, TRUE, reason, priority_delay, data_set); } } // @TODO xpaths can't handle templates, rules, or id-refs // nvpair with provides or requires set to unfencing #define XPATH_UNFENCING_NVPAIR XML_CIB_TAG_NVPAIR \ "[(@" XML_NVPAIR_ATTR_NAME "='" PCMK_STONITH_PROVIDES "'" \ "or @" XML_NVPAIR_ATTR_NAME "='" XML_RSC_ATTR_REQUIRES "') " \ "and @" XML_NVPAIR_ATTR_VALUE "='unfencing']" // unfencing in rsc_defaults or any resource #define XPATH_ENABLE_UNFENCING \ "/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RESOURCES \ "//" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR \ "|/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RSCCONFIG \ "/" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR static void set_if_xpath(uint64_t flag, const char *xpath, pe_working_set_t *data_set) { xmlXPathObjectPtr result = NULL; if (!pcmk_is_set(data_set->flags, flag)) { result = xpath_search(data_set->input, xpath); if (result && (numXpathResults(result) > 0)) { pe__set_working_set_flags(data_set, flag); } freeXpathObject(result); } } gboolean unpack_config(xmlNode * config, pe_working_set_t * data_set) { const char *value = NULL; GHashTable *config_hash = pcmk__strkey_table(free, free); pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; data_set->config_hash = config_hash; pe__unpack_dataset_nvpairs(config, XML_CIB_TAG_PROPSET, &rule_data, config_hash, CIB_OPTIONS_FIRST, FALSE, data_set); verify_pe_options(data_set->config_hash); set_config_flag(data_set, "enable-startup-probes", pe_flag_startup_probes); if (!pcmk_is_set(data_set->flags, pe_flag_startup_probes)) { crm_info("Startup probes: disabled (dangerous)"); } value = pe_pref(data_set->config_hash, XML_ATTR_HAVE_WATCHDOG); if (value && crm_is_true(value)) { crm_info("Watchdog-based self-fencing will be performed via SBD if " "fencing is required and stonith-watchdog-timeout is nonzero"); pe__set_working_set_flags(data_set, pe_flag_have_stonith_resource); } /* Set certain flags via xpath here, so they can be used before the relevant * configuration sections are unpacked. */ set_if_xpath(pe_flag_enable_unfencing, XPATH_ENABLE_UNFENCING, data_set); value = pe_pref(data_set->config_hash, "stonith-timeout"); data_set->stonith_timeout = (int) crm_parse_interval_spec(value); crm_debug("STONITH timeout: %d", data_set->stonith_timeout); set_config_flag(data_set, "stonith-enabled", pe_flag_stonith_enabled); crm_debug("STONITH of failed nodes is %s", pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)? "enabled" : "disabled"); data_set->stonith_action = pe_pref(data_set->config_hash, "stonith-action"); if (!strcmp(data_set->stonith_action, "poweroff")) { pe_warn_once(pe_wo_poweroff, "Support for stonith-action of 'poweroff' is deprecated " "and will be removed in a future release (use 'off' instead)"); data_set->stonith_action = "off"; } crm_trace("STONITH will %s nodes", data_set->stonith_action); set_config_flag(data_set, "concurrent-fencing", pe_flag_concurrent_fencing); crm_debug("Concurrent fencing is %s", pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)? "enabled" : "disabled"); value = pe_pref(data_set->config_hash, XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY); if (value) { data_set->priority_fencing_delay = crm_parse_interval_spec(value) / 1000; crm_trace("Priority fencing delay is %ds", data_set->priority_fencing_delay); } set_config_flag(data_set, "stop-all-resources", pe_flag_stop_everything); crm_debug("Stop all active resources: %s", pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_stop_everything))); set_config_flag(data_set, "symmetric-cluster", pe_flag_symmetric_cluster); if (pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster)) { crm_debug("Cluster is symmetric" " - resources can run anywhere by default"); } value = pe_pref(data_set->config_hash, "no-quorum-policy"); if (pcmk__str_eq(value, "ignore", pcmk__str_casei)) { data_set->no_quorum_policy = no_quorum_ignore; } else if (pcmk__str_eq(value, "freeze", pcmk__str_casei)) { data_set->no_quorum_policy = no_quorum_freeze; } else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) { data_set->no_quorum_policy = no_quorum_demote; } else if (pcmk__str_eq(value, "suicide", pcmk__str_casei)) { if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { int do_panic = 0; crm_element_value_int(data_set->input, XML_ATTR_QUORUM_PANIC, &do_panic); if (do_panic || pcmk_is_set(data_set->flags, pe_flag_have_quorum)) { data_set->no_quorum_policy = no_quorum_suicide; } else { crm_notice("Resetting no-quorum-policy to 'stop': cluster has never had quorum"); data_set->no_quorum_policy = no_quorum_stop; } } else { pcmk__config_err("Resetting no-quorum-policy to 'stop' because " "fencing is disabled"); data_set->no_quorum_policy = no_quorum_stop; } } else { data_set->no_quorum_policy = no_quorum_stop; } switch (data_set->no_quorum_policy) { case no_quorum_freeze: crm_debug("On loss of quorum: Freeze resources"); break; case no_quorum_stop: crm_debug("On loss of quorum: Stop ALL resources"); break; case no_quorum_demote: crm_debug("On loss of quorum: " "Demote promotable resources and stop other resources"); break; case no_quorum_suicide: crm_notice("On loss of quorum: Fence all remaining nodes"); break; case no_quorum_ignore: crm_notice("On loss of quorum: Ignore"); break; } set_config_flag(data_set, "stop-orphan-resources", pe_flag_stop_rsc_orphans); crm_trace("Orphan resources are %s", pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)? "stopped" : "ignored"); set_config_flag(data_set, "stop-orphan-actions", pe_flag_stop_action_orphans); crm_trace("Orphan resource actions are %s", pcmk_is_set(data_set->flags, pe_flag_stop_action_orphans)? "stopped" : "ignored"); value = pe_pref(data_set->config_hash, "remove-after-stop"); if (value != NULL) { if (crm_is_true(value)) { pe__set_working_set_flags(data_set, pe_flag_remove_after_stop); #ifndef PCMK__COMPAT_2_0 pe_warn_once(pe_wo_remove_after, "Support for the remove-after-stop cluster property is" " deprecated and will be removed in a future release"); #endif } else { pe__clear_working_set_flags(data_set, pe_flag_remove_after_stop); } } set_config_flag(data_set, "maintenance-mode", pe_flag_maintenance_mode); crm_trace("Maintenance mode: %s", pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_maintenance_mode))); set_config_flag(data_set, "start-failure-is-fatal", pe_flag_start_failure_fatal); crm_trace("Start failures are %s", pcmk_is_set(data_set->flags, pe_flag_start_failure_fatal)? "always fatal" : "handled by failcount"); if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { set_config_flag(data_set, "startup-fencing", pe_flag_startup_fencing); } if (pcmk_is_set(data_set->flags, pe_flag_startup_fencing)) { crm_trace("Unseen nodes will be fenced"); } else { pe_warn_once(pe_wo_blind, "Blind faith: not fencing unseen nodes"); } pcmk__score_red = char2score(pe_pref(data_set->config_hash, "node-health-red")); pcmk__score_green = char2score(pe_pref(data_set->config_hash, "node-health-green")); pcmk__score_yellow = char2score(pe_pref(data_set->config_hash, "node-health-yellow")); crm_debug("Node scores: 'red' = %s, 'yellow' = %s, 'green' = %s", pe_pref(data_set->config_hash, "node-health-red"), pe_pref(data_set->config_hash, "node-health-yellow"), pe_pref(data_set->config_hash, "node-health-green")); data_set->placement_strategy = pe_pref(data_set->config_hash, "placement-strategy"); crm_trace("Placement strategy: %s", data_set->placement_strategy); set_config_flag(data_set, "shutdown-lock", pe_flag_shutdown_lock); crm_trace("Resources will%s be locked to cleanly shut down nodes", (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)? "" : " not")); if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { value = pe_pref(data_set->config_hash, XML_CONFIG_ATTR_SHUTDOWN_LOCK_LIMIT); data_set->shutdown_lock = crm_parse_interval_spec(value) / 1000; crm_trace("Shutdown locks expire after %us", data_set->shutdown_lock); } return TRUE; } pe_node_t * pe_create_node(const char *id, const char *uname, const char *type, const char *score, pe_working_set_t * data_set) { pe_node_t *new_node = NULL; if (pe_find_node(data_set->nodes, uname) != NULL) { pcmk__config_warn("More than one node entry has name '%s'", uname); } new_node = calloc(1, sizeof(pe_node_t)); if (new_node == NULL) { return NULL; } new_node->weight = char2score(score); new_node->fixed = FALSE; new_node->details = calloc(1, sizeof(struct pe_node_shared_s)); if (new_node->details == NULL) { free(new_node); 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->type = node_ping; if (pcmk__str_eq(type, "remote", pcmk__str_casei)) { new_node->details->type = node_remote; pe__set_working_set_flags(data_set, pe_flag_have_remote_nodes); } else if (pcmk__str_eq(type, "member", pcmk__str_null_matches | pcmk__str_casei)) { new_node->details->type = node_member; } new_node->details->attrs = pcmk__strkey_table(free, free); if (pe__is_guest_or_remote_node(new_node)) { g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("remote")); } else { g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("cluster")); } new_node->details->utilization = pcmk__strkey_table(free, free); new_node->details->digest_cache = pcmk__strkey_table(free, pe__free_digests); data_set->nodes = g_list_insert_sorted(data_set->nodes, new_node, sort_node_uname); return new_node; } static const char * expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pe_working_set_t *data) { xmlNode *attr_set = NULL; xmlNode *attr = NULL; const char *container_id = 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__str_eq((const char *)attr_set->name, XML_TAG_META_SETS, pcmk__str_casei)) { continue; } for (attr = pcmk__xe_first_child(attr_set); attr != NULL; attr = pcmk__xe_next(attr)) { const char *value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE); const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME); if (pcmk__str_eq(name, XML_RSC_ATTR_REMOTE_NODE, pcmk__str_casei)) { remote_name = value; } else if (pcmk__str_eq(name, "remote-addr", pcmk__str_casei)) { remote_server = value; } else if (pcmk__str_eq(name, "remote-port", pcmk__str_casei)) { remote_port = value; } else if (pcmk__str_eq(name, "remote-connect-timeout", pcmk__str_casei)) { connect_timeout = value; } else if (pcmk__str_eq(name, "remote-allow-migrate", pcmk__str_casei)) { remote_allow_migrate=value; } else if (pcmk__str_eq(name, XML_RSC_ATTR_MANAGED, pcmk__str_casei)) { 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(pe_working_set_t *data_set, pe_node_t *new_node) { if ((new_node->details->type == node_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(data_set->flags, pe_flag_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, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; pe_node_t *new_node = NULL; const char *id = NULL; const char *uname = NULL; const char *type = NULL; const char *score = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; for (xml_obj = pcmk__xe_first_child(xml_nodes); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_NODE, pcmk__str_none)) { new_node = NULL; id = crm_element_value(xml_obj, XML_ATTR_ID); uname = crm_element_value(xml_obj, XML_ATTR_UNAME); type = crm_element_value(xml_obj, XML_ATTR_TYPE); score = crm_element_value(xml_obj, XML_RULE_ATTR_SCORE); crm_trace("Processing