diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c index a24bf1a217..f6b623ee83 100644 --- a/daemons/controld/controld_execd.c +++ b/daemons/controld/controld_execd.c @@ -1,2446 +1,2446 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include // lrmd_event_data_t, lrmd_rsc_info_t, etc. #include #include #include #include #include #include #include #define START_DELAY_THRESHOLD 5 * 60 * 1000 #define MAX_LRM_REG_FAILS 30 struct delete_event_s { int rc; const char *rsc; lrm_state_t *lrm_state; }; static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id); static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list); static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data); static lrmd_event_data_t *construct_op(const lrm_state_t *lrm_state, const xmlNode *rsc_op, const char *rsc_id, const char *operation); static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, xmlNode *msg, struct ra_metadata_s *md); static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level); static void lrm_connection_destroy(void) { if (pcmk_is_set(controld_globals.fsa_input_register, R_LRM_CONNECTED)) { crm_crit("Lost connection to local executor"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_LRM_CONNECTED); } } static char * make_stop_id(const char *rsc, int call_id) { return crm_strdup_printf("%s:%d", rsc, call_id); } static void copy_instance_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") == NULL) { g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value)); } } static void copy_meta_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") != NULL) { g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value)); } } /*! * \internal * \brief Remove a recurring operation from a resource's history * * \param[in,out] history Resource history to modify * \param[in] op Operation to remove * * \return TRUE if the operation was found and removed, FALSE otherwise */ static gboolean history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_event_data_t *existing = iter->data; if ((op->interval_ms == existing->interval_ms) && pcmk__str_eq(op->rsc_id, existing->rsc_id, pcmk__str_none) && pcmk__str_eq(op->op_type, existing->op_type, pcmk__str_casei)) { history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter); lrmd_free_event(existing); return TRUE; } } return FALSE; } /*! * \internal * \brief Free all recurring operations in resource history * * \param[in,out] history Resource history to modify */ static void history_free_recurring_ops(rsc_history_t *history) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_free_event(iter->data); } g_list_free(history->recurring_op_list); history->recurring_op_list = NULL; } /*! * \internal * \brief Free resource history * * \param[in,out] history Resource history to free */ void history_free(gpointer data) { rsc_history_t *history = (rsc_history_t*)data; if (history->stop_params) { g_hash_table_destroy(history->stop_params); } /* Don't need to free history->rsc.id because it's set to history->id */ free(history->rsc.type); free(history->rsc.standard); free(history->rsc.provider); lrmd_free_event(history->failed); lrmd_free_event(history->last); free(history->id); history_free_recurring_ops(history); free(history); } static void update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op) { int target_rc = 0; rsc_history_t *entry = NULL; if (op->rsc_deleted) { crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type); controld_delete_resource_history(op->rsc_id, lrm_state->node_name, NULL, crmd_cib_smart_opt()); return; } if (pcmk__str_eq(op->op_type, PCMK_ACTION_NOTIFY, pcmk__str_casei)) { return; } crm_debug("Updating history for '%s' with %s op", op->rsc_id, op->op_type); entry = g_hash_table_lookup(lrm_state->resource_history, op->rsc_id); if (entry == NULL && rsc) { entry = calloc(1, sizeof(rsc_history_t)); entry->id = strdup(op->rsc_id); g_hash_table_insert(lrm_state->resource_history, entry->id, entry); entry->rsc.id = entry->id; entry->rsc.type = strdup(rsc->type); entry->rsc.standard = strdup(rsc->standard); pcmk__str_update(&entry->rsc.provider, rsc->provider); } else if (entry == NULL) { crm_info("Resource %s no longer exists, not updating cache", op->rsc_id); return; } entry->last_callid = op->call_id; target_rc = rsc_op_expected_rc(op); if (op->op_status == PCMK_EXEC_CANCELLED) { if (op->interval_ms > 0) { crm_trace("Removing cancelled recurring op: " PCMK__OP_FMT, op->rsc_id, op->op_type, op->interval_ms); history_remove_recurring_op(entry, op); return; } else { crm_trace("Skipping " PCMK__OP_FMT " rc=%d, status=%d", op->rsc_id, op->op_type, op->interval_ms, op->rc, op->op_status); } } else if (did_rsc_op_fail(op, target_rc)) { /* Store failed monitors here, otherwise the block below will cause them * to be forgotten when a stop happens. */ if (entry->failed) { lrmd_free_event(entry->failed); } entry->failed = lrmd_copy_event(op); } else if (op->interval_ms == 0) { if (entry->last) { lrmd_free_event(entry->last); } entry->last = lrmd_copy_event(op); if (op->params && pcmk__strcase_any_of(op->op_type, PCMK_ACTION_START, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, PCMK_ACTION_MONITOR, NULL)) { if (entry->stop_params) { g_hash_table_destroy(entry->stop_params); } entry->stop_params = pcmk__strkey_table(free, free); g_hash_table_foreach(op->params, copy_instance_keys, entry->stop_params); } } if (op->interval_ms > 0) { /* Ensure there are no duplicates */ history_remove_recurring_op(entry, op); crm_trace("Adding recurring op: " PCMK__OP_FMT, op->rsc_id, op->op_type, op->interval_ms); entry->recurring_op_list = g_list_prepend(entry->recurring_op_list, lrmd_copy_event(op)); } else if ((entry->recurring_op_list != NULL) && !pcmk__str_eq(op->op_type, PCMK_ACTION_MONITOR, pcmk__str_casei)) { crm_trace("Dropping %d recurring ops because of: " PCMK__OP_FMT, g_list_length(entry->recurring_op_list), op->rsc_id, op->op_type, op->interval_ms); history_free_recurring_ops(entry); } } /*! * \internal * \brief Send a direct OK ack for a resource task * * \param[in] lrm_state LRM connection * \param[in] input Input message being ack'ed * \param[in] rsc_id ID of affected resource * \param[in] rsc Affected resource (if available) * \param[in] task Operation task being ack'ed * \param[in] ack_host Name of host to send ack to * \param[in] ack_sys IPC system name to ack */ static void send_task_ok_ack(const lrm_state_t *lrm_state, const ha_msg_input_t *input, const char *rsc_id, const lrmd_rsc_info_t *rsc, const char *task, const char *ack_host, const char *ack_sys) { lrmd_event_data_t *op = construct_op(lrm_state, input->xml, rsc_id, task); lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); controld_ack_event_directly(ack_host, ack_sys, rsc, op, rsc_id); lrmd_free_event(op); } static inline const char * op_node_name(lrmd_event_data_t *op) { return pcmk__s(op->remote_nodename, controld_globals.our_nodename); } void lrm_op_callback(lrmd_event_data_t * op) { CRM_CHECK(op != NULL, return); switch (op->type) { case lrmd_event_disconnect: if (op->remote_nodename == NULL) { /* If this is the local executor IPC connection, set the right * bits in the controller when the connection goes down. */ lrm_connection_destroy(); } break; case lrmd_event_exec_complete: { lrm_state_t *lrm_state = lrm_state_find(op_node_name(op)); CRM_ASSERT(lrm_state != NULL); process_lrm_event(lrm_state, op, NULL, NULL); } break; default: break; } } static void try_local_executor_connect(long long action, fsa_data_t *msg_data, lrm_state_t *lrm_state) { int rc = pcmk_rc_ok; crm_debug("Connecting to the local executor"); // If we can connect, great rc = controld_connect_local_executor(lrm_state); if (rc == pcmk_rc_ok) { controld_set_fsa_input_flags(R_LRM_CONNECTED); crm_info("Connection to the local executor established"); return; } // Otherwise, if we can try again, set a timer to do so if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) { crm_warn("Failed to connect to the local executor %d time%s " "(%d max): %s", lrm_state->num_lrm_register_fails, pcmk__plural_s(lrm_state->num_lrm_register_fails), MAX_LRM_REG_FAILS, pcmk_rc_str(rc)); controld_start_wait_timer(); crmd_fsa_stall(FALSE); return; } // Otherwise give up crm_err("Failed to connect to the executor the max allowed " "%d time%s: %s", lrm_state->num_lrm_register_fails, pcmk__plural_s(lrm_state->num_lrm_register_fails), pcmk_rc_str(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } /* A_LRM_CONNECT */ void do_lrm_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) { /* This only pertains to local executor connections. Remote connections are * handled as resources within the scheduler. Connecting and disconnecting * from remote executor instances is handled differently. */ lrm_state_t *lrm_state = NULL; if (controld_globals.our_nodename == NULL) { return; /* Nothing to do */ } lrm_state = lrm_state_find_or_create(controld_globals.our_nodename); if (lrm_state == NULL) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } if (action & A_LRM_DISCONNECT) { if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) { if (action == A_LRM_DISCONNECT) { crmd_fsa_stall(FALSE); return; } } controld_clear_fsa_input_flags(R_LRM_CONNECTED); lrm_state_disconnect(lrm_state); lrm_state_reset_tables(lrm_state, FALSE); } if (action & A_LRM_CONNECT) { try_local_executor_connect(action, msg_data, lrm_state); } if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __func__); } } static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level) { int counter = 0; gboolean rc = TRUE; const char *when = "lrm disconnect"; GHashTableIter gIter; const char *key = NULL; rsc_history_t *entry = NULL; active_op_t *pending = NULL; crm_debug("Checking for active resources before exit"); if (cur_state == S_TERMINATE) { log_level = LOG_ERR; when = "shutdown"; } else if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { when = "shutdown... waiting"; } if ((lrm_state->active_ops != NULL) && lrm_state_is_connected(lrm_state)) { guint removed = g_hash_table_foreach_remove(lrm_state->active_ops, stop_recurring_actions, lrm_state); guint nremaining = g_hash_table_size(lrm_state->active_ops); if (removed || nremaining) { crm_notice("Stopped %u recurring operation%s at %s (%u remaining)", removed, pcmk__plural_s(removed), when, nremaining); } } if (lrm_state->active_ops != NULL) { g_hash_table_iter_init(&gIter, lrm_state->active_ops); while (g_hash_table_iter_next(&gIter, NULL, (void **)&pending)) { /* Ignore recurring actions in the shutdown calculations */ if (pending->interval_ms == 0) { counter++; } } } if (counter > 0) { do_crm_log(log_level, "%d pending executor operation%s at %s", counter, pcmk__plural_s(counter), when); if ((cur_state == S_TERMINATE) || !pcmk_is_set(controld_globals.fsa_input_register, R_SENT_RSC_STOP)) { g_hash_table_iter_init(&gIter, lrm_state->active_ops); while (g_hash_table_iter_next(&gIter, (gpointer*)&key, (gpointer*)&pending)) { do_crm_log(log_level, "Pending action: %s (%s)", key, pending->op_key); } } else { rc = FALSE; } return rc; } if (lrm_state->resource_history == NULL) { return rc; } if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { /* At this point we're not waiting, we're just shutting down */ when = "shutdown"; } counter = 0; g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (gpointer*)&entry)) { if (is_rsc_active(lrm_state, entry->id) == FALSE) { continue; } counter++; if (log_level == LOG_ERR) { crm_info("Found %s active at %s", entry->id, when); } else { crm_trace("Found %s active at %s", entry->id, when); } if (lrm_state->active_ops != NULL) { GHashTableIter hIter; g_hash_table_iter_init(&hIter, lrm_state->active_ops); while (g_hash_table_iter_next(&hIter, (gpointer*)&key, (gpointer*)&pending)) { if (pcmk__str_eq(entry->id, pending->rsc_id, pcmk__str_none)) { crm_notice("%sction %s (%s) incomplete at %s", pending->interval_ms == 0 ? "A" : "Recurring a", key, pending->op_key, when); } } } } if (counter) { crm_err("%d resource%s active at %s", counter, (counter == 1)? " was" : "s were", when); } return rc; } static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id) { rsc_history_t *entry = NULL; entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->last == NULL) { return FALSE; } crm_trace("Processing %s: %s.%d=%d", rsc_id, entry->last->op_type, entry->last->interval_ms, entry->last->rc); if ((entry->last->rc == PCMK_OCF_OK) && pcmk__str_eq(entry->last->op_type, PCMK_ACTION_STOP, pcmk__str_casei)) { return FALSE; } else if (entry->last->rc == PCMK_OCF_OK && pcmk__str_eq(entry->last->op_type, PCMK_ACTION_MIGRATE_TO, pcmk__str_casei)) { // A stricter check is too complex ... leave that to the scheduler return FALSE; } else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) { return FALSE; } else if ((entry->last->interval_ms == 0) && (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) { /* Badly configured resources can't be reliably stopped */ return FALSE; } return TRUE; } static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list) { GHashTableIter iter; rsc_history_t *entry = NULL; g_hash_table_iter_init(&iter, lrm_state->resource_history); while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) { GList *gIter = NULL; xmlNode *xml_rsc = create_xml_node(rsc_list, XML_LRM_TAG_RESOURCE); crm_xml_add(xml_rsc, PCMK_XA_ID, entry->id); crm_xml_add(xml_rsc, XML_ATTR_TYPE, entry->rsc.type); crm_xml_add(xml_rsc, XML_AGENT_ATTR_CLASS, entry->rsc.standard); crm_xml_add(xml_rsc, XML_AGENT_ATTR_PROVIDER, entry->rsc.provider); if (entry->last && entry->last->params) { static const char *name = CRM_META "_" PCMK__META_CONTAINER; const char *container = g_hash_table_lookup(entry->last->params, name); if (container) { crm_trace("Resource %s is a part of container resource %s", entry->id, container); crm_xml_add(xml_rsc, PCMK__META_CONTAINER, container); } } controld_add_resource_history_xml(xml_rsc, &(entry->rsc), entry->failed, lrm_state->node_name); controld_add_resource_history_xml(xml_rsc, &(entry->rsc), entry->last, lrm_state->node_name); for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) { controld_add_resource_history_xml(xml_rsc, &(entry->rsc), gIter->data, lrm_state->node_name); } } return FALSE; } xmlNode * controld_query_executor_state(void) { xmlNode *xml_state = NULL; xmlNode *xml_data = NULL; xmlNode *rsc_list = NULL; crm_node_t *peer = NULL; lrm_state_t *lrm_state = lrm_state_find(controld_globals.our_nodename); if (!lrm_state) { crm_err("Could not find executor state for node %s", controld_globals.our_nodename); return NULL; } peer = crm_get_peer_full(0, lrm_state->node_name, CRM_GET_PEER_ANY); CRM_CHECK(peer != NULL, return NULL); xml_state = create_node_state_update(peer, node_update_cluster|node_update_peer, NULL, __func__); if (xml_state == NULL) { return NULL; } xml_data = create_xml_node(xml_state, XML_CIB_TAG_LRM); crm_xml_add(xml_data, PCMK_XA_ID, peer->uuid); rsc_list = create_xml_node(xml_data, XML_LRM_TAG_RESOURCES); /* Build a list of active (not always running) resources */ build_active_RAs(lrm_state, rsc_list); crm_log_xml_trace(xml_state, "Current executor state"); return xml_state; } /*! * \internal * \brief Map standard Pacemaker return code to operation status and OCF code * * \param[out] event Executor event whose status and return code should be set * \param[in] rc Standard Pacemaker return code */ void controld_rc2event(lrmd_event_data_t *event, int rc) { /* This is called for cleanup requests from controller peers/clients, not * for resource actions, so no exit reason is needed. */ switch (rc) { case pcmk_rc_ok: lrmd__set_result(event, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); break; case EACCES: lrmd__set_result(event, PCMK_OCF_INSUFFICIENT_PRIV, PCMK_EXEC_ERROR, NULL); break; default: lrmd__set_result(event, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, NULL); break; } } /*! * \internal * \brief Trigger a new transition after CIB status was deleted * * If a CIB status delete was not expected (as part of the transition graph), * trigger a new transition by updating the (arbitrary) "last-lrm-refresh" * cluster property. * * \param[in] from_sys IPC name that requested the delete * \param[in] rsc_id Resource whose status was deleted (for logging only) */ void controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id) { if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) { char *now_s = crm_strdup_printf("%lld", (long long) time(NULL)); crm_debug("Triggering a refresh after %s cleaned %s", from_sys, rsc_id); cib__update_node_attr(controld_globals.logger_out, controld_globals.cib_conn, cib_none, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, "last-lrm-refresh", now_s, NULL, NULL); free(now_s); } } static void notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc) { lrmd_event_data_t *op = NULL; const char *from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM); const char *from_host = crm_element_value(input->msg, F_CRM_HOST_FROM); crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "localhost"), rsc_id, ((rc == pcmk_ok)? "" : " not")); op = construct_op(lrm_state, input->xml, rsc_id, PCMK_ACTION_DELETE); controld_rc2event(op, pcmk_legacy2rc(rc)); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); controld_trigger_delete_refresh(from_sys, rsc_id); } static gboolean lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data) { struct delete_event_s *event = user_data; struct pending_deletion_op_s *op = value; if (pcmk__str_eq(event->rsc, op->rsc, pcmk__str_none)) { notify_deleted(event->lrm_state, op->input, event->rsc, event->rc); return TRUE; } return FALSE; } static gboolean lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data) { const char *rsc = user_data; active_op_t *pending = value; if (pcmk__str_eq(rsc, pending->rsc_id, pcmk__str_none)) { crm_info("Removing op %s:%d for deleted resource %s", pending->op_key, pending->call_id, rsc); return TRUE; } return FALSE; } static void delete_rsc_entry(lrm_state_t *lrm_state, ha_msg_input_t *input, const char *rsc_id, GHashTableIter *rsc_iter, int rc, const char *user_name, bool from_cib) { struct delete_event_s event; CRM_CHECK(rsc_id != NULL, return); if (rc == pcmk_ok) { char *rsc_id_copy = strdup(rsc_id); if (rsc_iter) { g_hash_table_iter_remove(rsc_iter); } else { g_hash_table_remove(lrm_state->resource_history, rsc_id_copy); } if (from_cib) { controld_delete_resource_history(rsc_id_copy, lrm_state->node_name, user_name, crmd_cib_smart_opt()); } g_hash_table_foreach_remove(lrm_state->active_ops, lrm_remove_deleted_op, rsc_id_copy); free(rsc_id_copy); } if (input) { notify_deleted(lrm_state, input, rsc_id, rc); } event.rc = rc; event.rsc = rsc_id; event.lrm_state = lrm_state; g_hash_table_foreach_remove(lrm_state->deletion_ops, lrm_remove_deleted_rsc, &event); } static inline gboolean last_failed_matches_op(rsc_history_t *entry, const char *op, guint interval_ms) { if (entry == NULL) { return FALSE; } if (op == NULL) { return TRUE; } return (pcmk__str_eq(op, entry->failed->op_type, pcmk__str_casei) && (interval_ms == entry->failed->interval_ms)); } /*! * \internal * \brief Clear a resource's last failure * * Erase a resource's last failure on a particular node from both the * LRM resource history in the CIB, and the resource history remembered * for the LRM state. * * \param[in] rsc_id Resource name * \param[in] node_name Node name * \param[in] operation If specified, only clear if matching this operation * \param[in] interval_ms If operation is specified, it has this interval */ void lrm_clear_last_failure(const char *rsc_id, const char *node_name, const char *operation, guint interval_ms) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { return; } if (lrm_state->resource_history != NULL) { rsc_history_t *entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (last_failed_matches_op(entry, operation, interval_ms)) { lrmd_free_event(entry->failed); entry->failed = NULL; } } } /* Returns: gboolean - cancellation is in progress */ static gboolean cancel_op(lrm_state_t * lrm_state, const char *rsc_id, const char *key, int op, gboolean remove) { int rc = pcmk_ok; char *local_key = NULL; active_op_t *pending = NULL; CRM_CHECK(op != 0, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); if (key == NULL) { local_key = make_stop_id(rsc_id, op); key = local_key; } pending = g_hash_table_lookup(lrm_state->active_ops, key); if (pending) { if (remove && !pcmk_is_set(pending->flags, active_op_remove)) { controld_set_active_op_flags(pending, active_op_remove); crm_debug("Scheduling %s for removal", key); } if (pcmk_is_set(pending->flags, active_op_cancelled)) { crm_debug("Operation %s already cancelled", key); free(local_key); return FALSE; } controld_set_active_op_flags(pending, active_op_cancelled); } else { crm_info("No pending op found for %s", key); free(local_key); return FALSE; } crm_debug("Cancelling op %d for %s (%s)", op, rsc_id, key); rc = lrm_state_cancel(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms); if (rc == pcmk_ok) { crm_debug("Op %d for %s (%s): cancelled", op, rsc_id, key); free(local_key); return TRUE; } crm_debug("Op %d for %s (%s): Nothing to cancel", op, rsc_id, key); /* The caller needs to make sure the entry is * removed from the active operations list * * Usually by returning TRUE inside the worker function * supplied to g_hash_table_foreach_remove() * * Not removing the entry from active operations will block * the node from shutting down */ free(local_key); return FALSE; } struct cancel_data { gboolean done; gboolean remove; const char *key; lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean cancel_action_by_key(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct cancel_data *data = user_data; active_op_t *op = value; if (pcmk__str_eq(op->op_key, data->key, pcmk__str_none)) { data->done = TRUE; remove = !cancel_op(data->lrm_state, data->rsc->id, key, op->call_id, data->remove); } return remove; } static gboolean cancel_op_key(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *key, gboolean remove) { guint removed = 0; struct cancel_data data; CRM_CHECK(rsc != NULL, return FALSE); CRM_CHECK(key != NULL, return FALSE); data.key = key; data.rsc = rsc; data.done = FALSE; data.remove = remove; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove(lrm_state->active_ops, cancel_action_by_key, &data); crm_trace("Removed %u op cache entries, new size: %u", removed, g_hash_table_size(lrm_state->active_ops)); return data.done; } /*! * \internal * \brief Retrieve resource information from LRM * * \param[in,out] lrm_state Executor connection state to use * \param[in] rsc_xml XML containing resource configuration * \param[in] do_create If true, register resource if not already * \param[out] rsc_info Where to store information obtained from executor * * \retval pcmk_ok Success (and rsc_info holds newly allocated result) * \retval -EINVAL Required information is missing from arguments * \retval -ENOTCONN No active connection to LRM * \retval -ENODEV Resource not found * \retval -errno Error communicating with executor when registering resource * * \note Caller is responsible for freeing result on success. */ static int get_lrm_resource(lrm_state_t *lrm_state, const xmlNode *rsc_xml, gboolean do_create, lrmd_rsc_info_t **rsc_info) { const char *id = ID(rsc_xml); CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL); CRM_CHECK(id, return -EINVAL); if (lrm_state_is_connected(lrm_state) == FALSE) { return -ENOTCONN; } crm_trace("Retrieving resource information for %s from the executor", id); *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); // If resource isn't known by ID, try clone name, if provided if (!*rsc_info) { - const char *long_id = crm_element_value(rsc_xml, XML_ATTR_ID_LONG); + const char *long_id = crm_element_value(rsc_xml, PCMK__XA_LONG_ID); if (long_id) { *rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0); } } if ((*rsc_info == NULL) && do_create) { const char *class = crm_element_value(rsc_xml, XML_AGENT_ATTR_CLASS); const char *provider = crm_element_value(rsc_xml, XML_AGENT_ATTR_PROVIDER); const char *type = crm_element_value(rsc_xml, XML_ATTR_TYPE); int rc; crm_trace("Registering resource %s with the executor", id); rc = lrm_state_register_rsc(lrm_state, id, class, provider, type, lrmd_opt_drop_recurring); if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Could not register resource %s with the executor on %s: %s " CRM_XS " rc=%d", id, lrm_state->node_name, pcmk_strerror(rc), rc); /* Register this as an internal error if this involves the local * executor. Otherwise, we're likely dealing with an unresponsive * remote node, which is not an FSA failure. */ if (lrm_state_is_local(lrm_state) == TRUE) { register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } return rc; } *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); } return *rsc_info? pcmk_ok : -ENODEV; } static void delete_resource(lrm_state_t *lrm_state, const char *id, lrmd_rsc_info_t *rsc, GHashTableIter *iter, const char *sys, const char *user, ha_msg_input_t *request, bool unregister, bool from_cib) { int rc = pcmk_ok; crm_info("Removing resource %s from executor for %s%s%s", id, sys, (user? " as " : ""), (user? user : "")); if (rsc && unregister) { rc = lrm_state_unregister_rsc(lrm_state, id, 0); } if (rc == pcmk_ok) { crm_trace("Resource %s deleted from executor", id); } else if (rc == -EINPROGRESS) { crm_info("Deletion of resource '%s' from executor is pending", id); if (request) { struct pending_deletion_op_s *op = NULL; char *ref = crm_element_value_copy(request->msg, XML_ATTR_REFERENCE); op = calloc(1, sizeof(struct pending_deletion_op_s)); op->rsc = strdup(rsc->id); op->input = copy_ha_msg_input(request); g_hash_table_insert(lrm_state->deletion_ops, ref, op); } return; } else { crm_warn("Could not delete '%s' from executor for %s%s%s: %s " CRM_XS " rc=%d", id, sys, (user? " as " : ""), (user? user : ""), pcmk_strerror(rc), rc); } delete_rsc_entry(lrm_state, request, id, iter, rc, user, from_cib); } static int get_fake_call_id(lrm_state_t *lrm_state, const char *rsc_id) { int call_id = 999999999; rsc_history_t *entry = NULL; if(lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* Make sure the call id is greater than the last successful operation, * otherwise the failure will not result in a possible recovery of the resource * as it could appear the failure occurred before the successful start */ if (entry) { call_id = entry->last_callid + 1; } if (call_id < 0) { call_id = 1; } return call_id; } static void fake_op_status(lrm_state_t *lrm_state, lrmd_event_data_t *op, int op_status, enum ocf_exitcode op_exitcode, const char *exit_reason) { op->call_id = get_fake_call_id(lrm_state, op->rsc_id); op->t_run = time(NULL); op->t_rcchange = op->t_run; lrmd__set_result(op, op_exitcode, op_status, exit_reason); } static void force_reprobe(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node, bool reprobe_all_nodes) { GHashTableIter gIter; rsc_history_t *entry = NULL; crm_info("Clearing resource history on node %s", lrm_state->node_name); g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { /* only unregister the resource during a reprobe if it is not a remote connection * resource. otherwise unregistering the connection will terminate remote-node * membership */ bool unregister = true; if (is_remote_lrmd_ra(NULL, NULL, entry->id)) { unregister = false; if (reprobe_all_nodes) { lrm_state_t *remote_lrm_state = lrm_state_find(entry->id); if (remote_lrm_state != NULL) { /* If reprobing all nodes, be sure to reprobe the remote * node before clearing its connection resource */ force_reprobe(remote_lrm_state, from_sys, from_host, user_name, TRUE, reprobe_all_nodes); } } } /* Don't delete from the CIB, since we'll delete the whole node's LRM * state from the CIB soon */ delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys, user_name, NULL, unregister, false); } /* Now delete the copy in the CIB */ controld_delete_node_state(lrm_state->node_name, controld_section_lrm, cib_scope_local); // @COMPAT DCs < 1.1.14 need this deleted (in case it was explicitly false) update_attrd(lrm_state->node_name, CRM_OP_PROBED, NULL, user_name, is_remote_node); } /*! * \internal * \brief Fail a requested action without actually executing it * * For an action that can't be executed, process it similarly to an actual * execution result, with specified error status (except for notify actions, * which will always be treated as successful). * * \param[in,out] lrm_state Executor connection that action is for * \param[in] action Action XML from request * \param[in] rc Desired return code to use * \param[in] op_status Desired operation status to use * \param[in] exit_reason Human-friendly detail, if error */ static void synthesize_lrmd_failure(lrm_state_t *lrm_state, const xmlNode *action, int op_status, enum ocf_exitcode rc, const char *exit_reason) { lrmd_event_data_t *op = NULL; const char *operation = crm_element_value(action, XML_LRM_ATTR_TASK); const char *target_node = crm_element_value(action, XML_LRM_ATTR_TARGET); xmlNode *xml_rsc = find_xml_node(action, XML_CIB_TAG_RESOURCE, TRUE); if ((xml_rsc == NULL) || (ID(xml_rsc) == NULL)) { /* @TODO Should we do something else, like direct ack? */ crm_info("Can't fake %s failure (%d) on %s without resource configuration", crm_element_value(action, XML_LRM_ATTR_TASK_KEY), rc, target_node); return; } else if(operation == NULL) { /* This probably came from crm_resource -C, nothing to do */ crm_info("Can't fake %s failure (%d) on %s without operation", ID(xml_rsc), rc, target_node); return; } op = construct_op(lrm_state, action, ID(xml_rsc), operation); if (pcmk__str_eq(operation, PCMK_ACTION_NOTIFY, pcmk__str_casei)) { // Notifications can't fail fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_OK, NULL); } else { fake_op_status(lrm_state, op, op_status, rc, exit_reason); } crm_info("Faking " PCMK__OP_FMT " result (%d) on %s", op->rsc_id, op->op_type, op->interval_ms, op->rc, target_node); // Process the result as if it came from the LRM process_lrm_event(lrm_state, op, NULL, action); lrmd_free_event(op); } /*! * \internal * \brief Get target of an LRM operation (replacing \p NULL with local node * name) * * \param[in] xml LRM operation data XML * * \return LRM operation target node name (local node or Pacemaker Remote node) */ static const char * lrm_op_target(const xmlNode *xml) { const char *target = NULL; if (xml) { target = crm_element_value(xml, XML_LRM_ATTR_TARGET); } if (target == NULL) { target = controld_globals.our_nodename; } return target; } static void fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { lrmd_event_data_t *op = NULL; lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = find_xml_node(xml, XML_CIB_TAG_RESOURCE, TRUE); CRM_CHECK(xml_rsc != NULL, return); /* The executor simply executes operations and reports the results, without * any concept of success or failure, so to fail a resource, we must fake * what a failure looks like. * * To do this, we create a fake executor operation event for the resource, * and pass that event to the executor client callback so it will be * processed as if it came from the executor. */ op = construct_op(lrm_state, xml, ID(xml_rsc), "asyncmon"); free((char*) op->user_data); op->user_data = NULL; op->interval_ms = 0; if (user_name && !pcmk__is_privileged(user_name)) { crm_err("%s does not have permission to fail %s", user_name, ID(xml_rsc)); fake_op_status(lrm_state, op, PCMK_EXEC_ERROR, PCMK_OCF_INSUFFICIENT_PRIV, "Unprivileged user cannot fail resources"); controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc)); lrmd_free_event(op); return; } if (get_lrm_resource(lrm_state, xml_rsc, TRUE, &rsc) == pcmk_ok) { crm_info("Failing resource %s...", rsc->id); fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_UNKNOWN_ERROR, "Simulated failure"); process_lrm_event(lrm_state, op, NULL, xml); op->rc = PCMK_OCF_OK; // The request to fail the resource succeeded lrmd_free_rsc_info(rsc); } else { crm_info("Cannot find/create resource in order to fail it..."); crm_log_xml_warn(xml, "bad input"); fake_op_status(lrm_state, op, PCMK_EXEC_ERROR, PCMK_OCF_UNKNOWN_ERROR, "Cannot fail unknown resource"); } controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc)); lrmd_free_event(op); } static void handle_reprobe_op(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node, bool reprobe_all_nodes) { crm_notice("Forcing the status of all resources to be redetected"); force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node, reprobe_all_nodes); if (!pcmk__strcase_any_of(from_sys, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) { xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, NULL, from_host, from_sys, CRM_SYSTEM_LRMD, controld_globals.our_uuid); crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(reply); } } static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys) { char *op_key = NULL; char *meta_key = NULL; int call = 0; const char *call_id = NULL; const char *op_task = NULL; guint interval_ms = 0; gboolean in_progress = FALSE; xmlNode *params = find_xml_node(input->xml, XML_TAG_ATTRS, TRUE); CRM_CHECK(params != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_TASK); op_task = crm_element_value(params, meta_key); free(meta_key); CRM_CHECK(op_task != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_INTERVAL_MS); if (crm_element_value_ms(params, meta_key, &interval_ms) != pcmk_ok) { free(meta_key); return FALSE; } free(meta_key); op_key = pcmk__op_key(rsc->id, op_task, interval_ms); meta_key = crm_meta_name(XML_LRM_ATTR_CALLID); call_id = crm_element_value(params, meta_key); free(meta_key); crm_debug("Scheduler requested op %s (call=%s) be cancelled", op_key, (call_id? call_id : "NA")); pcmk__scan_min_int(call_id, &call, 0); if (call == 0) { // Normal case when the scheduler cancels a recurring op in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE); } else { // Normal case when the scheduler cancels an orphan op in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE); } // Acknowledge cancellation operation if for a remote connection resource if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) { char *op_id = make_stop_id(rsc->id, call); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) { crm_info("Nothing known about operation %d for %s", call, op_key); } controld_delete_action_history_by_key(rsc->id, lrm_state->node_name, op_key, call); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); /* needed at least for cancellation of a remote operation */ if (lrm_state->active_ops != NULL) { g_hash_table_remove(lrm_state->active_ops, op_id); } free(op_id); } else { /* No ack is needed since abcdaa8, but peers with older versions * in a rolling upgrade need one. We didn't bump the feature set * at that commit, so we can only compare against the previous * CRM version (3.0.8). If any peers have feature set 3.0.9 but * not abcdaa8, they will time out waiting for the ack (no * released versions of Pacemaker are affected). */ const char *peer_version = crm_element_value(params, PCMK_XA_CRM_FEATURE_SET); if (compare_version(peer_version, "3.0.8") <= 0) { crm_info("Sending compatibility ack for %s cancellation to %s (CRM version %s)", op_key, from_host, peer_version); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); } } free(op_key); return TRUE; } static void do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host, bool crm_rsc_delete, const char *user_name) { bool unregister = true; int cib_rc = controld_delete_resource_history(rsc->id, lrm_state->node_name, user_name, cib_dryrun|cib_sync_call); if (cib_rc != pcmk_rc_ok) { lrmd_event_data_t *op = NULL; op = construct_op(lrm_state, input->xml, rsc->id, PCMK_ACTION_DELETE); /* These are resource clean-ups, not actions, so no exit reason is * needed. */ lrmd__set_result(op, pcmk_rc2ocf(cib_rc), PCMK_EXEC_ERROR, NULL); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc->id); lrmd_free_event(op); return; } if (crm_rsc_delete && is_remote_lrmd_ra(NULL, NULL, rsc->id)) { unregister = false; } delete_resource(lrm_state, rsc->id, rsc, NULL, from_sys, user_name, input, unregister, true); } // User data for asynchronous metadata execution struct metadata_cb_data { lrmd_rsc_info_t *rsc; // Copy of resource information xmlNode *input_xml; // Copy of FSA input XML }; static struct metadata_cb_data * new_metadata_cb_data(lrmd_rsc_info_t *rsc, xmlNode *input_xml) { struct metadata_cb_data *data = NULL; data = calloc(1, sizeof(struct metadata_cb_data)); CRM_ASSERT(data != NULL); data->input_xml = copy_xml(input_xml); data->rsc = lrmd_copy_rsc_info(rsc); return data; } static void free_metadata_cb_data(struct metadata_cb_data *data) { lrmd_free_rsc_info(data->rsc); free_xml(data->input_xml); free(data); } /*! * \internal * \brief Execute an action after metadata has been retrieved * * \param[in] pid Ignored * \param[in] result Result of metadata action * \param[in] user_data Metadata callback data */ static void metadata_complete(int pid, const pcmk__action_result_t *result, void *user_data) { struct metadata_cb_data *data = (struct metadata_cb_data *) user_data; struct ra_metadata_s *md = NULL; lrm_state_t *lrm_state = lrm_state_find(lrm_op_target(data->input_xml)); if ((lrm_state != NULL) && pcmk__result_ok(result)) { md = controld_cache_metadata(lrm_state->metadata_cache, data->rsc, result->action_stdout); } if (!pcmk_is_set(controld_globals.fsa_input_register, R_HA_DISCONNECTED)) { do_lrm_rsc_op(lrm_state, data->rsc, data->input_xml, md); } free_metadata_cb_data(data); } /* A_LRM_INVOKE */ void do_lrm_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) { lrm_state_t *lrm_state = NULL; const char *crm_op = NULL; const char *from_sys = NULL; const char *from_host = NULL; const char *operation = NULL; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *user_name = NULL; const char *target_node = lrm_op_target(input->xml); gboolean is_remote_node = FALSE; bool crm_rsc_delete = FALSE; // Message routed to the local node is targeting a specific, non-local node is_remote_node = !pcmk__str_eq(target_node, controld_globals.our_nodename, pcmk__str_casei); lrm_state = lrm_state_find(target_node); if ((lrm_state == NULL) && is_remote_node) { crm_err("Failing action because local node has never had connection to remote node %s", target_node); synthesize_lrmd_failure(NULL, input->xml, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, "Local node has no connection to remote"); return; } CRM_ASSERT(lrm_state != NULL); user_name = pcmk__update_acl_user(input->msg, F_CRM_USER, NULL); crm_op = crm_element_value(input->msg, F_CRM_TASK); from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM); if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) { from_host = crm_element_value(input->msg, F_CRM_HOST_FROM); } if (pcmk__str_eq(crm_op, PCMK_ACTION_LRM_DELETE, pcmk__str_none)) { if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) { crm_rsc_delete = TRUE; // from crm_resource } operation = PCMK_ACTION_DELETE; } else if (input->xml != NULL) { operation = crm_element_value(input->xml, XML_LRM_ATTR_TASK); } CRM_CHECK(!pcmk__str_empty(crm_op) || !pcmk__str_empty(operation), return); crm_trace("'%s' execution request from %s as %s user", pcmk__s(crm_op, operation), pcmk__s(from_sys, "unknown subsystem"), pcmk__s(user_name, "current")); if (pcmk__str_eq(crm_op, CRM_OP_LRM_FAIL, pcmk__str_none)) { fail_lrm_resource(input->xml, lrm_state, user_name, from_host, from_sys); } else if (pcmk__str_eq(crm_op, CRM_OP_LRM_REFRESH, pcmk__str_none)) { /* @COMPAT This can only be sent by crm_resource --refresh on a * Pacemaker Remote node running Pacemaker 1.1.9, which is extremely * unlikely. It previously would cause the controller to re-write its * resource history to the CIB. Just ignore it. */ crm_notice("Ignoring refresh request from Pacemaker Remote 1.1.9 node"); // @COMPAT DCs <1.1.14 in a rolling upgrade might schedule this op } else if (pcmk__str_eq(operation, CRM_OP_PROBED, pcmk__str_none)) { update_attrd(lrm_state->node_name, CRM_OP_PROBED, XML_BOOLEAN_TRUE, user_name, is_remote_node); } else if (pcmk__str_eq(crm_op, CRM_OP_REPROBE, pcmk__str_none) || pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_none)) { const char *raw_target = NULL; if (input->xml != NULL) { // For CRM_OP_REPROBE, a NULL target means we're targeting all nodes raw_target = crm_element_value(input->xml, XML_LRM_ATTR_TARGET); } handle_reprobe_op(lrm_state, from_sys, from_host, user_name, is_remote_node, (raw_target == NULL)); } else if (operation != NULL) { lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = find_xml_node(input->xml, XML_CIB_TAG_RESOURCE, TRUE); gboolean create_rsc = !pcmk__str_eq(operation, PCMK_ACTION_DELETE, pcmk__str_none); int rc; // We can't return anything meaningful without a resource ID CRM_CHECK(xml_rsc && ID(xml_rsc), return); rc = get_lrm_resource(lrm_state, xml_rsc, create_rsc, &rsc); if (rc == -ENOTCONN) { synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, "Not connected to remote executor"); return; } else if ((rc < 0) && !create_rsc) { /* Delete of malformed or nonexistent resource * (deleting something that does not exist is a success) */ crm_notice("Not registering resource '%s' for a %s event " CRM_XS " get-rc=%d (%s) transition-key=%s", ID(xml_rsc), operation, rc, pcmk_strerror(rc), ID(input->xml)); delete_rsc_entry(lrm_state, input, ID(xml_rsc), NULL, pcmk_ok, user_name, true); return; } else if (rc == -EINVAL) { // Resource operation on malformed resource crm_err("Invalid resource definition for %s", ID(xml_rsc)); crm_log_xml_warn(input->msg, "invalid resource"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR, PCMK_OCF_NOT_CONFIGURED, // fatal error "Invalid resource definition"); return; } else if (rc < 0) { // Error communicating with the executor crm_err("Could not register resource '%s' with executor: %s " CRM_XS " rc=%d", ID(xml_rsc), pcmk_strerror(rc), rc); crm_log_xml_warn(input->msg, "failed registration"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR, PCMK_OCF_INVALID_PARAM, // hard error "Could not register resource with executor"); return; } if (pcmk__str_eq(operation, PCMK_ACTION_CANCEL, pcmk__str_none)) { if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) { crm_log_xml_warn(input->xml, "Bad command"); } } else if (pcmk__str_eq(operation, PCMK_ACTION_DELETE, pcmk__str_none)) { do_lrm_delete(input, lrm_state, rsc, from_sys, from_host, crm_rsc_delete, user_name); } else { struct ra_metadata_s *md = NULL; /* Getting metadata from cache is OK except for start actions -- * always refresh from the agent for those, in case the resource * agent was updated. * * @TODO Only refresh metadata for starts if the agent actually * changed (using something like inotify, or a hash or modification * time of the agent executable). */ if (strcmp(operation, PCMK_ACTION_START) != 0) { md = controld_get_rsc_metadata(lrm_state, rsc, controld_metadata_from_cache); } if ((md == NULL) && crm_op_needs_metadata(rsc->standard, operation)) { /* Most likely, we'll need the agent metadata to record the * pending operation and the operation result. Get it now rather * than wait until then, so the metadata action doesn't eat into * the real action's timeout. * * @TODO Metadata is retrieved via direct execution of the * agent, which has a couple of related issues: the executor * should execute agents, not the controller; and metadata for * Pacemaker Remote nodes should be collected on those nodes, * not locally. */ struct metadata_cb_data *data = NULL; data = new_metadata_cb_data(rsc, input->xml); crm_info("Retrieving metadata for %s (%s%s%s:%s) asynchronously", rsc->id, rsc->standard, ((rsc->provider == NULL)? "" : ":"), ((rsc->provider == NULL)? "" : rsc->provider), rsc->type); (void) lrmd__metadata_async(rsc, metadata_complete, (void *) data); } else { do_lrm_rsc_op(lrm_state, rsc, input->xml, md); } } lrmd_free_rsc_info(rsc); } else { crm_err("Invalid execution request: unknown command '%s' (bug?)", crm_op); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } static lrmd_event_data_t * construct_op(const lrm_state_t *lrm_state, const xmlNode *rsc_op, const char *rsc_id, const char *operation) { lrmd_event_data_t *op = NULL; const char *op_delay = NULL; const char *op_timeout = NULL; GHashTable *params = NULL; xmlNode *primitive = NULL; const char *class = NULL; const char *transition = NULL; CRM_ASSERT(rsc_id && operation); op = lrmd_new_event(rsc_id, operation, 0); op->type = lrmd_event_exec_complete; op->timeout = 0; op->start_delay = 0; lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL); if (rsc_op == NULL) { CRM_LOG_ASSERT(pcmk__str_eq(operation, PCMK_ACTION_STOP, pcmk__str_casei)); op->user_data = NULL; /* the stop_all_resources() case * by definition there is no DC (or they'd be shutting * us down). * So we should put our version here. */ op->params = pcmk__strkey_table(free, free); g_hash_table_insert(op->params, strdup(PCMK_XA_CRM_FEATURE_SET), strdup(CRM_FEATURE_SET)); crm_trace("Constructed %s op for %s", operation, rsc_id); return op; } params = xml2list(rsc_op); g_hash_table_remove(params, CRM_META "_op_target_rc"); op_delay = crm_meta_value(params, XML_OP_ATTR_START_DELAY); pcmk__scan_min_int(op_delay, &op->start_delay, 0); op_timeout = crm_meta_value(params, XML_ATTR_TIMEOUT); pcmk__scan_min_int(op_timeout, &op->timeout, 0); if (pcmk__guint_from_hash(params, CRM_META "_" XML_LRM_ATTR_INTERVAL_MS, 0, &(op->interval_ms)) != pcmk_rc_ok) { op->interval_ms = 0; } /* Use pcmk_monitor_timeout instead of meta timeout for stonith recurring monitor, if set */ primitive = find_xml_node(rsc_op, XML_CIB_TAG_RESOURCE, FALSE); class = crm_element_value(primitive, XML_AGENT_ATTR_CLASS); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_fence_params) && pcmk__str_eq(operation, PCMK_ACTION_MONITOR, pcmk__str_casei) && (op->interval_ms > 0)) { op_timeout = g_hash_table_lookup(params, "pcmk_monitor_timeout"); if (op_timeout != NULL) { op->timeout = crm_get_msec(op_timeout); } } if (!pcmk__str_eq(operation, PCMK_ACTION_STOP, pcmk__str_casei)) { op->params = params; } else { rsc_history_t *entry = NULL; if (lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* If we do not have stop parameters cached, use * whatever we are given */ if (!entry || !entry->stop_params) { op->params = params; } else { /* Copy the cached parameter list so that we stop the resource * with the old attributes, not the new ones */ op->params = pcmk__strkey_table(free, free); g_hash_table_foreach(params, copy_meta_keys, op->params); g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params); g_hash_table_destroy(params); params = NULL; } } /* sanity */ if (op->timeout <= 0) { op->timeout = op->interval_ms; } if (op->start_delay < 0) { op->start_delay = 0; } transition = crm_element_value(rsc_op, XML_ATTR_TRANSITION_KEY); CRM_CHECK(transition != NULL, return op); op->user_data = strdup(transition); if (op->interval_ms != 0) { if (pcmk__strcase_any_of(operation, PCMK_ACTION_START, PCMK_ACTION_STOP, NULL)) { crm_err("Start and Stop actions cannot have an interval: %u", op->interval_ms); op->interval_ms = 0; } } crm_trace("Constructed %s op for %s: interval=%u", operation, rsc_id, op->interval_ms); return op; } /*! * \internal * \brief Send a (synthesized) event result * * Reply with a synthesized event result directly, as opposed to going through * the executor. * * \param[in] to_host Host to send result to * \param[in] to_sys IPC name to send result (NULL for transition engine) * \param[in] rsc Type information about resource the result is for * \param[in,out] op Event with result to send * \param[in] rsc_id ID of resource the result is for */ void controld_ack_event_directly(const char *to_host, const char *to_sys, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *rsc_id) { xmlNode *reply = NULL; xmlNode *update, *iter; crm_node_t *peer = NULL; CRM_CHECK(op != NULL, return); if (op->rsc_id == NULL) { CRM_ASSERT(rsc_id != NULL); op->rsc_id = strdup(rsc_id); } if (to_sys == NULL) { to_sys = CRM_SYSTEM_TENGINE; } peer = crm_get_peer(0, controld_globals.our_nodename); update = create_node_state_update(peer, node_update_none, NULL, __func__); iter = create_xml_node(update, XML_CIB_TAG_LRM); crm_xml_add(iter, PCMK_XA_ID, controld_globals.our_uuid); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE); crm_xml_add(iter, PCMK_XA_ID, op->rsc_id); controld_add_resource_history_xml(iter, rsc, op, controld_globals.our_nodename); reply = create_request(CRM_OP_INVOKE_LRM, update, to_host, to_sys, CRM_SYSTEM_LRMD, NULL); crm_log_xml_trace(update, "[direct ACK]"); crm_debug("ACK'ing resource op " PCMK__OP_FMT " from %s: %s", op->rsc_id, op->op_type, op->interval_ms, op->user_data, crm_element_value(reply, XML_ATTR_REFERENCE)); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(update); free_xml(reply); } gboolean verify_stopped(enum crmd_fsa_state cur_state, int log_level) { gboolean res = TRUE; GList *lrm_state_list = lrm_state_get_list(); GList *state_entry; for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) { lrm_state_t *lrm_state = state_entry->data; if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) { /* keep iterating through all even when false is returned */ res = FALSE; } } controld_set_fsa_input_flags(R_SENT_RSC_STOP); g_list_free(lrm_state_list); lrm_state_list = NULL; return res; } struct stop_recurring_action_s { lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct stop_recurring_action_s *event = user_data; active_op_t *op = value; if ((op->interval_ms != 0) && pcmk__str_eq(op->rsc_id, event->rsc->id, pcmk__str_none)) { crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key); remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE); } return remove; } static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; lrm_state_t *lrm_state = user_data; active_op_t *op = value; if (op->interval_ms != 0) { crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (const char *) key); remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE); } return remove; } /*! * \internal * \brief Check whether recurring actions should be cancelled before an action * * \param[in] rsc_id Resource that action is for * \param[in] action Action being performed * \param[in] interval_ms Operation interval of \p action (in milliseconds) * * \return true if recurring actions should be cancelled, otherwise false */ static bool should_cancel_recurring(const char *rsc_id, const char *action, guint interval_ms) { if (is_remote_lrmd_ra(NULL, NULL, rsc_id) && (interval_ms == 0) && (strcmp(action, PCMK_ACTION_MIGRATE_TO) == 0)) { /* Don't stop monitoring a migrating Pacemaker Remote connection * resource until the entire migration has completed. We must detect if * the connection is unexpectedly severed, even during a migration. */ return false; } // Cancel recurring actions before changing resource state return (interval_ms == 0) && !pcmk__str_any_of(action, PCMK_ACTION_MONITOR, PCMK_ACTION_NOTIFY, NULL); } /*! * \internal * \brief Check whether an action should not be performed at this time * * \param[in] operation Action to be performed * * \return Readable description of why action should not be performed, * or NULL if it should be performed */ static const char * should_nack_action(const char *action) { if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN) && pcmk__str_eq(action, PCMK_ACTION_START, pcmk__str_none)) { register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); return "Not attempting start due to shutdown in progress"; } switch (controld_globals.fsa_state) { case S_NOT_DC: case S_POLICY_ENGINE: // Recalculating case S_TRANSITION_ENGINE: break; default: if (!pcmk__str_eq(action, PCMK_ACTION_STOP, pcmk__str_none)) { return "Controller cannot attempt actions at this time"; } break; } return NULL; } static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, xmlNode *msg, struct ra_metadata_s *md) { int rc; int call_id = 0; char *op_id = NULL; lrmd_event_data_t *op = NULL; fsa_data_t *msg_data = NULL; const char *transition = NULL; const char *operation = NULL; const char *nack_reason = NULL; CRM_CHECK((rsc != NULL) && (msg != NULL), return); operation = crm_element_value(msg, XML_LRM_ATTR_TASK); CRM_CHECK(!pcmk__str_empty(operation), return); transition = crm_element_value(msg, XML_ATTR_TRANSITION_KEY); if (pcmk__str_empty(transition)) { crm_log_xml_err(msg, "Missing transition number"); } if (lrm_state == NULL) { // This shouldn't be possible, but provide a failsafe just in case crm_err("Cannot execute %s of %s: No executor connection " CRM_XS " transition_key=%s", operation, rsc->id, pcmk__s(transition, "")); synthesize_lrmd_failure(NULL, msg, PCMK_EXEC_INVALID, PCMK_OCF_UNKNOWN_ERROR, "No executor connection"); return; } if (pcmk__str_any_of(operation, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, NULL)) { /* Pre-2.1.0 DCs will schedule reload actions only, and 2.1.0+ DCs * will schedule reload-agent actions only. In either case, we need * to map that to whatever the resource agent actually supports. * Default to the OCF 1.1 name. */ if ((md != NULL) && pcmk_is_set(md->ra_flags, ra_supports_legacy_reload)) { operation = PCMK_ACTION_RELOAD; } else { operation = PCMK_ACTION_RELOAD_AGENT; } } op = construct_op(lrm_state, msg, rsc->id, operation); CRM_CHECK(op != NULL, return); if (should_cancel_recurring(rsc->id, operation, op->interval_ms)) { guint removed = 0; struct stop_recurring_action_s data; data.rsc = rsc; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove(lrm_state->active_ops, stop_recurring_action_by_rsc, &data); if (removed) { crm_debug("Stopped %u recurring operation%s in preparation for " PCMK__OP_FMT, removed, pcmk__plural_s(removed), rsc->id, operation, op->interval_ms); } } /* now do the op */ crm_notice("Requesting local execution of %s operation for %s on %s " CRM_XS " transition_key=%s op_key=" PCMK__OP_FMT, pcmk__readable_action(op->op_type, op->interval_ms), rsc->id, lrm_state->node_name, pcmk__s(transition, ""), rsc->id, operation, op->interval_ms); nack_reason = should_nack_action(operation); if (nack_reason != NULL) { crm_notice("Discarding attempt to perform action %s on %s in state %s " "(shutdown=%s)", operation, rsc->id, fsa_state2string(controld_globals.fsa_state), pcmk__btoa(pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN))); lrmd__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_INVALID, nack_reason); controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id); lrmd_free_event(op); free(op_id); return; } controld_record_pending_op(lrm_state->node_name, rsc, op); op_id = pcmk__op_key(rsc->id, op->op_type, op->interval_ms); if (op->interval_ms > 0) { /* cancel it so we can then restart it without conflict */ cancel_op_key(lrm_state, rsc, op_id, FALSE); } rc = controld_execute_resource_agent(lrm_state, rsc->id, op->op_type, op->user_data, op->interval_ms, op->timeout, op->start_delay, op->params, &call_id); if (rc == pcmk_rc_ok) { /* record all operations so we can wait * for them to complete during shutdown */ char *call_id_s = make_stop_id(rsc->id, call_id); active_op_t *pending = NULL; pending = calloc(1, sizeof(active_op_t)); crm_trace("Recording pending op: %d - %s %s", call_id, op_id, call_id_s); pending->call_id = call_id; pending->interval_ms = op->interval_ms; pending->op_type = strdup(operation); pending->op_key = strdup(op_id); pending->rsc_id = strdup(rsc->id); pending->start_time = time(NULL); pcmk__str_update(&pending->user_data, op->user_data); if (crm_element_value_epoch(msg, PCMK_OPT_SHUTDOWN_LOCK, &(pending->lock_time)) != pcmk_ok) { pending->lock_time = 0; } g_hash_table_replace(lrm_state->active_ops, call_id_s, pending); if ((op->interval_ms > 0) && (op->start_delay > START_DELAY_THRESHOLD)) { int target_rc = PCMK_OCF_OK; crm_info("Faking confirmation of %s: execution postponed for over 5 minutes", op_id); decode_transition_key(op->user_data, NULL, NULL, NULL, &target_rc); lrmd__set_result(op, target_rc, PCMK_EXEC_DONE, NULL); controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id); } pending->params = op->params; op->params = NULL; } else if (lrm_state_is_local(lrm_state)) { crm_err("Could not initiate %s action for resource %s locally: %s " CRM_XS " rc=%d", operation, rsc->id, pcmk_rc_str(rc), rc); fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc)); process_lrm_event(lrm_state, op, NULL, NULL); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } else { crm_err("Could not initiate %s action for resource %s remotely on %s: " "%s " CRM_XS " rc=%d", operation, rsc->id, lrm_state->node_name, pcmk_rc_str(rc), rc); fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc)); process_lrm_event(lrm_state, op, NULL, NULL); } free(op_id); lrmd_free_event(op); } void do_lrm_event(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input cur_input, fsa_data_t * msg_data) { CRM_CHECK(FALSE, return); } static char * unescape_newlines(const char *string) { char *pch = NULL; char *ret = NULL; static const char *escaped_newline = "\\n"; if (!string) { return NULL; } ret = strdup(string); pch = strstr(ret, escaped_newline); while (pch != NULL) { /* Replace newline escape pattern with actual newline (and a space so we * don't have to shuffle the rest of the buffer) */ pch[0] = '\n'; pch[1] = ' '; pch = strstr(pch, escaped_newline); } return ret; } static bool did_lrm_rsc_op_fail(lrm_state_t *lrm_state, const char * rsc_id, const char * op_type, guint interval_ms) { rsc_history_t *entry = NULL; CRM_CHECK(lrm_state != NULL, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); CRM_CHECK(op_type != NULL, return FALSE); entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->failed == NULL) { return FALSE; } if (pcmk__str_eq(entry->failed->rsc_id, rsc_id, pcmk__str_none) && pcmk__str_eq(entry->failed->op_type, op_type, pcmk__str_casei) && entry->failed->interval_ms == interval_ms) { return TRUE; } return FALSE; } /*! * \internal * \brief Log the result of an executor action (actual or synthesized) * * \param[in] op Executor action to log result for * \param[in] op_key Operation key for action * \param[in] node_name Name of node action was performed on, if known * \param[in] confirmed Whether to log that graph action was confirmed */ static void log_executor_event(const lrmd_event_data_t *op, const char *op_key, const char *node_name, gboolean confirmed) { int log_level = LOG_ERR; GString *str = g_string_sized_new(100); // reasonable starting size pcmk__g_strcat(str, "Result of ", pcmk__readable_action(op->op_type, op->interval_ms), " operation for ", op->rsc_id, NULL); if (node_name != NULL) { pcmk__g_strcat(str, " on ", node_name, NULL); } switch (op->op_status) { case PCMK_EXEC_DONE: log_level = LOG_NOTICE; pcmk__g_strcat(str, ": ", services_ocf_exitcode_str(op->rc), NULL); break; case PCMK_EXEC_TIMEOUT: pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status), " after ", pcmk__readable_interval(op->timeout), NULL); break; case PCMK_EXEC_CANCELLED: log_level = LOG_INFO; /* order of __attribute__ and Fall through comment is IMPORTANT! * do not change it without proper testing with both clang and gcc * in multiple versions. * the clang check allows to build with all versions of clang. * the has_c_attribute check is to workaround a bug in clang version * in rhel7. has_attribute would happily return "YES SIR WE GOT IT" * and fail the build the next line. */ #ifdef __clang__ #ifdef __has_c_attribute #if __has_attribute(fallthrough) __attribute__((fallthrough)); #endif #endif #endif // Fall through default: pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status), NULL); } if ((op->exit_reason != NULL) && ((op->op_status != PCMK_EXEC_DONE) || (op->rc != PCMK_OCF_OK))) { pcmk__g_strcat(str, " (", op->exit_reason, ")", NULL); } g_string_append(str, " " CRM_XS); g_string_append_printf(str, " graph action %sconfirmed; call=%d key=%s", (confirmed? "" : "un"), op->call_id, op_key); if (op->op_status == PCMK_EXEC_DONE) { g_string_append_printf(str, " rc=%d", op->rc); } do_crm_log(log_level, "%s", str->str); g_string_free(str, TRUE); /* The services library has already logged the output at info or debug * level, so just raise to notice if it looks like a failure. */ if ((op->output != NULL) && (op->rc != PCMK_OCF_OK)) { char *prefix = crm_strdup_printf(PCMK__OP_FMT "@%s output", op->rsc_id, op->op_type, op->interval_ms, node_name); crm_log_output(LOG_NOTICE, prefix, op->output); free(prefix); } } void process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op, active_op_t *pending, const xmlNode *action_xml) { char *op_id = NULL; char *op_key = NULL; gboolean remove = FALSE; gboolean removed = FALSE; bool need_direct_ack = FALSE; lrmd_rsc_info_t *rsc = NULL; const char *node_name = NULL; CRM_CHECK(op != NULL, return); CRM_CHECK(op->rsc_id != NULL, return); // Remap new status codes for older DCs if (compare_version(controld_globals.dc_version, "3.2.0") < 0) { switch (op->op_status) { case PCMK_EXEC_NOT_CONNECTED: lrmd__set_result(op, PCMK_OCF_CONNECTION_DIED, PCMK_EXEC_ERROR, op->exit_reason); break; case PCMK_EXEC_INVALID: lrmd__set_result(op, CRM_DIRECT_NACK_RC, PCMK_EXEC_ERROR, op->exit_reason); break; default: break; } } op_id = make_stop_id(op->rsc_id, op->call_id); op_key = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); // Get resource info if available (from executor state or action XML) if (lrm_state) { rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0); } if ((rsc == NULL) && action_xml) { xmlNode *xml = find_xml_node(action_xml, XML_CIB_TAG_RESOURCE, TRUE); const char *standard = crm_element_value(xml, XML_AGENT_ATTR_CLASS); const char *provider = crm_element_value(xml, XML_AGENT_ATTR_PROVIDER); const char *type = crm_element_value(xml, XML_ATTR_TYPE); if (standard && type) { crm_info("%s agent information not cached, using %s%s%s:%s from action XML", op->rsc_id, standard, (provider? ":" : ""), (provider? provider : ""), type); rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type); } else { crm_err("Can't process %s result because %s agent information not cached or in XML", op_key, op->rsc_id); } } // Get node name if available (from executor state or action XML) if (lrm_state) { node_name = lrm_state->node_name; } else if (action_xml) { node_name = crm_element_value(action_xml, XML_LRM_ATTR_TARGET); } if(pending == NULL) { remove = TRUE; if (lrm_state) { pending = g_hash_table_lookup(lrm_state->active_ops, op_id); } } if (op->op_status == PCMK_EXEC_ERROR) { switch(op->rc) { case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_PROMOTED: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: // Leave it to the TE/scheduler to decide if this is an error op->op_status = PCMK_EXEC_DONE; break; default: /* Nothing to do */ break; } } if (op->op_status != PCMK_EXEC_CANCELLED) { /* We might not record the result, so directly acknowledge it to the * originator instead, so it doesn't time out waiting for the result * (especially important if part of a transition). */ need_direct_ack = TRUE; if (controld_action_is_recordable(op->op_type)) { if (node_name && rsc) { // We should record the result, and happily, we can time_t lock_time = (pending == NULL)? 0 : pending->lock_time; controld_update_resource_history(node_name, rsc, op, lock_time); need_direct_ack = FALSE; } else if (op->rsc_deleted) { /* We shouldn't record the result (likely the resource was * refreshed, cleaned, or removed while this operation was * in flight). */ crm_notice("Not recording %s result in CIB because " "resource information was removed since it was initiated", op_key); } else { /* This shouldn't be possible; the executor didn't consider the * resource deleted, but we couldn't find resource or node * information. */ crm_err("Unable to record %s result in CIB: %s", op_key, (node_name? "No resource information" : "No node name")); } } } else if (op->interval_ms == 0) { /* A non-recurring operation was cancelled. Most likely, the * never-initiated action was removed from the executor's pending * operations list upon resource removal. */ need_direct_ack = TRUE; } else if (pending == NULL) { /* This recurring operation was cancelled, but was not pending. No * transition actions are waiting on it, nothing needs to be done. */ } else if (op->user_data == NULL) { /* This recurring operation was cancelled and pending, but we don't * have a transition key. This should never happen. */ crm_err("Recurring operation %s was cancelled without transition information", op_key); } else if (pcmk_is_set(pending->flags, active_op_remove)) { /* This recurring operation was cancelled (by us) and pending, and we * have been waiting for it to finish. */ if (lrm_state) { controld_delete_action_history(op); } /* Directly acknowledge failed recurring actions here. The above call to * controld_delete_action_history() will not erase any corresponding * last_failure entry, which means that the DC won't confirm the * cancellation via process_op_deletion(), and the transition would * otherwise wait for the action timer to pop. */ if (did_lrm_rsc_op_fail(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms)) { need_direct_ack = TRUE; } } else if (op->rsc_deleted) { /* This recurring operation was cancelled (but not by us, and the * executor does not have resource information, likely due to resource * cleanup, refresh, or removal) and pending. */ crm_debug("Recurring op %s was cancelled due to resource deletion", op_key); need_direct_ack = TRUE; } else { /* This recurring operation was cancelled (but not by us, likely by the * executor before stopping the resource) and pending. We don't need to * do anything special. */ } if (need_direct_ack) { controld_ack_event_directly(NULL, NULL, NULL, op, op->rsc_id); } if(remove == FALSE) { /* The caller will do this afterwards, but keep the logging consistent */ removed = TRUE; } else if (lrm_state && ((op->interval_ms == 0) || (op->op_status == PCMK_EXEC_CANCELLED))) { gboolean found = g_hash_table_remove(lrm_state->active_ops, op_id); if (op->interval_ms != 0) { removed = TRUE; } else if (found) { removed = TRUE; crm_trace("Op %s (call=%d, stop-id=%s, remaining=%u): Confirmed", op_key, op->call_id, op_id, g_hash_table_size(lrm_state->active_ops)); } } log_executor_event(op, op_key, node_name, removed); if (lrm_state) { if (!pcmk__str_eq(op->op_type, PCMK_ACTION_META_DATA, pcmk__str_casei)) { crmd_alert_resource_op(lrm_state->node_name, op); } else if (rsc && (op->rc == PCMK_OCF_OK)) { char *metadata = unescape_newlines(op->output); controld_cache_metadata(lrm_state->metadata_cache, rsc, metadata); free(metadata); } } if (op->rsc_deleted) { crm_info("Deletion of resource '%s' complete after %s", op->rsc_id, op_key); if (lrm_state) { delete_rsc_entry(lrm_state, NULL, op->rsc_id, NULL, pcmk_ok, NULL, true); } } /* If a shutdown was escalated while operations were pending, * then the FSA will be stalled right now... allow it to continue */ controld_trigger_fsa(); if (lrm_state && rsc) { update_history_cache(lrm_state, rsc, op); } lrmd_free_rsc_info(rsc); free(op_key); free(op_id); } diff --git a/include/crm_internal.h b/include/crm_internal.h index 17e5803c7a..a2f1e467ce 100644 --- a/include/crm_internal.h +++ b/include/crm_internal.h @@ -1,129 +1,130 @@ /* * Copyright 2006-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef CRM_INTERNAL__H # define CRM_INTERNAL__H # ifndef PCMK__CONFIG_H # define PCMK__CONFIG_H # include # endif # include /* Our minimum glib dependency is 2.42. Define that as both the minimum and * maximum glib APIs that are allowed (i.e. APIs that were already deprecated * in 2.42, and APIs introduced after 2.42, cannot be used by Pacemaker code). */ #define GLIB_VERSION_MIN_REQUIRED GLIB_VERSION_2_42 #define GLIB_VERSION_MAX_ALLOWED GLIB_VERSION_2_42 # include # include # include /* Public API headers can guard including deprecated API headers with this * symbol, thus preventing internal code (which includes this header) from using * deprecated APIs, while still allowing external code to use them by default. */ #define PCMK_ALLOW_DEPRECATED 0 # include # include # include # include # include # include # include # include # include # include #define N_(String) (String) #ifdef ENABLE_NLS # define _(String) gettext(String) #else # define _(String) (String) #endif /* * XML attribute names used only by internal code */ #define PCMK__XA_ATTR_DAMPENING "attr_dampening" #define PCMK__XA_ATTR_FORCE "attrd_is_force_write" #define PCMK__XA_ATTR_INTERVAL "attr_clear_interval" #define PCMK__XA_ATTR_IS_PRIVATE "attr_is_private" #define PCMK__XA_ATTR_IS_REMOTE "attr_is_remote" #define PCMK__XA_ATTR_NAME "attr_name" #define PCMK__XA_ATTR_NODE_ID "attr_host_id" #define PCMK__XA_ATTR_NODE_NAME "attr_host" #define PCMK__XA_ATTR_OPERATION "attr_clear_operation" #define PCMK__XA_ATTR_PATTERN "attr_regex" #define PCMK__XA_ATTR_RESOURCE "attr_resource" #define PCMK__XA_ATTR_SECTION "attr_section" #define PCMK__XA_ATTR_SET "attr_set" #define PCMK__XA_ATTR_SET_TYPE "attr_set_type" #define PCMK__XA_ATTR_SYNC_POINT "attr_sync_point" #define PCMK__XA_ATTR_USER "attr_user" #define PCMK__XA_ATTR_UUID "attr_key" #define PCMK__XA_ATTR_VALUE "attr_value" #define PCMK__XA_ATTR_VERSION "attr_version" #define PCMK__XA_ATTR_WRITER "attr_writer" #define PCMK__XA_CONFIG_ERRORS "config-errors" #define PCMK__XA_CONFIG_WARNINGS "config-warnings" #define PCMK__XA_CONFIRM "confirm" #define PCMK__XA_CONN_HOST "connection_host" #define PCMK__XA_CRMD "crmd" #define PCMK__XA_DIGEST "digest" #define PCMK__XA_EXPECTED "expected" #define PCMK__XA_FILE "file" #define PCMK__XA_GRAPH_ERRORS "graph-errors" #define PCMK__XA_GRAPH_WARNINGS "graph-warnings" #define PCMK__XA_IN_CCM "in_ccm" #define PCMK__XA_JOIN "join" +#define PCMK__XA_LONG_ID "long-id" #define PCMK__XA_MODE "mode" #define PCMK__XA_NODE_START_STATE "node_start_state" #define PCMK__XA_PATH "path" #define PCMK__XA_REAP "reap" #define PCMK__XA_SCHEMA "schema" #define PCMK__XA_SCHEMAS "schemas" #define PCMK__XA_TASK "task" #define PCMK__XA_UPTIME "uptime" /* * IPC service names that are only used internally */ # define PCMK__SERVER_BASED_RO "cib_ro" # define PCMK__SERVER_BASED_RW "cib_rw" # define PCMK__SERVER_BASED_SHM "cib_shm" /* * IPC commands that can be sent to Pacemaker daemons */ #define PCMK__ATTRD_CMD_PEER_REMOVE "peer-remove" #define PCMK__ATTRD_CMD_UPDATE "update" #define PCMK__ATTRD_CMD_UPDATE_BOTH "update-both" #define PCMK__ATTRD_CMD_UPDATE_DELAY "update-delay" #define PCMK__ATTRD_CMD_QUERY "query" #define PCMK__ATTRD_CMD_REFRESH "refresh" #define PCMK__ATTRD_CMD_FLUSH "flush" #define PCMK__ATTRD_CMD_SYNC "sync" #define PCMK__ATTRD_CMD_SYNC_RESPONSE "sync-response" #define PCMK__ATTRD_CMD_CLEAR_FAILURE "clear-failure" #define PCMK__ATTRD_CMD_CONFIRM "confirm" #define PCMK__CONTROLD_CMD_NODES "list-nodes" #endif /* CRM_INTERNAL__H */ diff --git a/lib/common/ipc_controld.c b/lib/common/ipc_controld.c index 5198c9dd73..d46fff3e64 100644 --- a/lib/common/ipc_controld.c +++ b/lib/common/ipc_controld.c @@ -1,657 +1,657 @@ /* * Copyright 2020-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include "crmcommon_private.h" struct controld_api_private_s { char *client_uuid; unsigned int replies_expected; }; /*! * \internal * \brief Get a string representation of a controller API reply type * * \param[in] reply Controller API reply type * * \return String representation of a controller API reply type */ const char * pcmk__controld_api_reply2str(enum pcmk_controld_api_reply reply) { switch (reply) { case pcmk_controld_reply_reprobe: return "reprobe"; case pcmk_controld_reply_info: return "info"; case pcmk_controld_reply_resource: return "resource"; case pcmk_controld_reply_ping: return "ping"; case pcmk_controld_reply_nodes: return "nodes"; default: return "unknown"; } } // \return Standard Pacemaker return code static int new_data(pcmk_ipc_api_t *api) { struct controld_api_private_s *private = NULL; api->api_data = calloc(1, sizeof(struct controld_api_private_s)); if (api->api_data == NULL) { return errno; } private = api->api_data; /* This is set to the PID because that's how it was always done, but PIDs * are not unique because clients can be remote. The value appears to be * unused other than as part of F_CRM_SYS_FROM in IPC requests, which is * only compared against the internal system names (CRM_SYSTEM_TENGINE, * etc.), so it shouldn't be a problem. */ private->client_uuid = pcmk__getpid_s(); /* @TODO Implement a call ID model similar to the CIB, executor, and fencer * IPC APIs, so that requests and replies can be matched, and * duplicate replies can be discarded. */ return pcmk_rc_ok; } static void free_data(void *data) { free(((struct controld_api_private_s *) data)->client_uuid); free(data); } // \return Standard Pacemaker return code static int post_connect(pcmk_ipc_api_t *api) { /* The controller currently requires clients to register via a hello * request, but does not reply back. */ struct controld_api_private_s *private = api->api_data; const char *client_name = crm_system_name? crm_system_name : "client"; xmlNode *hello; int rc; hello = create_hello_message(private->client_uuid, client_name, PCMK__CONTROLD_API_MAJOR, PCMK__CONTROLD_API_MINOR); rc = pcmk__send_ipc_request(api, hello); free_xml(hello); if (rc != pcmk_rc_ok) { crm_info("Could not send IPC hello to %s: %s " CRM_XS " rc=%s", pcmk_ipc_name(api, true), pcmk_rc_str(rc), rc); } else { crm_debug("Sent IPC hello to %s", pcmk_ipc_name(api, true)); } return rc; } static void set_node_info_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data) { data->reply_type = pcmk_controld_reply_info; if (msg_data == NULL) { return; } data->data.node_info.have_quorum = pcmk__xe_attr_is_true(msg_data, PCMK_XA_HAVE_QUORUM); data->data.node_info.is_remote = pcmk__xe_attr_is_true(msg_data, XML_NODE_IS_REMOTE); /* Integer node_info.id is currently valid only for Corosync nodes. * * @TODO: Improve handling after