diff --git a/daemons/controld/controld_membership.c b/daemons/controld/controld_membership.c index eaee71e8b2..f7a2b17841 100644 --- a/daemons/controld/controld_membership.c +++ b/daemons/controld/controld_membership.c @@ -1,468 +1,468 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ /* put these first so that uuid_t is defined without conflicts */ #include #include #include #include #include #include #include #include void post_cache_update(int instance); extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); static void reap_dead_nodes(gpointer key, gpointer value, gpointer user_data) { crm_node_t *node = value; if (crm_is_peer_active(node) == FALSE) { crm_update_peer_join(__func__, node, crm_join_none); if(node && node->uname) { if (pcmk__str_eq(controld_globals.our_nodename, node->uname, pcmk__str_casei)) { crm_err("We're not part of the cluster anymore"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); } else if (!AM_I_DC && pcmk__str_eq(node->uname, controld_globals.dc_name, pcmk__str_casei)) { crm_warn("Our DC node (%s) left the cluster", node->uname); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } } if ((controld_globals.fsa_state == S_INTEGRATION) || (controld_globals.fsa_state == S_FINALIZE_JOIN)) { check_join_state(controld_globals.fsa_state, __func__); } if ((node != NULL) && (node->uuid != NULL)) { fail_incompletable_actions(controld_globals.transition_graph, node->uuid); } } } void post_cache_update(int instance) { xmlNode *no_op = NULL; crm_peer_seq = instance; crm_debug("Updated cache after membership event %d.", instance); g_hash_table_foreach(crm_peer_cache, reap_dead_nodes, NULL); controld_set_fsa_input_flags(R_MEMBERSHIP); if (AM_I_DC) { populate_cib_nodes(node_update_quick | node_update_cluster | node_update_peer | node_update_expected, __func__); } /* * If we lost nodes, we should re-check the election status * Safe to call outside of an election */ controld_set_fsa_action_flags(A_ELECTION_CHECK); controld_trigger_fsa(); /* Membership changed, remind everyone we're here. * This will aid detection of duplicate DCs */ no_op = create_request(CRM_OP_NOOP, NULL, NULL, CRM_SYSTEM_CRMD, AM_I_DC ? CRM_SYSTEM_DC : CRM_SYSTEM_CRMD, NULL); send_cluster_message(NULL, crm_msg_crmd, no_op, FALSE); free_xml(no_op); } static void crmd_node_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if (rc == pcmk_ok) { crm_trace("Node update %d complete", call_id); } else if(call_id < pcmk_ok) { crm_err("Node update failed: %s (%d)", pcmk_strerror(call_id), call_id); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else { crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } /*! * \internal * \brief Create an XML node state tag with updates * * \param[in,out] node Node whose state will be used for update * \param[in] flags Bitmask of node_update_flags indicating what to update * \param[in,out] parent XML node to contain update (or NULL) * \param[in] source Who requested the update (only used for logging) * * \return Pointer to created node state tag */ xmlNode * create_node_state_update(crm_node_t *node, int flags, xmlNode *parent, const char *source) { const char *value = NULL; xmlNode *node_state; if (!node->state) { crm_info("Node update for %s cancelled: no state, not seen yet", node->uname); return NULL; } node_state = create_xml_node(parent, PCMK__XE_NODE_STATE); if (pcmk_is_set(node->flags, crm_remote_node)) { pcmk__xe_set_bool_attr(node_state, PCMK_XA_REMOTE_NODE, true); } if (crm_xml_add(node_state, PCMK_XA_ID, crm_peer_uuid(node)) == NULL) { crm_info("Node update for %s cancelled: no ID", node->uname); free_xml(node_state); return NULL; } crm_xml_add(node_state, PCMK_XA_UNAME, node->uname); if ((flags & node_update_cluster) && node->state) { if (compare_version(controld_globals.dc_version, "3.18.0") >= 0) { // A value 0 means the node is not a cluster member. crm_xml_add_ll(node_state, PCMK__XA_IN_CCM, node->when_member); } else { pcmk__xe_set_bool_attr(node_state, PCMK__XA_IN_CCM, pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei)); } } if (!pcmk_is_set(node->flags, crm_remote_node)) { if (flags & node_update_peer) { if (compare_version(controld_globals.dc_version, "3.18.0") >= 0) { // A value 0 means the peer is offline in CPG. crm_xml_add_ll(node_state, PCMK_XA_CRMD, node->when_online); } else { // @COMPAT DCs < 2.1.7 use online/offline rather than timestamp - value = OFFLINESTATUS; + value = PCMK_VALUE_OFFLINE; if (pcmk_is_set(node->processes, crm_get_cluster_proc())) { value = PCMK_VALUE_ONLINE; } crm_xml_add(node_state, PCMK_XA_CRMD, value); } } if (flags & node_update_join) { if (node->join <= crm_join_none) { value = CRMD_JOINSTATE_DOWN; } else { value = CRMD_JOINSTATE_MEMBER; } crm_xml_add(node_state, PCMK__XA_JOIN, value); } if (flags & node_update_expected) { crm_xml_add(node_state, PCMK_XA_EXPECTED, node->expected); } } crm_xml_add(node_state, PCMK_XA_CRM_DEBUG_ORIGIN, source); return node_state; } static void remove_conflicting_node_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { char *node_uuid = user_data; do_crm_log_unlikely(rc == 0 ? LOG_DEBUG : LOG_NOTICE, "Deletion of the unknown conflicting node \"%s\": %s (rc=%d)", node_uuid, pcmk_strerror(rc), rc); } static void search_conflicting_node_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { char *new_node_uuid = user_data; xmlNode *node_xml = NULL; if (rc != pcmk_ok) { if (rc != -ENXIO) { crm_notice("Searching conflicting nodes for %s failed: %s (%d)", new_node_uuid, pcmk_strerror(rc), rc); } return; } else if (output == NULL) { return; } if (pcmk__xe_is(output, PCMK_XE_NODE)) { node_xml = output; } else { node_xml = pcmk__xml_first_child(output); } for (; node_xml != NULL; node_xml = pcmk__xml_next(node_xml)) { const char *node_uuid = NULL; const char *node_uname = NULL; GHashTableIter iter; crm_node_t *node = NULL; gboolean known = FALSE; if (!pcmk__xe_is(node_xml, PCMK_XE_NODE)) { continue; } node_uuid = crm_element_value(node_xml, PCMK_XA_ID); node_uname = crm_element_value(node_xml, PCMK_XA_UNAME); if (node_uuid == NULL || node_uname == NULL) { continue; } g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (node->uuid && pcmk__str_eq(node->uuid, node_uuid, pcmk__str_casei) && node->uname && pcmk__str_eq(node->uname, node_uname, pcmk__str_casei)) { known = TRUE; break; } } if (known == FALSE) { cib_t *cib_conn = controld_globals.cib_conn; int delete_call_id = 0; xmlNode *node_state_xml = NULL; crm_notice("Deleting unknown node %s/%s which has conflicting uname with %s", node_uuid, node_uname, new_node_uuid); delete_call_id = cib_conn->cmds->remove(cib_conn, PCMK_XE_NODES, node_xml, cib_scope_local); fsa_register_cib_callback(delete_call_id, strdup(node_uuid), remove_conflicting_node_callback); node_state_xml = create_xml_node(NULL, PCMK__XE_NODE_STATE); crm_xml_add(node_state_xml, PCMK_XA_ID, node_uuid); crm_xml_add(node_state_xml, PCMK_XA_UNAME, node_uname); delete_call_id = cib_conn->cmds->remove(cib_conn, PCMK_XE_STATUS, node_state_xml, cib_scope_local); fsa_register_cib_callback(delete_call_id, strdup(node_uuid), remove_conflicting_node_callback); free_xml(node_state_xml); } } } static void node_list_update_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if(call_id < pcmk_ok) { crm_err("Node list update failed: %s (%d)", pcmk_strerror(call_id), call_id); crm_log_xml_debug(msg, "update:failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else if(rc < pcmk_ok) { crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "update:failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } void populate_cib_nodes(enum node_update_flags flags, const char *source) { cib_t *cib_conn = controld_globals.cib_conn; int call_id = 0; gboolean from_hashtable = TRUE; xmlNode *node_list = create_xml_node(NULL, PCMK_XE_NODES); #if SUPPORT_COROSYNC if (!pcmk_is_set(flags, node_update_quick) && is_corosync_cluster()) { from_hashtable = pcmk__corosync_add_nodes(node_list); } #endif if (from_hashtable) { GHashTableIter iter; crm_node_t *node = NULL; GString *xpath = NULL; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *new_node = NULL; if ((node->uuid != NULL) && (node->uname != NULL)) { crm_trace("Creating node entry for %s/%s", node->uname, node->uuid); if (xpath == NULL) { xpath = g_string_sized_new(512); } else { g_string_truncate(xpath, 0); } /* We need both to be valid */ new_node = create_xml_node(node_list, PCMK_XE_NODE); crm_xml_add(new_node, PCMK_XA_ID, node->uuid); crm_xml_add(new_node, PCMK_XA_UNAME, node->uname); /* Search and remove unknown nodes with the conflicting uname from CIB */ pcmk__g_strcat(xpath, "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_NODES "/" PCMK_XE_NODE "[@" PCMK_XA_UNAME "='", node->uname, "']" "[@" PCMK_XA_ID "!='", node->uuid, "']", NULL); call_id = cib_conn->cmds->query(cib_conn, (const char *) xpath->str, NULL, cib_scope_local|cib_xpath); fsa_register_cib_callback(call_id, strdup(node->uuid), search_conflicting_node_callback); } } if (xpath != NULL) { g_string_free(xpath, TRUE); } } crm_trace("Populating section from %s", from_hashtable ? "hashtable" : "cluster"); if ((controld_update_cib(PCMK_XE_NODES, node_list, cib_scope_local, node_list_update_callback) == pcmk_rc_ok) && (crm_peer_cache != NULL) && AM_I_DC) { /* * There is no need to update the local CIB with our values if * we've not seen valid membership data */ GHashTableIter iter; crm_node_t *node = NULL; free_xml(node_list); node_list = create_xml_node(NULL, PCMK_XE_STATUS); g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { create_node_state_update(node, flags, node_list, source); } if (crm_remote_peer_cache) { g_hash_table_iter_init(&iter, crm_remote_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { create_node_state_update(node, flags, node_list, source); } } controld_update_cib(PCMK_XE_STATUS, node_list, cib_scope_local, crmd_node_update_complete); } free_xml(node_list); } static void cib_quorum_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if (rc == pcmk_ok) { crm_trace("Quorum update %d complete", call_id); } else { crm_err("Quorum update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } void crm_update_quorum(gboolean quorum, gboolean force_update) { bool has_quorum = pcmk_is_set(controld_globals.flags, controld_has_quorum); if (quorum) { controld_set_global_flags(controld_ever_had_quorum); } else if (pcmk_all_flags_set(controld_globals.flags, controld_ever_had_quorum |controld_no_quorum_suicide)) { pcmk__panic(__func__); } if (AM_I_DC && ((has_quorum && !quorum) || (!has_quorum && quorum) || force_update)) { xmlNode *update = NULL; update = create_xml_node(NULL, PCMK_XE_CIB); crm_xml_add_int(update, PCMK_XA_HAVE_QUORUM, quorum); crm_xml_add(update, PCMK_XA_DC_UUID, controld_globals.our_uuid); crm_debug("Updating quorum status to %s", pcmk__btoa(quorum)); controld_update_cib(PCMK_XE_CIB, update, cib_scope_local, cib_quorum_update_complete); free_xml(update); /* Quorum changes usually cause a new transition via other activity: * quorum gained via a node joining will abort via the node join, * and quorum lost via a node leaving will usually abort via resource * activity and/or fencing. * * However, it is possible that nothing else causes a transition (e.g. * someone forces quorum via corosync-cmaptcl, or quorum is lost due to * a node in standby shutting down cleanly), so here ensure a new * transition is triggered. */ if (quorum) { /* If quorum was gained, abort after a short delay, in case multiple * nodes are joining around the same time, so the one that brings us * to quorum doesn't cause all the remaining ones to be fenced. */ abort_after_delay(INFINITY, pcmk__graph_restart, "Quorum gained", 5000); } else { abort_transition(INFINITY, pcmk__graph_restart, "Quorum lost", NULL); } } if (quorum) { controld_set_global_flags(controld_has_quorum); } else { controld_clear_global_flags(controld_has_quorum); } } diff --git a/include/crm/common/options.h b/include/crm/common/options.h index e5c0a7f5b4..0983230319 100644 --- a/include/crm/common/options.h +++ b/include/crm/common/options.h @@ -1,198 +1,199 @@ /* * Copyright 2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_OPTIONS__H # define PCMK__CRM_COMMON_OPTIONS__H #ifdef __cplusplus extern "C" { #endif /** * \file * \brief API related to options * \ingroup core */ /* * Cluster options */ #define PCMK_OPT_BATCH_LIMIT "batch-limit" #define PCMK_OPT_CLUSTER_DELAY "cluster-delay" #define PCMK_OPT_CLUSTER_INFRASTRUCTURE "cluster-infrastructure" #define PCMK_OPT_CLUSTER_IPC_LIMIT "cluster-ipc-limit" #define PCMK_OPT_CLUSTER_NAME "cluster-name" #define PCMK_OPT_CLUSTER_RECHECK_INTERVAL "cluster-recheck-interval" #define PCMK_OPT_CONCURRENT_FENCING "concurrent-fencing" #define PCMK_OPT_DC_DEADTIME "dc-deadtime" #define PCMK_OPT_DC_VERSION "dc-version" #define PCMK_OPT_ELECTION_TIMEOUT "election-timeout" #define PCMK_OPT_ENABLE_ACL "enable-acl" #define PCMK_OPT_ENABLE_STARTUP_PROBES "enable-startup-probes" #define PCMK_OPT_FENCE_REACTION "fence-reaction" #define PCMK_OPT_HAVE_WATCHDOG "have-watchdog" #define PCMK_OPT_JOIN_FINALIZATION_TIMEOUT "join-finalization-timeout" #define PCMK_OPT_JOIN_INTEGRATION_TIMEOUT "join-integration-timeout" #define PCMK_OPT_LOAD_THRESHOLD "load-threshold" #define PCMK_OPT_MAINTENANCE_MODE "maintenance-mode" #define PCMK_OPT_MIGRATION_LIMIT "migration-limit" #define PCMK_OPT_NO_QUORUM_POLICY "no-quorum-policy" #define PCMK_OPT_NODE_ACTION_LIMIT "node-action-limit" #define PCMK_OPT_NODE_HEALTH_BASE "node-health-base" #define PCMK_OPT_NODE_HEALTH_GREEN "node-health-green" #define PCMK_OPT_NODE_HEALTH_RED "node-health-red" #define PCMK_OPT_NODE_HEALTH_STRATEGY "node-health-strategy" #define PCMK_OPT_NODE_HEALTH_YELLOW "node-health-yellow" #define PCMK_OPT_NODE_PENDING_TIMEOUT "node-pending-timeout" #define PCMK_OPT_PE_ERROR_SERIES_MAX "pe-error-series-max" #define PCMK_OPT_PE_INPUT_SERIES_MAX "pe-input-series-max" #define PCMK_OPT_PE_WARN_SERIES_MAX "pe-warn-series-max" #define PCMK_OPT_PLACEMENT_STRATEGY "placement-strategy" #define PCMK_OPT_PRIORITY_FENCING_DELAY "priority-fencing-delay" #define PCMK_OPT_SHUTDOWN_ESCALATION "shutdown-escalation" #define PCMK_OPT_SHUTDOWN_LOCK "shutdown-lock" #define PCMK_OPT_SHUTDOWN_LOCK_LIMIT "shutdown-lock-limit" #define PCMK_OPT_START_FAILURE_IS_FATAL "start-failure-is-fatal" #define PCMK_OPT_STARTUP_FENCING "startup-fencing" #define PCMK_OPT_STONITH_ACTION "stonith-action" #define PCMK_OPT_STONITH_ENABLED "stonith-enabled" #define PCMK_OPT_STONITH_MAX_ATTEMPTS "stonith-max-attempts" #define PCMK_OPT_STONITH_TIMEOUT "stonith-timeout" #define PCMK_OPT_STONITH_WATCHDOG_TIMEOUT "stonith-watchdog-timeout" #define PCMK_OPT_STOP_ALL_RESOURCES "stop-all-resources" #define PCMK_OPT_STOP_ORPHAN_ACTIONS "stop-orphan-actions" #define PCMK_OPT_STOP_ORPHAN_RESOURCES "stop-orphan-resources" #define PCMK_OPT_SYMMETRIC_CLUSTER "symmetric-cluster" #define PCMK_OPT_TRANSITION_DELAY "transition-delay" /* * Meta-attributes */ #define PCMK_META_ALLOW_MIGRATE "allow-migrate" #define PCMK_META_ALLOW_UNHEALTHY_NODES "allow-unhealthy-nodes" #define PCMK_META_CLONE_MAX "clone-max" #define PCMK_META_CLONE_MIN "clone-min" #define PCMK_META_CLONE_NODE_MAX "clone-node-max" #define PCMK_META_CONTAINER_ATTR_TARGET "container-attribute-target" #define PCMK_META_CRITICAL "critical" #define PCMK_META_ENABLED "enabled" #define PCMK_META_FAILURE_TIMEOUT "failure-timeout" #define PCMK_META_GLOBALLY_UNIQUE "globally-unique" #define PCMK_META_INTERLEAVE "interleave" #define PCMK_META_INTERVAL "interval" #define PCMK_META_IS_MANAGED "is-managed" #define PCMK_META_INTERVAL_ORIGIN "interval-origin" #define PCMK_META_MAINTENANCE "maintenance" #define PCMK_META_MIGRATION_THRESHOLD "migration-threshold" #define PCMK_META_MULTIPLE_ACTIVE "multiple-active" #define PCMK_META_NOTIFY "notify" #define PCMK_META_ON_FAIL "on-fail" #define PCMK_META_ORDERED "ordered" #define PCMK_META_PRIORITY "priority" #define PCMK_META_PROMOTABLE "promotable" #define PCMK_META_PROMOTED_MAX "promoted-max" #define PCMK_META_PROMOTED_NODE_MAX "promoted-node-max" #define PCMK_META_RECORD_PENDING "record-pending" #define PCMK_META_REMOTE_ADDR "remote-addr" #define PCMK_META_REMOTE_ALLOW_MIGRATE "remote-allow-migrate" #define PCMK_META_REMOTE_CONNECT_TIMEOUT "remote-connect-timeout" #define PCMK_META_REMOTE_NODE "remote-node" #define PCMK_META_REMOTE_PORT "remote-port" #define PCMK_META_REQUIRES "requires" #define PCMK_META_RESOURCE_STICKINESS "resource-stickiness" #define PCMK_META_START_DELAY "start-delay" #define PCMK_META_TARGET_ROLE "target-role" #define PCMK_META_TIMEOUT "timeout" #define PCMK_META_TIMESTAMP_FORMAT "timestamp-format" /* * Remote resource instance attributes */ #define PCMK_REMOTE_RA_ADDR "addr" #define PCMK_REMOTE_RA_PORT "port" #define PCMK_REMOTE_RA_RECONNECT_INTERVAL "reconnect_interval" #define PCMK_REMOTE_RA_SERVER "server" /* * Enumerated values */ #define PCMK_VALUE_ALWAYS "always" #define PCMK_VALUE_AND "and" #define PCMK_VALUE_BALANCED "balanced" #define PCMK_VALUE_BLOCK "block" #define PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS "cib-bootstrap-options" #define PCMK_VALUE_DATE_SPEC "date_spec" #define PCMK_VALUE_DEFAULT "default" #define PCMK_VALUE_DEFINED "defined" #define PCMK_VALUE_DEMOTE "demote" #define PCMK_VALUE_DENY "deny" #define PCMK_VALUE_EQ "eq" #define PCMK_VALUE_EXCLUSIVE "exclusive" #define PCMK_VALUE_FAILED "failed" #define PCMK_VALUE_FALSE "false" #define PCMK_VALUE_FENCE "fence" #define PCMK_VALUE_FREEZE "freeze" #define PCMK_VALUE_GRANTED "granted" #define PCMK_VALUE_GREEN "green" #define PCMK_VALUE_GT "gt" #define PCMK_VALUE_GTE "gte" #define PCMK_VALUE_HOST "host" #define PCMK_VALUE_IGNORE "ignore" #define PCMK_VALUE_INTEGER "integer" #define PCMK_VALUE_LITERAL "literal" #define PCMK_VALUE_LT "lt" #define PCMK_VALUE_LTE "lte" #define PCMK_VALUE_MANDATORY "Mandatory" #define PCMK_VALUE_MEMBER "member" #define PCMK_VALUE_META "meta" #define PCMK_VALUE_MINIMAL "minimal" #define PCMK_VALUE_NE "ne" #define PCMK_VALUE_NEVER "never" #define PCMK_VALUE_NOT_DEFINED "not_defined" #define PCMK_VALUE_NUMBER "number" +#define PCMK_VALUE_OFFLINE "offline" #define PCMK_VALUE_ONLINE "online" #define PCMK_VALUE_OPTIONAL "Optional" #define PCMK_VALUE_OR "or" #define PCMK_VALUE_PANIC "panic" #define PCMK_VALUE_PARAM "param" #define PCMK_VALUE_PENDING "pending" #define PCMK_VALUE_READ "read" #define PCMK_VALUE_RED "red" #define PCMK_VALUE_REMOTE "remote" #define PCMK_VALUE_RESTART "restart" #define PCMK_VALUE_RESTART_CONTAINER "restart-container" #define PCMK_VALUE_REVOKED "revoked" #define PCMK_VALUE_SERIALIZE "Serialize" #define PCMK_VALUE_STANDBY "standby" #define PCMK_VALUE_STRING "string" #define PCMK_VALUE_STOP "stop" #define PCMK_VALUE_SUCCESS "success" #define PCMK_VALUE_TRUE "true" #define PCMK_VALUE_UNKNOWN "unknown" #define PCMK_VALUE_UTILIZATION "utilization" #define PCMK_VALUE_VERSION "version" #define PCMK_VALUE_WRITE "write" #define PCMK_VALUE_YELLOW "yellow" // @COMPAT This will become a deprecated alias for PCMK_VALUE_FENCE (see T279) #define PCMK_VALUE_FENCE_LEGACY "suicide" #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_OPTIONS__H diff --git a/lib/cluster/cpg.c b/lib/cluster/cpg.c index bec90bd2cb..45435fe872 100644 --- a/lib/cluster/cpg.c +++ b/lib/cluster/cpg.c @@ -1,1095 +1,1095 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* PCMK__SPECIAL_PID* */ #include "crmcluster_private.h" /* @TODO Once we can update the public API to require crm_cluster_t* in more * functions, we can ditch this in favor of cluster->cpg_handle. */ static cpg_handle_t pcmk_cpg_handle = 0; // @TODO These could be moved to crm_cluster_t* at that time as well static bool cpg_evicted = false; static GList *cs_message_queue = NULL; static int cs_message_timer = 0; struct pcmk__cpg_host_s { uint32_t id; uint32_t pid; gboolean local; enum crm_ais_msg_types type; uint32_t size; char uname[MAX_NAME]; } __attribute__ ((packed)); typedef struct pcmk__cpg_host_s pcmk__cpg_host_t; struct pcmk__cpg_msg_s { struct qb_ipc_response_header header __attribute__ ((aligned(8))); uint32_t id; gboolean is_compressed; pcmk__cpg_host_t host; pcmk__cpg_host_t sender; uint32_t size; uint32_t compressed_size; /* 584 bytes */ char data[0]; } __attribute__ ((packed)); typedef struct pcmk__cpg_msg_s pcmk__cpg_msg_t; static void crm_cs_flush(gpointer data); #define msg_data_len(msg) (msg->is_compressed?msg->compressed_size:msg->size) #define cs_repeat(rc, counter, max, code) do { \ rc = code; \ if ((rc == CS_ERR_TRY_AGAIN) || (rc == CS_ERR_QUEUE_FULL)) { \ counter++; \ crm_debug("Retrying operation after %ds", counter); \ sleep(counter); \ } else { \ break; \ } \ } while (counter < max) /*! * \brief Disconnect from Corosync CPG * * \param[in,out] cluster Cluster to disconnect */ void cluster_disconnect_cpg(crm_cluster_t *cluster) { pcmk_cpg_handle = 0; if (cluster->cpg_handle) { crm_trace("Disconnecting CPG"); cpg_leave(cluster->cpg_handle, &cluster->group); cpg_finalize(cluster->cpg_handle); cluster->cpg_handle = 0; } else { crm_info("No CPG connection"); } } /*! * \brief Get the local Corosync node ID (via CPG) * * \param[in] handle CPG connection to use (or 0 to use new connection) * * \return Corosync ID of local node (or 0 if not known) */ uint32_t get_local_nodeid(cpg_handle_t handle) { cs_error_t rc = CS_OK; int retries = 0; static uint32_t local_nodeid = 0; cpg_handle_t local_handle = handle; cpg_model_v1_data_t cpg_model_info = {CPG_MODEL_V1, NULL, NULL, NULL, 0}; int fd = -1; uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; if(local_nodeid != 0) { return local_nodeid; } if(handle == 0) { crm_trace("Creating connection"); cs_repeat(rc, retries, 5, cpg_model_initialize(&local_handle, CPG_MODEL_V1, (cpg_model_data_t *)&cpg_model_info, NULL)); if (rc != CS_OK) { crm_err("Could not connect to the CPG API: %s (%d)", cs_strerror(rc), rc); return 0; } rc = cpg_fd_get(local_handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CPG API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CPG provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_err("CPG provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CPG provider: %s (%d)", strerror(-rv), -rv); goto bail; } } if (rc == CS_OK) { retries = 0; crm_trace("Performing lookup"); cs_repeat(rc, retries, 5, cpg_local_get(local_handle, &local_nodeid)); } if (rc != CS_OK) { crm_err("Could not get local node id from the CPG API: %s (%d)", pcmk__cs_err_str(rc), rc); } bail: if(handle == 0) { crm_trace("Closing connection"); cpg_finalize(local_handle); } crm_debug("Local nodeid is %u", local_nodeid); return local_nodeid; } /*! * \internal * \brief Callback function for Corosync message queue timer * * \param[in] data CPG handle * * \return FALSE (to indicate to glib that timer should not be removed) */ static gboolean crm_cs_flush_cb(gpointer data) { cs_message_timer = 0; crm_cs_flush(data); return FALSE; } // Send no more than this many CPG messages in one flush #define CS_SEND_MAX 200 /*! * \internal * \brief Send messages in Corosync CPG message queue * * \param[in] data CPG handle */ static void crm_cs_flush(gpointer data) { unsigned int sent = 0; guint queue_len = 0; cs_error_t rc = 0; cpg_handle_t *handle = (cpg_handle_t *) data; if (*handle == 0) { crm_trace("Connection is dead"); return; } queue_len = g_list_length(cs_message_queue); if (((queue_len % 1000) == 0) && (queue_len > 1)) { crm_err("CPG queue has grown to %d", queue_len); } else if (queue_len == CS_SEND_MAX) { crm_warn("CPG queue has grown to %d", queue_len); } if (cs_message_timer != 0) { /* There is already a timer, wait until it goes off */ crm_trace("Timer active %d", cs_message_timer); return; } while ((cs_message_queue != NULL) && (sent < CS_SEND_MAX)) { struct iovec *iov = cs_message_queue->data; rc = cpg_mcast_joined(*handle, CPG_TYPE_AGREED, iov, 1); if (rc != CS_OK) { break; } sent++; crm_trace("CPG message sent, size=%llu", (unsigned long long) iov->iov_len); cs_message_queue = g_list_remove(cs_message_queue, iov); free(iov->iov_base); free(iov); } queue_len -= sent; do_crm_log((queue_len > 5)? LOG_INFO : LOG_TRACE, "Sent %u CPG message%s (%d still queued): %s (rc=%d)", sent, pcmk__plural_s(sent), queue_len, pcmk__cs_err_str(rc), (int) rc); if (cs_message_queue) { uint32_t delay_ms = 100; if (rc != CS_OK) { /* Proportionally more if sending failed but cap at 1s */ delay_ms = QB_MIN(1000, CS_SEND_MAX + (10 * queue_len)); } cs_message_timer = g_timeout_add(delay_ms, crm_cs_flush_cb, data); } } /*! * \internal * \brief Dispatch function for CPG handle * * \param[in,out] user_data Cluster object * * \return 0 on success, -1 on error (per mainloop_io_t interface) */ static int pcmk_cpg_dispatch(gpointer user_data) { cs_error_t rc = CS_OK; crm_cluster_t *cluster = (crm_cluster_t *) user_data; rc = cpg_dispatch(cluster->cpg_handle, CS_DISPATCH_ONE); if (rc != CS_OK) { crm_err("Connection to the CPG API failed: %s (%d)", pcmk__cs_err_str(rc), rc); cpg_finalize(cluster->cpg_handle); cluster->cpg_handle = 0; return -1; } else if (cpg_evicted) { crm_err("Evicted from CPG membership"); return -1; } return 0; } static inline const char * ais_dest(const pcmk__cpg_host_t *host) { if (host->local) { return "local"; } else if (host->size > 0) { return host->uname; } else { return ""; } } static inline const char * msg_type2text(enum crm_ais_msg_types type) { const char *text = "unknown"; switch (type) { case crm_msg_none: text = "unknown"; break; case crm_msg_ais: text = "ais"; break; case crm_msg_cib: text = "cib"; break; case crm_msg_crmd: text = "crmd"; break; case crm_msg_pe: text = "pengine"; break; case crm_msg_te: text = "tengine"; break; case crm_msg_lrmd: text = "lrmd"; break; case crm_msg_attrd: text = "attrd"; break; case crm_msg_stonithd: text = "stonithd"; break; case crm_msg_stonith_ng: text = "stonith-ng"; break; } return text; } /*! * \internal * \brief Check whether a Corosync CPG message is valid * * \param[in] msg Corosync CPG message to check * * \return true if \p msg is valid, otherwise false */ static bool check_message_sanity(const pcmk__cpg_msg_t *msg) { int32_t payload_size = msg->header.size - sizeof(pcmk__cpg_msg_t); if (payload_size < 1) { crm_err("%sCPG message %d from %s invalid: " "Claimed size of %d bytes is too small " CRM_XS " from %s[%u] to %s@%s", (msg->is_compressed? "Compressed " : ""), msg->id, ais_dest(&(msg->sender)), (int) msg->header.size, msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } if (msg->header.error != CS_OK) { crm_err("%sCPG message %d from %s invalid: " "Sender indicated error %d " CRM_XS " from %s[%u] to %s@%s", (msg->is_compressed? "Compressed " : ""), msg->id, ais_dest(&(msg->sender)), msg->header.error, msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } if (msg_data_len(msg) != payload_size) { crm_err("%sCPG message %d from %s invalid: " "Total size %d inconsistent with payload size %d " CRM_XS " from %s[%u] to %s@%s", (msg->is_compressed? "Compressed " : ""), msg->id, ais_dest(&(msg->sender)), (int) msg->header.size, (int) msg_data_len(msg), msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } if (!msg->is_compressed && /* msg->size != (strlen(msg->data) + 1) would be a stronger check, * but checking the last byte or two should be quick */ (((msg->size > 1) && (msg->data[msg->size - 2] == '\0')) || (msg->data[msg->size - 1] != '\0'))) { crm_err("CPG message %d from %s invalid: " "Payload does not end at byte %llu " CRM_XS " from %s[%u] to %s@%s", msg->id, ais_dest(&(msg->sender)), (unsigned long long) msg->size, msg_type2text(msg->sender.type), msg->sender.pid, msg_type2text(msg->host.type), ais_dest(&(msg->host))); return false; } crm_trace("Verified %d-byte %sCPG message %d from %s[%u]@%s to %s@%s", (int) msg->header.size, (msg->is_compressed? "compressed " : ""), msg->id, msg_type2text(msg->sender.type), msg->sender.pid, ais_dest(&(msg->sender)), msg_type2text(msg->host.type), ais_dest(&(msg->host))); return true; } /*! * \brief Extract text data from a Corosync CPG message * * \param[in] handle CPG connection (to get local node ID if not known) * \param[in] nodeid Corosync ID of node that sent message * \param[in] pid Process ID of message sender (for logging only) * \param[in,out] content CPG message * \param[out] kind If not NULL, will be set to CPG header ID * (which should be an enum crm_ais_msg_class value, * currently always crm_class_cluster) * \param[out] from If not NULL, will be set to sender uname * (valid for the lifetime of \p content) * * \return Newly allocated string with message data * \note It is the caller's responsibility to free the return value with free(). */ char * pcmk_message_common_cs(cpg_handle_t handle, uint32_t nodeid, uint32_t pid, void *content, uint32_t *kind, const char **from) { char *data = NULL; pcmk__cpg_msg_t *msg = (pcmk__cpg_msg_t *) content; if(handle) { // Do filtering and field massaging uint32_t local_nodeid = get_local_nodeid(handle); const char *local_name = get_local_node_name(); if (msg->sender.id > 0 && msg->sender.id != nodeid) { crm_err("Nodeid mismatch from %d.