node %s/%s", uname, id); if (id == NULL) { pcmk__config_err("Ignoring <" XML_CIB_TAG_NODE "> entry in configuration without id"); continue; } new_node = pe_create_node(id, uname, type, score, data_set); if (new_node == NULL) { return FALSE; } /* if(data_set->have_quorum == FALSE */ /* && data_set->no_quorum_policy == no_quorum_stop) { */ /* /\* start shutting resources down *\/ */ /* new_node->weight = -INFINITY; */ /* } */ handle_startup_fencing(data_set, new_node); add_node_attrs(xml_obj, new_node, FALSE, data_set); pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_UTILIZATION, &rule_data, new_node->details->utilization, NULL, FALSE, data_set); crm_trace("Done with node %s", crm_element_value(xml_obj, XML_ATTR_UNAME)); } } if (data_set->localhost && pe_find_node(data_set->nodes, data_set->localhost) == NULL) { crm_info("Creating a fake local node"); pe_create_node(data_set->localhost, data_set->localhost, NULL, 0, data_set); } return TRUE; } static void setup_container(pe_resource_t * rsc, pe_working_set_t * data_set) { const char *container_id = NULL; if (rsc->children) { g_list_foreach(rsc->children, (GFunc) setup_container, data_set); return; } container_id = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_CONTAINER); if (container_id && !pcmk__str_eq(container_id, rsc->id, pcmk__str_casei)) { pe_resource_t *container = pe_find_resource(data_set->resources, container_id); if (container) { rsc->container = container; pe__set_resource_flags(container, pe_rsc_is_container); container->fillers = g_list_append(container->fillers, rsc); pe_rsc_trace(rsc, "Resource %s's container is %s", rsc->id, container_id); } else { pe_err("Resource %s: Unknown resource container (%s)", rsc->id, container_id); } } } gboolean unpack_remote_nodes(xmlNode * xml_resources, pe_working_set_t * data_set) { 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 = 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(data_set->nodes, new_node_id) == NULL) { crm_trace("Found remote node %s defined by resource %s", new_node_id, ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } 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__str_eq((const char *)xml_obj->name, XML_CIB_TAG_RESOURCE, pcmk__str_none)) { /* 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, data_set); if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found guest node %s in resource %s", new_node_id, ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } continue; } /* Check for guest nodes inside a group. Clones are currently not * supported as guest nodes. */ if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_GROUP, pcmk__str_none)) { 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, data_set); if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found guest node %s in resource %s inside group %s", new_node_id, ID(xml_obj2), ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } } } } 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(pe_working_set_t *data_set, pe_resource_t *new_rsc) { pe_node_t *remote_node = NULL; if (new_rsc->is_remote_node == FALSE) { return; } if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) { /* remote_nodes and remote_resources are not linked in quick location calculations */ return; } remote_node = pe_find_node(data_set->nodes, new_rsc->id); - CRM_CHECK(remote_node != NULL, return;); + CRM_CHECK(remote_node != NULL, return); pe_rsc_trace(new_rsc, "Linking remote connection resource %s to node %s", new_rsc->id, remote_node->details->uname); 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(data_set, 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. */ g_hash_table_replace(remote_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("container")); } } static void destroy_tag(gpointer data) { pe_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] data_set Where to put resource information * * \return TRUE * * \note unpack_remote_nodes() MUST be called before this, so that the nodes can * be used when common_unpack() calls resource_location() */ gboolean unpack_resources(xmlNode * xml_resources, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; GList *gIter = NULL; data_set->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)) { pe_resource_t *new_rsc = NULL; if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_RSC_TEMPLATE, pcmk__str_none)) { const char *template_id = ID(xml_obj); if (template_id && g_hash_table_lookup_extended(data_set->template_rsc_sets, template_id, NULL, NULL) == FALSE) { /* Record the template's ID for the knowledge of its existence anyway. */ g_hash_table_insert(data_set->template_rsc_sets, strdup(template_id), NULL); } continue; } crm_trace("Beginning unpack... <%s id=%s... >", crm_element_name(xml_obj), ID(xml_obj)); if (common_unpack(xml_obj, &new_rsc, NULL, data_set) && (new_rsc != NULL)) { data_set->resources = g_list_append(data_set->resources, new_rsc); pe_rsc_trace(new_rsc, "Added resource %s", new_rsc->id); } else { pcmk__config_err("Ignoring <%s> resource '%s' " "because configuration is invalid", crm_element_name(xml_obj), crm_str(ID(xml_obj))); if (new_rsc != NULL && new_rsc->fns != NULL) { new_rsc->fns->free(new_rsc); } } } for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; setup_container(rsc, data_set); link_rsc2remotenode(data_set, rsc); } data_set->resources = g_list_sort(data_set->resources, sort_rsc_priority); if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) { /* Ignore */ } else if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled) && !pcmk_is_set(data_set->flags, pe_flag_have_stonith_resource)) { 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 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, pe_working_set_t * data_set) { xmlNode *xml_tag = NULL; data_set->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 = ID(xml_tag); if (!pcmk__str_eq((const char *)xml_tag->name, XML_CIB_TAG_TAG, pcmk__str_none)) { continue; } if (tag_id == NULL) { pcmk__config_err("Ignoring <%s> without " XML_ATTR_ID, crm_element_name(xml_tag)); 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 = ID(xml_obj_ref); if (!pcmk__str_eq((const char *)xml_obj_ref->name, XML_CIB_TAG_OBJ_REF, pcmk__str_none)) { continue; } if (obj_ref == NULL) { pcmk__config_err("Ignoring <%s> for tag '%s' without " XML_ATTR_ID, crm_element_name(xml_obj_ref), tag_id); continue; } if (add_tag_ref(data_set->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, pe_working_set_t * data_set) { const char *ticket_id = NULL; const char *granted = NULL; const char *last_granted = NULL; const char *standby = NULL; xmlAttrPtr xIter = NULL; pe_ticket_t *ticket = NULL; ticket_id = 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(data_set->tickets, ticket_id); if (ticket == NULL) { ticket = ticket_new(ticket_id, data_set); 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 = crm_element_value(xml_ticket, prop_name); if (pcmk__str_eq(prop_name, XML_ATTR_ID, pcmk__str_none)) { continue; } g_hash_table_replace(ticket->state, strdup(prop_name), strdup(prop_value)); } granted = g_hash_table_lookup(ticket->state, "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, "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, "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, pe_working_set_t * data_set) { 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__str_eq((const char *)xml_obj->name, XML_CIB_TAG_TICKET_STATE, pcmk__str_none)) { continue; } unpack_ticket_state(xml_obj, data_set); } return TRUE; } static void unpack_handle_remote_attrs(pe_node_t *this_node, xmlNode *state, pe_working_set_t * data_set) { const char *resource_discovery_enabled = NULL; xmlNode *attrs = NULL; pe_resource_t *rsc = NULL; if (!pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { return; } if ((this_node == NULL) || !pe__is_guest_or_remote_node(this_node)) { return; } crm_trace("Processing remote node id=%s, uname=%s", this_node->details->id, this_node->details->uname); pcmk__scan_min_int(crm_element_value(state, XML_NODE_IS_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, XML_TAG_TRANSIENT_NODEATTRS, FALSE); add_node_attrs(attrs, this_node, TRUE, data_set); if (pe__shutdown_requested(this_node)) { crm_info("Node %s is shutting down", this_node->details->uname); this_node->details->shutdown = TRUE; } if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) { crm_info("Node %s is in standby-mode", this_node->details->uname); this_node->details->standby = TRUE; } if (crm_is_true(pe_node_attribute_raw(this_node, "maintenance")) || ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_managed))) { crm_info("Node %s is in maintenance-mode", this_node->details->uname); this_node->details->maintenance = TRUE; } resource_discovery_enabled = pe_node_attribute_raw(this_node, XML_NODE_ATTR_RSC_DISCOVERY); if (resource_discovery_enabled && !crm_is_true(resource_discovery_enabled)) { if (pe__is_remote_node(this_node) && !pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { crm_warn("Ignoring %s attribute on remote node %s because stonith is disabled", XML_NODE_ATTR_RSC_DISCOVERY, this_node->details->uname); } 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("Node %s has resource discovery disabled", this_node->details->uname); 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] node Cluster node whose attributes are being unpacked * \param[in] data_set Cluster working set */ static void unpack_transient_attributes(xmlNode *state, pe_node_t *node, pe_working_set_t *data_set) { const char *discovery = NULL; xmlNode *attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE); add_node_attrs(attrs, node, TRUE, data_set); if (crm_is_true(pe_node_attribute_raw(node, "standby"))) { crm_info("Node %s is in standby-mode", node->details->uname); node->details->standby = TRUE; } if (crm_is_true(pe_node_attribute_raw(node, "maintenance"))) { crm_info("Node %s is in maintenance-mode", node->details->uname); node->details->maintenance = TRUE; } discovery = pe_node_attribute_raw(node, XML_NODE_ATTR_RSC_DISCOVERY); if ((discovery != NULL) && !crm_is_true(discovery)) { crm_warn("Ignoring %s attribute for node %s because disabling " "resource discovery is not allowed for cluster nodes", XML_NODE_ATTR_RSC_DISCOVERY, node->details->uname); } } /*! * \internal * \brief Unpack a node state entry (first pass) * * Unpack one node state entry from status. This unpacks information from the * 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] data_set Cluster working set */ static void unpack_node_state(xmlNode *state, pe_working_set_t *data_set) { const char *id = NULL; const char *uname = NULL; pe_node_t *this_node = NULL; id = crm_element_value(state, XML_ATTR_ID); if (id == NULL) { crm_warn("Ignoring malformed " XML_CIB_TAG_STATE " entry without " XML_ATTR_ID); return; } uname = crm_element_value(state, XML_ATTR_UNAME); if (uname == NULL) { crm_warn("Ignoring malformed " XML_CIB_TAG_STATE " entry without " XML_ATTR_UNAME); return; } this_node = pe_find_node_any(data_set->nodes, id, uname); if (this_node == NULL) { pcmk__config_warn("Ignoring recorded node state for '%s' because " "it is no longer in the configuration", uname); return; } if (pe__is_guest_or_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, XML_NODE_IS_FENCED), &(this_node->details->remote_was_fenced), 