crm_node_t is refactored to handle layer- * specific data better. */ crm_element_value_int(msg_data, PCMK_XA_ID, &(data->data.node_info.id)); data->data.node_info.uuid = crm_element_value(msg_data, PCMK_XA_ID); data->data.node_info.uname = crm_element_value(msg_data, XML_ATTR_UNAME); data->data.node_info.state = crm_element_value(msg_data, PCMK__XA_CRMD); } static void set_ping_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data) { data->reply_type = pcmk_controld_reply_ping; if (msg_data == NULL) { return; } data->data.ping.sys_from = crm_element_value(msg_data, XML_PING_ATTR_SYSFROM); data->data.ping.fsa_state = crm_element_value(msg_data, XML_PING_ATTR_CRMDSTATE); data->data.ping.result = crm_element_value(msg_data, XML_PING_ATTR_STATUS); } static void set_nodes_data(pcmk_controld_api_reply_t *data, xmlNode *msg_data) { pcmk_controld_api_node_t *node_info; data->reply_type = pcmk_controld_reply_nodes; for (xmlNode *node = first_named_child(msg_data, XML_CIB_TAG_NODE); node != NULL; node = crm_next_same_xml(node)) { long long id_ll = 0; node_info = calloc(1, sizeof(pcmk_controld_api_node_t)); crm_element_value_ll(node, PCMK_XA_ID, &id_ll); if (id_ll > 0) { node_info->id = id_ll; } node_info->uname = crm_element_value(node, XML_ATTR_UNAME); node_info->state = crm_element_value(node, PCMK__XA_IN_CCM); data->data.nodes = g_list_prepend(data->data.nodes, node_info); } } static bool reply_expected(pcmk_ipc_api_t *api, const xmlNode *request) { // We only need to handle commands that API functions can send return pcmk__str_any_of(crm_element_value(request, F_CRM_TASK), PCMK__CONTROLD_CMD_NODES, CRM_OP_LRM_DELETE, CRM_OP_LRM_FAIL, CRM_OP_NODE_INFO, CRM_OP_PING, CRM_OP_REPROBE, CRM_OP_RM_NODE_CACHE, NULL); } static bool dispatch(pcmk_ipc_api_t *api, xmlNode *reply) { struct controld_api_private_s *private = api->api_data; crm_exit_t status = CRM_EX_OK; xmlNode *msg_data = NULL; const char *value = NULL; pcmk_controld_api_reply_t reply_data = { pcmk_controld_reply_unknown, NULL, NULL, }; if (pcmk__xe_is(reply, "ack")) { /* ACKs are trivial responses that do not count toward expected replies, * and do not have all the fields that validation requires, so skip that * processing. */ return private->replies_expected > 0; } if (private->replies_expected > 0) { private->replies_expected--; } // Do some basic validation of the reply /* @TODO We should be able to verify that value is always a response, but * currently the controller doesn't always properly set the type. Even * if we fix the controller, we'll still need to handle replies from * old versions (feature set could be used to differentiate). */ value = crm_element_value(reply, F_CRM_MSG_TYPE); if (pcmk__str_empty(value) || !pcmk__str_any_of(value, XML_ATTR_REQUEST, XML_ATTR_RESPONSE, NULL)) { crm_info("Unrecognizable message from controller: " "invalid message type '%s'", pcmk__s(value, "")); status = CRM_EX_PROTOCOL; goto done; } if (pcmk__str_empty(crm_element_value(reply, XML_ATTR_REFERENCE))) { crm_info("Unrecognizable message from controller: no reference"); status = CRM_EX_PROTOCOL; goto done; } value = crm_element_value(reply, F_CRM_TASK); if (pcmk__str_empty(value)) { crm_info("Unrecognizable message from controller: no command name"); status = CRM_EX_PROTOCOL; goto done; } // Parse useful info from reply reply_data.feature_set = crm_element_value(reply, PCMK_XA_VERSION); reply_data.host_from = crm_element_value(reply, F_CRM_HOST_FROM); msg_data = get_message_xml(reply, F_CRM_DATA); if (!strcmp(value, CRM_OP_REPROBE)) { reply_data.reply_type = pcmk_controld_reply_reprobe; } else if (!strcmp(value, CRM_OP_NODE_INFO)) { set_node_info_data(&reply_data, msg_data); } else if (!strcmp(value, CRM_OP_INVOKE_LRM)) { reply_data.reply_type = pcmk_controld_reply_resource; reply_data.data.resource.node_state = msg_data; } else if (!strcmp(value, CRM_OP_PING)) { set_ping_data(&reply_data, msg_data); } else if (!strcmp(value, PCMK__CONTROLD_CMD_NODES)) { set_nodes_data(&reply_data, msg_data); } else { crm_info("Unrecognizable message from controller: unknown command '%s'", value); status = CRM_EX_PROTOCOL; } done: pcmk__call_ipc_callback(api, pcmk_ipc_event_reply, status, &reply_data); // Free any reply data that was allocated if (pcmk__str_eq(value, PCMK__CONTROLD_CMD_NODES, pcmk__str_casei)) { g_list_free_full(reply_data.data.nodes, free); } return false; // No further replies needed } pcmk__ipc_methods_t * pcmk__controld_api_methods(void) { pcmk__ipc_methods_t *cmds = calloc(1, sizeof(pcmk__ipc_methods_t)); if (cmds != NULL) { cmds->new_data = new_data; cmds->free_data = free_data; cmds->post_connect = post_connect; cmds->reply_expected = reply_expected; cmds->dispatch = dispatch; } return cmds; } /*! * \internal * \brief Create XML for a controller IPC request * * \param[in] api Controller connection * \param[in] op Controller IPC command name * \param[in] node Node name to set as destination host * \param[in] msg_data XML to attach to request as message data * * \return Newly allocated XML for request */ static xmlNode * create_controller_request(const pcmk_ipc_api_t *api, const char *op, const char *node, xmlNode *msg_data) { struct controld_api_private_s *private = NULL; const char *sys_to = NULL; if (api == NULL) { return NULL; } private = api->api_data; if ((node == NULL) && !strcmp(op, CRM_OP_PING)) { sys_to = CRM_SYSTEM_DC; } else { sys_to = CRM_SYSTEM_CRMD; } return create_request(op, msg_data, node, sys_to, (crm_system_name? crm_system_name : "client"), private->client_uuid); } // \return Standard Pacemaker return code static int send_controller_request(pcmk_ipc_api_t *api, const xmlNode *request, bool reply_is_expected) { if (crm_element_value(request, XML_ATTR_REFERENCE) == NULL) { return EINVAL; } if (reply_is_expected) { struct controld_api_private_s *private = api->api_data; private->replies_expected++; } return pcmk__send_ipc_request(api, request); } static xmlNode * create_reprobe_message_data(const char *target_node, const char *router_node) { xmlNode *msg_data; msg_data = create_xml_node(NULL, "data_for_" CRM_OP_REPROBE); crm_xml_add(msg_data, XML_LRM_ATTR_TARGET, target_node); if ((router_node != NULL) && !pcmk__str_eq(router_node, target_node, pcmk__str_casei)) { crm_xml_add(msg_data, XML_LRM_ATTR_ROUTER_NODE, router_node); } return msg_data; } /*! * \brief Send a reprobe controller operation * * \param[in,out] api Controller connection * \param[in] target_node Name of node to reprobe * \param[in] router_node Router node for host * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_reprobe. */ int pcmk_controld_api_reprobe(pcmk_ipc_api_t *api, const char *target_node, const char *router_node) { xmlNode *request; xmlNode *msg_data; int rc = pcmk_rc_ok; if (api == NULL) { return EINVAL; } if (router_node == NULL) { router_node = target_node; } crm_debug("Sending %s IPC request to reprobe %s via %s", pcmk_ipc_name(api, true), pcmk__s(target_node, "local node"), pcmk__s(router_node, "local node")); msg_data = create_reprobe_message_data(target_node, router_node); request = create_controller_request(api, CRM_OP_REPROBE, router_node, msg_data); rc = send_controller_request(api, request, true); free_xml(msg_data); free_xml(request); return rc; } /*! * \brief Send a "node info" controller operation * * \param[in,out] api Controller connection * \param[in] nodeid ID of node to get info for (or 0 for local node) * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_info. */ int pcmk_controld_api_node_info(pcmk_ipc_api_t *api, uint32_t nodeid) { xmlNode *request; int rc = pcmk_rc_ok; request = create_controller_request(api, CRM_OP_NODE_INFO, NULL, NULL); if (request == NULL) { return EINVAL; } if (nodeid > 0) { crm_xml_set_id(request, "%lu", (unsigned long) nodeid); } rc = send_controller_request(api, request, true); free_xml(request); return rc; } /*! * \brief Ask the controller for status * * \param[in,out] api Controller connection * \param[in] node_name Name of node whose status is desired (NULL for DC) * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_ping. */ int pcmk_controld_api_ping(pcmk_ipc_api_t *api, const char *node_name) { xmlNode *request; int rc = pcmk_rc_ok; request = create_controller_request(api, CRM_OP_PING, node_name, NULL); if (request == NULL) { return EINVAL; } rc = send_controller_request(api, request, true); free_xml(request); return rc; } /*! * \brief Ask the controller for cluster information * * \param[in,out] api Controller connection * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_nodes. */ int pcmk_controld_api_list_nodes(pcmk_ipc_api_t *api) { xmlNode *request; int rc = EINVAL; request = create_controller_request(api, PCMK__CONTROLD_CMD_NODES, NULL, NULL); if (request != NULL) { rc = send_controller_request(api, request, true); free_xml(request); } return rc; } // \return Standard Pacemaker return code static int controller_resource_op(pcmk_ipc_api_t *api, const char *op, const char *target_node, const char *router_node, bool cib_only, const char *rsc_id, const char *rsc_long_id, const char *standard, const char *provider, const char *type) { int rc = pcmk_rc_ok; char *key; xmlNode *request, *msg_data, *xml_rsc, *params; if (api == NULL) { return EINVAL; } if (router_node == NULL) { router_node = target_node; } msg_data = create_xml_node(NULL, XML_GRAPH_TAG_RSC_OP); /* The controller logs the transition key from resource op requests, so we * need to have *something* for it. * @TODO don't use "crm-resource" */ key = pcmk__transition_key(0, getpid(), 0, "xxxxxxxx-xrsc-opxx-xcrm-resourcexxxx"); crm_xml_add(msg_data, XML_ATTR_TRANSITION_KEY, key); free(key); crm_xml_add(msg_data, XML_LRM_ATTR_TARGET, target_node); if (!pcmk__str_eq(router_node, target_node, pcmk__str_casei)) { crm_xml_add(msg_data, XML_LRM_ATTR_ROUTER_NODE, router_node); } if (cib_only) { // Indicate that only the CIB needs to be cleaned crm_xml_add(msg_data, PCMK__XA_MODE, XML_TAG_CIB); } xml_rsc = create_xml_node(msg_data, XML_CIB_TAG_RESOURCE); crm_xml_add(xml_rsc, PCMK_XA_ID, rsc_id); - crm_xml_add(xml_rsc, XML_ATTR_ID_LONG, rsc_long_id); + crm_xml_add(xml_rsc, PCMK__XA_LONG_ID, rsc_long_id); crm_xml_add(xml_rsc, XML_AGENT_ATTR_CLASS, standard); crm_xml_add(xml_rsc, XML_AGENT_ATTR_PROVIDER, provider); crm_xml_add(xml_rsc, XML_ATTR_TYPE, type); params = create_xml_node(msg_data, XML_TAG_ATTRS); crm_xml_add(params, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); // The controller parses the timeout from the request key = crm_meta_name(XML_ATTR_TIMEOUT); crm_xml_add(params, key, "60000"); /* 1 minute */ //@TODO pass as arg free(key); request = create_controller_request(api, op, router_node, msg_data); rc = send_controller_request(api, request, true); free_xml(msg_data); free_xml(request); return rc; } /*! * \brief Ask the controller to fail a resource * * \param[in,out] api Controller connection * \param[in] target_node Name of node resource is on * \param[in] router_node Router node for target * \param[in] rsc_id ID of resource to fail * \param[in] rsc_long_id Long ID of resource (if any) * \param[in] standard Standard of resource * \param[in] provider Provider of resource (if any) * \param[in] type Type of resource to fail * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_resource. */ int pcmk_controld_api_fail(pcmk_ipc_api_t *api, const char *target_node, const char *router_node, const char *rsc_id, const char *rsc_long_id, const char *standard, const char *provider, const char *type) { crm_debug("Sending %s IPC request to fail %s (a.k.a. %s) on %s via %s", pcmk_ipc_name(api, true), pcmk__s(rsc_id, "unknown resource"), pcmk__s(rsc_long_id, "no other names"), pcmk__s(target_node, "unspecified node"), pcmk__s(router_node, "unspecified node")); return controller_resource_op(api, CRM_OP_LRM_FAIL, target_node, router_node, false, rsc_id, rsc_long_id, standard, provider, type); } /*! * \brief Ask the controller to refresh a resource * * \param[in,out] api Controller connection * \param[in] target_node Name of node resource is on * \param[in] router_node Router node for target * \param[in] rsc_id ID of resource to refresh * \param[in] rsc_long_id Long ID of resource (if any) * \param[in] standard Standard of resource * \param[in] provider Provider of resource (if any) * \param[in] type Type of resource * \param[in] cib_only If true, clean resource from CIB only * * \return Standard Pacemaker return code * \note Event callback will get a reply of type pcmk_controld_reply_resource. */ int pcmk_controld_api_refresh(pcmk_ipc_api_t *api, const char *target_node, const char *router_node, const char *rsc_id, const char *rsc_long_id, const char *standard, const char *provider, const char *type, bool cib_only) { crm_debug("Sending %s IPC request to refresh %s (a.k.a. %s) on %s via %s", pcmk_ipc_name(api, true), pcmk__s(rsc_id, "unknown resource"), pcmk__s(rsc_long_id, "no other names"), pcmk__s(target_node, "unspecified node"), pcmk__s(router_node, "unspecified node")); return controller_resource_op(api, CRM_OP_LRM_DELETE, target_node, router_node, cib_only, rsc_id, rsc_long_id, standard, provider, type); } /*! * \brief Get the number of IPC replies currently expected from the controller * * \param[in] api Controller IPC API connection * * \return Number of replies expected */ unsigned int pcmk_controld_api_replies_expected(const pcmk_ipc_api_t *api) { struct controld_api_private_s *private = api->api_data; return private->replies_expected; } /*! * \brief Create XML for a controller IPC "hello" message * * \deprecated This function is deprecated as part of the public C API. */ // \todo make this static to this file when breaking API backward compatibility xmlNode * create_hello_message(const char *uuid, const char *client_name, const char *major_version, const char *minor_version) { xmlNode *hello_node = NULL; xmlNode *hello = NULL; if (pcmk__str_empty(uuid) || pcmk__str_empty(client_name) || pcmk__str_empty(major_version) || pcmk__str_empty(minor_version)) { crm_err("Could not create IPC hello message from %s (UUID %s): " "missing information", client_name? client_name : "unknown client", uuid? uuid : "unknown"); return NULL; } hello_node = create_xml_node(NULL, XML_TAG_OPTIONS); if (hello_node == NULL) { crm_err("Could not create IPC hello message from %s (UUID %s): " "Message data creation failed", client_name, uuid); return NULL; } crm_xml_add(hello_node, "major_version", major_version); crm_xml_add(hello_node, "minor_version", minor_version); crm_xml_add(hello_node, "client_name", client_name); crm_xml_add(hello_node, "client_uuid", uuid); hello = create_request(CRM_OP_HELLO, hello_node, NULL, NULL, client_name, uuid); if (hello == NULL) { crm_err("Could not create IPC hello message from %s (UUID %s): " "Request creation failed", client_name, uuid); return NULL; } free_xml(hello_node); crm_trace("Created hello message from %s (UUID %s)", client_name, uuid); return hello; } diff --git a/lib/pacemaker/pcmk_graph_producer.c b/lib/pacemaker/pcmk_graph_producer.c index 9299909b79..3753d64218 100644 --- a/lib/pacemaker/pcmk_graph_producer.c +++ b/lib/pacemaker/pcmk_graph_producer.c @@ -1,1097 +1,1097 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include "libpacemaker_private.h" // Convenience macros for logging action properties #define action_type_str(flags) \ (pcmk_is_set((flags), pcmk_action_pseudo)? "pseudo-action" : "action") #define action_optional_str(flags) \ (pcmk_is_set((flags), pcmk_action_optional)? "optional" : "required") #define action_runnable_str(flags) \ (pcmk_is_set((flags), pcmk_action_runnable)? "runnable" : "unrunnable") #define action_node_str(a) \ (((a)->node == NULL)? "no node" : (a)->node->details->uname) /*! * \internal * \brief Add an XML node tag for a specified ID * * \param[in] id Node UUID to add * \param[in,out] xml Parent XML tag to add to */ static xmlNode* add_node_to_xml_by_id(const char *id, xmlNode *xml) { xmlNode *node_xml; node_xml = create_xml_node(xml, XML_CIB_TAG_NODE); crm_xml_add(node_xml, PCMK_XA_ID, id); return node_xml; } /*! * \internal * \brief Add an XML node tag for a specified node * * \param[in] node Node to add * \param[in,out] xml XML to add node to */ static void add_node_to_xml(const pcmk_node_t *node, void *xml) { add_node_to_xml_by_id(node->details->id, (xmlNode *) xml); } /*! * \internal * \brief Count (optionally add to XML) nodes needing maintenance state update * * \param[in,out] xml Parent XML tag to add to, if any * \param[in] scheduler Scheduler data * * \return Count of nodes added * \note Only Pacemaker Remote nodes are considered currently */ static int add_maintenance_nodes(xmlNode *xml, const pcmk_scheduler_t *scheduler) { xmlNode *maintenance = NULL; int count = 0; if (xml != NULL) { maintenance = create_xml_node(xml, XML_GRAPH_TAG_MAINTENANCE); } for (const GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { const pcmk_node_t *node = iter->data; if (pe__is_guest_or_remote_node(node) && (node->details->maintenance != node->details->remote_maintenance)) { if (maintenance != NULL) { crm_xml_add(add_node_to_xml_by_id(node->details->id, maintenance), XML_NODE_IS_MAINTENANCE, (node->details->maintenance? "1" : "0")); } count++; } } crm_trace("%s %d nodes in need of maintenance mode update in state", ((maintenance == NULL)? "Counted" : "Added"), count); return count; } /*! * \internal * \brief Add pseudo action with nodes needing maintenance state update * * \param[in,out] scheduler Scheduler data */ static void add_maintenance_update(pcmk_scheduler_t *scheduler) { pcmk_action_t *action = NULL; if (add_maintenance_nodes(NULL, scheduler) != 0) { action = get_pseudo_op(PCMK_ACTION_MAINTENANCE_NODES, scheduler); pcmk__set_action_flags(action, pcmk_action_always_in_graph); } } /*! * \internal * \brief Add XML with nodes that an action is expected to bring down * * If a specified action is expected to bring any nodes down, add an XML block * with their UUIDs. When a node is lost, this allows the controller to * determine whether it was expected. * * \param[in,out] xml Parent XML tag to add to * \param[in] action Action to check for downed nodes */ static void add_downed_nodes(xmlNode *xml, const pcmk_action_t *action) { CRM_CHECK((xml != NULL) && (action != NULL) && (action->node != NULL), return); if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { /* Shutdown makes the action's node down */ xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED); add_node_to_xml_by_id(action->node->details->id, downed); } else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { /* Fencing makes the action's node and any hosted guest nodes down */ const char *fence = g_hash_table_lookup(action->meta, "stonith_action"); if (pcmk__is_fencing_action(fence)) { xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED); add_node_to_xml_by_id(action->node->details->id, downed); pe_foreach_guest_node(action->node->details->data_set, action->node, add_node_to_xml, downed); } } else if (action->rsc && action->rsc->is_remote_node && pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_none)) { /* Stopping a remote connection resource makes connected node down, * unless it's part of a migration */ GList *iter; pcmk_action_t *input; bool migrating = false; for (iter = action->actions_before; iter != NULL; iter = iter->next) { input = ((pcmk__related_action_t *) iter->data)->action; if ((input->rsc != NULL) && pcmk__str_eq(action->rsc->id, input->rsc->id, pcmk__str_none) && pcmk__str_eq(input->task, PCMK_ACTION_MIGRATE_FROM, pcmk__str_none)) { migrating = true; break; } } if (!migrating) { xmlNode *downed = create_xml_node(xml, XML_GRAPH_TAG_DOWNED); add_node_to_xml_by_id(action->rsc->id, downed); } } } /*! * \internal * \brief Create a transition graph operation key for a clone action * * \param[in] action Clone action * \param[in] interval_ms Action interval in milliseconds * * \return Newly allocated string with transition graph operation key */ static char * clone_op_key(const pcmk_action_t *action, guint interval_ms) { if (pcmk__str_eq(action->task, PCMK_ACTION_NOTIFY, pcmk__str_none)) { const char *n_type = g_hash_table_lookup(action->meta, "notify_type"); const char *n_task = g_hash_table_lookup(action->meta, "notify_operation"); CRM_LOG_ASSERT((n_type != NULL) && (n_task != NULL)); return pcmk__notify_key(action->rsc->clone_name, n_type, n_task); } else if (action->cancel_task != NULL) { return pcmk__op_key(action->rsc->clone_name, action->cancel_task, interval_ms); } else { return pcmk__op_key(action->rsc->clone_name, action->task, interval_ms); } } /*! * \internal * \brief Add node details to transition graph action XML * * \param[in] action Scheduled action * \param[in,out] xml Transition graph action XML for \p action */ static void add_node_details(const pcmk_action_t *action, xmlNode *xml) { pcmk_node_t *router_node = pcmk__connection_host_for_action(action); crm_xml_add(xml, XML_LRM_ATTR_TARGET, action->node->details->uname); crm_xml_add(xml, XML_LRM_ATTR_TARGET_UUID, action->node->details->id); if (router_node != NULL) { crm_xml_add(xml, XML_LRM_ATTR_ROUTER_NODE, router_node->details->uname); } } /*! * \internal * \brief Add resource details to transition graph action XML * * \param[in] action Scheduled action * \param[in,out] action_xml Transition graph action XML for \p action */ static void add_resource_details(const pcmk_action_t *action, xmlNode *action_xml) { xmlNode *rsc_xml = NULL; const char *attr_list[] = { XML_AGENT_ATTR_CLASS, XML_AGENT_ATTR_PROVIDER, XML_ATTR_TYPE }; /* If a resource is locked to a node via PCMK_OPT_SHUTDOWN_LOCK, mark its * actions so the controller can preserve the lock when the action * completes. */ if (pcmk__action_locks_rsc_to_node(action)) { crm_xml_add_ll(action_xml, PCMK_OPT_SHUTDOWN_LOCK, (long long) action->rsc->lock_time); } // List affected resource rsc_xml = create_xml_node(action_xml, (const char *) action->rsc->xml->name); if (pcmk_is_set(action->rsc->flags, pcmk_rsc_removed) && (action->rsc->clone_name != NULL)) { /* Use the numbered instance name here, because if there is more * than one instance on a node, we need to make sure the command * goes to the right one. * * This is important even for anonymous clones, because the clone's * unique meta-attribute might have just been toggled from on to * off. */ crm_debug("Using orphan clone name %s instead of %s", action->rsc->id, action->rsc->clone_name); crm_xml_add(rsc_xml, PCMK_XA_ID, action->rsc->clone_name); - crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->id); + crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->id); } else if (!pcmk_is_set(action->rsc->flags, pcmk_rsc_unique)) { const char *xml_id = ID(action->rsc->xml); crm_debug("Using anonymous clone name %s for %s (aka %s)", xml_id, action->rsc->id, action->rsc->clone_name); /* ID is what we'd like client to use - * ID_LONG is what they might know it as instead + * LONG_ID is what they might know it as instead * - * ID_LONG is only strictly needed /here/ during the + * LONG_ID is only strictly needed /here/ during the * transition period until all nodes in the cluster * are running the new software /and/ have rebooted * once (meaning that they've only ever spoken to a DC * supporting this feature). * * If anyone toggles the unique flag to 'on', the * 'instance free' name will correspond to an orphan * and fall into the clause above instead */ crm_xml_add(rsc_xml, PCMK_XA_ID, xml_id); if ((action->rsc->clone_name != NULL) && !pcmk__str_eq(xml_id, action->rsc->clone_name, pcmk__str_none)) { - crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->clone_name); + crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->clone_name); } else { - crm_xml_add(rsc_xml, XML_ATTR_ID_LONG, action->rsc->id); + crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->id); } } else { CRM_ASSERT(action->rsc->clone_name == NULL); crm_xml_add(rsc_xml, PCMK_XA_ID, action->rsc->id); } for (int lpc = 0; lpc < PCMK__NELEM(attr_list); lpc++) { crm_xml_add(rsc_xml, attr_list[lpc], g_hash_table_lookup(action->rsc->meta, attr_list[lpc])); } } /*! * \internal * \brief Add action attributes to transition graph action XML * * \param[in,out] action Scheduled action * \param[in,out] action_xml Transition graph action XML for \p action */ static void add_action_attributes(pcmk_action_t *action, xmlNode *action_xml) { xmlNode *args_xml = NULL; /* We create free-standing XML to start, so we can sort the attributes * before adding it to action_xml, which keeps the scheduler regression * test graphs comparable. */ args_xml = create_xml_node(NULL, XML_TAG_ATTRS); crm_xml_add(args_xml, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); g_hash_table_foreach(action->extra, hash2field, args_xml); if ((action->rsc != NULL) && (action->node != NULL)) { // Get the resource instance attributes, evaluated properly for node GHashTable *params = pe_rsc_params(action->rsc, action->node, action->rsc->cluster); pcmk__substitute_remote_addr(action->rsc, params); g_hash_table_foreach(params, hash2smartfield, args_xml); } else if ((action->rsc != NULL) && (action->rsc->variant <= pcmk_rsc_variant_primitive)) { GHashTable *params = pe_rsc_params(action->rsc, NULL, action->rsc->cluster); g_hash_table_foreach(params, hash2smartfield, args_xml); } g_hash_table_foreach(action->meta, hash2metafield, args_xml); if (action->rsc != NULL) { pcmk_resource_t *parent = action->rsc; while (parent != NULL) { parent->cmds->add_graph_meta(parent, args_xml); parent = parent->parent; } pcmk__add_guest_meta_to_xml(args_xml, action); } else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none) && (action->node != NULL)) { /* Pass the node's attributes as meta-attributes. * * @TODO: Determine whether it is still necessary to do this. It was * added in 33d99707, probably for the libfence-based implementation in * c9a90bd, which is no longer used. */ g_hash_table_foreach(action->node->details->attrs, hash2metafield, args_xml); } sorted_xml(args_xml, action_xml, FALSE); free_xml(args_xml); } /*! * \internal * \brief Create the transition graph XML for a scheduled action * * \param[in,out] parent Parent XML element to add action to * \param[in,out] action Scheduled action * \param[in] skip_details If false, add action details as sub-elements * \param[in] scheduler Scheduler data */ static void create_graph_action(xmlNode *parent, pcmk_action_t *action, bool skip_details, const pcmk_scheduler_t *scheduler) { bool needs_node_info = true; bool needs_maintenance_info = false; xmlNode *action_xml = NULL; if ((action == NULL) || (scheduler == NULL)) { return; } // Create the top-level element based on task if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { /* All fences need node info; guest node fences are pseudo-events */ if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { action_xml = create_xml_node(parent, XML_GRAPH_TAG_PSEUDO_EVENT); } else { action_xml = create_xml_node(parent, XML_GRAPH_TAG_CRM_EVENT); } } else if (pcmk__str_any_of(action->task, PCMK_ACTION_DO_SHUTDOWN, PCMK_ACTION_CLEAR_FAILCOUNT, NULL)) { action_xml = create_xml_node(parent, XML_GRAPH_TAG_CRM_EVENT); } else if (pcmk__str_eq(action->task, PCMK_ACTION_LRM_DELETE, pcmk__str_none)) { // CIB-only clean-up for shutdown locks action_xml = create_xml_node(parent, XML_GRAPH_TAG_CRM_EVENT); crm_xml_add(action_xml, PCMK__XA_MODE, XML_TAG_CIB); } else if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { if (pcmk__str_eq(action->task, PCMK_ACTION_MAINTENANCE_NODES, pcmk__str_none)) { needs_maintenance_info = true; } action_xml = create_xml_node(parent, XML_GRAPH_TAG_PSEUDO_EVENT); needs_node_info = false; } else { action_xml = create_xml_node(parent, XML_GRAPH_TAG_RSC_OP); } crm_xml_add_int(action_xml, PCMK_XA_ID, action->id); crm_xml_add(action_xml, XML_LRM_ATTR_TASK, action->task); if ((action->rsc != NULL) && (action->rsc->clone_name != NULL)) { char *clone_key = NULL; guint interval_ms; if (pcmk__guint_from_hash(action->meta, XML_LRM_ATTR_INTERVAL_MS, 0, &interval_ms) != pcmk_rc_ok) { interval_ms = 0; } clone_key = clone_op_key(action, interval_ms); crm_xml_add(action_xml, XML_LRM_ATTR_TASK_KEY, clone_key); crm_xml_add(action_xml, "internal_" XML_LRM_ATTR_TASK_KEY, action->uuid); free(clone_key); } else { crm_xml_add(action_xml, XML_LRM_ATTR_TASK_KEY, action->uuid); } if (needs_node_info && (action->node != NULL)) { add_node_details(action, action_xml); g_hash_table_insert(action->meta, strdup(XML_LRM_ATTR_TARGET), strdup(action->node->details->uname)); g_hash_table_insert(action->meta, strdup(XML_LRM_ATTR_TARGET_UUID), strdup(action->node->details->id)); } if (skip_details) { return; } if ((action->rsc != NULL) && !pcmk_is_set(action->flags, pcmk_action_pseudo)) { // This is a real resource action, so add resource details add_resource_details(action, action_xml); } /* List any attributes in effect */ add_action_attributes(action, action_xml); /* List any nodes this action is expected to make down */ if (needs_node_info && (action->node != NULL)) { add_downed_nodes(action_xml, action); } if (needs_maintenance_info) { add_maintenance_nodes(action_xml, scheduler); } } /*! * \internal * \brief Check whether an action should be added to the transition graph * * \param[in] action Action to check * * \return true if action should be added to graph, otherwise false */ static bool should_add_action_to_graph(const pcmk_action_t *action) { if (!pcmk_is_set(action->flags, pcmk_action_runnable)) { crm_trace("Ignoring action %s (%d): unrunnable", action->uuid, action->id); return false; } if (pcmk_is_set(action->flags, pcmk_action_optional) && !pcmk_is_set(action->flags, pcmk_action_always_in_graph)) { crm_trace("Ignoring action %s (%d): optional", action->uuid, action->id); return false; } /* Actions for unmanaged resources should be excluded from the graph, * with the exception of monitors and cancellation of recurring monitors. */ if ((action->rsc != NULL) && !pcmk_is_set(action->rsc->flags, pcmk_rsc_managed) && !pcmk__str_eq(action->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { const char *interval_ms_s; /* A cancellation of a recurring monitor will get here because the task * is cancel rather than monitor, but the interval can still be used to * recognize it. The interval has been normalized to milliseconds by * this point, so a string comparison is sufficient. */ interval_ms_s = g_hash_table_lookup(action->meta, XML_LRM_ATTR_INTERVAL_MS); if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches)) { crm_trace("Ignoring action %s (%d): for unmanaged resource (%s)", action->uuid, action->id, action->rsc->id); return false; } } /* Always add pseudo-actions, fence actions, and shutdown actions (already * determined to be required and runnable by this point) */ if (pcmk_is_set(action->flags, pcmk_action_pseudo) || pcmk__strcase_any_of(action->task, PCMK_ACTION_STONITH, PCMK_ACTION_DO_SHUTDOWN, NULL)) { return true; } if (action->node == NULL) { pcmk__sched_err("Skipping action %s (%d) " "because it was not assigned to a node (bug?)", action->uuid, action->id); pcmk__log_action("Unassigned", action, false); return false; } if (pcmk_is_set(action->flags, pcmk_action_on_dc)) { crm_trace("Action %s (%d) should be dumped: " "can run on DC instead of %s", action->uuid, action->id, pe__node_name(action->node)); } else if (pe__is_guest_node(action->node) && !action->node->details->remote_requires_reset) { crm_trace("Action %s (%d) should be dumped: " "assuming will be runnable on guest %s", action->uuid, action->id, pe__node_name(action->node)); } else if (!action->node->details->online) { pcmk__sched_err("Skipping action %s (%d) " "because it was scheduled for offline node (bug?)", action->uuid, action->id); pcmk__log_action("Offline node", action, false); return false; } else if (action->node->details->unclean) { pcmk__sched_err("Skipping action %s (%d) " "because it was scheduled for unclean node (bug?)", action->uuid, action->id); pcmk__log_action("Unclean node", action, false); return false; } return true; } /*! * \internal * \brief Check whether an ordering's flags can change an action * * \param[in] ordering Ordering to check * * \return true if ordering has flags that can change an action, false otherwise */ static bool ordering_can_change_actions(const pcmk__related_action_t *ordering) { return pcmk_any_flags_set(ordering->type, ~(pcmk__ar_then_implies_first_graphed |pcmk__ar_first_implies_then_graphed |pcmk__ar_ordered)); } /*! * \internal * \brief Check whether an action input should be in the transition graph * * \param[in] action Action to check * \param[in,out] input Action input to check * * \return true if input should be in graph, false otherwise * \note This function may not only check an input, but disable it under certian * circumstances (load or anti-colocation orderings that are not needed). */ static bool should_add_input_to_graph(const pcmk_action_t *action, pcmk__related_action_t *input) { if (input->state == pe_link_dumped) { return true; } if ((uint32_t) input->type == pcmk__ar_none) { crm_trace("Ignoring %s (%d) input %s (%d): " "ordering disabled", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (!pcmk_is_set(input->action->flags, pcmk_action_runnable) && !ordering_can_change_actions(input)) { crm_trace("Ignoring %s (%d) input %s (%d): " "optional and input unrunnable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (!pcmk_is_set(input->action->flags, pcmk_action_runnable) && pcmk_is_set(input->type, pcmk__ar_min_runnable)) { crm_trace("Ignoring %s (%d) input %s (%d): " "minimum number of instances required but input unrunnable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (pcmk_is_set(input->type, pcmk__ar_unmigratable_then_blocks) && !pcmk_is_set(input->action->flags, pcmk_action_runnable)) { crm_trace("Ignoring %s (%d) input %s (%d): " "input blocked if 'then' unmigratable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (pcmk_is_set(input->type, pcmk__ar_if_first_unmigratable) && pcmk_is_set(input->action->flags, pcmk_action_migratable)) { crm_trace("Ignoring %s (%d) input %s (%d): ordering applies " "only if input is unmigratable, but it is migratable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (((uint32_t) input->type == pcmk__ar_ordered) && pcmk_is_set(input->action->flags, pcmk_action_migratable) && pcmk__ends_with(input->action->uuid, "_stop_0")) { crm_trace("Ignoring %s (%d) input %s (%d): " "optional but stop in migration", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if ((uint32_t) input->type == pcmk__ar_if_on_same_node_or_target) { pcmk_node_t *input_node = input->action->node; if ((action->rsc != NULL) && pcmk__str_eq(action->task, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)) { pcmk_node_t *assigned = action->rsc->allocated_to; /* For load_stopped -> migrate_to orderings, we care about where * the resource has been assigned, not where migrate_to will be * executed. */ if (!pe__same_node(input_node, assigned)) { crm_trace("Ignoring %s (%d) input %s (%d): " "migration target %s is not same as input node %s", action->uuid, action->id, input->action->uuid, input->action->id, (assigned? assigned->details->uname : ""), (input_node? input_node->details->uname : "")); input->type = (enum pe_ordering) pcmk__ar_none; return false; } } else if (!pe__same_node(input_node, action->node)) { crm_trace("Ignoring %s (%d) input %s (%d): " "not on same node (%s vs %s)", action->uuid, action->id, input->action->uuid, input->action->id, (action->node? action->node->details->uname : ""), (input_node? input_node->details->uname : "")); input->type = (enum pe_ordering) pcmk__ar_none; return false; } else if (pcmk_is_set(input->action->flags, pcmk_action_optional)) { crm_trace("Ignoring %s (%d) input %s (%d): " "ordering optional", action->uuid, action->id, input->action->uuid, input->action->id); input->type = (enum pe_ordering) pcmk__ar_none; return false; } } else if ((uint32_t) input->type == pcmk__ar_if_required_on_same_node) { if (input->action->node && action->node && !pe__same_node(input->action->node, action->node)) { crm_trace("Ignoring %s (%d) input %s (%d): " "not on same node (%s vs %s)", action->uuid, action->id, input->action->uuid, input->action->id, pe__node_name(action->node), pe__node_name(input->action->node)); input->type = (enum pe_ordering) pcmk__ar_none; return false; } else if (pcmk_is_set(input->action->flags, pcmk_action_optional)) { crm_trace("Ignoring %s (%d) input %s (%d): optional", action->uuid, action->id, input->action->uuid, input->action->id); input->type = (enum pe_ordering) pcmk__ar_none; return false; } } else if (input->action->rsc && input->action->rsc != action->rsc && pcmk_is_set(input->action->rsc->flags, pcmk_rsc_failed) && !pcmk_is_set(input->action->rsc->flags, pcmk_rsc_managed) && pcmk__ends_with(input->action->uuid, "_stop_0") && action->rsc && pe_rsc_is_clone(action->rsc)) { crm_warn("Ignoring requirement that %s complete before %s:" " unmanaged failed resources cannot prevent clone shutdown", input->action->uuid, action->uuid); return false; } else if (pcmk_is_set(input->action->flags, pcmk_action_optional) && !pcmk_any_flags_set(input->action->flags, pcmk_action_always_in_graph |pcmk_action_added_to_graph) && !should_add_action_to_graph(input->action)) { crm_trace("Ignoring %s (%d) input %s (%d): " "input optional", action->uuid, action->id, input->action->uuid, input->action->id); return false; } crm_trace("%s (%d) input %s %s (%d) on %s should be dumped: %s %s %#.6x", action->uuid, action->id, action_type_str(input->action->flags), input->action->uuid, input->action->id, action_node_str(input->action), action_runnable_str(input->action->flags), action_optional_str(input->action->flags), input->type); return true; } /*! * \internal * \brief Check whether an ordering creates an ordering loop * * \param[in] init_action "First" action in ordering * \param[in] action Callers should always set this the same as * \p init_action (this function may use a different * value for recursive calls) * \param[in,out] input Action wrapper for "then" action in ordering * * \return true if the ordering creates a loop, otherwise false */ bool pcmk__graph_has_loop(const pcmk_action_t *init_action, const pcmk_action_t *action, pcmk__related_action_t *input) { bool has_loop = false; if (pcmk_is_set(input->action->flags, pcmk_action_detect_loop)) { crm_trace("Breaking tracking loop: %s@%s -> %s@%s (%#.6x)", input->action->uuid, input->action->node? input->action->node->details->uname : "", action->uuid, action->node? action->node->details->uname : "", input->type); return false; } // Don't need to check inputs that won't be used if (!should_add_input_to_graph(action, input)) { return false; } if (input->action == init_action) { crm_debug("Input loop found in %s@%s ->...