%d: claimed nodeid=%u", nodeid, pid, msg->sender.id); return NULL; } else if (msg->host.id != 0 && (local_nodeid != msg->host.id)) { /* Not for us */ crm_trace("Not for us: %u != %u", msg->host.id, local_nodeid); return NULL; } else if (msg->host.size != 0 && !pcmk__str_eq(msg->host.uname, local_name, pcmk__str_casei)) { /* Not for us */ crm_trace("Not for us: %s != %s", msg->host.uname, local_name); return NULL; } msg->sender.id = nodeid; if (msg->sender.size == 0) { crm_node_t *peer = pcmk__get_node(nodeid, NULL, NULL, pcmk__node_search_cluster); if (peer == NULL) { crm_err("Peer with nodeid=%u is unknown", nodeid); } else if (peer->uname == NULL) { crm_err("No uname for peer with nodeid=%u", nodeid); } else { crm_notice("Fixing uname for peer with nodeid=%u", nodeid); msg->sender.size = strlen(peer->uname); memset(msg->sender.uname, 0, MAX_NAME); memcpy(msg->sender.uname, peer->uname, msg->sender.size); } } } crm_trace("Got new%s message (size=%d, %d, %d)", msg->is_compressed ? " compressed" : "", msg_data_len(msg), msg->size, msg->compressed_size); if (kind != NULL) { *kind = msg->header.id; } if (from != NULL) { *from = msg->sender.uname; } if (msg->is_compressed && msg->size > 0) { int rc = BZ_OK; char *uncompressed = NULL; unsigned int new_size = msg->size + 1; if (!check_message_sanity(msg)) { goto badmsg; } crm_trace("Decompressing message data"); uncompressed = calloc(1, new_size); rc = BZ2_bzBuffToBuffDecompress(uncompressed, &new_size, msg->data, msg->compressed_size, 1, 0); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_err("Decompression failed: %s " CRM_XS " rc=%d", pcmk_rc_str(rc), rc); free(uncompressed); goto badmsg; } CRM_ASSERT(new_size == msg->size); data = uncompressed; } else if (!check_message_sanity(msg)) { goto badmsg; } else { data = strdup(msg->data); } // Is this necessary? pcmk__get_node(msg->sender.id, msg->sender.uname, NULL, pcmk__node_search_cluster); crm_trace("Payload: %.200s", data); return data; badmsg: crm_err("Invalid message (id=%d, dest=%s:%s, from=%s:%s.%d):" " min=%d, total=%d, size=%d, bz2_size=%d", msg->id, ais_dest(&(msg->host)), msg_type2text(msg->host.type), ais_dest(&(msg->sender)), msg_type2text(msg->sender.type), msg->sender.pid, (int)sizeof(pcmk__cpg_msg_t), msg->header.size, msg->size, msg->compressed_size); free(data); return NULL; } /*! * \internal * \brief Compare cpg_address objects by node ID * * \param[in] first First cpg_address structure to compare * \param[in] second Second cpg_address structure to compare * * \return Negative number if first's node ID is lower, * positive number if first's node ID is greater, * or 0 if both node IDs are equal */ static int cmp_member_list_nodeid(const void *first, const void *second) { const struct cpg_address *const a = *((const struct cpg_address **) first), *const b = *((const struct cpg_address **) second); if (a->nodeid < b->nodeid) { return -1; } else if (a->nodeid > b->nodeid) { return 1; } /* don't bother with "reason" nor "pid" */ return 0; } /*! * \internal * \brief Get a readable string equivalent of a cpg_reason_t value * * \param[in] reason CPG reason value * * \return Readable string suitable for logging */ static const char * cpgreason2str(cpg_reason_t reason) { switch (reason) { case CPG_REASON_JOIN: return " via cpg_join"; case CPG_REASON_LEAVE: return " via cpg_leave"; case CPG_REASON_NODEDOWN: return " via cluster exit"; case CPG_REASON_NODEUP: return " via cluster join"; case CPG_REASON_PROCDOWN: return " for unknown reason"; default: break; } return ""; } /*! * \internal * \brief Get a log-friendly node name * * \param[in] peer Node to check * * \return Node's uname, or readable string if not known */ static inline const char * peer_name(const crm_node_t *peer) { if (peer == NULL) { return "unknown node"; } else if (peer->uname == NULL) { return "peer node"; } else { return peer->uname; } } /*! * \internal * \brief Process a CPG peer's leaving the cluster * * \param[in] cpg_group_name CPG group name (for logging) * \param[in] event_counter Event number (for logging) * \param[in] local_nodeid Node ID of local node * \param[in] cpg_peer CPG peer that left * \param[in] sorted_member_list List of remaining members, qsort()-ed by ID * \param[in] member_list_entries Number of entries in \p sorted_member_list */ static void node_left(const char *cpg_group_name, int event_counter, uint32_t local_nodeid, const struct cpg_address *cpg_peer, const struct cpg_address **sorted_member_list, size_t member_list_entries) { crm_node_t *peer = pcmk__search_node_caches(cpg_peer->nodeid, NULL, pcmk__node_search_cluster); const struct cpg_address **rival = NULL; /* Most CPG-related Pacemaker code assumes that only one process on a node * can be in the process group, but Corosync does not impose this * limitation, and more than one can be a member in practice due to a * daemon attempting to start while another instance is already running. * * Check for any such duplicate instances, because we don't want to process * their leaving as if our actual peer left. If the peer that left still has * an entry in sorted_member_list (with a different PID), we will ignore the * leaving. * * @TODO Track CPG members' PIDs so we can tell exactly who left. */ if (peer != NULL) { rival = bsearch(&cpg_peer, sorted_member_list, member_list_entries, sizeof(const struct cpg_address *), cmp_member_list_nodeid); } if (rival == NULL) { crm_info("Group %s event %d: %s (node %u pid %u) left%s", cpg_group_name, event_counter, peer_name(peer), cpg_peer->nodeid, cpg_peer->pid, cpgreason2str(cpg_peer->reason)); if (peer != NULL) { crm_update_peer_proc(__func__, peer, crm_proc_cpg, - OFFLINESTATUS); + PCMK_VALUE_OFFLINE); } } else if (cpg_peer->nodeid == local_nodeid) { crm_warn("Group %s event %d: duplicate local pid %u left%s", cpg_group_name, event_counter, cpg_peer->pid, cpgreason2str(cpg_peer->reason)); } else { crm_warn("Group %s event %d: " "%s (node %u) duplicate pid %u left%s (%u remains)", cpg_group_name, event_counter, peer_name(peer), cpg_peer->nodeid, cpg_peer->pid, cpgreason2str(cpg_peer->reason), (*rival)->pid); } } /*! * \brief Handle a CPG configuration change event * * \param[in] handle CPG connection * \param[in] cpg_name CPG group name * \param[in] member_list List of current CPG members * \param[in] member_list_entries Number of entries in \p member_list * \param[in] left_list List of CPG members that left * \param[in] left_list_entries Number of entries in \p left_list * \param[in] joined_list List of CPG members that joined * \param[in] joined_list_entries Number of entries in \p joined_list */ void pcmk_cpg_membership(cpg_handle_t handle, const struct cpg_name *groupName, const struct cpg_address *member_list, size_t member_list_entries, const struct cpg_address *left_list, size_t left_list_entries, const struct cpg_address *joined_list, size_t joined_list_entries) { int i; gboolean found = FALSE; static int counter = 0; uint32_t local_nodeid = get_local_nodeid(handle); const struct cpg_address **sorted; sorted = malloc(member_list_entries * sizeof(const struct cpg_address *)); CRM_ASSERT(sorted != NULL); for (size_t iter = 0; iter < member_list_entries; iter++) { sorted[iter] = member_list + iter; } /* so that the cross-matching multiply-subscribed nodes is then cheap */ qsort(sorted, member_list_entries, sizeof(const struct cpg_address *), cmp_member_list_nodeid); for (i = 0; i < left_list_entries; i++) { node_left(groupName->value, counter, local_nodeid, &left_list[i], sorted, member_list_entries); } free(sorted); sorted = NULL; for (i = 0; i < joined_list_entries; i++) { crm_info("Group %s event %d: node %u pid %u joined%s", groupName->value, counter, joined_list[i].nodeid, joined_list[i].pid, cpgreason2str(joined_list[i].reason)); } for (i = 0; i < member_list_entries; i++) { crm_node_t *peer = pcmk__get_node(member_list[i].nodeid, NULL, NULL, pcmk__node_search_cluster); if (member_list[i].nodeid == local_nodeid && member_list[i].pid != getpid()) { // See the note in node_left() crm_warn("Group %s event %d: detected duplicate local pid %u", groupName->value, counter, member_list[i].pid); continue; } crm_info("Group %s event %d: %s (node %u pid %u) is member", groupName->value, counter, peer_name(peer), member_list[i].nodeid, member_list[i].pid); /* If the caller left auto-reaping enabled, this will also update the * state to member. */ peer = crm_update_peer_proc(__func__, peer, crm_proc_cpg, PCMK_VALUE_ONLINE); if (peer && peer->state && strcmp(peer->state, CRM_NODE_MEMBER)) { /* The node is a CPG member, but we currently think it's not a * cluster member. This is possible only if auto-reaping was * disabled. The node may be joining, and we happened to get the CPG * notification before the quorum notification; or the node may have * just died, and we are processing its final messages; or a bug * has affected the peer cache. */ time_t now = time(NULL); if (peer->when_lost == 0) { // Track when we first got into this contradictory state peer->when_lost = now; } else if (now > (peer->when_lost + 60)) { // If it persists for more than a minute, update the state crm_warn("Node %u is member of group %s but was believed offline", member_list[i].nodeid, groupName->value); pcmk__update_peer_state(__func__, peer, CRM_NODE_MEMBER, 0); } } if (local_nodeid == member_list[i].nodeid) { found = TRUE; } } if (!found) { crm_err("Local node was evicted from group %s", groupName->value); cpg_evicted = true; } counter++; } /*! * \brief Connect to Corosync CPG * * \param[in,out] cluster Cluster object * * \return TRUE on success, otherwise FALSE */ gboolean cluster_connect_cpg(crm_cluster_t *cluster) { cs_error_t rc; int fd = -1; int retries = 0; uint32_t id = 0; crm_node_t *peer = NULL; cpg_handle_t handle = 0; const char *message_name = pcmk__message_name(crm_system_name); uid_t found_uid = 0; gid_t found_gid = 0; pid_t found_pid = 0; int rv; struct mainloop_fd_callbacks cpg_fd_callbacks = { .dispatch = pcmk_cpg_dispatch, .destroy = cluster->destroy, }; cpg_model_v1_data_t cpg_model_info = { .model = CPG_MODEL_V1, .cpg_deliver_fn = cluster->cpg.cpg_deliver_fn, .cpg_confchg_fn = cluster->cpg.cpg_confchg_fn, .cpg_totem_confchg_fn = NULL, .flags = 0, }; cpg_evicted = false; cluster->group.length = 0; cluster->group.value[0] = 0; /* group.value is char[128] */ strncpy(cluster->group.value, message_name, 127); cluster->group.value[127] = 0; cluster->group.length = 1 + QB_MIN(127, strlen(cluster->group.value)); cs_repeat(rc, retries, 30, cpg_model_initialize(&handle, CPG_MODEL_V1, (cpg_model_data_t *)&cpg_model_info, NULL)); if (rc != CS_OK) { crm_err("Could not connect to the CPG API: %s (%d)", cs_strerror(rc), rc); goto bail; } rc = cpg_fd_get(handle, &fd); if (rc != CS_OK) { crm_err("Could not obtain the CPG API connection: %s (%d)", cs_strerror(rc), rc); goto bail; } /* CPG provider run as root (in given user namespace, anyway)? */ if (!(rv = crm_ipc_is_authentic_process(fd, (uid_t) 0,(gid_t) 0, &found_pid, &found_uid, &found_gid))) { crm_err("CPG provider is not authentic:" " process %lld (uid: %lld, gid: %lld)", (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); rc = CS_ERR_ACCESS; goto bail; } else if (rv < 0) { crm_err("Could not verify authenticity of CPG provider: %s (%d)", strerror(-rv), -rv); rc = CS_ERR_ACCESS; goto bail; } id = get_local_nodeid(handle); if (id == 0) { crm_err("Could not get local node id from the CPG API"); goto bail; } cluster->nodeid = id; retries = 0; cs_repeat(rc, retries, 30, cpg_join(handle, &cluster->group)); if (rc != CS_OK) { crm_err("Could not join the CPG group '%s': %d", message_name, rc); goto bail; } pcmk_cpg_handle = handle; cluster->cpg_handle = handle; mainloop_add_fd("corosync-cpg", G_PRIORITY_MEDIUM, fd, cluster, &cpg_fd_callbacks); bail: if (rc != CS_OK) { cpg_finalize(handle); return FALSE; } peer = pcmk__get_node(id, NULL, NULL, pcmk__node_search_cluster); crm_update_peer_proc(__func__, peer, crm_proc_cpg, PCMK_VALUE_ONLINE); return TRUE; } /*! * \internal * \brief Send an XML message via Corosync CPG * * \param[in] msg XML message to send * \param[in] node Cluster node to send message to * \param[in] dest Type of message to send * * \return TRUE on success, otherwise FALSE */ bool pcmk__cpg_send_xml(const xmlNode *msg, const crm_node_t *node, enum crm_ais_msg_types dest) { bool rc = true; char *data = NULL; data = dump_xml_unformatted(msg); rc = send_cluster_text(crm_class_cluster, data, FALSE, node, dest); free(data); return rc; } /*! * \internal * \brief Send string data via Corosync CPG * * \param[in] msg_class Message class (to set as CPG header ID) * \param[in] data Data to send * \param[in] local What to set as host "local" value (which is never used) * \param[in] node Cluster node to send message to * \param[in] dest Type of message to send * * \return TRUE on success, otherwise FALSE */ gboolean send_cluster_text(enum crm_ais_msg_class msg_class, const char *data, gboolean local, const crm_node_t *node, enum crm_ais_msg_types dest) { static int msg_id = 0; static int local_pid = 0; static int local_name_len = 0; static const char *local_name = NULL; char *target = NULL; struct iovec *iov; pcmk__cpg_msg_t *msg = NULL; enum crm_ais_msg_types sender = text2msg_type(crm_system_name); switch (msg_class) { case crm_class_cluster: break; default: crm_err("Invalid message class: %d", msg_class); return FALSE; } CRM_CHECK(dest != crm_msg_ais, return FALSE); if (local_name == NULL) { local_name = get_local_node_name(); } if ((local_name_len == 0) && (local_name != NULL)) { local_name_len = strlen(local_name); } if (data == NULL) { data = ""; } if (local_pid == 0) { local_pid = getpid(); } if (sender == crm_msg_none) { sender = local_pid; } msg = calloc(1, sizeof(pcmk__cpg_msg_t)); msg_id++; msg->id = msg_id; msg->header.id = msg_class; msg->header.error = CS_OK; msg->host.type = dest; msg->host.local = local; if (node) { if (node->uname) { target = strdup(node->uname); msg->host.size = strlen(node->uname); memset(msg->host.uname, 0, MAX_NAME); memcpy(msg->host.uname, node->uname, msg->host.size); } else { target = crm_strdup_printf("%u", node->id); } msg->host.id = node->id; } else { target = strdup("all"); } msg->sender.id = 0; msg->sender.type = sender; msg->sender.pid = local_pid; msg->sender.size = local_name_len; memset(msg->sender.uname, 0, MAX_NAME); if ((local_name != NULL) && (msg->sender.size != 0)) { memcpy(msg->sender.uname, local_name, msg->sender.size); } msg->size = 1 + strlen(data); msg->header.size = sizeof(pcmk__cpg_msg_t) + msg->size; if (msg->size < CRM_BZ2_THRESHOLD) { msg = pcmk__realloc(msg, msg->header.size); memcpy(msg->data, data, msg->size); } else { char *compressed = NULL; unsigned int new_size = 0; char *uncompressed = strdup(data); if (pcmk__compress(uncompressed, (unsigned int) msg->size, 0, &compressed, &new_size) == pcmk_rc_ok) { msg->header.size = sizeof(pcmk__cpg_msg_t) + new_size; msg = pcmk__realloc(msg, msg->header.size); memcpy(msg->data, compressed, new_size); msg->is_compressed = TRUE; msg->compressed_size = new_size; } else { // cppcheck seems not to understand the abort logic in pcmk__realloc // cppcheck-suppress memleak msg = pcmk__realloc(msg, msg->header.size); memcpy(msg->data, data, msg->size); } free(uncompressed); free(compressed); } iov = calloc(1, sizeof(struct iovec)); iov->iov_base = msg; iov->iov_len = msg->header.size; if (msg->compressed_size) { crm_trace("Queueing CPG message %u to %s (%llu bytes, %d bytes compressed payload): %.200s", msg->id, target, (unsigned long long) iov->iov_len, msg->compressed_size, data); } else { crm_trace("Queueing CPG message %u to %s (%llu bytes, %d bytes payload): %.200s", msg->id, target, (unsigned