0); return; } unpack_transient_attributes(state, this_node, data_set); /* 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)", this_node->details->uname, id); determine_online_status(state, this_node, data_set); if (!pcmk_is_set(data_set->flags, pe_flag_have_quorum) && this_node->details->online && (data_set->no_quorum_policy == no_quorum_suicide)) { /* Everything else should flow from this automatically * (at least until the scheduler becomes able to migrate off * healthy resources) */ pe_fence_node(data_set, 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] data_set Cluster working set * * \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(xmlNode *status, bool fence, pe_working_set_t *data_set) { int rc = pcmk_rc_ok; // Loop through all node_state entries in CIB status for (xmlNode *state = first_named_child(status, XML_CIB_TAG_STATE); state != NULL; state = crm_next_same_xml(state)) { const char *id = ID(state); const char *uname = crm_element_value(state, XML_ATTR_UNAME); pe_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 " XML_CIB_TAG_STATE " without id and/or uname"); continue; } this_node = pe_find_node_any(data_set->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 (pe__is_guest_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. */ pe_resource_t *rsc = this_node->details->remote_rsc; if ((rsc == NULL) || (rsc->role != RSC_ROLE_STARTED) || (rsc->container->role != RSC_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 (pe__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. */ pe_resource_t *rsc = this_node->details->remote_rsc; if ((rsc == NULL) || (!pcmk_is_set(data_set->flags, pe_flag_shutdown_lock) && (rsc->role != RSC_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(data_set->flags, pe_flag_stonith_enabled |pe_flag_shutdown_lock) && !this_node->details->online) { crm_trace("Not unpacking resource history for offline " "cluster node %s", id); continue; } if (pe__is_guest_or_remote_node(this_node)) { determine_remote_online_status(data_set, this_node); unpack_handle_remote_attrs(this_node, state, data_set); } crm_trace("Unpacking resource history for %snode %s", (fence? "unseen " : ""), id); this_node->details->unpacked = TRUE; unpack_node_lrm(this_node, state, data_set); 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, pe_working_set_t * data_set) { xmlNode *state = NULL; crm_trace("Beginning unpack"); if (data_set->tickets == NULL) { data_set->tickets = pcmk__strkey_table(free, destroy_ticket); } for (state = pcmk__xe_first_child(status); state != NULL; state = pcmk__xe_next(state)) { if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_TICKETS, pcmk__str_none)) { unpack_tickets_state((xmlNode *) state, data_set); } else if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { unpack_node_state(state, data_set); } } while (unpack_node_history(status, FALSE, data_set) == 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(data_set->flags, pe_flag_stonith_enabled), data_set); /* Now that we know where resources are, we can schedule stops of containers * with failed bundle connections */ if (data_set->stop_needed != NULL) { for (GList *item = data_set->stop_needed; item; item = item->next) { pe_resource_t *container = item->data; pe_node_t *node = pe__current_node(container); if (node) { stop_action(container, node, FALSE); } } g_list_free(data_set->stop_needed); data_set->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 = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *this_node = gIter->data; if (!pe__is_guest_or_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, RSC_ROLE_STOPPED, "remote shutdown"); } if (!this_node->details->unpacked) { determine_remote_online_status(data_set, this_node); } } return TRUE; } static gboolean determine_online_status_no_fencing(pe_working_set_t * data_set, xmlNode * node_state, pe_node_t * this_node) { gboolean online = FALSE; const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE); const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER); const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER); const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); if (!crm_is_true(in_cluster)) { crm_trace("Node is down: in_cluster=%s", crm_str(in_cluster)); } else if (pcmk__str_eq(is_peer, ONLINESTATUS, pcmk__str_casei)) { if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { online = TRUE; } else { crm_debug("Node is not ready to run resources: %s", join); } } else if (this_node->details->expected_up == FALSE) { crm_trace("Controller is down: in_cluster=%s", crm_str(in_cluster)); crm_trace("\tis_peer=%s, join=%s, expected=%s", crm_str(is_peer), crm_str(join), crm_str(exp_state)); } else { /* mark it unclean */ pe_fence_node(data_set, this_node, "peer is unexpectedly down", FALSE); crm_info("\tin_cluster=%s, is_peer=%s, join=%s, expected=%s", crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state)); } return online; } static gboolean determine_online_status_fencing(pe_working_set_t * data_set, xmlNode * node_state, pe_node_t * this_node) { gboolean online = FALSE; gboolean do_terminate = FALSE; bool crmd_online = FALSE; const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE); const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER); const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER); const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); const char *terminate = pe_node_attribute_raw(this_node, "terminate"); /* - XML_NODE_IN_CLUSTER ::= true|false - XML_NODE_IS_PEER ::= online|offline - XML_NODE_JOIN_STATE ::= member|down|pending|banned - XML_NODE_EXPECTED ::= member|down */ if (crm_is_true(terminate)) { do_terminate = TRUE; } else if (terminate != NULL && strlen(terminate) > 0) { /* could be a time() value */ char t = terminate[0]; if (t != '0' && isdigit(t)) { do_terminate = TRUE; } } crm_trace("%s: in_cluster=%s, is_peer=%s, join=%s, expected=%s, term=%d", this_node->details->uname, crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state), do_terminate); online = crm_is_true(in_cluster); crmd_online = pcmk__str_eq(is_peer, ONLINESTATUS, pcmk__str_casei); if (exp_state == NULL) { exp_state = CRMD_JOINSTATE_DOWN; } if (this_node->details->shutdown) { crm_debug("%s is shutting down", this_node->details->uname); /* Slightly different criteria since we can't shut down a dead peer */ online = crmd_online; } else if (in_cluster == NULL) { pe_fence_node(data_set, this_node, "peer has not been seen by the cluster", FALSE); } else if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_casei)) { pe_fence_node(data_set, this_node, "peer failed Pacemaker membership criteria", FALSE); } else if (do_terminate == FALSE && pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN, pcmk__str_casei)) { if (crm_is_true(in_cluster) || crmd_online) { crm_info("- Node %s is not ready to run resources", this_node->details->uname); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { crm_trace("%s is down or still coming up", this_node->details->uname); } } else if (do_terminate && pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_casei) && crm_is_true(in_cluster) == FALSE && !crmd_online) { crm_info("Node %s was just shot", this_node->details->uname); online = FALSE; } else if (crm_is_true(in_cluster) == FALSE) { // Consider `priority-fencing-delay` for lost nodes pe_fence_node(data_set, this_node, "peer is no longer part of the cluster", TRUE); } else if (!crmd_online) { pe_fence_node(data_set, this_node, "peer process is no longer available", FALSE); /* Everything is running at this point, now check join state */ } else if (do_terminate) { pe_fence_node(data_set, this_node, "termination was requested", FALSE); } else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { crm_info("Node %s is active", this_node->details->uname); } else if (pcmk__strcase_any_of(join, CRMD_JOINSTATE_PENDING, CRMD_JOINSTATE_DOWN, NULL)) { crm_info("Node %s is not ready to run resources", this_node->details->uname); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { pe_fence_node(data_set, this_node, "peer was in an unknown state", FALSE); crm_warn("%s: in-cluster=%s, is-peer=%s, join=%s, expected=%s, term=%d, shutdown=%d", this_node->details->uname, crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state), do_terminate, this_node->details->shutdown); } return online; } static void determine_remote_online_status(pe_working_set_t * data_set, pe_node_t * this_node) { pe_resource_t *rsc = this_node->details->remote_rsc; pe_resource_t *container = NULL; pe_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 == RSC_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 == RSC_ROLE_STARTED && rsc->next_role == RSC_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, pe_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, pe_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 == RSC_ROLE_STOPPED || (container && container->role == RSC_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(xmlNode * node_state, pe_node_t * this_node, pe_working_set_t * data_set) { gboolean online = FALSE; const char *exp_state = crm_element_value(node_state, XML_NODE_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(data_set->flags, pe_flag_stonith_enabled)) { online = determine_online_status_no_fencing(data_set, node_state, this_node); } else { online = determine_online_status_fencing(data_set, node_state, this_node); } if (online) { this_node->details->online = TRUE; } else { /* remove node from contention */ this_node->fixed = TRUE; this_node->weight = -INFINITY; } if (online && this_node->details->shutdown) { /* don't run resources here */ this_node->fixed = TRUE; this_node->weight = -INFINITY; } if (this_node->details->type == node_ping) { crm_info("Node %s is not a Pacemaker node", this_node->details->uname); } else if (this_node->details->unclean) { pe_proc_warn("Node %s is unclean", this_node->details->uname); } else if (this_node->details->online) { crm_info("Node %s is %s", this_node->details->uname, 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("Node %s is offline", this_node->details->uname); } } /*! * \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 pe_resource_t * create_fake_resource(const char *rsc_id, xmlNode * rsc_entry, pe_working_set_t * data_set) { pe_resource_t *rsc = NULL; xmlNode *xml_rsc = create_xml_node(NULL, XML_CIB_TAG_RESOURCE); copy_in_properties(xml_rsc, rsc_entry); crm_xml_add(xml_rsc, XML_ATTR_ID, rsc_id); crm_log_xml_debug(xml_rsc, "Orphan resource"); if (!common_unpack(xml_rsc, &rsc, NULL, data_set)) { return NULL; } if (xml_contains_remote_node(xml_rsc)) { pe_node_t *node; crm_debug("Detected orphaned remote node %s", rsc_id); node = pe_find_node(data_set->nodes, rsc_id); if (node == NULL) { node = pe_create_node(rsc_id, rsc_id, "remote", NULL, data_set); } link_rsc2remotenode(data_set, 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, XML_RSC_ATTR_CONTAINER)) { /* This orphaned rsc needs to be mapped to a container. */ crm_trace("Detected orphaned container filler %s", rsc_id); pe__set_resource_flags(rsc, pe_rsc_orphan_container_filler); } pe__set_resource_flags(rsc, pe_rsc_orphan); data_set->resources = g_list_append(data_set->resources, rsc); return