-> %s@%s", action->uuid, action->node? action->node->details->uname : "", init_action->uuid, init_action->node? init_action->node->details->uname : ""); return true; } pcmk__set_action_flags(input->action, pcmk_action_detect_loop); crm_trace("Checking inputs of action %s@%s input %s@%s (%#.6x)" "for graph loop with %s@%s ", action->uuid, action->node? action->node->details->uname : "", input->action->uuid, input->action->node? input->action->node->details->uname : "", input->type, init_action->uuid, init_action->node? init_action->node->details->uname : ""); // Recursively check input itself for loops for (GList *iter = input->action->actions_before; iter != NULL; iter = iter->next) { if (pcmk__graph_has_loop(init_action, input->action, (pcmk__related_action_t *) iter->data)) { // Recursive call already logged a debug message has_loop = true; break; } } pcmk__clear_action_flags(input->action, pcmk_action_detect_loop); if (!has_loop) { crm_trace("No input loop found in %s@%s -> %s@%s (%#.6x)", input->action->uuid, input->action->node? input->action->node->details->uname : "", action->uuid, action->node? action->node->details->uname : "", input->type); } return has_loop; } /*! * \internal * \brief Create a synapse XML element for a transition graph * * \param[in] action Action that synapse is for * \param[in,out] scheduler Scheduler data containing graph * * \return Newly added XML element for new graph synapse */ static xmlNode * create_graph_synapse(const pcmk_action_t *action, pcmk_scheduler_t *scheduler) { int synapse_priority = 0; xmlNode *syn = create_xml_node(scheduler->graph, "synapse"); crm_xml_add_int(syn, PCMK_XA_ID, scheduler->num_synapse); scheduler->num_synapse++; if (action->rsc != NULL) { synapse_priority = action->rsc->priority; } if (action->priority > synapse_priority) { synapse_priority = action->priority; } if (synapse_priority > 0) { crm_xml_add_int(syn, XML_CIB_ATTR_PRIORITY, synapse_priority); } return syn; } /*! * \internal * \brief Add an action to the transition graph XML if appropriate * * \param[in,out] data Action to possibly add * \param[in,out] user_data Scheduler data * * \note This will de-duplicate the action inputs, meaning that the * pcmk__related_action_t:type flags can no longer be relied on to retain * their original settings. That means this MUST be called after * pcmk__apply_orderings() is complete, and nothing after this should rely * on those type flags. (For example, some code looks for type equal to * some flag rather than whether the flag is set, and some code looks for * particular combinations of flags -- such code must be done before * pcmk__create_graph().) */ static void add_action_to_graph(gpointer data, gpointer user_data) { pcmk_action_t *action = (pcmk_action_t *) data; pcmk_scheduler_t *scheduler = (pcmk_scheduler_t *) user_data; xmlNode *syn = NULL; xmlNode *set = NULL; xmlNode *in = NULL; /* If we haven't already, de-duplicate inputs (even if we won't be adding * the action to the graph, so that crm_simulate's dot graphs don't have * duplicates). */ if (!pcmk_is_set(action->flags, pcmk_action_inputs_deduplicated)) { pcmk__deduplicate_action_inputs(action); pcmk__set_action_flags(action, pcmk_action_inputs_deduplicated); } if (pcmk_is_set(action->flags, pcmk_action_added_to_graph) || !should_add_action_to_graph(action)) { return; // Already added, or shouldn't be } pcmk__set_action_flags(action, pcmk_action_added_to_graph); crm_trace("Adding action %d (%s%s%s) to graph", action->id, action->uuid, ((action->node == NULL)? "" : " on "), ((action->node == NULL)? "" : action->node->details->uname)); syn = create_graph_synapse(action, scheduler); set = create_xml_node(syn, "action_set"); in = create_xml_node(syn, "inputs"); create_graph_action(set, action, false, scheduler); for (GList *lpc = action->actions_before; lpc != NULL; lpc = lpc->next) { pcmk__related_action_t *input = lpc->data; if (should_add_input_to_graph(action, input)) { xmlNode *input_xml = create_xml_node(in, "trigger"); input->state = pe_link_dumped; create_graph_action(input_xml, input->action, true, scheduler); } } } static int transition_id = -1; /*! * \internal * \brief Log a message after calculating a transition * * \param[in] filename Where transition input is stored */ void pcmk__log_transition_summary(const char *filename) { if (was_processing_error || crm_config_error) { crm_err("Calculated transition %d (with errors)%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } else if (was_processing_warning || crm_config_warning) { crm_warn("Calculated transition %d (with warnings)%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } else { crm_notice("Calculated transition %d%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } if (crm_config_error) { crm_notice("Configuration errors found during scheduler processing," " please run \"crm_verify -L\" to identify issues"); } } /*! * \internal * \brief Add a resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void pcmk__add_rsc_actions_to_graph(pcmk_resource_t *rsc) { GList *iter = NULL; CRM_ASSERT(rsc != NULL); pcmk__rsc_trace(rsc, "Adding actions for %s to graph", rsc->id); // First add the resource's own actions g_list_foreach(rsc->actions, add_action_to_graph, rsc->cluster); // Then recursively add its children's actions (appropriate to variant) for (iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) iter->data; child_rsc->cmds->add_actions_to_graph(child_rsc); } } /*! * \internal * \brief Create a transition graph with all cluster actions needed * * \param[in,out] scheduler Scheduler data */ void pcmk__create_graph(pcmk_scheduler_t *scheduler) { GList *iter = NULL; const char *value = NULL; long long limit = 0LL; transition_id++; crm_trace("Creating transition graph %d", transition_id); scheduler->graph = create_xml_node(NULL, XML_TAG_GRAPH); value = pe_pref(scheduler->config_hash, PCMK_OPT_CLUSTER_DELAY); crm_xml_add(scheduler->graph, PCMK_OPT_CLUSTER_DELAY, value); value = pe_pref(scheduler->config_hash, PCMK_OPT_STONITH_TIMEOUT); crm_xml_add(scheduler->graph, PCMK_OPT_STONITH_TIMEOUT, value); crm_xml_add(scheduler->graph, "failed-stop-offset", "INFINITY"); if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) { crm_xml_add(scheduler->graph, "failed-start-offset", "INFINITY"); } else { crm_xml_add(scheduler->graph, "failed-start-offset", "1"); } value = pe_pref(scheduler->config_hash, PCMK_OPT_BATCH_LIMIT); crm_xml_add(scheduler->graph, PCMK_OPT_BATCH_LIMIT, value); crm_xml_add_int(scheduler->graph, "transition_id", transition_id); value = pe_pref(scheduler->config_hash, PCMK_OPT_MIGRATION_LIMIT); if ((pcmk__scan_ll(value, &limit, 0LL) == pcmk_rc_ok) && (limit > 0)) { crm_xml_add(scheduler->graph, PCMK_OPT_MIGRATION_LIMIT, value); } if (scheduler->recheck_by > 0) { char *recheck_epoch = NULL; recheck_epoch = crm_strdup_printf("%llu", (long long) scheduler->recheck_by); crm_xml_add(scheduler->graph, "recheck-by", recheck_epoch); free(recheck_epoch); } /* The following code will de-duplicate action inputs, so nothing past this * should rely on the action input type flags retaining their original * values. */ // Add resource actions to graph for (iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; pcmk__rsc_trace(rsc, "Processing actions for %s", rsc->id); rsc->cmds->add_actions_to_graph(rsc); } // Add pseudo-action for list of nodes with maintenance state update add_maintenance_update(scheduler); // Add non-resource (node) actions for (iter = scheduler->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = (pcmk_action_t *) iter->data; if ((action->rsc != NULL) && (action->node != NULL) && action->node->details->shutdown && !pcmk_is_set(action->rsc->flags, pcmk_rsc_maintenance) && !pcmk_any_flags_set(action->flags, pcmk_action_optional|pcmk_action_runnable) && pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_none)) { /* Eventually we should just ignore the 'fence' case, but for now * it's the best way to detect (in CTS) when CIB resource updates * are being lost. */ if (pcmk_is_set(scheduler->flags, pcmk_sched_quorate) || (scheduler->no_quorum_policy == pcmk_no_quorum_ignore)) { const bool managed = pcmk_is_set(action->rsc->flags, pcmk_rsc_managed); const bool failed = pcmk_is_set(action->rsc->flags, pcmk_rsc_failed); crm_crit("Cannot %s %s because of %s:%s%s (%s)", action->node->details->unclean? "fence" : "shut down", pe__node_name(action->node), action->rsc->id, (managed? " blocked" : " unmanaged"), (failed? " failed" : ""), action->uuid); } } add_action_to_graph((gpointer) action, (gpointer) scheduler); } crm_log_xml_trace(scheduler->graph, "graph"); } diff --git a/lib/pacemaker/pcmk_simulate.c b/lib/pacemaker/pcmk_simulate.c index 75b9d3bd6c..a69ac3310a 100644 --- a/lib/pacemaker/pcmk_simulate.c +++ b/lib/pacemaker/pcmk_simulate.c @@ -1,1006 +1,1006 @@ /* * Copyright 2021-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include "libpacemaker_private.h" static pcmk__output_t *out = NULL; static cib_t *fake_cib = NULL; static GList *fake_resource_list = NULL; static const GList *fake_op_fail_list = NULL; static void set_effective_date(pcmk_scheduler_t *scheduler, bool print_original, const char *use_date); /*! * \internal * \brief Create an action name for use in a dot graph * * \param[in] action Action to create name for * \param[in] verbose If true, add action ID to name * * \return Newly allocated string with action name * \note It is the caller's responsibility to free the result. */ static char * create_action_name(const pcmk_action_t *action, bool verbose) { char *action_name = NULL; const char *prefix = ""; const char *action_host = NULL; const char *clone_name = NULL; const char *task = action->task; if (action->node != NULL) { action_host = action->node->details->uname; } else if (!pcmk_is_set(action->flags, pcmk_action_pseudo)) { action_host = ""; } if (pcmk__str_eq(action->task, PCMK_ACTION_CANCEL, pcmk__str_none)) { prefix = "Cancel "; task = action->cancel_task; } if (action->rsc != NULL) { clone_name = action->rsc->clone_name; } if (clone_name != NULL) { char *key = NULL; guint interval_ms = 0; if (pcmk__guint_from_hash(action->meta, XML_LRM_ATTR_INTERVAL_MS, 0, &interval_ms) != pcmk_rc_ok) { interval_ms = 0; } if (pcmk__strcase_any_of(action->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_NOTIFIED, NULL)) { const char *n_type = g_hash_table_lookup(action->meta, "notify_key_type"); const char *n_task = g_hash_table_lookup(action->meta, "notify_key_operation"); CRM_ASSERT(n_type != NULL); CRM_ASSERT(n_task != NULL); key = pcmk__notify_key(clone_name, n_type, n_task); } else { key = pcmk__op_key(clone_name, task, interval_ms); } if (action_host != NULL) { action_name = crm_strdup_printf("%s%s %s", prefix, key, action_host); } else { action_name = crm_strdup_printf("%s%s", prefix, key); } free(key); } else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { const char *op = g_hash_table_lookup(action->meta, "stonith_action"); action_name = crm_strdup_printf("%s%s '%s' %s", prefix, action->task, op, action_host); } else if (action->rsc && action_host) { action_name = crm_strdup_printf("%s%s %s", prefix, action->uuid, action_host); } else if (action_host) { action_name = crm_strdup_printf("%s%s %s", prefix, action->task, action_host); } else { action_name = crm_strdup_printf("%s", action->uuid); } if (verbose) { char *with_id = crm_strdup_printf("%s (%d)", action_name, action->id); free(action_name); action_name = with_id; } return action_name; } /*! * \internal * \brief Display the status of a cluster * * \param[in,out] scheduler Scheduler data * \param[in] show_opts How to modify display (as pcmk_show_opt_e flags) * \param[in] section_opts Sections to display (as pcmk_section_e flags) * \param[in] title What to use as list title * \param[in] print_spacer Whether to display a spacer first */ static void print_cluster_status(pcmk_scheduler_t *scheduler, uint32_t show_opts, uint32_t section_opts, const char *title, bool print_spacer) { pcmk__output_t *out = scheduler->priv; GList *all = NULL; crm_exit_t stonith_rc = 0; enum pcmk_pacemakerd_state state = pcmk_pacemakerd_state_invalid; section_opts |= pcmk_section_nodes | pcmk_section_resources; show_opts |= pcmk_show_inactive_rscs | pcmk_show_failed_detail; all = g_list_prepend(all, (gpointer) "*"); PCMK__OUTPUT_SPACER_IF(out, print_spacer); out->begin_list(out, NULL, NULL, "%s", title); out->message(out, "cluster-status", scheduler, state, stonith_rc, NULL, false, section_opts, show_opts, NULL, all, all); out->end_list(out); g_list_free(all); } /*! * \internal * \brief Display a summary of all actions scheduled in a transition * * \param[in,out] scheduler Scheduler data (fully scheduled) * \param[in] print_spacer Whether to display a spacer first */ static void print_transition_summary(pcmk_scheduler_t *scheduler, bool print_spacer) { pcmk__output_t *out = scheduler->priv; PCMK__OUTPUT_SPACER_IF(out, print_spacer); out->begin_list(out, NULL, NULL, "Transition Summary"); pcmk__output_actions(scheduler); out->end_list(out); } /*! * \internal * \brief Reset scheduler input, output, date, and flags * * \param[in,out] scheduler Scheduler data * \param[in] input What to set as cluster input * \param[in] out What to set as cluster output object * \param[in] use_date What to set as cluster's current timestamp * \param[in] flags Group of enum pcmk_scheduler_flags to set */ static void reset(pcmk_scheduler_t *scheduler, xmlNodePtr input, pcmk__output_t *out, const char *use_date, unsigned int flags) { scheduler->input = input; scheduler->priv = out; set_effective_date(scheduler, true, use_date); if (pcmk_is_set(flags, pcmk_sim_sanitized)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_sanitized); } if (pcmk_is_set(flags, pcmk_sim_show_scores)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_output_scores); } if (pcmk_is_set(flags, pcmk_sim_show_utilization)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_show_utilization); } } /*! * \brief Write out a file in dot(1) format describing the actions that will * be taken by the scheduler in response to an input CIB file. * * \param[in,out] scheduler Scheduler data * \param[in] dot_file The filename to write * \param[in] all_actions Write all actions, even those that are optional * or are on unmanaged resources * \param[in] verbose Add extra information, such as action IDs, to the * output * * \return Standard Pacemaker return code */ static int write_sim_dotfile(pcmk_scheduler_t *scheduler, const char *dot_file, bool all_actions, bool verbose) { GList *iter = NULL; FILE *dot_strm = fopen(dot_file, "w"); if (dot_strm == NULL) { return errno; } fprintf(dot_strm, " digraph \"g\" {\n"); for (iter = scheduler->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = (pcmk_action_t *) iter->data; const char *style = "dashed"; const char *font = "black"; const char *color = "black"; char *action_name = create_action_name(action, verbose); if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { font = "orange"; } if (pcmk_is_set(action->flags, pcmk_action_added_to_graph)) { style = "bold"; color = "green"; } else if ((action->rsc != NULL) && !pcmk_is_set(action->rsc->flags, pcmk_rsc_managed)) { color = "red"; font = "purple"; if (!all_actions) { goto do_not_write; } } else if (pcmk_is_set(action->flags, pcmk_action_optional)) { color = "blue"; if (!all_actions) { goto do_not_write; } } else { color = "red"; CRM_LOG_ASSERT(!pcmk_is_set(action->flags, pcmk_action_runnable)); } pcmk__set_action_flags(action, pcmk_action_added_to_graph); fprintf(dot_strm, "\"%s\" [ style=%s color=\"%s\" fontcolor=\"%s\"]\n", action_name, style, color, font); do_not_write: free(action_name); } for (iter = scheduler->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = (pcmk_action_t *) iter->data; for (GList *before_iter = action->actions_before; before_iter != NULL; before_iter = before_iter->next) { pcmk__related_action_t *before = before_iter->data; char *before_name = NULL; char *after_name = NULL; const char *style = "dashed"; bool optional = true; if (before->state == pe_link_dumped) { optional = false; style = "bold"; } else if ((uint32_t) before->type == pcmk__ar_none) { continue; } else if (pcmk_is_set(before->action->flags, pcmk_action_added_to_graph) && pcmk_is_set(action->flags, pcmk_action_added_to_graph) && (uint32_t) before->type != pcmk__ar_if_on_same_node_or_target) { optional = false; } if (all_actions || !optional) { before_name = create_action_name(before->action, verbose); after_name = create_action_name(action, verbose); fprintf(dot_strm, "\"%s\" -> \"%s\" [ style = %s]\n", before_name, after_name, style); free(before_name); free(after_name); } } } fprintf(dot_strm, "}\n"); fflush(dot_strm); fclose(dot_strm); return