long long) iov->iov_len, msg->size, data); } free(target); cs_message_queue = g_list_append(cs_message_queue, iov); crm_cs_flush(&pcmk_cpg_handle); return TRUE; } /*! * \brief Get the message type equivalent of a string * * \param[in] text String of message type * * \return Message type equivalent of \p text */ enum crm_ais_msg_types text2msg_type(const char *text) { int type = crm_msg_none; CRM_CHECK(text != NULL, return type); text = pcmk__message_name(text); if (pcmk__str_eq(text, "ais", pcmk__str_casei)) { type = crm_msg_ais; } else if (pcmk__str_eq(text, CRM_SYSTEM_CIB, pcmk__str_casei)) { type = crm_msg_cib; } else if (pcmk__strcase_any_of(text, CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL)) { type = crm_msg_crmd; } else if (pcmk__str_eq(text, CRM_SYSTEM_TENGINE, pcmk__str_casei)) { type = crm_msg_te; } else if (pcmk__str_eq(text, CRM_SYSTEM_PENGINE, pcmk__str_casei)) { type = crm_msg_pe; } else if (pcmk__str_eq(text, CRM_SYSTEM_LRMD, pcmk__str_casei)) { type = crm_msg_lrmd; } else if (pcmk__str_eq(text, CRM_SYSTEM_STONITHD, pcmk__str_casei)) { type = crm_msg_stonithd; } else if (pcmk__str_eq(text, "stonith-ng", pcmk__str_casei)) { type = crm_msg_stonith_ng; } else if (pcmk__str_eq(text, "attrd", pcmk__str_casei)) { type = crm_msg_attrd; } else { /* This will normally be a transient client rather than * a cluster daemon. Set the type to the pid of the client */ int scan_rc = sscanf(text, "%d", &type); if (scan_rc != 1 || type <= crm_msg_stonith_ng) { /* Ensure it's sane */ type = crm_msg_none; } } return type; } diff --git a/lib/pacemaker/pcmk_injections.c b/lib/pacemaker/pcmk_injections.c index 12986470d0..c5325c66e1 100644 --- a/lib/pacemaker/pcmk_injections.c +++ b/lib/pacemaker/pcmk_injections.c @@ -1,796 +1,796 @@ /* * Copyright 2009-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include // lrmd_event_data_t, etc. #include #include #include #include "libpacemaker_private.h" bool pcmk__simulate_node_config = false; #define XPATH_NODE_CONFIG "//" PCMK_XE_NODE "[@" PCMK_XA_UNAME "='%s']" #define XPATH_NODE_STATE "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" #define XPATH_NODE_STATE_BY_ID "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_ID "='%s']" #define XPATH_RSC_HISTORY XPATH_NODE_STATE \ "//" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" /*! * \internal * \brief Inject a fictitious transient node attribute into scheduler input * * \param[in,out] out Output object for displaying error messages * \param[in,out] cib_node \c PCMK__XE_NODE_STATE XML to inject attribute into * \param[in] name Transient node attribute name to inject * \param[in] value Transient node attribute value to inject */ static void inject_transient_attr(pcmk__output_t *out, xmlNode *cib_node, const char *name, const char *value) { xmlNode *attrs = NULL; xmlNode *instance_attrs = NULL; const char *node_uuid = ID(cib_node); out->message(out, "inject-attr", name, value, cib_node); attrs = first_named_child(cib_node, PCMK__XE_TRANSIENT_ATTRIBUTES); if (attrs == NULL) { attrs = create_xml_node(cib_node, PCMK__XE_TRANSIENT_ATTRIBUTES); crm_xml_add(attrs, PCMK_XA_ID, node_uuid); } instance_attrs = first_named_child(attrs, PCMK_XE_INSTANCE_ATTRIBUTES); if (instance_attrs == NULL) { instance_attrs = create_xml_node(attrs, PCMK_XE_INSTANCE_ATTRIBUTES); crm_xml_add(instance_attrs, PCMK_XA_ID, node_uuid); } crm_create_nvpair_xml(instance_attrs, NULL, name, value); } /*! * \internal * \brief Inject a fictitious fail count into a scheduler input * * \param[in,out] out Output object for displaying error messages * \param[in,out] cib_node Node state XML to inject into * \param[in] resource ID of resource for fail count to inject * \param[in] task Action name for fail count to inject * \param[in] interval_ms Action interval (in milliseconds) for fail count * \param[in] rc Action result for fail count to inject (if 0, or * 7 when interval_ms is 0, inject nothing) */ void pcmk__inject_failcount(pcmk__output_t *out, xmlNode *cib_node, const char *resource, const char *task, guint interval_ms, int rc) { if (rc == 0) { return; } else if ((rc == 7) && (interval_ms == 0)) { return; } else { char *name = NULL; char *now = pcmk__ttoa(time(NULL)); name = pcmk__failcount_name(resource, task, interval_ms); inject_transient_attr(out, cib_node, name, "value++"); free(name); name = pcmk__lastfailure_name(resource, task, interval_ms); inject_transient_attr(out, cib_node, name, now); free(name); free(now); } } /*! * \internal * \brief Create a CIB configuration entry for a fictitious node * * \param[in,out] cib_conn CIB object to use * \param[in] node Node name to use */ static void create_node_entry(cib_t *cib_conn, const char *node) { int rc = pcmk_ok; char *xpath = crm_strdup_printf(XPATH_NODE_CONFIG, node); rc = cib_conn->cmds->query(cib_conn, xpath, NULL, cib_xpath|cib_sync_call|cib_scope_local); if (rc == -ENXIO) { // Only add if not already existing xmlNode *cib_object = create_xml_node(NULL, PCMK_XE_NODE); crm_xml_add(cib_object, PCMK_XA_ID, node); // Use node name as ID crm_xml_add(cib_object, PCMK_XA_UNAME, node); cib_conn->cmds->create(cib_conn, PCMK_XE_NODES, cib_object, cib_sync_call|cib_scope_local); /* Not bothering with subsequent query to see if it exists, we'll bomb out later in the call to query_node_uuid()... */ free_xml(cib_object); } free(xpath); } /*! * \internal * \brief Synthesize a fake executor event for an action * * \param[in] cib_resource XML for any existing resource action history * \param[in] task Name of action to synthesize * \param[in] interval_ms Interval of action to synthesize * \param[in] outcome Result of action to synthesize * * \return Newly allocated executor event * \note It is the caller's responsibility to free the result with * lrmd_free_event(). */ static lrmd_event_data_t * create_op(const xmlNode *cib_resource, const char *task, guint interval_ms, int outcome) { lrmd_event_data_t *op = NULL; xmlNode *xop = NULL; op = lrmd_new_event(ID(cib_resource), task, interval_ms); lrmd__set_result(op, outcome, PCMK_EXEC_DONE, "Simulated action result"); op->params = NULL; // Not needed for simulation purposes op->t_run = (unsigned int) time(NULL); op->t_rcchange = op->t_run; // Use a call ID higher than any existing history entries op->call_id = 0; for (xop = pcmk__xe_first_child(cib_resource); xop != NULL; xop = pcmk__xe_next(xop)) { int tmp = 0; crm_element_value_int(xop, PCMK__XA_CALL_ID, &tmp); if (tmp > op->call_id) { op->call_id = tmp; } } op->call_id++; return op; } /*! * \internal * \brief Inject a fictitious resource history entry into a scheduler input * * \param[in,out] cib_resource Resource history XML to inject entry into * \param[in,out] op Action result to inject * \param[in] target_rc Expected result for action to inject * * \return XML of injected resource history entry */ xmlNode * pcmk__inject_action_result(xmlNode *cib_resource, lrmd_event_data_t *op, int target_rc) { return pcmk__create_history_xml(cib_resource, op, CRM_FEATURE_SET, target_rc, NULL, crm_system_name); } /*! * \internal * \brief Inject a fictitious node into a scheduler input * * \param[in,out] cib_conn Scheduler input CIB to inject node into * \param[in] node Name of node to inject * \param[in] uuid UUID of node to inject * * \return XML of \c PCMK__XE_NODE_STATE entry for new node * \note If the global pcmk__simulate_node_config has been set to true, a * node entry in the configuration section will be added, as well as a * node state entry in the status section. */ xmlNode * pcmk__inject_node(cib_t *cib_conn, const char *node, const char *uuid) { int rc = pcmk_ok; xmlNode *cib_object = NULL; char *xpath = crm_strdup_printf(XPATH_NODE_STATE, node); bool duplicate = false; char *found_uuid = NULL; if (pcmk__simulate_node_config) { create_node_entry(cib_conn, node); } rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object, cib_xpath|cib_sync_call|cib_scope_local); if ((cib_object != NULL) && (ID(cib_object) == NULL)) { crm_err("Detected multiple " PCMK__XE_NODE_STATE " entries for " "xpath=%s, bailing", xpath); duplicate = true; goto done; } if (rc == -ENXIO) { if (uuid == NULL) { query_node_uuid(cib_conn, node, &found_uuid, NULL); } else { found_uuid = strdup(uuid); } if (found_uuid) { char *xpath_by_uuid = crm_strdup_printf(XPATH_NODE_STATE_BY_ID, found_uuid); /* It's possible that a PCMK__XE_NODE_STATE entry doesn't have a * PCMK_XA_UNAME yet */ rc = cib_conn->cmds->query(cib_conn, xpath_by_uuid, &cib_object, cib_xpath|cib_sync_call|cib_scope_local); if ((cib_object != NULL) && (ID(cib_object) == NULL)) { crm_err("Can't inject node state for %s because multiple " "state entries found for ID %s", node, found_uuid); duplicate = true; free(xpath_by_uuid); goto done; } else if (cib_object != NULL) { crm_xml_add(cib_object, PCMK_XA_UNAME, node); rc = cib_conn->cmds->modify(cib_conn, PCMK_XE_STATUS, cib_object, cib_sync_call|cib_scope_local); } free(xpath_by_uuid); } } if (rc == -ENXIO) { cib_object = create_xml_node(NULL, PCMK__XE_NODE_STATE); crm_xml_add(cib_object, PCMK_XA_ID, found_uuid); crm_xml_add(cib_object, PCMK_XA_UNAME, node); cib_conn->cmds->create(cib_conn, PCMK_XE_STATUS, cib_object, cib_sync_call|cib_scope_local); free_xml(cib_object); rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object, cib_xpath|cib_sync_call|cib_scope_local); crm_trace("Injecting node state for %s (rc=%d)", node, rc); } done: free(found_uuid); free(xpath); if (duplicate) { crm_log_xml_warn(cib_object, "Duplicates"); crm_exit(CRM_EX_SOFTWARE); return NULL; // not reached, but makes static analysis happy } CRM_ASSERT(rc == pcmk_ok); return cib_object; } /*! * \internal * \brief Inject a fictitious node state change into a scheduler input * * \param[in,out] cib_conn Scheduler input CIB to inject into * \param[in] node Name of node to inject change for * \param[in] up If true, change state to online, otherwise offline * * \return XML of changed (or added) node state entry */ xmlNode * pcmk__inject_node_state_change(cib_t *cib_conn, const char *node, bool up) { xmlNode *cib_node = pcmk__inject_node(cib_conn, node, NULL); if (up) { pcmk__xe_set_props(cib_node, PCMK__XA_IN_CCM, PCMK_VALUE_TRUE, PCMK_XA_CRMD, PCMK_VALUE_ONLINE, PCMK__XA_JOIN, CRMD_JOINSTATE_MEMBER, PCMK_XA_EXPECTED, CRMD_JOINSTATE_MEMBER, NULL); } else { pcmk__xe_set_props(cib_node, PCMK__XA_IN_CCM, PCMK_VALUE_FALSE, - PCMK_XA_CRMD, OFFLINESTATUS, + PCMK_XA_CRMD, PCMK_VALUE_OFFLINE, PCMK__XA_JOIN, CRMD_JOINSTATE_DOWN, PCMK_XA_EXPECTED, CRMD_JOINSTATE_DOWN, NULL); } crm_xml_add(cib_node, PCMK_XA_CRM_DEBUG_ORIGIN, crm_system_name); return cib_node; } /*! * \internal * \brief Check whether a node has history for a given resource * * \param[in,out] cib_node Node state XML to check * \param[in] resource Resource name to check for * * \return Resource's \c PCMK__XE_LRM_RESOURCE XML entry beneath \p cib_node if * found, otherwise \c NULL */ static xmlNode * find_resource_xml(xmlNode *cib_node, const char *resource) { const char *node = crm_element_value(cib_node, PCMK_XA_UNAME); char *xpath = crm_strdup_printf(XPATH_RSC_HISTORY, node, resource); xmlNode *match = get_xpath_object(xpath, cib_node, LOG_TRACE); free(xpath); return match; } /*! * \internal * \brief Inject a resource history element into a scheduler input * * \param[in,out] out Output object for displaying error messages * \param[in,out] cib_node Node state XML to inject resource history entry into * \param[in] resource ID (in configuration) of resource to inject * \param[in] lrm_name ID as used in history (could be clone instance) * \param[in] rclass Resource agent class of resource to inject * \param[in] rtype Resource agent type of resource to inject * \param[in] rprovider Resource agent provider of resource to inject * * \return XML of injected resource history element * \note If a history element already exists under either \p resource or * \p lrm_name, this will return it rather than injecting a new one. */ xmlNode * pcmk__inject_resource_history(pcmk__output_t *out, xmlNode *cib_node, const char *resource, const char *lrm_name, const char *rclass, const char *rtype, const char *rprovider) { xmlNode *lrm = NULL; xmlNode *container = NULL; xmlNode *cib_resource = NULL; cib_resource = find_resource_xml(cib_node, resource); if (cib_resource != NULL) { /* If an existing LRM history entry uses the resource name, * continue using it, even if lrm_name is different. */ return cib_resource; } // Check for history entry under preferred name if (strcmp(resource, lrm_name) != 0) { cib_resource = find_resource_xml(cib_node, lrm_name); if (cib_resource != NULL) { return cib_resource; } } if ((rclass == NULL) || (rtype == NULL)) { // @TODO query configuration for class, provider, type out->err(out, "Resource %s not found in the status section of %s " "(supply class and type to continue)", resource, ID(cib_node)); return NULL; } else if (!pcmk__strcase_any_of(rclass, PCMK_RESOURCE_CLASS_OCF, PCMK_RESOURCE_CLASS_STONITH, PCMK_RESOURCE_CLASS_SERVICE, PCMK_RESOURCE_CLASS_UPSTART, PCMK_RESOURCE_CLASS_SYSTEMD, PCMK_RESOURCE_CLASS_LSB, NULL)) { out->err(out, "Invalid class for %s: %s", resource, rclass); return NULL; } else if (pcmk_is_set(pcmk_get_ra_caps(rclass), pcmk_ra_cap_provider) && (rprovider == NULL)) { // @TODO query configuration for provider out->err(out, "Please specify the provider for resource %s", resource); return NULL; } crm_info("Injecting new resource %s into node state '%s'", lrm_name, ID(cib_node)); lrm = first_named_child(cib_node, PCMK__XE_LRM); if (lrm == NULL) { const char *node_uuid = ID(cib_node); lrm = create_xml_node(cib_node, PCMK__XE_LRM); crm_xml_add(lrm, PCMK_XA_ID, node_uuid); } container = first_named_child(lrm, PCMK__XE_LRM_RESOURCES); if (container == NULL) { container = create_xml_node(lrm, PCMK__XE_LRM_RESOURCES); } cib_resource = create_xml_node(container, PCMK__XE_LRM_RESOURCE); // If we're creating a new entry, use the preferred name crm_xml_add(cib_resource, PCMK_XA_ID, lrm_name); crm_xml_add(cib_resource, PCMK_XA_CLASS, rclass); crm_xml_add(cib_resource, PCMK_XA_PROVIDER, rprovider); crm_xml_add(cib_resource, PCMK_XA_TYPE, rtype); return cib_resource; } static int find_ticket_state(pcmk__output_t *out, cib_t *the_cib, const char *ticket_id, xmlNode **ticket_state_xml) { int rc = pcmk_ok; xmlNode *xml_search = NULL; GString *xpath = g_string_sized_new(256); CRM_ASSERT(ticket_state_xml != NULL); *ticket_state_xml = NULL; g_string_append(xpath, "/" PCMK_XE_CIB "/" PCMK_XE_STATUS "/" PCMK_XE_TICKETS); if (ticket_id) { pcmk__g_strcat(xpath, "/" PCMK__XE_TICKET_STATE "[@" PCMK_XA_ID "=\"", ticket_id, "\"]", NULL); } rc = the_cib->cmds->query(the_cib, (const char *) xpath->str, &xml_search, cib_sync_call|cib_scope_local|cib_xpath); g_string_free(xpath, TRUE); if (rc != pcmk_ok) { return rc; } crm_log_xml_debug(xml_search, "Match"); if ((xml_search->children != NULL) && (ticket_id != NULL)) { out->err(out, "Multiple " PCMK__XE_TICKET_STATE "s match ticket_id=%s", ticket_id); } *ticket_state_xml = xml_search; return rc; } /*! * \internal * \brief Inject a ticket attribute into ticket state * * \param[in,out] out Output object for displaying error messages * \param[in] ticket_id Ticket whose state should be changed * \param[in] attr_name Ticket attribute name to inject * \param[in] attr_value Boolean value of ticket attribute to inject * \param[in,out] cib CIB object to use * * \return Standard Pacemaker return code */ static int set_ticket_state_attr(pcmk__output_t *out, const char *ticket_id, const char *attr_name, bool attr_value, cib_t *cib) { int rc = pcmk_rc_ok; xmlNode *xml_top = NULL; xmlNode *ticket_state_xml = NULL; // Check for an existing ticket state entry rc = find_ticket_state(out, cib, ticket_id, &ticket_state_xml); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { // Ticket state found, use it crm_debug("Injecting attribute into existing ticket state %s", ticket_id); xml_top = ticket_state_xml; } else if (rc == ENXIO) { // No ticket state, create it xmlNode *xml_obj = NULL; xml_top = create_xml_node(NULL, PCMK_XE_STATUS); xml_obj = create_xml_node(xml_top, PCMK_XE_TICKETS); ticket_state_xml = create_xml_node(xml_obj, PCMK__XE_TICKET_STATE); crm_xml_add(ticket_state_xml, PCMK_XA_ID, ticket_id); } else { // Error return rc; } // Add the attribute to the ticket state pcmk__xe_set_bool_attr(ticket_state_xml, attr_name, attr_value); crm_log_xml_debug(xml_top, "Update"); // Commit the change to the CIB rc = cib->cmds->modify(cib, PCMK_XE_STATUS, xml_top, cib_sync_call|cib_scope_local); rc = pcmk_legacy2rc(rc); free_xml(xml_top); return rc; } /*! * \internal * \brief Inject a fictitious action into the cluster * * \param[in,out] out Output object for displaying error messages * \param[in] spec Action specification to inject * \param[in,out] cib CIB object for scheduler input * \param[in] scheduler Scheduler data */ static void inject_action(pcmk__output_t *out, const char *spec, cib_t *cib, const pcmk_scheduler_t *scheduler) { int rc; int outcome = PCMK_OCF_OK; guint interval_ms = 0; char *key = NULL; char *node = NULL; char *task = NULL; char *resource = NULL; const char *rtype = NULL; const char *rclass = NULL; const char *rprovider = NULL; xmlNode *cib_op = NULL; xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; const pcmk_resource_t *rsc = NULL; lrmd_event_data_t *op = NULL; out->message(out, "inject-spec", spec); key = calloc(1, strlen(spec) + 1); node = calloc(1, strlen(spec) + 1); rc = sscanf(spec, "%[^@]@%[^=]=%d", key, node, &outcome); if (rc != 3) { out->err(out, "Invalid operation spec: %s. Only found %d fields", spec, rc); goto done; } parse_op_key(key, &resource, &task, &interval_ms); rsc = pe_find_resource(scheduler->resources, resource); if (rsc == NULL) { out->err(out, "Invalid resource name: %s", resource); goto done; } rclass = crm_element_value(rsc->xml, PCMK_XA_CLASS); rtype = crm_element_value(rsc->xml, PCMK_XA_TYPE); rprovider = crm_element_value(rsc->xml, PCMK_XA_PROVIDER); cib_node = pcmk__inject_node(cib, node, NULL); CRM_ASSERT(cib_node != NULL); pcmk__inject_failcount(out, cib_node, resource, task, interval_ms, outcome); cib_resource = pcmk__inject_resource_history(out, cib_node, resource, resource, rclass, rtype, rprovider); CRM_ASSERT(cib_resource != NULL); op = create_op(cib_resource, task, interval_ms, outcome); CRM_ASSERT(op != NULL); cib_op = pcmk__inject_action_result(cib_resource, op, 0); CRM_ASSERT(cib_op != NULL); lrmd_free_event(op); rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); done: free(task); free(node); free(key); } /*! * \internal * \brief Inject fictitious scheduler inputs * * \param[in,out] scheduler Scheduler data * \param[in,out] cib CIB object for scheduler input to modify * \param[in] injections Injections to apply */ void pcmk__inject_scheduler_input(pcmk_scheduler_t *scheduler, cib_t *cib, const pcmk_injections_t *injections) { int rc = pcmk_ok; const GList *iter = NULL; xmlNode *cib_node = NULL; pcmk__output_t *out = scheduler->priv; out->message(out, "inject-modify-config", injections->quorum, injections->watchdog); if (injections->quorum != NULL) { xmlNode *top = create_xml_node(NULL, PCMK_XE_CIB); /* crm_xml_add(top, PCMK_XA_DC_UUID, dc_uuid); */ crm_xml_add(top, PCMK_XA_HAVE_QUORUM, injections->quorum); rc = cib->cmds->modify(cib, NULL, top, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } if (injections->watchdog != NULL) { rc = cib__update_node_attr(out, cib, cib_sync_call|cib_scope_local, PCMK_XE_CRM_CONFIG, NULL, NULL, NULL, NULL, PCMK_OPT_HAVE_WATCHDOG, injections->watchdog, NULL, NULL); CRM_ASSERT(rc == pcmk_rc_ok); } for (iter = injections->node_up; iter != NULL; iter = iter->next) { const char *node = (const char *) iter->data; out->message(out, "inject-modify-node", "Online", node); cib_node = pcmk__inject_node_state_change(cib, node, true); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); } for (iter = injections->node_down; iter != NULL; iter = iter->next) { const char *node = (const char *) iter->data; char *xpath = NULL; out->message(out, "inject-modify-node", "Offline", node); cib_node = pcmk__inject_node_state_change(cib, node, false); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); xpath = crm_strdup_printf("//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" "/" PCMK__XE_LRM, node); cib->cmds->remove(cib, xpath, NULL, cib_xpath|cib_sync_call|cib_scope_local); free(xpath); xpath = crm_strdup_printf("//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" "/" PCMK__XE_TRANSIENT_ATTRIBUTES, node); cib->cmds->remove(cib, xpath, NULL, cib_xpath|cib_sync_call|cib_scope_local); free(xpath); } for (iter = injections->node_fail; iter != NULL; iter = iter->next) { const char *node = (const char *) iter->data; out->message(out, "inject-modify-node", "Failing", node); cib_node = pcmk__inject_node_state_change(cib, node, true); crm_xml_add(cib_node, PCMK__XA_IN_CCM, PCMK_VALUE_FALSE); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, PCMK_XE_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); } for (iter = injections->ticket_grant; iter != NULL; iter = iter->next) { const char *ticket_id = (const char *) iter->data; out->message(out, "inject-modify-ticket", "Granting", ticket_id); rc = set_ticket_state_attr(out, ticket_id, PCMK__XA_GRANTED, true, cib); CRM_ASSERT(rc == pcmk_rc_ok); } for (iter = injections->ticket_revoke; iter != NULL; iter = iter->next) { const char *ticket_id = (const char *) iter->data; out->message(out, "inject-modify-ticket", "Revoking", ticket_id); rc = set_ticket_state_attr(out, ticket_id, PCMK__XA_GRANTED, false, cib); CRM_ASSERT(rc == pcmk_rc_ok); } for (iter = injections->ticket_standby; iter != NULL; iter = iter->next) { const char *ticket_id = (const char *) iter->data; out->message(out, "inject-modify-ticket", "Standby", ticket_id); rc = set_ticket_state_attr(out, ticket_id, PCMK_XA_STANDBY, true, cib); CRM_ASSERT(rc == pcmk_rc_ok); } for (iter = injections->ticket_activate; iter != NULL; iter = iter->next) { const char *ticket_id = (const char *) iter->data; out->message(out, "inject-modify-ticket", "Activating", ticket_id); rc = set_ticket_state_attr(out, ticket_id, PCMK_XA_STANDBY, false, cib); CRM_ASSERT(rc == pcmk_rc_ok); } for (iter = injections->op_inject; iter != NULL; iter = iter->next) { inject_action(out, (const char *) iter->data, cib, scheduler); } if (!out->is_quiet(out)) { out->end_list(out); } } void pcmk_free_injections(pcmk_injections_t *injections) { if (injections == NULL) { return; } g_list_free_full(injections->node_up, g_free); g_list_free_full(injections->node_down, g_free); g_list_free_full(injections->node_fail, g_free); g_list_free_full(injections->op_fail, g_free); g_list_free_full(injections->op_inject, g_free); g_list_free_full(injections->ticket_grant, g_free); g_list_free_full(injections->ticket_revoke, g_free); g_list_free_full(injections->ticket_standby, g_free); g_list_free_full(injections->ticket_activate, g_free); free(injections->quorum); free(injections->watchdog); free(injections); } diff --git a/lib/pengine/pe_output.c b/lib/pengine/pe_output.c index 71fad1fc93..42aa4de259 100644 --- a/lib/pengine/pe_output.c +++ b/lib/pengine/pe_output.c @@ -1,3284 +1,3285 @@ /* * Copyright 2019-2024 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 const char * pe__resource_description(const pcmk_resource_t *rsc, uint32_t show_opts) { const char * desc = NULL; // User-supplied description if (pcmk_any_flags_set(show_opts, pcmk_show_rsc_only|pcmk_show_description)) { desc = crm_element_value(rsc->xml, PCMK_XA_DESCRIPTION); } return desc; } /* Never display node attributes whose name starts with one of these prefixes */ #define FILTER_STR { PCMK__FAIL_COUNT_PREFIX, PCMK__LAST_FAILURE_PREFIX, \ "shutdown", PCMK_NODE_ATTR_TERMINATE, "standby", "#", \ NULL } static int compare_attribute(gconstpointer a, gconstpointer b) { int rc; rc = strcmp((const char *)a, (const char *)b); return rc; } /*! * \internal * \brief Determine whether extended information about an attribute should be added. * * \param[in] node Node that ran this resource * \param[in,out] rsc_list List of resources for this node * \param[in,out] scheduler Scheduler data * \param[in] attrname Attribute to find * \param[out] expected_score Expected value for this attribute * * \return true if extended information should be printed, false otherwise * \note Currently, extended information is only supported for ping/pingd * resources, for which a message will be printed if connectivity is lost * or degraded. */ static bool add_extra_info(const pcmk_node_t *node, GList *rsc_list, pcmk_scheduler_t *scheduler, const char *attrname, int *expected_score) { GList *gIter = NULL; for (gIter = rsc_list; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; const char *type = g_hash_table_lookup(rsc->meta, PCMK_XA_TYPE); const char *name = NULL; GHashTable *params = NULL; if (rsc->children != NULL) { if (add_extra_info(node, rsc->children, scheduler, attrname, expected_score)) { return true; } } if (!pcmk__strcase_any_of(type, "ping", "pingd", NULL)) { continue; } params = pe_rsc_params(rsc, node, scheduler); name = g_hash_table_lookup(params, PCMK_XA_NAME); if (name == NULL) { name = "pingd"; } /* To identify the resource with the attribute name. */ if (pcmk__str_eq(name, attrname, pcmk__str_casei)) { int host_list_num = 0; const char *hosts = g_hash_table_lookup(params, "host_list"); const char *multiplier = g_hash_table_lookup(params, "multiplier"); int multiplier_i; if (hosts) { char **host_list = g_strsplit(hosts, " ", 0); host_list_num = g_strv_length(host_list); g_strfreev(host_list); } if ((multiplier == NULL) || (pcmk__scan_min_int(multiplier, &multiplier_i, INT_MIN) != pcmk_rc_ok)) { /* The ocf:pacemaker:ping resource agent defaults multiplier to * 1. The agent currently does not handle invalid text, but it * should, and this would be a reasonable choice ... */ multiplier_i = 1; } *expected_score = host_list_num * multiplier_i; return true; } } return false; } static GList * filter_attr_list(GList *attr_list, char *name) { int i; const char *filt_str[] = FILTER_STR; CRM_CHECK(name != NULL, return attr_list); /* filtering automatic attributes */ for (i = 0; filt_str[i] != NULL; i++) { if (g_str_has_prefix(name, filt_str[i])) { return attr_list; } } return g_list_insert_sorted(attr_list, name, compare_attribute); } static GList * get_operation_list(xmlNode *rsc_entry) { GList *op_list = NULL; xmlNode *rsc_op = NULL; for (rsc_op = pcmk__xe_first_child(rsc_entry); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) { const char *task = crm_element_value(rsc_op, PCMK_XA_OPERATION); const char *interval_ms_s = crm_element_value(rsc_op, PCMK_META_INTERVAL); const char *op_rc = crm_element_value(rsc_op, PCMK__XA_RC_CODE); int op_rc_i; pcmk__scan_min_int(op_rc, &op_rc_i, 0); /* Display 0-interval monitors as "probe" */ if (pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei) && pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches | pcmk__str_casei)) { task = "probe"; } /* Ignore notifies and some probes */ if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none) || (pcmk__str_eq(task, "probe", pcmk__str_none) && (op_rc_i == CRM_EX_NOT_RUNNING))) { continue; } if (pcmk__str_eq((const char *) rsc_op->name, PCMK__XE_LRM_RSC_OP, pcmk__str_none)) { op_list = g_list_append(op_list, rsc_op); } } op_list = g_list_sort(op_list, sort_op_by_callid); return op_list; } static void add_dump_node(gpointer key, gpointer value, gpointer user_data) { xmlNodePtr node = user_data; pcmk_create_xml_text_node(node, (const char *) key, (const char *) value); } static void append_dump_text(gpointer key, gpointer value, gpointer user_data) { char **dump_text = user_data; char *new_text = crm_strdup_printf("%s %s=%s", *dump_text, (char *)key, (char *)value); free(*dump_text); *dump_text = new_text; } #define XPATH_STACK "//" PCMK_XE_NVPAIR \ "[@" PCMK_XA_NAME "='" \ PCMK_OPT_CLUSTER_INFRASTRUCTURE "']" static const char * get_cluster_stack(pcmk_scheduler_t *scheduler) { xmlNode *stack = get_xpath_object(XPATH_STACK, scheduler->input, LOG_DEBUG); if (stack != NULL) { return crm_element_value(stack, PCMK_XA_VALUE); } return PCMK_VALUE_UNKNOWN; } static char * last_changed_string(const char *last_written, const char *user, const char *client, const char *origin) { if (last_written != NULL || user != NULL || client != NULL || origin != NULL) { return crm_strdup_printf("%s%s%s%s%s%s%s", last_written ? last_written : "", user ? " by " : "", user ? user : "", client ? " via " : "", client ? client : "", origin ? " on " : "", origin ? origin : ""); } else { return strdup(""); } } static char * op_history_string(xmlNode *xml_op, const char *task, const char *interval_ms_s, int rc, bool print_timing) { const char *call = crm_element_value(xml_op, PCMK__XA_CALL_ID); char *interval_str = NULL; char *buf = NULL; if (interval_ms_s && !pcmk__str_eq(interval_ms_s, "0", pcmk__str_casei)) { char *pair = pcmk__format_nvpair(PCMK_XA_INTERVAL, interval_ms_s, "ms"); interval_str = crm_strdup_printf(" %s", pair); free(pair); } if (print_timing) { char *last_change_str = NULL; char *exec_str = NULL; char *queue_str = NULL; const char *value = NULL; time_t epoch = 0; if ((crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &epoch) == pcmk_ok) && (epoch > 0)) { char *epoch_str = pcmk__epoch2str(&epoch, 0); last_change_str = crm_strdup_printf(" %s=\"%s\"", PCMK_XA_LAST_RC_CHANGE, pcmk__s(epoch_str, "")); free(epoch_str); } value = crm_element_value(xml_op, PCMK_XA_EXEC_TIME); if (value) { char *pair = pcmk__format_nvpair(PCMK_XA_EXEC_TIME, value, "ms"); exec_str = crm_strdup_printf(" %s", pair); free(pair); } value = crm_element_value(xml_op, PCMK_XA_QUEUE_TIME); if (value) { char *pair = pcmk__format_nvpair(PCMK_XA_QUEUE_TIME, value, "ms"); queue_str = crm_strdup_printf(" %s", pair); free(pair); } buf = crm_strdup_printf("(%s) %s:%s%s%s%s rc=%d (%s)", call, task, interval_str ? interval_str : "", last_change_str ? last_change_str : "", exec_str ? exec_str : "", queue_str ? queue_str : "", rc, services_ocf_exitcode_str(rc)); if (last_change_str) { free(last_change_str); } if (exec_str) { free(exec_str); } if (queue_str) { free(queue_str); } } else { buf = crm_strdup_printf("(%s) %s%s%s", call, task, interval_str ? ":" : "", interval_str ? interval_str : ""); } if (interval_str) { free(interval_str); } return buf; } static char * resource_history_string(pcmk_resource_t *rsc, const char *rsc_id, bool all, int failcount, time_t last_failure) { char *buf = NULL; if (rsc == NULL) { buf = crm_strdup_printf("%s: orphan", rsc_id); } else if (all || failcount || last_failure > 0) { char *failcount_s = NULL; char *lastfail_s = NULL; if (failcount > 0) { failcount_s = crm_strdup_printf(" %s=%d", PCMK_XA_FAIL_COUNT, failcount); } else { failcount_s = strdup(""); } if (last_failure > 0) { buf = pcmk__epoch2str(&last_failure, 0); lastfail_s = crm_strdup_printf(" %s='%s'", PCMK_XA_LAST_FAILURE, buf); free(buf); } buf = crm_strdup_printf("%s: " PCMK_META_MIGRATION_THRESHOLD "=%d%s%s", rsc_id, rsc->migration_threshold, failcount_s, lastfail_s? lastfail_s : ""); free(failcount_s); free(lastfail_s); } else { buf = crm_strdup_printf("%s:", rsc_id); } return buf; } /*! * \internal * \brief Get a node's feature set for status display purposes * * \param[in] node Node to check * * \return String representation of feature set if the node is fully up (using * "<3.15.1" for older nodes that don't set the #feature-set attribute), * otherwise NULL */ static const char * get_node_feature_set(const pcmk_node_t *node) { if (node->details->online && node->details->expected_up && !pe__is_guest_or_remote_node(node)) { const char *feature_set = g_hash_table_lookup(node->details->attrs, CRM_ATTR_FEATURE_SET); /* The feature set attribute is present since 3.15.1. If it is missing, * then the node must be running an earlier version. */ return pcmk__s(feature_set, "<3.15.1"); } return NULL; } static bool is_mixed_version(pcmk_scheduler_t *scheduler) { const char *feature_set = NULL; for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = gIter->data; const char *node_feature_set = get_node_feature_set(node); if (node_feature_set != NULL) { if (feature_set == NULL) { feature_set = node_feature_set; } else if (strcmp(feature_set, node_feature_set) != 0) { return true; } } } return false; } static char * formatted_xml_buf(const pcmk_resource_t *rsc, bool raw) { if (raw) { return dump_xml_formatted(rsc->orig_xml ? rsc->orig_xml : rsc->xml); } else { return dump_xml_formatted(rsc->xml); } } #define XPATH_DC_VERSION "//" PCMK_XE_NVPAIR \ "[@" PCMK_XA_NAME "='" PCMK_OPT_DC_VERSION "']" PCMK__OUTPUT_ARGS("cluster-summary", "pcmk_scheduler_t *", "enum pcmk_pacemakerd_state", "uint32_t", "uint32_t") static int cluster_summary(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); enum pcmk_pacemakerd_state pcmkd_state = (enum pcmk_pacemakerd_state) va_arg(args, int); uint32_t section_opts = va_arg(args, uint32_t); uint32_t show_opts = va_arg(args, uint32_t); int rc = pcmk_rc_no_output; const char *stack_s = get_cluster_stack(scheduler); if (pcmk_is_set(section_opts, pcmk_section_stack)) { PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary"); out->message(out, "cluster-stack", stack_s, pcmkd_state); } if (pcmk_is_set(section_opts, pcmk_section_dc)) { xmlNode *dc_version = get_xpath_object(XPATH_DC_VERSION, scheduler->input, LOG_DEBUG); const char *dc_version_s = dc_version? crm_element_value(dc_version, PCMK_XA_VALUE) : NULL; const char *quorum = crm_element_value(scheduler->input, PCMK_XA_HAVE_QUORUM); char *dc_name = scheduler->dc_node? pe__node_display_name(scheduler->dc_node, pcmk_is_set(show_opts, pcmk_show_node_id)) : NULL; bool mixed_version = is_mixed_version(scheduler); PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary"); out->message(out, "cluster-dc", scheduler->dc_node, quorum, dc_version_s, dc_name, mixed_version); free(dc_name); } if (pcmk_is_set(section_opts, pcmk_section_times)) { const char *last_written = crm_element_value(scheduler->input, PCMK_XA_CIB_LAST_WRITTEN); const char *user = crm_element_value(scheduler->input, PCMK_XA_UPDATE_USER); const char *client = crm_element_value(scheduler->input, PCMK_XA_UPDATE_CLIENT); const char *origin = crm_element_value(scheduler->input, PCMK_XA_UPDATE_ORIGIN); PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary"); out->message(out, "cluster-times", scheduler->localhost, last_written, user, client, origin); } if (pcmk_is_set(section_opts, pcmk_section_counts)) { PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary"); out->message(out, "cluster-counts", g_list_length(scheduler->nodes), scheduler->ninstances, scheduler->disabled_resources, scheduler->blocked_resources); } if (pcmk_is_set(section_opts, pcmk_section_options)) { PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary"); out->message(out, "cluster-options", scheduler); } PCMK__OUTPUT_LIST_FOOTER(out, rc); if (pcmk_is_set(section_opts, pcmk_section_maint_mode)) { if (out->message(out, "maint-mode", scheduler->flags) == pcmk_rc_ok) { rc = pcmk_rc_ok; } } return rc; } PCMK__OUTPUT_ARGS("cluster-summary", "pcmk_scheduler_t *", "enum pcmk_pacemakerd_state", "uint32_t", "uint32_t") static int cluster_summary_html(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); enum pcmk_pacemakerd_state pcmkd_state = (enum pcmk_pacemakerd_state) va_arg(args, int); uint32_t section_opts = va_arg(args, uint32_t); uint32_t show_opts = va_arg(args, uint32_t); int rc = pcmk_rc_no_output; const char *stack_s = get_cluster_stack(scheduler); if (pcmk_is_set(section_opts, pcmk_section_stack)) { PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary"); out->message(out, "cluster-stack", stack_s, pcmkd_state); } /* Always print DC if none, even if not requested */ if ((scheduler->dc_node == NULL) || pcmk_is_set(section_opts, pcmk_section_dc)) { xmlNode *dc_version = get_xpath_object(XPATH_DC_VERSION, scheduler->input, LOG_DEBUG); const char *dc_version_s = dc_version? crm_element_value(dc_version, PCMK_XA_VALUE) : NULL; const char *quorum = crm_element_value(scheduler->input, PCMK_XA_HAVE_QUORUM); char *dc_name = scheduler->dc_node? pe__node_display_name(scheduler->dc_node, pcmk_is_set(show_opts, pcmk_show_node_id)) : NULL; bool mixed_version = is_mixed_version(scheduler); PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary"); out->message(out, "cluster-dc", scheduler->dc_node, quorum, dc_version_s, dc_name, mixed_version); free(dc_name); } if (pcmk_is_set(section_opts, pcmk_section_times)) { const char *last_written = crm_element_value(scheduler->input, PCMK_XA_CIB_LAST_WRITTEN); const char *user = crm_element_value(scheduler->input, PCMK_XA_UPDATE_USER); const char *client = crm_element_value(scheduler->input, PCMK_XA_UPDATE_CLIENT); const char *origin = crm_element_value(scheduler->input, PCMK_XA_UPDATE_ORIGIN); PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary"); out->message(out, "cluster-times", scheduler->localhost, last_written, user, client, origin); } if (pcmk_is_set(section_opts, pcmk_section_counts)) { PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Cluster Summary"); out->message(out, "cluster-counts", g_list_length(scheduler->nodes), scheduler->ninstances, scheduler->disabled_resources, scheduler->blocked_resources); } if (pcmk_is_set(section_opts, pcmk_section_options)) { /* Kind of a hack - close the list we may have opened earlier in this * function so we can put all the options into their own list. We * only want to do this on HTML output, though. */ PCMK__OUTPUT_LIST_FOOTER(out, rc); out->begin_list(out, NULL, NULL, "Config Options"); out->message(out, "cluster-options", scheduler); } PCMK__OUTPUT_LIST_FOOTER(out, rc); if (pcmk_is_set(section_opts, pcmk_section_maint_mode)) { if (out->message(out, "maint-mode", scheduler->flags) == pcmk_rc_ok) { rc = pcmk_rc_ok; } } return rc; } char * pe__node_display_name(pcmk_node_t *node, bool print_detail) { char *node_name; const char *node_host = NULL; const char *node_id = NULL; int name_len; CRM_ASSERT((node != NULL) && (node->details != NULL) && (node->details->uname != NULL)); /* Host is displayed only if this is a guest node and detail is requested */ if (print_detail && pe__is_guest_node(node)) { const pcmk_resource_t *container = node->details->remote_rsc->container; const pcmk_node_t *host_node = pcmk__current_node(container); if (host_node && host_node->details) { node_host = host_node->details->uname; } if (node_host == NULL) { node_host = ""; /* so we at least get "uname@" to indicate guest */ } } /* Node ID is displayed if different from uname and detail is requested */ if (print_detail && !pcmk__str_eq(node->details->uname, node->details->id, pcmk__str_casei)) { node_id = node->details->id; } /* Determine name length */ name_len = strlen(node->details->uname) + 1; if (node_host) { name_len += strlen(node_host) + 1; /* "@node_host" */ } if (node_id) { name_len += strlen(node_id) + 3; /* + " (node_id)" */ } /* Allocate and populate display name */ node_name = malloc(name_len); CRM_ASSERT(node_name != NULL); strcpy(node_name, node->details->uname); if (node_host) { strcat(node_name, "@"); strcat(node_name, node_host); } if (node_id) { strcat(node_name, " ("); strcat(node_name, node_id); strcat(node_name, ")"); } return node_name; } int pe__name_and_nvpairs_xml(pcmk__output_t *out, bool is_list, const char *tag_name , size_t pairs_count, ...) { xmlNodePtr xml_node = NULL; va_list args; CRM_ASSERT(tag_name != NULL); xml_node = pcmk__output_xml_peek_parent(out); CRM_ASSERT(xml_node != NULL); xml_node = create_xml_node(xml_node, tag_name); va_start(args, pairs_count); while(pairs_count--) { const char *param_name = va_arg(args, const char *); const char *param_value = va_arg(args, const char *); if (param_name && param_value) { crm_xml_add(xml_node, param_name, param_value); } }; va_end(args); if (is_list) { pcmk__output_xml_push_parent(out, xml_node); } return pcmk_rc_ok; } static const char * role_desc(enum rsc_role_e role) { if (role == pcmk_role_promoted) { #ifdef PCMK__COMPAT_2_0 return "as " PCMK__ROLE_PROMOTED_LEGACY " "; #else return "in " PCMK__ROLE_PROMOTED " role "; #endif } return ""; } PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t") static int ban_html(pcmk__output_t *out, va_list args) { pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *); pcmk__location_t *location = va_arg(args, pcmk__location_t *); uint32_t show_opts = va_arg(args, uint32_t); char *node_name = pe__node_display_name(pe_node, pcmk_is_set(show_opts, pcmk_show_node_id)); char *buf = crm_strdup_printf("%s\tprevents %s from running %son %s", location->id, location->rsc->id, role_desc(location->role_filter), node_name); pcmk__output_create_html_node(out, "li", NULL, NULL, buf); free(node_name); free(buf); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t") static int ban_text(pcmk__output_t *out, va_list args) { pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *); pcmk__location_t *location = va_arg(args, pcmk__location_t *); uint32_t show_opts = va_arg(args, uint32_t); char *node_name = pe__node_display_name(pe_node, pcmk_is_set(show_opts, pcmk_show_node_id)); out->list_item(out, NULL, "%s\tprevents %s from running %son %s", location->id, location->rsc->id, role_desc(location->role_filter), node_name); free(node_name); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ban", "pcmk_node_t *", "pcmk__location_t *", "uint32_t") static int ban_xml(pcmk__output_t *out, va_list args) { pcmk_node_t *pe_node = va_arg(args, pcmk_node_t *); pcmk__location_t *location = va_arg(args, pcmk__location_t *); uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t); const char *promoted_only = pcmk__btoa(location->role_filter == pcmk_role_promoted); char *weight_s = pcmk__itoa(pe_node->weight); pcmk__output_create_xml_node(out, PCMK_XE_BAN, PCMK_XA_ID, location->id, PCMK_XA_RESOURCE, location->rsc->id, PCMK_XA_NODE, pe_node->details->uname, PCMK_XA_WEIGHT, weight_s, PCMK_XA_PROMOTED_ONLY, promoted_only, /* This is a deprecated alias for * promoted_only. Removing it will break * backward compatibility of the API schema, * which will require an API schema major * version bump. */ PCMK__XA_PROMOTED_ONLY_LEGACY, promoted_only, NULL); free(weight_s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ban-list", "pcmk_scheduler_t *", "const char *", "GList *", "uint32_t", "bool") static int ban_list(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); const char *prefix = va_arg(args, const char *); GList *only_rsc = va_arg(args, GList *); uint32_t show_opts = va_arg(args, uint32_t); bool print_spacer = va_arg(args, int); GList *gIter, *gIter2; int rc = pcmk_rc_no_output; /* Print each ban */ for (gIter = scheduler->placement_constraints; gIter != NULL; gIter = gIter->next) { pcmk__location_t *location = gIter->data; const pcmk_resource_t *rsc = location->rsc; if (prefix != NULL && !g_str_has_prefix(location->id, prefix)) { continue; } if (!pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) && !pcmk__str_in_list(rsc_printable_id(pe__const_top_resource(rsc, false)), only_rsc, pcmk__str_star_matches)) { continue; } for (gIter2 = location->nodes; gIter2 != NULL; gIter2 = gIter2->next) { pcmk_node_t *node = (pcmk_node_t *) gIter2->data; if (node->weight < 0) { PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Negative Location Constraints"); out->message(out, "ban", node, location, show_opts); } } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int") static int cluster_counts_html(pcmk__output_t *out, va_list args) { unsigned int nnodes = va_arg(args, unsigned int); int nresources = va_arg(args, int); int ndisabled = va_arg(args, int); int nblocked = va_arg(args, int); xmlNodePtr nodes_node = pcmk__output_create_xml_node(out, "li", NULL); xmlNodePtr resources_node = pcmk__output_create_xml_node(out, "li", NULL); char *nnodes_str = crm_strdup_printf("%d node%s configured", nnodes, pcmk__plural_s(nnodes)); pcmk_create_html_node(nodes_node, "span", NULL, NULL, nnodes_str); free(nnodes_str); if (ndisabled && nblocked) { char *s = crm_strdup_printf("%d resource instance%s configured (%d ", nresources, pcmk__plural_s(nresources), ndisabled); pcmk_create_html_node(resources_node, "span", NULL, NULL, s); free(s); pcmk_create_html_node(resources_node, "span", NULL, "bold", "DISABLED"); s = crm_strdup_printf(", %d ", nblocked); pcmk_create_html_node(resources_node, "span", NULL, NULL, s); free(s); pcmk_create_html_node(resources_node, "span", NULL, "bold", "BLOCKED"); pcmk_create_html_node(resources_node, "span", NULL, NULL, " from further action due to failure)"); } else if (ndisabled && !nblocked) { char *s = crm_strdup_printf("%d resource instance%s configured (%d ", nresources, pcmk__plural_s(nresources), ndisabled); pcmk_create_html_node(resources_node, "span", NULL, NULL, s); free(s); pcmk_create_html_node(resources_node, "span", NULL, "bold", "DISABLED"); pcmk_create_html_node(resources_node, "span", NULL, NULL, ")"); } else if (!ndisabled && nblocked) { char *s = crm_strdup_printf("%d resource instance%s configured (%d ", nresources, pcmk__plural_s(nresources), nblocked); pcmk_create_html_node(resources_node, "span", NULL, NULL, s); free(s); pcmk_create_html_node(resources_node, "span", NULL, "bold", "BLOCKED"); pcmk_create_html_node(resources_node, "span", NULL, NULL, " from further action due to failure)"); } else { char *s = crm_strdup_printf("%d resource instance%s configured", nresources, pcmk__plural_s(nresources)); pcmk_create_html_node(resources_node, "span", NULL, NULL, s); free(s); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int") static int cluster_counts_text(pcmk__output_t *out, va_list args) { unsigned int nnodes = va_arg(args, unsigned int); int nresources = va_arg(args, int); int ndisabled = va_arg(args, int); int nblocked = va_arg(args, int); out->list_item(out, NULL, "%d node%s configured", nnodes, pcmk__plural_s(nnodes)); if (ndisabled && nblocked) { out->list_item(out, NULL, "%d resource instance%s configured " "(%d DISABLED, %d BLOCKED from " "further action due to failure)", nresources, pcmk__plural_s(nresources), ndisabled, nblocked); } else if (ndisabled && !nblocked) { out->list_item(out, NULL, "%d resource instance%s configured " "(%d DISABLED)", nresources, pcmk__plural_s(nresources), ndisabled); } else if (!ndisabled && nblocked) { out->list_item(out, NULL, "%d resource instance%s configured " "(%d BLOCKED from further action " "due to failure)", nresources, pcmk__plural_s(nresources), nblocked); } else { out->list_item(out, NULL, "%d resource instance%s configured", nresources, pcmk__plural_s(nresources)); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-counts", "unsigned int", "int", "int", "int") static int cluster_counts_xml(pcmk__output_t *out, va_list args) { unsigned int nnodes = va_arg(args, unsigned int); int nresources = va_arg(args, int); int ndisabled = va_arg(args, int); int nblocked = va_arg(args, int); xmlNodePtr nodes_node = NULL; xmlNodePtr resources_node = NULL; char *s = NULL; nodes_node = pcmk__output_create_xml_node(out, PCMK_XE_NODES_CONFIGURED, NULL); resources_node = pcmk__output_create_xml_node(out, PCMK_XE_RESOURCES_CONFIGURED, NULL); s = pcmk__itoa(nnodes); crm_xml_add(nodes_node, PCMK_XA_NUMBER, s); free(s); s = pcmk__itoa(nresources); crm_xml_add(resources_node, PCMK_XA_NUMBER, s); free(s); s = pcmk__itoa(ndisabled); crm_xml_add(resources_node, PCMK_XA_DISABLED, s); free(s); s = pcmk__itoa(nblocked); crm_xml_add(resources_node, PCMK_XA_BLOCKED, s); free(s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *", "char *", "int") static int cluster_dc_html(pcmk__output_t *out, va_list args) { pcmk_node_t *dc = va_arg(args, pcmk_node_t *); const char *quorum = va_arg(args, const char *); const char *dc_version_s = va_arg(args, const char *); char *dc_name = va_arg(args, char *); bool mixed_version = va_arg(args, int); xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL); pcmk_create_html_node(node, "span", NULL, "bold", "Current DC: "); if (dc) { char *buf = crm_strdup_printf("%s (version %s) -", dc_name, dc_version_s ? dc_version_s : "unknown"); pcmk_create_html_node(node, "span", NULL, NULL, buf); free(buf); if (mixed_version) { pcmk_create_html_node(node, "span", NULL, "warning", " MIXED-VERSION"); } pcmk_create_html_node(node, "span", NULL, NULL, " partition"); if (crm_is_true(quorum)) { pcmk_create_html_node(node, "span", NULL, NULL, " with"); } else { pcmk_create_html_node(node, "span", NULL, "warning", " WITHOUT"); } pcmk_create_html_node(node, "span", NULL, NULL, " quorum"); } else { pcmk_create_html_node(node, "span", NULL, "warning", "NONE"); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *", "char *", "int") static int cluster_dc_text(pcmk__output_t *out, va_list args) { pcmk_node_t *dc = va_arg(args, pcmk_node_t *); const char *quorum = va_arg(args, const char *); const char *dc_version_s = va_arg(args, const char *); char *dc_name = va_arg(args, char *); bool mixed_version = va_arg(args, int); if (dc) { out->list_item(out, "Current DC", "%s (version %s) - %spartition %s quorum", dc_name, dc_version_s ? dc_version_s : "unknown", mixed_version ? "MIXED-VERSION " : "", crm_is_true(quorum) ? "with" : "WITHOUT"); } else { out->list_item(out, "Current DC", "NONE"); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-dc", "pcmk_node_t *", "const char *", "const char *", "char *", "int") static int cluster_dc_xml(pcmk__output_t *out, va_list args) { pcmk_node_t *dc = va_arg(args, pcmk_node_t *); const char *quorum = va_arg(args, const char *); const char *dc_version_s = va_arg(args, const char *); char *dc_name G_GNUC_UNUSED = va_arg(args, char *); bool mixed_version = va_arg(args, int); if (dc) { const char *with_quorum = pcmk__btoa(crm_is_true(quorum)); const char *mixed_version_s = pcmk__btoa(mixed_version); pcmk__output_create_xml_node(out, PCMK_XE_CURRENT_DC, PCMK_XA_PRESENT, PCMK_VALUE_TRUE, PCMK_XA_VERSION, pcmk__s(dc_version_s, ""), PCMK_XA_NAME, dc->details->uname, PCMK_XA_ID, dc->details->id, PCMK_XA_WITH_QUORUM, with_quorum, PCMK_XA_MIXED_VERSION, mixed_version_s, NULL); } else { pcmk__output_create_xml_node(out, PCMK_XE_CURRENT_DC, PCMK_XA_PRESENT, PCMK_VALUE_FALSE, NULL); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("maint-mode", "unsigned long long int") static int cluster_maint_mode_text(pcmk__output_t *out, va_list args) { unsigned long long flags = va_arg(args, unsigned long long); if (pcmk_is_set(flags, pcmk_sched_in_maintenance)) { pcmk__formatted_printf(out, "\n *** Resource management is DISABLED ***\n"); pcmk__formatted_printf(out, " The cluster will not attempt to start, stop or recover services\n"); return pcmk_rc_ok; } else if (pcmk_is_set(flags, pcmk_sched_stop_all)) { pcmk__formatted_printf(out, "\n *** Resource management is DISABLED ***\n"); pcmk__formatted_printf(out, " The cluster will keep all resources stopped\n"); return pcmk_rc_ok; } else { return pcmk_rc_no_output; } } PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *") static int cluster_options_html(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { out->list_item(out, NULL, "STONITH of failed nodes enabled"); } else { out->list_item(out, NULL, "STONITH of failed nodes disabled"); } if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) { out->list_item(out, NULL, "Cluster is symmetric"); } else { out->list_item(out, NULL, "Cluster is asymmetric"); } switch (scheduler->no_quorum_policy) { case pcmk_no_quorum_freeze: out->list_item(out, NULL, "No quorum policy: Freeze resources"); break; case pcmk_no_quorum_stop: out->list_item(out, NULL, "No quorum policy: Stop ALL resources"); break; case pcmk_no_quorum_demote: out->list_item(out, NULL, "No quorum policy: Demote promotable " "resources and stop all other resources"); break; case pcmk_no_quorum_ignore: out->list_item(out, NULL, "No quorum policy: Ignore"); break; case pcmk_no_quorum_fence: out->list_item(out, NULL, "No quorum policy: Suicide"); break; } if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) { xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL); pcmk_create_html_node(node, "span", NULL, NULL, "Resource management: "); pcmk_create_html_node(node, "span", NULL, "bold", "DISABLED"); pcmk_create_html_node(node, "span", NULL, NULL, " (the cluster will not attempt to start, stop, or recover services)"); } else if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_all)) { xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL); pcmk_create_html_node(node, "span", NULL, NULL, "Resource management: "); pcmk_create_html_node(node, "span", NULL, "bold", "STOPPED"); pcmk_create_html_node(node, "span", NULL, NULL, " (the cluster will keep all resources stopped)"); } else { out->list_item(out, NULL, "Resource management: enabled"); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *") static int cluster_options_log(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) { return out->info(out, "Resource management is DISABLED. The cluster will not attempt to start, stop or recover services."); } else if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_all)) { return out->info(out, "Resource management is DISABLED. The cluster has stopped all resources."); } else { return pcmk_rc_no_output; } } PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *") static int cluster_options_text(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { out->list_item(out, NULL, "STONITH of failed nodes enabled"); } else { out->list_item(out, NULL, "STONITH of failed nodes disabled"); } if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) { out->list_item(out, NULL, "Cluster is symmetric"); } else { out->list_item(out, NULL, "Cluster is asymmetric"); } switch (scheduler->no_quorum_policy) { case pcmk_no_quorum_freeze: out->list_item(out, NULL, "No quorum policy: Freeze resources"); break; case pcmk_no_quorum_stop: out->list_item(out, NULL, "No quorum policy: Stop ALL resources"); break; case pcmk_no_quorum_demote: out->list_item(out, NULL, "No quorum policy: Demote promotable " "resources and stop all other resources"); break; case pcmk_no_quorum_ignore: out->list_item(out, NULL, "No quorum policy: Ignore"); break; case pcmk_no_quorum_fence: out->list_item(out, NULL, "No quorum policy: Suicide"); break; } return pcmk_rc_ok; } /*! * \internal * \brief Get readable string representation of a no-quorum policy * * \param[in] policy No-quorum policy * * \return String representation of \p policy */ static const char * no_quorum_policy_text(enum pe_quorum_policy policy) { switch (policy) { case pcmk_no_quorum_freeze: return PCMK_VALUE_FREEZE; case pcmk_no_quorum_stop: return PCMK_VALUE_STOP; case pcmk_no_quorum_demote: return PCMK_VALUE_DEMOTE; case pcmk_no_quorum_ignore: return PCMK_VALUE_IGNORE; case pcmk_no_quorum_fence: return PCMK_VALUE_FENCE_LEGACY; default: return PCMK_VALUE_UNKNOWN; } } PCMK__OUTPUT_ARGS("cluster-options", "pcmk_scheduler_t *") static int cluster_options_xml(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); const char *stonith_enabled = pcmk__flag_text(scheduler->flags, pcmk_sched_fencing_enabled); const char *symmetric_cluster = pcmk__flag_text(scheduler->flags, pcmk_sched_symmetric_cluster); const char *no_quorum_policy = no_quorum_policy_text(scheduler->no_quorum_policy); const char *maintenance_mode = pcmk__flag_text(scheduler->flags, pcmk_sched_in_maintenance); const char *stop_all_resources = pcmk__flag_text(scheduler->flags, pcmk_sched_stop_all); char *stonith_timeout_ms_s = pcmk__itoa(scheduler->stonith_timeout); char *priority_fencing_delay_ms_s = pcmk__itoa(scheduler->priority_fencing_delay * 1000); pcmk__output_create_xml_node(out, PCMK_XE_CLUSTER_OPTIONS, PCMK_XA_STONITH_ENABLED, stonith_enabled, PCMK_XA_SYMMETRIC_CLUSTER, symmetric_cluster, PCMK_XA_NO_QUORUM_POLICY, no_quorum_policy, PCMK_XA_MAINTENANCE_MODE, maintenance_mode, PCMK_XA_STOP_ALL_RESOURCES, stop_all_resources, PCMK_XA_STONITH_TIMEOUT_MS, stonith_timeout_ms_s, PCMK_XA_PRIORITY_FENCING_DELAY_MS, priority_fencing_delay_ms_s, NULL); free(stonith_timeout_ms_s); free(priority_fencing_delay_ms_s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state") static int cluster_stack_html(pcmk__output_t *out, va_list args) { const char *stack_s = va_arg(args, const char *); enum pcmk_pacemakerd_state pcmkd_state = (enum pcmk_pacemakerd_state) va_arg(args, int); xmlNodePtr node = pcmk__output_create_xml_node(out, "li", NULL); pcmk_create_html_node(node, "span", NULL, "bold", "Stack: "); pcmk_create_html_node(node, "span", NULL, NULL, stack_s); if (pcmkd_state != pcmk_pacemakerd_state_invalid) { pcmk_create_html_node(node, "span", NULL, NULL, " ("); pcmk_create_html_node(node, "span", NULL, NULL, pcmk__pcmkd_state_enum2friendly(pcmkd_state)); pcmk_create_html_node(node, "span", NULL, NULL, ")"); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state") static int cluster_stack_text(pcmk__output_t *out, va_list args) { const char *stack_s = va_arg(args, const char *); enum pcmk_pacemakerd_state pcmkd_state = (enum pcmk_pacemakerd_state) va_arg(args, int); if (pcmkd_state != pcmk_pacemakerd_state_invalid) { out->list_item(out, "Stack", "%s (%s)", stack_s, pcmk__pcmkd_state_enum2friendly(pcmkd_state)); } else { out->list_item(out, "Stack", "%s", stack_s); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-stack", "const char *", "enum pcmk_pacemakerd_state") static int cluster_stack_xml(pcmk__output_t *out, va_list args) { const char *stack_s = va_arg(args, const char *); enum pcmk_pacemakerd_state pcmkd_state = (enum pcmk_pacemakerd_state) va_arg(args, int); const char *state_s = NULL; if (pcmkd_state != pcmk_pacemakerd_state_invalid) { state_s = pcmk_pacemakerd_api_daemon_state_enum2text(pcmkd_state); } pcmk__output_create_xml_node(out, "stack", PCMK_XA_TYPE, stack_s, "pacemakerd-state", state_s, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *", "const char *", "const char *", "const char *") static int cluster_times_html(pcmk__output_t *out, va_list args) { const char *our_nodename = va_arg(args, const char *); const char *last_written = va_arg(args, const char *); const char *user = va_arg(args, const char *); const char *client = va_arg(args, const char *); const char *origin = va_arg(args, const char *); xmlNodePtr updated_node = pcmk__output_create_xml_node(out, "li", NULL); xmlNodePtr changed_node = pcmk__output_create_xml_node(out, "li", NULL); char *time_s = pcmk__epoch2str(NULL, 0); pcmk_create_html_node(updated_node, "span", NULL, "bold", "Last updated: "); pcmk_create_html_node(updated_node, "span", NULL, NULL, time_s); if (our_nodename != NULL) { pcmk_create_html_node(updated_node, "span", NULL, NULL, " on "); pcmk_create_html_node(updated_node, "span", NULL, NULL, our_nodename); } free(time_s); time_s = last_changed_string(last_written, user, client, origin); pcmk_create_html_node(changed_node, "span", NULL, "bold", "Last change: "); pcmk_create_html_node(changed_node, "span", NULL, NULL, time_s); free(time_s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *", "const char *", "const char *", "const char *") static int cluster_times_xml(pcmk__output_t *out, va_list args) { const char *our_nodename = va_arg(args, const char *); const char *last_written = va_arg(args, const char *); const char *user = va_arg(args, const char *); const char *client = va_arg(args, const char *); const char *origin = va_arg(args, const char *); char *time_s = pcmk__epoch2str(NULL, 