rsc; } /*! * \internal * \brief Create orphan instance for anonymous clone resource history */ static pe_resource_t * create_anonymous_orphan(pe_resource_t *parent, const char *rsc_id, pe_node_t *node, pe_working_set_t *data_set) { pe_resource_t *top = pe__create_clone_child(parent, data_set); // find_rsc() because we might be a cloned group pe_resource_t *orphan = top->fns->find_rsc(top, rsc_id, NULL, pe_find_clone); pe_rsc_debug(parent, "Created orphan %s for %s: %s on %s", top->id, parent->id, rsc_id, node->details->uname); 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 clone-max instances); * (3) a newly created orphan (i.e. clone-max instances are already active). * * \param[in] data_set Cluster information * \param[in] node Node on which to check for instance * \param[in] parent Clone to check * \param[in] rsc_id Name of cloned resource in history (without instance) */ static pe_resource_t * find_anonymous_clone(pe_working_set_t * data_set, pe_node_t * node, pe_resource_t * parent, const char *rsc_id) { GList *rIter = NULL; pe_resource_t *rsc = NULL; pe_resource_t *inactive_instance = NULL; gboolean skip_inactive = FALSE; CRM_ASSERT(parent != NULL); CRM_ASSERT(pe_rsc_is_clone(parent)); CRM_ASSERT(!pcmk_is_set(parent->flags, pe_rsc_unique)); // Check for active (or partially active, for cloned groups) instance pe_rsc_trace(parent, "Looking for %s on %s in %s", rsc_id, node->details->uname, parent->id); for (rIter = parent->children; rsc == NULL && rIter; rIter = rIter->next) { GList *locations = NULL; pe_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 globally-unique * was flipped from true to false); and * (3) when we re-run calculations on the same data set 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 (((pe_node_t *)locations->data)->details == node->details) { /* 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, pe_find_clone); if (rsc) { /* If there are multiple instance history entries for an * anonymous clone in a single node's history (which can * happen if 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, node->details->uname); skip_inactive = TRUE; rsc = NULL; } else { pe_rsc_trace(parent, "Resource %s, active", rsc->id); } } } g_list_free(locations); } else { pe_rsc_trace(parent, "Resource %s, skip inactive", child->id); if (!skip_inactive && !inactive_instance && !pcmk_is_set(child->flags, pe_rsc_block)) { // Remember one inactive instance in case we don't find active inactive_instance = parent->fns->find_rsc(child, rsc_id, NULL, pe_find_clone); /* ... but don't use it if it was already associated with a * pending action on another node */ if (inactive_instance && inactive_instance->pending_node && (inactive_instance->pending_node->details != node->details)) { inactive_instance = NULL; } } } } if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) { pe_rsc_trace(parent, "Resource %s, empty slot", inactive_instance->id); rsc = inactive_instance; } /* If the resource has "requires" set to "quorum" or "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, pe_rsc_needs_fencing) && (!node->details->online || node->details->unclean) && !pe__is_guest_node(node) && !pe__is_universal_clone(parent, data_set)) { rsc = NULL; } if (rsc == NULL) { rsc = create_anonymous_orphan(parent, rsc_id, node, data_set); pe_rsc_trace(parent, "Resource %s, orphan", rsc->id); } return rsc; } static pe_resource_t * unpack_find_resource(pe_working_set_t * data_set, pe_node_t * node, const char *rsc_id, xmlNode * rsc_entry) { pe_resource_t *rsc = NULL; pe_resource_t *parent = NULL; crm_trace("looking for %s", rsc_id); rsc = pe_find_resource(data_set->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 clone-max=0, we create * a single :0 orphan to match against here. */ char *clone0_id = clone_zero(rsc_id); pe_resource_t *clone0 = pe_find_resource(data_set->resources, clone0_id); if (clone0 && !pcmk_is_set(clone0->flags, pe_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 > pe_native) { crm_trace("Resource history for %s is orphaned because it is no longer primitive", rsc_id); return NULL; } else { parent = uber_parent(rsc); } if (pe_rsc_is_anon_clone(parent)) { if (pe_rsc_is_bundled(parent)) { rsc = pe__find_bundle_replica(parent->parent, node); } else { char *base = clone_strip(rsc_id); rsc = find_anonymous_clone(data_set, 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)) { free(rsc->clone_name); rsc->clone_name = strdup(rsc_id); pe_rsc_debug(rsc, "Internally renamed %s on %s to %s%s", rsc_id, node->details->uname, rsc->id, (pcmk_is_set(rsc->flags, pe_rsc_orphan)? " (ORPHAN)" : "")); } return rsc; } static pe_resource_t * process_orphan_resource(xmlNode * rsc_entry, pe_node_t * node, pe_working_set_t * data_set) { pe_resource_t *rsc = NULL; const char *rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID); crm_debug("Detected orphan resource %s on %s", rsc_id, node->details->uname); rsc = create_fake_resource(rsc_id, rsc_entry, data_set); if (rsc == NULL) { return NULL; } if (!pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)) { pe__clear_resource_flags(rsc, pe_rsc_managed); } else { CRM_CHECK(rsc != NULL, return NULL); pe_rsc_trace(rsc, "Added orphan %s", rsc->id); resource_location(rsc, NULL, -INFINITY, "__orphan_do_not_run__", data_set); } return rsc; } static void process_rsc_state(pe_resource_t * rsc, pe_node_t * node, enum action_fail_response on_fail, xmlNode * migrate_op, pe_working_set_t * data_set) { pe_node_t *tmpnode = NULL; char *reason = NULL; enum action_fail_response save_on_fail = action_fail_ignore; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s", rsc->id, role2text(rsc->role), node->details->uname, fail2text(on_fail)); /* process current state */ if (rsc->role != RSC_ROLE_UNKNOWN) { pe_resource_t *iter = rsc; while (iter) { if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) { pe_node_t *n = pe__copy_node(node); pe_rsc_trace(rsc, "%s (aka. %s) known on %s", rsc->id, rsc->clone_name, n->details->uname); g_hash_table_insert(iter->known_on, (gpointer) n->details->id, n); } if (pcmk_is_set(iter->flags, pe_rsc_unique)) { break; } iter = iter->parent; } } /* If a managed resource is believed to be running, but node is down ... */ if (rsc->role > RSC_ROLE_STOPPED && node->details->online == FALSE && node->details->maintenance == FALSE && pcmk_is_set(rsc->flags, pe_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 (pe__is_guest_node(node)) { pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); should_fence = TRUE; } else if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { if (pe__is_remote_node(node) && node->details->remote_rsc && !pcmk_is_set(node->details->remote_rsc->flags, pe_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(data_set, 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 = action_fail_ignore; } switch (on_fail) { case action_fail_ignore: /* nothing to do */ break; case action_fail_demote: pe__set_resource_flags(rsc, pe_rsc_failed); demote_action(rsc, node, FALSE); break; case action_fail_fence: /* 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(data_set, node, reason, FALSE); free(reason); break; case action_fail_standby: node->details->standby = TRUE; node->details->standby_onfail = TRUE; break; case action_fail_block: /* is_managed == FALSE will prevent any * actions being sent for the resource */ pe__clear_resource_flags(rsc, pe_rsc_managed); pe__set_resource_flags(rsc, pe_rsc_block); break; case action_fail_migrate: /* make sure it comes up somewhere else * or not at all */ resource_location(rsc, node, -INFINITY, "__action_migration_auto__", data_set); break; case action_fail_stop: pe__set_next_role(rsc, RSC_ROLE_STOPPED, "on-fail=stop"); break; case action_fail_recover: if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); stop_action(rsc, node, FALSE); } break; case action_fail_restart_container: pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); if (rsc->container && pe_rsc_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. */ data_set->stop_needed = g_list_prepend(data_set->stop_needed, rsc->container); } else if (rsc->container) { stop_action(rsc->container, node, FALSE); } else if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { stop_action(rsc, node, FALSE); } break; case action_fail_reset_remote: pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { tmpnode = NULL; if (rsc->is_remote_node) { tmpnode = pe_find_node(data_set->nodes, rsc->id); } if (tmpnode && pe__is_remote_node(tmpnode) && tmpnode->details->remote_was_fenced == 0) { /* The remote connection resource failed in a way that * should result in fencing the remote node. */ pe_fence_node(data_set, tmpnode, "remote connection is unrecoverable", FALSE); } } /* require the stop action regardless if fencing is occurring or not. */ if (rsc->role > RSC_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, RSC_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, pe_rsc_failed) && rsc->is_remote_node) { tmpnode = pe_find_node(data_set->nodes, rsc->id); if (tmpnode && tmpnode->details->unclean) { tmpnode->details->unseen = FALSE; } } if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) { if (pcmk_is_set(rsc->flags, pe_rsc_managed)) { pcmk__config_warn("Detected active orphan %s running on %s", rsc->id, node->details->uname); } else { pcmk__config_warn("Resource '%s' must be stopped manually on " "%s because cluster is configured not to " "stop active orphans", rsc->id, node->details->uname); } } native_add_running(rsc, node, data_set, (save_on_fail != action_fail_ignore)); switch (on_fail) { case action_fail_ignore: break; case action_fail_demote: case action_fail_block: pe__set_resource_flags(rsc, pe_rsc_failed); break; default: pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); 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 */ pe_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, RSC_STOP, FALSE); GList *gIter = possible_matches; for (; gIter != NULL; gIter = gIter->next) { pe_action_t *stop = (pe_action_t *) gIter->data; pe__set_action_flags(stop, pe_action_optional); } g_list_free(possible_matches); } } /* create active recurring operations as optional */ static void process_recurring(pe_node_t * node, pe_resource_t * rsc, int start_index, int stop_index, GList *sorted_op_list, pe_working_set_t * data_set) { int counter = -1; const char *task = NULL; const char *status = NULL; GList *gIter = sorted_op_list; CRM_ASSERT(rsc); pe_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 = ID(rsc_op); counter++; if (node->details->online == FALSE) { pe_rsc_trace(rsc, "Skipping %s/%s: node is offline", rsc->id, node->details->uname); break; /* Need to check if there's a monitor for role="Stopped" */ } else if (start_index < stop_index && counter <= stop_index) { pe_rsc_trace(rsc, "Skipping %s/%s: resource is not active", id, node->details->uname); continue; } else if (counter < start_index) { pe_rsc_trace(rsc, "Skipping %s/%s: old %d", id, node->details->uname, counter); continue; } crm_element_value_ms(rsc_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if (interval_ms == 0) { pe_rsc_trace(rsc, "Skipping %s/%s: non-recurring", id, node->details->uname); continue; } status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS); if (pcmk__str_eq(status, "-1", pcmk__str_casei)) { pe_rsc_trace(rsc, "Skipping %s/%s: status", id, node->details->uname); continue; } task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); /* create the action */ key = pcmk__op_key(rsc->id, task, interval_ms); pe_rsc_trace(rsc, "Creating %s/%s", key, node->details->uname); custom_action(rsc, key, task, node, TRUE, TRUE, data_set); } } void calculate_active_ops(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; GList *gIter = sorted_op_list; *stop_index = -1; *start_index = -1; for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS); if (pcmk__str_eq(task, CRMD_ACTION_STOP, pcmk__str_casei) && pcmk__str_eq(status, "0", pcmk__str_casei)) { *stop_index = counter; } else if (pcmk__strcase_any_of(task, CRMD_ACTION_START, CRMD_ACTION_MIGRATED, NULL)) { *start_index = counter; } else if ((implied_monitor_start <= *stop_index) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { const char *rc = crm_element_value(rsc_op, XML_LRM_ATTR_RC); if (pcmk__strcase_any_of(rc, "0", "8", NULL)) { implied_monitor_start = counter; } } else if (pcmk__strcase_any_of(task, CRMD_ACTION_PROMOTE, CRMD_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(xmlNode *rsc_entry, pe_resource_t *rsc, pe_node_t *node, pe_working_set_t *data_set) { time_t lock_time = 0; // When lock started (i.e. node shutdown time) if ((crm_element_value_epoch(rsc_entry, XML_CONFIG_ATTR_SHUTDOWN_LOCK, &lock_time) == pcmk_ok) && (lock_time != 0)) { if ((data_set->shutdown_lock > 0) && (get_effective_time(data_set) > (lock_time + data_set->shutdown_lock))) { pe_rsc_info(rsc, "Shutdown lock for %s on %s expired", rsc->id, node->details->uname); pe__clear_resource_history(rsc, node, data_set); } else { rsc->lock_node = node; rsc->lock_time = lock_time; } } } /*! * \internal * \brief Unpack one lrm_resource entry from a node's CIB status * * \param[in] node Node whose status is being unpacked * \param[in] rsc_entry lrm_resource XML being unpacked * \param[in] data_set Cluster working set * * \return Resource corresponding to the entry, or NULL if no operation history */ static pe_resource_t * unpack_lrm_resource(pe_node_t *node, xmlNode *lrm_resource, pe_working_set_t *data_set) { GList *gIter = NULL; int stop_index = -1; int start_index = -1; enum rsc_role_e req_role = RSC_ROLE_UNKNOWN; const char *task = NULL; const char *rsc_id = ID(lrm_resource); pe_resource_t *rsc = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; xmlNode *migrate_op = NULL; xmlNode *rsc_op = NULL; xmlNode *last_failure = NULL; enum action_fail_response on_fail = action_fail_ignore; enum rsc_role_e saved_role = RSC_ROLE_UNKNOWN; if (rsc_id == NULL) { crm_warn("Ignoring malformed " XML_LRM_TAG_RESOURCE " entry without id"); return NULL; } crm_trace("Unpacking " XML_LRM_TAG_RESOURCE " for %s on %s", rsc_id, node->details->uname); // Build a list of individual lrm_rsc_op entries, so we can sort them for (rsc_op = first_named_child(lrm_resource, XML_LRM_TAG_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(data_set->flags, pe_flag_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(data_set, node, rsc_id, lrm_resource); 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, data_set); } } CRM_ASSERT(rsc != NULL); // Check whether the resource is "shutdown-locked" to this node if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { unpack_shutdown_lock(lrm_resource, rsc, node, data_set); } /* process operations */ saved_role = rsc->role; rsc->role = RSC_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; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); if (pcmk__str_eq(task, CRMD_ACTION_MIGRATED, pcmk__str_casei)) { migrate_op = rsc_op; } unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail, data_set); } /* 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, data_set); /* no need to free the contents */ g_list_free(sorted_op_list); process_rsc_state(rsc, node, on_fail, migrate_op, data_set); if (get_target_role(rsc, &req_role)) { if (rsc->next_role == RSC_ROLE_UNKNOWN || req_role < rsc->next_role) { pe__set_next_role(rsc, req_role, XML_RSC_ATTR_TARGET_ROLE); } else if (req_role > rsc->next_role) { pe_rsc_info(rsc, "%s: Not overwriting calculated next role %s" " with requested next role %s", rsc->id, role2text(rsc->next_role), role2text(req_role)); } } if (saved_role > rsc->role) { rsc->role = saved_role; } return rsc; } static void handle_orphaned_container_fillers(xmlNode * lrm_rsc_list, pe_working_set_t * data_set) { xmlNode *rsc_entry = NULL; for (rsc_entry = pcmk__xe_first_child(lrm_rsc_list); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) { pe_resource_t *rsc; pe_resource_t *container; const char *rsc_id; const char *container_id; if (!pcmk__str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, pcmk__str_casei)) { continue; } container_id = crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER); rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID); if (container_id == NULL || rsc_id == NULL) { continue; } container = pe_find_resource(data_set->resources, container_id); if (container == NULL) { continue; } rsc = pe_find_resource(data_set->resources, rsc_id); if (rsc == NULL || !pcmk_is_set(rsc->flags, pe_rsc_orphan_container_filler) || rsc->container != NULL) { continue; } pe_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] node Node whose status is being unpacked * \param[in] xml CIB node state XML * \param[in] data_set Cluster working set */ static void unpack_node_lrm(pe_node_t *node, xmlNode *xml, pe_working_set_t *data_set) { bool found_orphaned_container_filler = false; // Drill down to lrm_resources section xml = find_xml_node(xml, XML_CIB_TAG_LRM, FALSE); if (xml == NULL) { return; } xml = find_xml_node(xml, XML_LRM_TAG_RESOURCES, FALSE); if (xml == NULL) { return; } // Unpack each lrm_resource entry for (xmlNode *rsc_entry = first_named_child(xml, XML_LRM_TAG_RESOURCE); rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) { pe_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, data_set); if ((rsc != NULL) && pcmk_is_set(rsc->flags, pe_rsc_orphan_container_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, data_set); } } static void set_active(pe_resource_t * rsc) { pe_resource_t *top = uber_parent(rsc); if (top && pcmk_is_set(top->flags, pe_rsc_promotable)) { rsc->role = RSC_ROLE_UNPROMOTED; } else { rsc->role = RSC_ROLE_STARTED; } } static void set_node_score(gpointer key, gpointer value, gpointer user_data) { pe_node_t *node = value; int *score = user_data; node->weight = *score; } #define STATUS_PATH_MAX 1024 static xmlNode * find_lrm_op(const char *resource, const char *op, const char *node, const char *source, bool success_only, pe_working_set_t *data_set) { int offset = 0; char xpath[STATUS_PATH_MAX]; xmlNode *xml = NULL; offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "//node_state[@uname='%s']", node); offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "//" XML_LRM_TAG_RESOURCE "[@id='%s']", resource); /* Need to check against transition_magic too? */ if (source && pcmk__str_eq(op, CRMD_ACTION_MIGRATE, pcmk__str_casei)) { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s' and @migrate_target='%s']", op, source); } else if (source && pcmk__str_eq(op, CRMD_ACTION_MIGRATED, pcmk__str_casei)) { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s' and @migrate_source='%s']", op, source); } else { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s']", op); } CRM_LOG_ASSERT(offset > 0); xml = get_xpath_object(xpath, data_set->input, LOG_DEBUG); if (xml && success_only) { int rc = PCMK_OCF_UNKNOWN_ERROR; int status = PCMK_EXEC_ERROR; crm_element_value_int(xml, XML_LRM_ATTR_RC, &rc); crm_element_value_int(xml, XML_LRM_ATTR_OPSTATUS, &status); if ((rc != PCMK_OCF_OK) || (status != PCMK_EXEC_DONE)) { return NULL; } } return xml; } static int pe__call_id(xmlNode *op_xml) { int id = 0; if (op_xml) { crm_element_value_int(op_xml, XML_LRM_ATTR_CALLID, &id); } return id; } /*! * \brief Check whether a stop happened on the same node after some event * * \param[in] rsc Resource being checked * \param[in] node Node being checked * \param[in] xml_op Event that stop is being compared to * \param[in] data_set Cluster working set * * \return TRUE if stop happened after event, FALSE otherwise * * \note This is really unnecessary, but kept as a safety mechanism. We * currently don't save more than one successful event in history, so this * only matters when processing really old CIB files that we don't * technically support anymore, or as preparation for logging an extended * history in the future. */ static bool stop_happened_after(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { xmlNode *stop_op = find_lrm_op(rsc->id, CRMD_ACTION_STOP, node->details->uname, NULL, TRUE, data_set); return (stop_op && (pe__call_id(stop_op) > pe__call_id(xml_op))); } static void unpack_migrate_to_success(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { /* A successful migration sequence is: * migrate_to on source node * migrate_from on target node * stop on source node * * If a migrate_to is followed by a stop, the entire migration (successful * or failed) is complete, and we don't care what happened on the target. * * If no migrate_from has happened, the migration is considered to be * "partial". If the migrate_from failed, make sure the resource gets * stopped on both source and target (if up). * * If the migrate_to and migrate_from both succeeded (which also implies the * resource is no longer running on the source), but there is no stop, the * migration is considered to be "dangling". Schedule a stop on the source * in this case. */ int from_rc = 0; int from_status = 0; pe_node_t *target_node = NULL; pe_node_t *source_node = NULL; xmlNode *migrate_from = NULL; const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); // Sanity check CRM_CHECK(source && target && !strcmp(source, node->details->uname), return); if (stop_happened_after(rsc, node, xml_op, data_set)) { return; } // Clones are not allowed to migrate, so role can't be promoted rsc->role = RSC_ROLE_STARTED; target_node = pe_find_node(data_set->nodes, target); source_node = pe_find_node(data_set->nodes, source); // Check whether there was a migrate_from action on the target migrate_from = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATED, target, source, FALSE, data_set); if (migrate_from) { crm_element_value_int(migrate_from, XML_LRM_ATTR_RC, &from_rc); crm_element_value_int(migrate_from, XML_LRM_ATTR_OPSTATUS, &from_status); pe_rsc_trace(rsc, "%s op on %s exited with status=%d, rc=%d", ID(migrate_from), target, from_status, from_rc); } if (migrate_from && from_rc == PCMK_OCF_OK && (from_status == PCMK_EXEC_DONE)) { /* The migrate_to and migrate_from both succeeded, so mark the migration * as "dangling". This will be used to schedule a stop action on the * source without affecting the target. */ pe_rsc_trace(rsc, "Detected dangling migration op: %s on %s", ID(xml_op), source); rsc->role = RSC_ROLE_STOPPED; rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, node); } else if (migrate_from && (from_status != PCMK_EXEC_PENDING)) { // Failed if (target_node && target_node->details->online) { pe_rsc_trace(rsc, "Marking active on %s %p %d", target, target_node, target_node->details->online); native_add_running(rsc, target_node, data_set, TRUE); } } else { // Pending, or complete but erased if (target_node && target_node->details->online) { pe_rsc_trace(rsc, "Marking active on %s %p %d", target, target_node, target_node->details->online); native_add_running(rsc, target_node, data_set, FALSE); if (source_node && 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. */ rsc->partial_migration_target = target_node; rsc->partial_migration_source = source_node; } } else { /* Consider it failed here - forces a restart, prevents migration */ pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); pe__clear_resource_flags(rsc, pe_rsc_allow_migrate); } } } // Is there an action_name in node_name's rsc history newer than call_id? static bool newer_op(pe_resource_t *rsc, const char *action_name, const char *node_name, int call_id, pe_working_set_t *data_set) { xmlNode *action = find_lrm_op(rsc->id, action_name, node_name, NULL, TRUE, data_set); return pe__call_id(action) > call_id; } static void unpack_migrate_to_failure(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { int target_stop_id = 0; int target_migrate_from_id = 0; xmlNode *target_stop = NULL; xmlNode *target_migrate_from = NULL; const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); // Sanity check CRM_CHECK(source && target && !strcmp(source, node->details->uname), 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. */ rsc->role = RSC_ROLE_STARTED; // Check for stop on the target target_stop = find_lrm_op(rsc->id, CRMD_ACTION_STOP, target, NULL, TRUE, data_set); target_stop_id = pe__call_id(target_stop); // Check for migrate_from on the target target_migrate_from = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATED, target, source, TRUE, data_set); target_migrate_from_id = pe__call_id(target_migrate_from); if ((target_stop == NULL) || (target_stop_id < target_migrate_from_id)) { /* There was no stop on the target, or a stop that happened before a * migrate_from, so assume the resource is still active on the target * (if it is up). */ pe_node_t *target_node = pe_find_node(data_set->nodes, target); pe_rsc_trace(rsc, "stop (%d) + migrate_from (%d)", target_stop_id, target_migrate_from_id); if (target_node && target_node->details->online) { native_add_running(rsc, target_node, data_set, FALSE); } } else if (target_migrate_from == NULL) { /* We know there was a stop on the target, but there may not have been a * migrate_from (the stop could have happened before migrate_from was * scheduled or attempted). * * That means this could be a "dangling" migration. But first, check * whether there is a newer successful stop, start, or migrate_from on * the source node -- it's possible the failed migration was followed by * a successful stop, full restart, or migration in the reverse * direction, in which case we don't want to force a stop. */ int source_migrate_to_id = pe__call_id(xml_op); if (newer_op(rsc, CRMD_ACTION_MIGRATED, source, source_migrate_to_id, data_set) || newer_op(rsc, CRMD_ACTION_START, source, source_migrate_to_id, data_set) || newer_op(rsc, CRMD_ACTION_STOP, source, source_migrate_to_id, data_set)) { return; } // Mark node as having dangling migration so we can force a stop later rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, node); } } static void unpack_migrate_from_failure(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { xmlNode *source_stop = NULL; xmlNode *source_migrate_to = NULL; const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); // Sanity check CRM_CHECK(source && target && !strcmp(target, node->details->uname), 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. */ rsc->role = RSC_ROLE_STARTED; // Check for a stop on the source source_stop = find_lrm_op(rsc->id, CRMD_ACTION_STOP, source, NULL, TRUE, data_set); // Check for a migrate_to on the source source_migrate_to = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATE, source, target, TRUE, data_set); if ((source_stop == NULL) || (pe__call_id(source_stop) < pe__call_id(source_migrate_to))) { /* There was no stop on the source, or a stop that happened before * migrate_to, so assume the resource is still active on the source (if * it is up). */ pe_node_t *source_node = pe_find_node(data_set->nodes, source); if (source_node && source_node->details->online) { native_add_running(rsc, source_node, data_set, TRUE); } } } static void record_failed_op(xmlNode *op, const pe_node_t *node, const pe_resource_t *rsc, pe_working_set_t *data_set) { xmlNode *xIter = NULL; const char *op_key = crm_element_value(op, XML_LRM_ATTR_TASK_KEY); if (node->details->online == FALSE) { return; } for (xIter = data_set->failed->children; xIter; xIter = xIter->next) { const char *key = crm_element_value(xIter, XML_LRM_ATTR_TASK_KEY); const char *uname = crm_element_value(xIter, XML_ATTR_UNAME); if(pcmk__str_eq(op_key, key, pcmk__str_casei) && pcmk__str_eq(uname, node->details->uname, pcmk__str_casei)) { crm_trace("Skipping duplicate entry %s on %s", op_key, node->details->uname); return; } } crm_trace("Adding entry %s on %s", op_key, node->details->uname); crm_xml_add(op, XML_ATTR_UNAME, node->details->uname); crm_xml_add(op, XML_LRM_ATTR_RSCID, rsc->id); add_node_copy(data_set->failed, op); } static const char *get_op_key(xmlNode *xml_op) { const char *key = crm_element_value(xml_op, XML_LRM_ATTR_TASK_KEY); if(key == NULL) { key = ID(xml_op); } return key; } static const char * last_change_str(xmlNode *xml_op) { time_t when; const char *when_s = NULL; if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE, &when) == pcmk_ok) { when_s = pcmk__epoch2str(&when); if (when_s) { // Skip day of week to make message shorter when_s = strchr(when_s, ' '); if (when_s) { ++when_s; } } } return ((when_s && *when_s)? when_s : "unknown time"); } /*! * \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 action_fail_demote: switch (second) { case action_fail_ignore: return 1; case action_fail_demote: return 0; default: return -1; } break; case action_fail_reset_remote: switch (second) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: return 1; case action_fail_reset_remote: return 0; default: return -1; } break; case action_fail_restart_container: switch (second) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: case action_fail_reset_remote: return 1; case action_fail_restart_container: return 0; default: return -1; } break; default: break; } switch (second) { case action_fail_demote: return (first == action_fail_ignore)? -1 : 1; case action_fail_reset_remote: switch (first) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: return -1; default: return 1; } break; case action_fail_restart_container: switch (first) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: case action_fail_reset_remote: return -1; default: return 1; } break; default: break; } return first - second; } static void unpack_rsc_op_failure(pe_resource_t * rsc, pe_node_t * node, int rc, xmlNode * xml_op, xmlNode ** last_failure, enum action_fail_response * on_fail, pe_working_set_t * data_set) { guint interval_ms = 0; bool is_probe = false; pe_action_t *action = NULL; const char *key = get_op_key(xml_op); const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON); CRM_ASSERT(rsc); CRM_CHECK(task != NULL, return); *last_failure = xml_op; crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((interval_ms == 0) && !strcmp(task, CRMD_ACTION_STATUS)) { is_probe = true; } if (exit_reason == NULL) { exit_reason = ""; } if (!pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster) && (rc == PCMK_OCF_NOT_INSTALLED)) { crm_trace("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " CRM_XS " rc=%d id=%s", services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, (is_probe? "probe" : task), rsc->id, node->details->uname, last_change_str(xml_op), rc, ID(xml_op)); } else { crm_warn("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " CRM_XS " rc=%d id=%s", services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, (is_probe? "probe" : task), rsc->id, node->details->uname, last_change_str(xml_op), rc, ID(xml_op)); if (is_probe && (rc != PCMK_OCF_OK) && (rc != PCMK_OCF_NOT_RUNNING) && (rc != 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 resource-discovery option for location constraints", rsc->id, node->details->uname); } record_failed_op(xml_op, node, rsc, data_set); } action = custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set); if (cmp_on_fail(*on_fail, action->on_fail) < 0) { pe_rsc_trace(rsc, "on-fail %s -> %s for %s (%s)", fail2text(*on_fail), fail2text(action->on_fail), action->uuid, key); *on_fail = action->on_fail; } if (!strcmp(task, CRMD_ACTION_STOP)) { resource_location(rsc, node, -INFINITY, "__stop_fail__", data_set); } else if (!strcmp(task, CRMD_ACTION_MIGRATE)) { unpack_migrate_to_failure(rsc, node, xml_op, data_set); } else if (!strcmp(task, CRMD_ACTION_MIGRATED)) { unpack_migrate_from_failure(rsc, node, xml_op, data_set); } else if (!strcmp(task, CRMD_ACTION_PROMOTE)) { rsc->role = RSC_ROLE_PROMOTED; } else if (!strcmp(task, CRMD_ACTION_DEMOTE)) { if (action->on_fail == action_fail_block) { rsc->role = RSC_ROLE_PROMOTED; pe__set_next_role(rsc, RSC_ROLE_STOPPED, "demote with on-fail=block"); } else if(rc == PCMK_OCF_NOT_RUNNING) { rsc->role = RSC_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. */ rsc->role = RSC_ROLE_UNPROMOTED; } } if(is_probe && rc == PCMK_OCF_NOT_INSTALLED) { /* leave stopped */ pe_rsc_trace(rsc, "Leaving %s stopped", rsc->id); rsc->role = RSC_ROLE_STOPPED; } else if (rsc->role < RSC_ROLE_STARTED) { pe_rsc_trace(rsc, "Setting %s active", rsc->id); set_active(rsc); } pe_rsc_trace(rsc, "Resource %s: role=%s, unclean=%s, on_fail=%s, fail_role=%s", rsc->id, role2text(rsc->role), pcmk__btoa(node->details->unclean), fail2text(action->on_fail), role2text(action->fail_role)); if (action->fail_role != RSC_ROLE_STARTED && rsc->next_role < action->fail_role) { pe__set_next_role(rsc, action->fail_role, "failure"); } if (action->fail_role == RSC_ROLE_STOPPED) { int score = -INFINITY; pe_resource_t *fail_rsc = rsc; if (fail_rsc->parent) { pe_resource_t *parent = uber_parent(fail_rsc); if (pe_rsc_is_clone(parent) && !pcmk_is_set(parent->flags, pe_rsc_unique)) { /* For clone resources, if a child fails on an operation * with on-fail = stop, all the resources fail. Do this by preventing * the parent from coming up again. */ fail_rsc = parent; } } crm_notice("%s will not be started under current conditions", fail_rsc->id); /* make sure it doesn't come up again */ if (fail_rsc->allowed_nodes != NULL) { g_hash_table_destroy(fail_rsc->allowed_nodes); } fail_rsc->allowed_nodes = pe__node_list2table(data_set->nodes); g_hash_table_foreach(fail_rsc->allowed_nodes, set_node_score, &score); } pe_free_action(action); } /*! * \internal * \brief Remap operation status based on action result * * Given an action result, determine an appropriate operation 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] rsc Resource that operation history entry is for * \param[in] rc Actual return code of operation * \param[in] target_rc Expected return code of operation * \param[in] node Node where operation was executed * \param[in] xml_op Operation history entry XML from CIB status * \param[in,out] on_fail What should be done about the result * \param[in] data_set Current cluster working set * * \return Operation status based on return code and action info * \note This may update the resource's current and next role. */ static int determine_op_status( pe_resource_t *rsc, int rc, int target_rc, pe_node_t * node, xmlNode * xml_op, enum action_fail_response * on_fail, pe_working_set_t * data_set) { guint interval_ms = 0; bool is_probe = false; int result = PCMK_EXEC_DONE; const char *key = get_op_key(xml_op); const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON); CRM_ASSERT(rsc); CRM_CHECK(task != NULL, return PCMK_EXEC_ERROR); if (exit_reason == NULL) { exit_reason = ""; } crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((interval_ms == 0) && !strcmp(task, CRMD_ACTION_STATUS)) { is_probe = true; task = "probe"; } if (target_rc < 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 * target_rc 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. */ result = PCMK_EXEC_ERROR; crm_warn("Expected result not found for %s on %s (corrupt or obsolete CIB?)", key, node->details->uname); } else if (target_rc != rc) { result = PCMK_EXEC_ERROR; pe_rsc_debug(rsc, "%s on %s: expected %d (%s), got %d (%s%s%s)", key, node->details->uname, target_rc, services_ocf_exitcode_str(target_rc), rc, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason); } switch (rc) { case PCMK_OCF_OK: if (is_probe && (target_rc == PCMK_OCF_NOT_RUNNING)) { result = PCMK_EXEC_DONE; pe_rsc_info(rsc, "Probe found %s active on %s at %s", rsc->id, node->details->uname, last_change_str(xml_op)); } break; case PCMK_OCF_NOT_RUNNING: if (is_probe || (target_rc == rc) || !pcmk_is_set(rsc->flags, pe_rsc_managed)) { result = PCMK_EXEC_DONE; rsc->role = RSC_ROLE_STOPPED; /* clear any previous failure actions */ *on_fail = action_fail_ignore; pe__set_next_role(rsc, RSC_ROLE_UNKNOWN, "not running"); } break; case PCMK_OCF_RUNNING_PROMOTED: if (is_probe && (rc != target_rc)) { result = PCMK_EXEC_DONE; pe_rsc_info(rsc, "Probe found %s active and promoted on %s at %s", rsc->id, node->details->uname, last_change_str(xml_op)); } rsc->role = RSC_ROLE_PROMOTED; break; case PCMK_OCF_DEGRADED_PROMOTED: case PCMK_OCF_FAILED_PROMOTED: rsc->role = RSC_ROLE_PROMOTED; result = PCMK_EXEC_ERROR; break; case PCMK_OCF_NOT_CONFIGURED: result = PCMK_EXEC_ERROR_FATAL; break; case PCMK_OCF_UNIMPLEMENT_FEATURE: if (interval_ms > 0) { result = PCMK_EXEC_NOT_SUPPORTED; break; } // fall through case PCMK_OCF_NOT_INSTALLED: case PCMK_OCF_INVALID_PARAM: case PCMK_OCF_INSUFFICIENT_PRIV: if (!pe_can_fence(data_set, node) && !strcmp(task, CRMD_ACTION_STOP)) { /* If a stop fails and we can't fence, there's nothing else we can do */ pe_proc_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", rsc->id, task, node->details->uname, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, last_change_str(xml_op), rc, ID(xml_op)); pe__clear_resource_flags(rsc, pe_rsc_managed); pe__set_resource_flags(rsc, pe_rsc_block); } result = PCMK_EXEC_ERROR_HARD; break; default: if (result == PCMK_EXEC_DONE) { crm_info("Treating unknown exit status %d from %s of %s " "on %s at %s as failure", rc, task, rsc->id, node->details->uname, last_change_str(xml_op)); result = PCMK_EXEC_ERROR; } break; } return result; } // return TRUE if start or monitor last failure but parameters changed static bool should_clear_for_param_change(xmlNode *xml_op, const char *task, pe_resource_t *rsc, pe_node_t *node, pe_working_set_t *data_set) { if (!strcmp(task, "start") || !strcmp(task, "monitor")) { if (pe__bundle_needs_remote_name(rsc, data_set)) { /* 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, pe_check_last_failure, data_set); } else { op_digest_cache_t *digest_data = NULL; digest_data = rsc_action_digest_cmp(rsc, xml_op, node, data_set); switch (digest_data->rc) { case RSC_DIGEST_UNKNOWN: crm_trace("Resource %s history entry %s on %s" " has no digest to compare", rsc->id, get_op_key(xml_op), node->details->id); break; case RSC_DIGEST_MATCH: break; default: return TRUE; } } } return FALSE; } // Order action after fencing of remote node, given connection rsc static void order_after_remote_fencing(pe_action_t *action, pe_resource_t *remote_conn, pe_working_set_t *data_set) { pe_node_t *remote_node = pe_find_node(data_set->nodes, remote_conn->id); if (remote_node) { pe_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL, FALSE, data_set); order_actions(fence, action, pe_order_implies_then); } } static bool should_ignore_failure_timeout(pe_resource_t *rsc, xmlNode *xml_op, const char *task, guint interval_ms, bool is_last_failure, pe_working_set_t *data_set) { /* 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(data_set->flags, pe_flag_stonith_enabled) && (interval_ms != 0) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { pe_node_t *remote_node = pe_find_node(data_set->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] rsc Resource that operation happened to * \param[in] node Node that operation happened on * \param[in] rc Actual result of operation * \param[in] xml_op Operation history entry XML * \param[in] data_set Current working set * * \return TRUE if operation history entry is expired, FALSE otherwise */ static bool check_operation_expiry(pe_resource_t *rsc, pe_node_t *node, int rc, xmlNode *xml_op, pe_working_set_t *data_set) { bool expired = FALSE; bool is_last_failure = pcmk__ends_with(ID(xml_op), "_last_failure_0"); time_t last_run = 0; guint interval_ms = 0; int unexpired_fail_count = 0; const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *clear_reason = NULL; crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((rsc->failure_timeout > 0) && (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE, &last_run) == 0)) { // Resource has a failure-timeout, and history entry has a timestamp time_t now = get_effective_time(data_set); time_t last_failure = 0; // Is this particular operation history older than the failure timeout? if ((now >= (last_run + rsc->failure_timeout)) && !should_ignore_failure_timeout(rsc, xml_op, task, interval_ms, is_last_failure, data_set)) { expired = TRUE; } // Does the resource as a whole have an unexpired fail count? unexpired_fail_count = pe_get_failcount(node, rsc, &last_failure, pe_fc_effective, xml_op, data_set); // Update scheduler recheck time according to *last* failure crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d timeout=%ds" " last-failure@%lld", ID(xml_op), (long long) last_run, (expired? "" : "not "), (long long) now, unexpired_fail_count, rsc->failure_timeout, (long long) last_failure); last_failure += rsc->failure_timeout + 1; if (unexpired_fail_count && (now < last_failure)) { pe__update_recheck_time(last_failure, data_set); } } if (expired) { if (pe_get_failcount(node, rsc, NULL, pe_fc_default, xml_op, data_set)) { // 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. */ expired = FALSE; } } else if (is_last_failure && rsc->remote_reconnect_ms) { /* 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(xml_op, task, rsc, node, data_set)) { clear_reason = "resource parameters have changed"; } if (clear_reason != NULL) { // Schedule clearing of the fail count pe_action_t *clear_op = pe__clear_failcount(rsc, node, clear_reason, data_set); if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled) && rsc->remote_reconnect_ms) { /* 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", task, rsc->id); order_after_remote_fencing(clear_op, rsc, data_set); } } if (expired && (interval_ms == 0) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { switch(rc) { 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 expired = FALSE; break; } } return expired; } int pe__target_rc_from_xml(xmlNode *xml_op) { int target_rc = 0; const char *key = crm_element_value(xml_op, XML_ATTR_TRANSITION_KEY); if (key == NULL) { return -1; } decode_transition_key(key, NULL, NULL, NULL, &target_rc); return target_rc; } static enum action_fail_response get_action_on_fail(pe_resource_t *rsc, const char *key, const char *task, pe_working_set_t * data_set) { enum action_fail_response result = action_fail_recover; pe_action_t *action = custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set); result = action->on_fail; pe_free_action(action); return result; } static void update_resource_state(pe_resource_t * rsc, pe_node_t * node, xmlNode * xml_op, const char * task, int rc, xmlNode * last_failure, enum action_fail_response * on_fail, pe_working_set_t * data_set) { gboolean clear_past_failure = FALSE; CRM_ASSERT(rsc); CRM_ASSERT(xml_op); if (rc == PCMK_OCF_NOT_RUNNING) { clear_past_failure = TRUE; } else if (rc == PCMK_OCF_NOT_INSTALLED) { rsc->role = RSC_ROLE_STOPPED; } else if (pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { if (last_failure) { const char *op_key = get_op_key(xml_op); const char *last_failure_key = get_op_key(last_failure); if (pcmk__str_eq(op_key, last_failure_key, pcmk__str_casei)) { clear_past_failure = TRUE; } } if (rsc->role < RSC_ROLE_STARTED) { set_active(rsc); } } else if (pcmk__str_eq(task, CRMD_ACTION_START, pcmk__str_casei)) { rsc->role = RSC_ROLE_STARTED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_STOP, pcmk__str_casei)) { rsc->role = RSC_ROLE_STOPPED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_PROMOTE, pcmk__str_casei)) { rsc->role = RSC_ROLE_PROMOTED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_DEMOTE, pcmk__str_casei)) { if (*on_fail == action_fail_demote) { // Demote clears an error only if on-fail=demote clear_past_failure = TRUE; } rsc->role = RSC_ROLE_UNPROMOTED; } else if (pcmk__str_eq(task, CRMD_ACTION_MIGRATED, pcmk__str_casei)) { rsc->role = RSC_ROLE_STARTED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_MIGRATE, pcmk__str_casei)) { unpack_migrate_to_success(rsc, node, xml_op, data_set); } else if (rsc->role < RSC_ROLE_STARTED) { pe_rsc_trace(rsc, "%s active on %s", rsc->id, node->details->uname); set_active(rsc); } /* clear any previous failure actions */ if (clear_past_failure) { switch (*on_fail) { case action_fail_stop: case action_fail_fence: case action_fail_migrate: case action_fail_standby: pe_rsc_trace(rsc, "%s.