pcmk_rc_ok; } /*! * \brief Profile the configuration updates and scheduler actions in a single * CIB file, printing the profiling timings. * * \note \p scheduler->priv must have been set to a valid \p pcmk__output_t * object before this function is called. * * \param[in] xml_file The CIB file to profile * \param[in] repeat Number of times to run * \param[in,out] scheduler Scheduler data * \param[in] use_date The date to set the cluster's time to (may be NULL) */ static void profile_file(const char *xml_file, long long repeat, pcmk_scheduler_t *scheduler, const char *use_date) { pcmk__output_t *out = scheduler->priv; xmlNode *cib_object = NULL; clock_t start = 0; clock_t end; unsigned long long scheduler_flags = pcmk_sched_no_compat; CRM_ASSERT(out != NULL); cib_object = filename2xml(xml_file); start = clock(); if (pcmk_find_cib_element(cib_object, 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; } if (pcmk_is_set(scheduler->flags, pcmk_sched_output_scores)) { scheduler_flags |= pcmk_sched_output_scores; } if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) { scheduler_flags |= pcmk_sched_show_utilization; } for (int i = 0; i < repeat; ++i) { xmlNode *input = (repeat == 1)? cib_object : copy_xml(cib_object); scheduler->input = input; set_effective_date(scheduler, false, use_date); pcmk__schedule_actions(input, scheduler_flags, scheduler); pe_reset_working_set(scheduler); } end = clock(); out->message(out, "profile", xml_file, start, end); } void pcmk__profile_dir(const char *dir, long long repeat, pcmk_scheduler_t *scheduler, const char *use_date) { pcmk__output_t *out = scheduler->priv; struct dirent **namelist; int file_num = scandir(dir, &namelist, 0, alphasort); CRM_ASSERT(out != NULL); if (file_num > 0) { struct stat prop; char buffer[FILENAME_MAX]; out->begin_list(out, NULL, NULL, "Timings"); while (file_num--) { if ('.' == namelist[file_num]->d_name[0]) { free(namelist[file_num]); continue; } else if (!pcmk__ends_with_ext(namelist[file_num]->d_name, ".xml")) { free(namelist[file_num]); continue; } snprintf(buffer, sizeof(buffer), "%s/%s", dir, namelist[file_num]->d_name); if (stat(buffer, &prop) == 0 && S_ISREG(prop.st_mode)) { profile_file(buffer, repeat, scheduler, use_date); } free(namelist[file_num]); } free(namelist); out->end_list(out); } } /*! * \brief Set the date of the cluster, either to the value given by * \p use_date, or to the "execution-date" value in the CIB. * * \note \p scheduler->priv must have been set to a valid \p pcmk__output_t * object before this function is called. * * \param[in,out] scheduler Scheduler data * \param[in] print_original If \p true, the "execution-date" should * also be printed * \param[in] use_date The date to set the cluster's time to * (may be NULL) */ static void set_effective_date(pcmk_scheduler_t *scheduler, bool print_original, const char *use_date) { pcmk__output_t *out = scheduler->priv; time_t original_date = 0; CRM_ASSERT(out != NULL); crm_element_value_epoch(scheduler->input, "execution-date", &original_date); if (use_date) { scheduler->now = crm_time_new(use_date); out->info(out, "Setting effective cluster time: %s", use_date); crm_time_log(LOG_NOTICE, "Pretending 'now' is", scheduler->now, crm_time_log_date | crm_time_log_timeofday); } else if (original_date != 0) { scheduler->now = pcmk__copy_timet(original_date); if (print_original) { char *when = crm_time_as_string(scheduler->now, crm_time_log_date|crm_time_log_timeofday); out->info(out, "Using the original execution date of: %s", when); free(when); } } } /*! * \internal * \brief Simulate successfully executing a pseudo-action in a graph * * \param[in,out] graph Graph to update with pseudo-action result * \param[in,out] action Pseudo-action to simulate executing * * \return Standard Pacemaker return code */ static int simulate_pseudo_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); out->message(out, "inject-pseudo-action", node, task); pcmk__update_graph(graph, action); return pcmk_rc_ok; } /*! * \internal * \brief Simulate executing a resource action in a graph * * \param[in,out] graph Graph to update with resource action result * \param[in,out] action Resource action to simulate executing * * \return Standard Pacemaker return code */ static int simulate_resource_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { int rc; lrmd_event_data_t *op = NULL; int target_outcome = PCMK_OCF_OK; const char *rtype = NULL; const char *rclass = NULL; const char *resource = NULL; const char *rprovider = NULL; const char *resource_config_name = NULL; const char *operation = crm_element_value(action->xml, "operation"); const char *target_rc_s = crm_meta_value(action->params, XML_ATTR_TE_TARGET_RC); xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; xmlNode *action_rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE); char *node = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET); char *uuid = NULL; const char *router_node = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); // Certain actions don't need to be displayed or history entries if (pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_none)) { crm_debug("No history injection for %s op on %s", operation, node); goto done; // Confirm action and update graph } if (action_rsc == NULL) { // Shouldn't be possible crm_log_xml_err(action->xml, "Bad"); free(node); return EPROTO; } /* A resource might be known by different names in the configuration and in * the action (for example, a clone instance). Grab the configuration name * (which is preferred when writing history), and if necessary, the instance * name. */ resource_config_name = crm_element_value(action_rsc, PCMK_XA_ID); if (resource_config_name == NULL) { // Shouldn't be possible crm_log_xml_err(action->xml, "No ID"); free(node); return EPROTO; } resource = resource_config_name; if (pe_find_resource(fake_resource_list, resource) == NULL) { - const char *longname = crm_element_value(action_rsc, XML_ATTR_ID_LONG); + const char *longname = crm_element_value(action_rsc, PCMK__XA_LONG_ID); if ((longname != NULL) && (pe_find_resource(fake_resource_list, longname) != NULL)) { resource = longname; } } // Certain actions need to be displayed but don't need history entries if (pcmk__strcase_any_of(operation, PCMK_ACTION_DELETE, PCMK_ACTION_META_DATA, NULL)) { out->message(out, "inject-rsc-action", resource, operation, node, (guint) 0); goto done; // Confirm action and update graph } rclass = crm_element_value(action_rsc, XML_AGENT_ATTR_CLASS); rtype = crm_element_value(action_rsc, XML_ATTR_TYPE); rprovider = crm_element_value(action_rsc, XML_AGENT_ATTR_PROVIDER); pcmk__scan_min_int(target_rc_s, &target_outcome, 0); CRM_ASSERT(fake_cib->cmds->query(fake_cib, NULL, NULL, cib_sync_call|cib_scope_local) == pcmk_ok); // Ensure the action node is in the CIB uuid = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET_UUID); cib_node = pcmk__inject_node(fake_cib, node, ((router_node == NULL)? uuid: node)); free(uuid); CRM_ASSERT(cib_node != NULL); // Add a history entry for the action cib_resource = pcmk__inject_resource_history(out, cib_node, resource, resource_config_name, rclass, rtype, rprovider); if (cib_resource == NULL) { crm_err("Could not simulate action %d history for resource %s", action->id, resource); free(node); free_xml(cib_node); return EINVAL; } // Simulate and display an executor event for the action result op = pcmk__event_from_graph_action(cib_resource, action, PCMK_EXEC_DONE, target_outcome, "User-injected result"); out->message(out, "inject-rsc-action", resource, op->op_type, node, op->interval_ms); // Check whether action is in a list of desired simulated failures for (const GList *iter = fake_op_fail_list; iter != NULL; iter = iter->next) { const char *spec = (const char *) iter->data; char *key = NULL; const char *match_name = NULL; // Allow user to specify anonymous clone with or without instance number key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = resource; } free(key); // If not found, try the resource's name in the configuration if ((match_name == NULL) && (strcmp(resource, resource_config_name) != 0)) { key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource_config_name, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = resource_config_name; } free(key); } if (match_name == NULL) { continue; // This failed action entry doesn't match } // ${match_name}_${task}_${interval_in_ms}@${node}=${rc} rc = sscanf(spec, "%*[^=]=%d", (int *) &op->rc); if (rc != 1) { out->err(out, "Invalid failed operation '%s' " "(result code must be integer)", spec); continue; // Keep checking other list entries } out->info(out, "Pretending action %d failed with rc=%d", action->id, op->rc); pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); graph->abort_priority = INFINITY; pcmk__inject_failcount(out, cib_node, match_name, op->op_type, op->interval_ms, op->rc); break; } pcmk__inject_action_result(cib_resource, op, target_outcome); lrmd_free_event(op); rc = fake_cib->cmds->modify(fake_cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); done: free(node); free_xml(cib_node); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); pcmk__update_graph(graph, action); return pcmk_rc_ok; } /*! * \internal * \brief Simulate successfully executing a cluster action * * \param[in,out] graph Graph to update with action result * \param[in,out] action Cluster action to simulate * * \return Standard Pacemaker return code */ static int simulate_cluster_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); xmlNode *rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); out->message(out, "inject-cluster-action", node, task, rsc); pcmk__update_graph(graph, action); return pcmk_rc_ok; } /*! * \internal * \brief Simulate successfully executing a fencing action * * \param[in,out] graph Graph to update with action result * \param[in,out] action Fencing action to simulate * * \return Standard Pacemaker return code */ static int simulate_fencing_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *op = crm_meta_value(action->params, "stonith_action"); char *target = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET); out->message(out, "inject-fencing-action", target, op); if (!pcmk__str_eq(op, PCMK_ACTION_ON, pcmk__str_casei)) { int rc = pcmk_ok; GString *xpath = g_string_sized_new(512); // Set node state to offline xmlNode *cib_node = pcmk__inject_node_state_change(fake_cib, target, false); CRM_ASSERT(cib_node != NULL); crm_xml_add(cib_node, PCMK_XA_CRM_DEBUG_ORIGIN, __func__); rc = fake_cib->cmds->replace(fake_cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); // Simulate controller clearing node's resource history and attributes pcmk__g_strcat(xpath, "//" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='", target, "']/" XML_CIB_TAG_LRM, NULL); fake_cib->cmds->remove(fake_cib, (const char *) xpath->str, NULL, cib_xpath|cib_sync_call|cib_scope_local); g_string_truncate(xpath, 0); pcmk__g_strcat(xpath, "//" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='", target, "']" "/" XML_TAG_TRANSIENT_NODEATTRS, NULL); fake_cib->cmds->remove(fake_cib, (const char *) xpath->str, NULL, cib_xpath|cib_sync_call|cib_scope_local); free_xml(cib_node); g_string_free(xpath, TRUE); } pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); pcmk__update_graph(graph, action); free(target); return pcmk_rc_ok; } enum pcmk__graph_status pcmk__simulate_transition(pcmk_scheduler_t *scheduler, cib_t *cib, const GList *op_fail_list) { pcmk__graph_t *transition = NULL; enum pcmk__graph_status graph_rc; pcmk__graph_functions_t simulation_fns = { simulate_pseudo_action, simulate_resource_action, simulate_cluster_action, simulate_fencing_action, }; out = scheduler->priv; fake_cib = cib; fake_op_fail_list = op_fail_list; if (!out->is_quiet(out)) { out->begin_list(out, NULL, NULL, "Executing Cluster Transition"); } pcmk__set_graph_functions(&simulation_fns); transition = pcmk__unpack_graph(scheduler->graph, crm_system_name); pcmk__log_graph(LOG_DEBUG, transition); fake_resource_list = scheduler->resources; do { graph_rc = pcmk__execute_graph(transition); } while (graph_rc == pcmk__graph_active); fake_resource_list = NULL; if (graph_rc != pcmk__graph_complete) { out->err(out, "Transition failed: %s", pcmk__graph_status2text(graph_rc)); pcmk__log_graph(LOG_ERR, transition); out->err(out, "An invalid transition was produced"); } pcmk__free_graph(transition); if (!out->is_quiet(out)) { // If not quiet, we'll need the resulting CIB for later display xmlNode *cib_object = NULL; int rc = fake_cib->cmds->query(fake_cib, NULL, &cib_object, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); pe_reset_working_set(scheduler); scheduler->input = cib_object; out->end_list(out); } return graph_rc; } int pcmk__simulate(pcmk_scheduler_t *scheduler, pcmk__output_t *out, const pcmk_injections_t *injections, unsigned int flags, uint32_t section_opts, const char *use_date, const char *input_file, const char *graph_file, const char *dot_file) { int printed = pcmk_rc_no_output; int rc = pcmk_rc_ok; xmlNodePtr input = NULL; cib_t *cib = NULL; rc = cib__signon_query(out, &cib, &input); if (rc != pcmk_rc_ok) { goto simulate_done; } reset(scheduler, input, out, use_date, flags); cluster_status(scheduler); if ((cib->variant == cib_native) && pcmk_is_set(section_opts, pcmk_section_times)) { if (pcmk__our_nodename == NULL) { // Currently used only in the times section pcmk__query_node_name(out, 0, &pcmk__our_nodename, 0); } scheduler->localhost = pcmk__our_nodename; } if (!out->is_quiet(out)) { const bool show_pending = pcmk_is_set(flags, pcmk_sim_show_pending); if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) { printed = out->message(out, "maint-mode", scheduler->flags); } if (scheduler->disabled_resources || scheduler->blocked_resources) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); printed = out->info(out, "%d of %d resource instances DISABLED and " "%d BLOCKED from further action due to failure", scheduler->disabled_resources, scheduler->ninstances, scheduler->blocked_resources); } /* Most formatted output headers use caps for each word, but this one * only has the first word capitalized for compatibility with pcs. */ print_cluster_status(scheduler, (show_pending? pcmk_show_pending : 0), section_opts, "Current cluster status", (printed == pcmk_rc_ok)); printed = pcmk_rc_ok; } // If the user requested any injections, handle them if ((injections->node_down != NULL) || (injections->node_fail != NULL) || (injections->node_up != NULL) || (injections->op_inject != NULL) || (injections->ticket_activate != NULL) || (injections->ticket_grant != NULL) || (injections->ticket_revoke != NULL) || (injections->ticket_standby != NULL) || (injections->watchdog != NULL)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); pcmk__inject_scheduler_input(scheduler, cib, injections); printed = pcmk_rc_ok; rc = cib->cmds->query(cib, NULL, &input, cib_sync_call); if (rc != pcmk_rc_ok) { rc = pcmk_legacy2rc(rc); goto simulate_done; } cleanup_calculations(scheduler); reset(scheduler, input, out, use_date, flags); cluster_status(scheduler); } if (input_file != NULL) { rc = write_xml_file(input, input_file, FALSE); if (rc < 0) { rc = pcmk_legacy2rc(rc); goto simulate_done; } } if (pcmk_any_flags_set(flags, pcmk_sim_process | pcmk_sim_simulate)) { pcmk__output_t *logger_out = NULL; unsigned long long scheduler_flags = pcmk_sched_no_compat; if (pcmk_is_set(scheduler->flags, pcmk_sched_output_scores)) { scheduler_flags |= pcmk_sched_output_scores; } if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) { scheduler_flags |= pcmk_sched_show_utilization; } if (pcmk_all_flags_set(scheduler->flags, pcmk_sched_output_scores |pcmk_sched_show_utilization)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Assignment Scores and Utilization Information"); printed = pcmk_rc_ok; } else if (pcmk_is_set(scheduler->flags, pcmk_sched_output_scores)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Assignment Scores"); printed = pcmk_rc_ok; } else if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Utilization Information"); printed = pcmk_rc_ok; } else { rc = pcmk__log_output_new(&logger_out); if (rc != pcmk_rc_ok) { goto simulate_done; } pe__register_messages(logger_out); pcmk__register_lib_messages(logger_out); scheduler->priv = logger_out; } pcmk__schedule_actions(input, scheduler_flags, scheduler); if (logger_out == NULL) { out->end_list(out); } else { logger_out->finish(logger_out, CRM_EX_OK, true, NULL); pcmk__output_free(logger_out); scheduler->priv = out; } input = NULL; /* Don't try and free it twice */ if (graph_file != NULL) { rc = write_xml_file(scheduler->graph, graph_file, FALSE); if (rc < 0) { rc = pcmk_rc_graph_error; goto simulate_done; } } if (dot_file != NULL) { rc = write_sim_dotfile(scheduler, dot_file, pcmk_is_set(flags, pcmk_sim_all_actions), pcmk_is_set(flags, pcmk_sim_verbose)); if (rc != pcmk_rc_ok) { rc = pcmk_rc_dot_error; goto simulate_done; } } if (!out->is_quiet(out)) { print_transition_summary(scheduler, printed == pcmk_rc_ok); } } rc = pcmk_rc_ok; if (!pcmk_is_set(flags, pcmk_sim_simulate)) { goto simulate_done; } PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); if (pcmk__simulate_transition(scheduler, cib, injections->op_fail) != pcmk__graph_complete) { rc = pcmk_rc_invalid_transition; } if (out->is_quiet(out)) { goto simulate_done; } set_effective_date(scheduler, true, use_date); if (pcmk_is_set(flags, pcmk_sim_show_scores)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_output_scores); } if (pcmk_is_set(flags, pcmk_sim_show_utilization)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_show_utilization); } cluster_status(scheduler); print_cluster_status(scheduler, 0, section_opts, "Revised Cluster Status", true); simulate_done: cib__clean_up_connection(&cib); return rc; } int pcmk_simulate(xmlNodePtr *xml, pcmk_scheduler_t *scheduler, const pcmk_injections_t *injections, unsigned int flags, unsigned int section_opts, const char *use_date, const char *input_file, const char *graph_file, const char *dot_file) { pcmk__output_t *out = NULL; int rc = pcmk_rc_ok; rc = pcmk__xml_output_new(&out, xml); if (rc != pcmk_rc_ok) { return rc; } pe__register_messages(out); pcmk__register_lib_messages(out); rc = pcmk__simulate(scheduler, out, injections, flags, section_opts, use_date, input_file, graph_file, dot_file); pcmk__xml_output_finish(out, pcmk_rc2exitc(rc), xml); return rc; }