%s is not cleared by a completed stop", rsc->id, fail2text(*on_fail)); break; case action_fail_block: case action_fail_ignore: case action_fail_demote: case action_fail_recover: case action_fail_restart_container: *on_fail = action_fail_ignore; pe__set_next_role(rsc, RSC_ROLE_UNKNOWN, "clear past failures"); break; case action_fail_reset_remote: if (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 = action_fail_ignore; pe__set_next_role(rsc, RSC_ROLE_UNKNOWN, "clear past failures and reset remote"); } break; } } } /*! * \internal * \brief Remap informational monitor results to usual values * * Certain OCF result 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. * * \param[in] rc Actual result of a monitor action * \param[in] xml_op Operation history XML * \param[in] node Node that operation happened on * \param[in] rsc Resource that operation happened to * \param[in] data_set Cluster working set * * \return Result code that pacemaker should use * * \note If the result is remapped, and the node is not shutting down or failed, * the operation will be recorded in the data set's list of failed * operations, to highlight it for the user. */ static int remap_monitor_rc(int rc, xmlNode *xml_op, const pe_node_t *node, const pe_resource_t *rsc, pe_working_set_t *data_set) { int remapped_rc = pcmk__effective_rc(rc); if (rc != remapped_rc) { crm_trace("Remapping monitor result %d to %d", rc, remapped_rc); if (!node->details->shutdown || node->details->online) { record_failed_op(xml_op, node, rsc, data_set); } } return remapped_rc; } static void unpack_rsc_op(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *on_fail, pe_working_set_t *data_set) { int rc = 0; int task_id = 0; int target_rc = 0; int status = PCMK_EXEC_UNKNOWN; guint interval_ms = 0; const char *task = NULL; const char *task_key = NULL; const char *exit_reason = NULL; bool expired = FALSE; pe_resource_t *parent = rsc; enum action_fail_response failure_strategy = action_fail_recover; CRM_CHECK(rsc && node && xml_op, return); target_rc = pe__target_rc_from_xml(xml_op); task_key = get_op_key(xml_op); task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON); if (exit_reason == NULL) { exit_reason = ""; } crm_element_value_int(xml_op, XML_LRM_ATTR_RC, &rc); crm_element_value_int(xml_op, XML_LRM_ATTR_CALLID, &task_id); crm_element_value_int(xml_op, XML_LRM_ATTR_OPSTATUS, &status); crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); CRM_CHECK(task != NULL, return); CRM_CHECK((status >= PCMK_EXEC_PENDING) && (status <= PCMK_EXEC_MAX), return); if (!strcmp(task, CRMD_ACTION_NOTIFY) || !strcmp(task, CRMD_ACTION_METADATA)) { /* safe to ignore these */ return; } if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) { parent = uber_parent(rsc); } pe_rsc_trace(rsc, "Unpacking task %s/%s (call_id=%d, status=%d, rc=%d) on %s (role=%s)", task_key, task, task_id, status, rc, node->details->uname, role2text(rsc->role)); if (node->details->unclean) { pe_rsc_trace(rsc, "Node %s (where %s is running) is unclean." " Further action depends on the value of the stop's on-fail attribute", node->details->uname, rsc->id); } /* It should be possible to call remap_monitor_rc() first then call * check_operation_expiry() only if rc != target_rc, because there should * never be a fail count without at least one unexpected result in the * resource history. That would be more efficient by avoiding having to call * check_operation_expiry() for expected results. * * However, we do have such configurations in the scheduler regression * tests, even if it shouldn't be possible with the current code. It's * probably a good idea anyway, but that would require updating the test * inputs to something currently possible. */ if ((status != PCMK_EXEC_NOT_INSTALLED) && check_operation_expiry(rsc, node, rc, xml_op, data_set)) { expired = TRUE; } if (!strcmp(task, CRMD_ACTION_STATUS)) { rc = remap_monitor_rc(rc, xml_op, node, rsc, data_set); } if (expired && (rc != target_rc)) { const char *magic = crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC); if (interval_ms == 0) { crm_notice("Ignoring expired %s failure on %s " CRM_XS " actual=%d expected=%d magic=%s", task_key, node->details->uname, rc, target_rc, magic); goto done; } else if(node->details->online && node->details->unclean == FALSE) { /* Reschedule the recurring monitor. CancelXmlOp() won't work at * this stage, so as a hacky workaround, forcibly change the restart * digest so check_action_definition() 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 an 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 after failure expired on %s " CRM_XS " actual=%d expected=%d magic=%s", task_key, node->details->uname, rc, target_rc, magic); crm_xml_add(xml_op, XML_LRM_ATTR_RESTART_DIGEST, "calculated-failure-timeout"); goto done; } } /* If the executor reported an operation 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. */ if(status == PCMK_EXEC_DONE || status == PCMK_EXEC_ERROR) { status = determine_op_status(rsc, rc, target_rc, node, xml_op, on_fail, data_set); pe_rsc_trace(rsc, "Remapped %s status to %d", task_key, status); } switch (status) { case PCMK_EXEC_CANCELLED: // Should never happen pe_err("Resource history contains cancellation '%s' " "(%s of %s on %s at %s)", ID(xml_op), task, rsc->id, node->details->uname, last_change_str(xml_op)); goto done; case PCMK_EXEC_PENDING: if (!strcmp(task, CRMD_ACTION_START)) { pe__set_resource_flags(rsc, pe_rsc_start_pending); set_active(rsc); } else if (!strcmp(task, CRMD_ACTION_PROMOTE)) { rsc->role = RSC_ROLE_PROMOTED; } else if (!strcmp(task, CRMD_ACTION_MIGRATE) && node->details->unclean) { /* If a pending migrate_to action is out on a unclean node, * we have to force the stop action on the target. */ const char *migrate_target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); pe_node_t *target = pe_find_node(data_set->nodes, migrate_target); if (target) { stop_action(rsc, target, FALSE); } } if (rsc->pending_task == NULL) { if ((interval_ms != 0) || strcmp(task, CRMD_ACTION_STATUS)) { rsc->pending_task = strdup(task); rsc->pending_node = node; } else { /* Pending probes are not printed, even if pending * operations are requested. If someone ever requests that * behavior, enable the below and the corresponding part of * native.c:native_pending_task(). */ #if 0 rsc->pending_task = strdup("probe"); rsc->pending_node = node; #endif } } goto done; case PCMK_EXEC_DONE: pe_rsc_trace(rsc, "%s of %s on %s completed at %s " CRM_XS " id=%s", task, rsc->id, node->details->uname, last_change_str(xml_op), ID(xml_op)); update_resource_state(rsc, node, xml_op, task, rc, *last_failure, on_fail, data_set); goto done; case PCMK_EXEC_NOT_INSTALLED: failure_strategy = get_action_on_fail(rsc, task_key, task, data_set); if (failure_strategy == action_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", task, rsc->id, node->details->uname, status, rc, ID(xml_op)); /* Also for printing it as "FAILED" by marking it as pe_rsc_failed later */ *on_fail = action_fail_migrate; } resource_location(parent, node, -INFINITY, "hard-error", data_set); unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set); goto done; case PCMK_EXEC_NOT_CONNECTED: if (pe__is_guest_or_remote_node(node) && pcmk_is_set(node->details->remote_rsc->flags, pe_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. */ pe__set_resource_flags(node->details->remote_rsc, pe_rsc_failed|pe_rsc_stop); } 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 /* These should only be possible in fence action results, not operation * history, but have some handling in place as a fail-safe. */ case PCMK_EXEC_NO_FENCE_DEVICE: case PCMK_EXEC_NO_SECRETS: status = PCMK_EXEC_ERROR_HARD; break; // Not done, do error handling } failure_strategy = get_action_on_fail(rsc, task_key, task, data_set); if ((failure_strategy == action_fail_ignore) || (failure_strategy == action_fail_restart_container && !strcmp(task, CRMD_ACTION_STOP))) { crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s " "succeeded " CRM_XS " rc=%d id=%s", task, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, rsc->id, node->details->uname, last_change_str(xml_op), rc, ID(xml_op)); update_resource_state(rsc, node, xml_op, task, target_rc, *last_failure, on_fail, data_set); crm_xml_add(xml_op, XML_ATTR_UNAME, node->details->uname); pe__set_resource_flags(rsc, pe_rsc_failure_ignored); record_failed_op(xml_op, node, rsc, data_set); if ((failure_strategy == action_fail_restart_container) && cmp_on_fail(*on_fail, action_fail_recover) <= 0) { *on_fail = failure_strategy; } } else { unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set); if (status == PCMK_EXEC_ERROR_HARD) { do_crm_log(rc != PCMK_OCF_NOT_INSTALLED?LOG_ERR:LOG_NOTICE, "Preventing %s from restarting on %s because " "of hard failure (%s%s%s)" CRM_XS " rc=%d id=%s", parent->id, node->details->uname, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, rc, ID(xml_op)); resource_location(parent, node, -INFINITY, "hard-error", data_set); } else if (status == PCMK_EXEC_ERROR_FATAL) { crm_err("Preventing %s from restarting anywhere because " "of fatal failure (%s%s%s) " CRM_XS " rc=%d id=%s", parent->id, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, rc, ID(xml_op)); resource_location(parent, NULL, -INFINITY, "fatal-error", data_set); } } done: pe_rsc_trace(rsc, "Resource %s after %s: role=%s, next=%s", rsc->id, task, role2text(rsc->role), role2text(rsc->next_role)); } static void add_node_attrs(xmlNode *xml_obj, pe_node_t *node, bool overwrite, pe_working_set_t *data_set) { const char *cluster_name = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_UNAME), strdup(node->details->uname)); g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_ID), strdup(node->details->id)); if (pcmk__str_eq(node->details->id, data_set->dc_uuid, pcmk__str_casei)) { data_set->dc_node = node; node->details->is_dc = TRUE; g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_TRUE)); } else { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_FALSE)); } cluster_name = g_hash_table_lookup(data_set->config_hash, "cluster-name"); if (cluster_name) { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_CLUSTER_NAME), strdup(cluster_name)); } pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_ATTR_SETS, &rule_data, node->details->attrs, NULL, overwrite, data_set); if (pe_node_attribute_raw(node, CRM_ATTR_SITE_NAME) == NULL) { const char *site_name = pe_node_attribute_raw(node, "site-name"); if (site_name) { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_SITE_NAME), strdup(site_name)); } else if (cluster_name) { /* Default to cluster-name if unset */ g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_SITE_NAME), strdup(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__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, pcmk__str_none)) { crm_xml_add(rsc_op, "resource", rsc); crm_xml_add(rsc_op, XML_ATTR_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", ID(rsc_entry)); break; } else if (counter < start_index) { crm_trace("Skipping %s: old", 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, pe_working_set_t * data_set) { GList *output = NULL; GList *intermediate = NULL; xmlNode *tmp = NULL; xmlNode *status = find_xml_node(data_set->input, XML_CIB_TAG_STATUS, TRUE); pe_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__str_eq((const char *)node_state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { const char *uname = crm_element_value(node_state, XML_ATTR_UNAME); if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) { continue; } this_node = pe_find_node(data_set->nodes, uname); if(this_node == NULL) { CRM_LOG_ASSERT(this_node != NULL); continue; } else if (pe__is_guest_or_remote_node(this_node)) { determine_remote_online_status(data_set, this_node); } else { determine_online_status(node_state, this_node, data_set); } if (this_node->details->online || pcmk_is_set(data_set->flags, pe_flag_stonith_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, XML_CIB_TAG_LRM, FALSE); tmp = find_xml_node(tmp, XML_LRM_TAG_RESOURCES, FALSE); for (lrm_rsc = pcmk__xe_first_child(tmp); lrm_rsc != NULL; lrm_rsc = pcmk__xe_next(lrm_rsc)) { if (pcmk__str_eq((const char *)lrm_rsc->name, XML_LRM_TAG_RESOURCE, pcmk__str_none)) { const char *rsc_id = crm_element_value(lrm_rsc, XML_ATTR_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; }