0); pcmk__output_create_xml_node(out, PCMK_XE_LAST_UPDATE, PCMK_XA_TIME, time_s, PCMK_XA_ORIGIN, our_nodename, NULL); pcmk__output_create_xml_node(out, PCMK_XE_LAST_CHANGE, PCMK_XA_TIME, pcmk__s(last_written, ""), PCMK_XA_USER, pcmk__s(user, ""), PCMK_XA_CLIENT, pcmk__s(client, ""), PCMK_XA_ORIGIN, pcmk__s(origin, ""), NULL); free(time_s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-times", "const char *", "const char *", "const char *", "const char *", "const char *") static int cluster_times_text(pcmk__output_t *out, va_list args) { const char *our_nodename = va_arg(args, const char *); const char *last_written = va_arg(args, const char *); const char *user = va_arg(args, const char *); const char *client = va_arg(args, const char *); const char *origin = va_arg(args, const char *); char *time_s = pcmk__epoch2str(NULL, 0); out->list_item(out, "Last updated", "%s%s%s", time_s, (our_nodename != NULL)? " on " : "", pcmk__s(our_nodename, "")); free(time_s); time_s = last_changed_string(last_written, user, client, origin); out->list_item(out, "Last change", " %s", time_s); free(time_s); return pcmk_rc_ok; } /*! * \internal * \brief Display a failed action in less-technical natural language * * \param[in,out] out Output object to use for display * \param[in] xml_op XML containing failed action * \param[in] op_key Operation key of failed action * \param[in] node_name Where failed action occurred * \param[in] rc OCF exit code of failed action * \param[in] status Execution status of failed action * \param[in] exit_reason Exit reason given for failed action * \param[in] exec_time String containing execution time in milliseconds */ static void failed_action_friendly(pcmk__output_t *out, const xmlNode *xml_op, const char *op_key, const char *node_name, int rc, int status, const char *exit_reason, const char *exec_time) { char *rsc_id = NULL; char *task = NULL; guint interval_ms = 0; time_t last_change_epoch = 0; GString *str = NULL; if (pcmk__str_empty(op_key) || !parse_op_key(op_key, &rsc_id, &task, &interval_ms)) { rsc_id = strdup("unknown resource"); task = strdup("unknown action"); interval_ms = 0; } CRM_ASSERT((rsc_id != NULL) && (task != NULL)); str = g_string_sized_new(256); // Should be sufficient for most messages pcmk__g_strcat(str, rsc_id, " ", NULL); if (interval_ms != 0) { pcmk__g_strcat(str, pcmk__readable_interval(interval_ms), "-interval ", NULL); } pcmk__g_strcat(str, pcmk__readable_action(task, interval_ms), " on ", node_name, NULL); if (status == PCMK_EXEC_DONE) { pcmk__g_strcat(str, " returned '", services_ocf_exitcode_str(rc), "'", NULL); if (!pcmk__str_empty(exit_reason)) { pcmk__g_strcat(str, " (", exit_reason, ")", NULL); } } else { pcmk__g_strcat(str, " could not be executed (", pcmk_exec_status_str(status), NULL); if (!pcmk__str_empty(exit_reason)) { pcmk__g_strcat(str, ": ", exit_reason, NULL); } g_string_append_c(str, ')'); } if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &last_change_epoch) == pcmk_ok) { char *s = pcmk__epoch2str(&last_change_epoch, 0); pcmk__g_strcat(str, " at ", s, NULL); free(s); } if (!pcmk__str_empty(exec_time)) { int exec_time_ms = 0; if ((pcmk__scan_min_int(exec_time, &exec_time_ms, 0) == pcmk_rc_ok) && (exec_time_ms > 0)) { pcmk__g_strcat(str, " after ", pcmk__readable_interval(exec_time_ms), NULL); } } out->list_item(out, NULL, "%s", str->str); g_string_free(str, TRUE); free(rsc_id); free(task); } /*! * \internal * \brief Display a failed action with technical details * * \param[in,out] out Output object to use for display * \param[in] xml_op XML containing failed action * \param[in] op_key Operation key of failed action * \param[in] node_name Where failed action occurred * \param[in] rc OCF exit code of failed action * \param[in] status Execution status of failed action * \param[in] exit_reason Exit reason given for failed action * \param[in] exec_time String containing execution time in milliseconds */ static void failed_action_technical(pcmk__output_t *out, const xmlNode *xml_op, const char *op_key, const char *node_name, int rc, int status, const char *exit_reason, const char *exec_time) { const char *call_id = crm_element_value(xml_op, PCMK__XA_CALL_ID); const char *queue_time = crm_element_value(xml_op, PCMK_XA_QUEUE_TIME); const char *exit_status = services_ocf_exitcode_str(rc); const char *lrm_status = pcmk_exec_status_str(status); time_t last_change_epoch = 0; GString *str = NULL; if (pcmk__str_empty(op_key)) { op_key = "unknown operation"; } if (pcmk__str_empty(exit_status)) { exit_status = "unknown exit status"; } if (pcmk__str_empty(call_id)) { call_id = "unknown"; } str = g_string_sized_new(256); g_string_append_printf(str, "%s on %s '%s' (%d): call=%s, status='%s'", op_key, node_name, exit_status, rc, call_id, lrm_status); if (!pcmk__str_empty(exit_reason)) { pcmk__g_strcat(str, ", exitreason='", exit_reason, "'", NULL); } if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &last_change_epoch) == pcmk_ok) { char *last_change_str = pcmk__epoch2str(&last_change_epoch, 0); pcmk__g_strcat(str, ", " PCMK_XA_LAST_RC_CHANGE "=" "'", last_change_str, "'", NULL); free(last_change_str); } if (!pcmk__str_empty(queue_time)) { pcmk__g_strcat(str, ", queued=", queue_time, "ms", NULL); } if (!pcmk__str_empty(exec_time)) { pcmk__g_strcat(str, ", exec=", exec_time, "ms", NULL); } out->list_item(out, NULL, "%s", str->str); g_string_free(str, TRUE); } PCMK__OUTPUT_ARGS("failed-action", "xmlNodePtr", "uint32_t") static int failed_action_default(pcmk__output_t *out, va_list args) { xmlNodePtr xml_op = va_arg(args, xmlNodePtr); uint32_t show_opts = va_arg(args, uint32_t); const char *op_key = pcmk__xe_history_key(xml_op); const char *node_name = crm_element_value(xml_op, PCMK_XA_UNAME); const char *exit_reason = crm_element_value(xml_op, PCMK_XA_EXIT_REASON); const char *exec_time = crm_element_value(xml_op, PCMK_XA_EXEC_TIME); int rc; int status; pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_RC_CODE), &rc, 0); pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS), &status, 0); if (pcmk__str_empty(node_name)) { node_name = "unknown node"; } if (pcmk_is_set(show_opts, pcmk_show_failed_detail)) { failed_action_technical(out, xml_op, op_key, node_name, rc, status, exit_reason, exec_time); } else { failed_action_friendly(out, xml_op, op_key, node_name, rc, status, exit_reason, exec_time); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("failed-action", "xmlNodePtr", "uint32_t") static int failed_action_xml(pcmk__output_t *out, va_list args) { xmlNodePtr xml_op = va_arg(args, xmlNodePtr); uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t); const char *op_key = pcmk__xe_history_key(xml_op); const char *op_key_name = "op_key"; int rc; int status; const char *uname = crm_element_value(xml_op, PCMK_XA_UNAME); const char *call_id = crm_element_value(xml_op, PCMK__XA_CALL_ID); const char *exitstatus = NULL; const char *exit_reason = crm_element_value(xml_op, PCMK_XA_EXIT_REASON); const char *status_s = NULL; time_t epoch = 0; char *rc_s = NULL; char *reason_s = crm_xml_escape(exit_reason ? exit_reason : "none"); xmlNodePtr node = NULL; exit_reason = pcmk__s(reason_s, ""); pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_RC_CODE), &rc, 0); pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS), &status, 0); if (crm_element_value(xml_op, PCMK__XA_OPERATION_KEY) == NULL) { op_key_name = PCMK_XA_ID; } exitstatus = services_ocf_exitcode_str(rc); rc_s = pcmk__itoa(rc); status_s = pcmk_exec_status_str(status); node = pcmk__output_create_xml_node(out, PCMK_XE_FAILURE, op_key_name, op_key, PCMK_XA_NODE, uname, PCMK_XA_EXITSTATUS, exitstatus, PCMK_XA_EXITREASON, exit_reason, PCMK_XA_EXITCODE, rc_s, PCMK_XA_CALL, call_id, PCMK_XA_STATUS, status_s, NULL); free(rc_s); if ((crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &epoch) == pcmk_ok) && (epoch > 0)) { const char *queue_time = crm_element_value(xml_op, PCMK_XA_QUEUE_TIME); const char *exec = crm_element_value(xml_op, PCMK_XA_EXEC_TIME); guint interval_ms = 0; char *interval_ms_s = NULL; char *rc_change = pcmk__epoch2str(&epoch, crm_time_log_date |crm_time_log_timeofday |crm_time_log_with_timezone); crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &interval_ms); interval_ms_s = crm_strdup_printf("%u", interval_ms); pcmk__xe_set_props(node, PCMK_XA_LAST_RC_CHANGE, rc_change, "queued", queue_time, PCMK_XA_EXEC, exec, PCMK_XA_INTERVAL, interval_ms_s, "task", crm_element_value(xml_op, PCMK_XA_OPERATION), NULL); free(interval_ms_s); free(rc_change); } free(reason_s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("failed-action-list", "pcmk_scheduler_t *", "GList *", "GList *", "uint32_t", "bool") static int failed_action_list(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); uint32_t show_opts = va_arg(args, uint32_t); bool print_spacer = va_arg(args, int); xmlNode *xml_op = NULL; int rc = pcmk_rc_no_output; if (xmlChildElementCount(scheduler->failed) == 0) { return rc; } for (xml_op = pcmk__xml_first_child(scheduler->failed); xml_op != NULL; xml_op = pcmk__xml_next(xml_op)) { char *rsc = NULL; if (!pcmk__str_in_list(crm_element_value(xml_op, PCMK_XA_UNAME), only_node, pcmk__str_star_matches|pcmk__str_casei)) { continue; } if (pcmk_xe_mask_probe_failure(xml_op)) { continue; } if (!parse_op_key(pcmk__xe_history_key(xml_op), &rsc, NULL, NULL)) { continue; } if (!pcmk__str_in_list(rsc, only_rsc, pcmk__str_star_matches)) { free(rsc); continue; } free(rsc); PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Failed Resource Actions"); out->message(out, "failed-action", xml_op, show_opts); } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } static void status_node(pcmk_node_t *node, xmlNodePtr parent, uint32_t show_opts) { int health = pe__node_health(node); // Cluster membership if (node->details->online) { pcmk_create_html_node(parent, "span", NULL, PCMK_VALUE_ONLINE, " online"); } else { - pcmk_create_html_node(parent, "span", NULL, "offline", " OFFLINE"); + pcmk_create_html_node(parent, "span", NULL, PCMK_VALUE_OFFLINE, + " OFFLINE"); } // Standby mode if (node->details->standby_onfail && (node->details->running_rsc != NULL)) { pcmk_create_html_node(parent, "span", NULL, PCMK_VALUE_STANDBY, " (in standby due to on-fail," " with active resources)"); } else if (node->details->standby_onfail) { pcmk_create_html_node(parent, "span", NULL, PCMK_VALUE_STANDBY, " (in standby due to on-fail)"); } else if (node->details->standby && (node->details->running_rsc != NULL)) { pcmk_create_html_node(parent, "span", NULL, PCMK_VALUE_STANDBY, " (in standby, with active resources)"); } else if (node->details->standby) { pcmk_create_html_node(parent, "span", NULL, PCMK_VALUE_STANDBY, " (in standby)"); } // Maintenance mode if (node->details->maintenance) { pcmk_create_html_node(parent, "span", NULL, "maint", " (in maintenance mode)"); } // Node health if (health < 0) { pcmk_create_html_node(parent, "span", NULL, "health_red", " (health is RED)"); } else if (health == 0) { pcmk_create_html_node(parent, "span", NULL, "health_yellow", " (health is YELLOW)"); } // Feature set if (pcmk_is_set(show_opts, pcmk_show_feature_set)) { const char *feature_set = get_node_feature_set(node); if (feature_set != NULL) { char *buf = crm_strdup_printf(", feature set %s", feature_set); pcmk_create_html_node(parent, "span", NULL, NULL, buf); free(buf); } } } PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool", "GList *", "GList *") static int node_html(pcmk__output_t *out, va_list args) { pcmk_node_t *node = va_arg(args, pcmk_node_t *); uint32_t show_opts = va_arg(args, uint32_t); bool full = va_arg(args, int); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id)); if (full) { xmlNodePtr item_node; if (pcmk_all_flags_set(show_opts, pcmk_show_brief | pcmk_show_rscs_by_node)) { GList *rscs = pe__filter_rsc_list(node->details->running_rsc, only_rsc); out->begin_list(out, NULL, NULL, "%s:", node_name); item_node = pcmk__output_xml_create_parent(out, "li", NULL); pcmk_create_html_node(item_node, "span", NULL, NULL, "Status:"); status_node(node, item_node, show_opts); if (rscs != NULL) { uint32_t new_show_opts = (show_opts | pcmk_show_rsc_only) & ~pcmk_show_inactive_rscs; out->begin_list(out, NULL, NULL, "Resources"); pe__rscs_brief_output(out, rscs, new_show_opts); out->end_list(out); } pcmk__output_xml_pop_parent(out); out->end_list(out); } else if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) { GList *lpc2 = NULL; int rc = pcmk_rc_no_output; out->begin_list(out, NULL, NULL, "%s:", node_name); item_node = pcmk__output_xml_create_parent(out, "li", NULL); pcmk_create_html_node(item_node, "span", NULL, NULL, "Status:"); status_node(node, item_node, show_opts); for (lpc2 = node->details->running_rsc; lpc2 != NULL; lpc2 = lpc2->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) lpc2->data; PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Resources"); show_opts |= pcmk_show_rsc_only; out->message(out, crm_map_element_name(rsc->xml), show_opts, rsc, only_node, only_rsc); } PCMK__OUTPUT_LIST_FOOTER(out, rc); pcmk__output_xml_pop_parent(out); out->end_list(out); } else { char *buf = crm_strdup_printf("%s:", node_name); item_node = pcmk__output_create_xml_node(out, "li", NULL); pcmk_create_html_node(item_node, "span", NULL, "bold", buf); status_node(node, item_node, show_opts); free(buf); } } else { out->begin_list(out, NULL, NULL, "%s:", node_name); } free(node_name); return pcmk_rc_ok; } /*! * \internal * \brief Get a human-friendly textual description of a node's status * * \param[in] node Node to check * * \return String representation of node's status */ static const char * node_text_status(const pcmk_node_t *node) { if (node->details->unclean) { if (node->details->online) { return "UNCLEAN (online)"; } else if (node->details->pending) { return "UNCLEAN (pending)"; } else { return "UNCLEAN (offline)"; } } else if (node->details->pending) { return "pending"; } else if (node->details->standby_onfail && node->details->online) { return "standby (on-fail)"; } else if (node->details->standby) { if (node->details->online) { if (node->details->running_rsc) { return "standby (with active resources)"; } else { return "standby"; } } else { return "OFFLINE (standby)"; } } else if (node->details->maintenance) { if (node->details->online) { return "maintenance"; } else { return "OFFLINE (maintenance)"; } } else if (node->details->online) { return "online"; } return "OFFLINE"; } PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool", "GList *", "GList *") static int node_text(pcmk__output_t *out, va_list args) { pcmk_node_t *node = va_arg(args, pcmk_node_t *); uint32_t show_opts = va_arg(args, uint32_t); bool full = va_arg(args, int); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); if (full) { char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id)); GString *str = g_string_sized_new(64); int health = pe__node_health(node); // Create a summary line with node type, name, and status if (pe__is_guest_node(node)) { g_string_append(str, "GuestNode"); } else if (pe__is_remote_node(node)) { g_string_append(str, "RemoteNode"); } else { g_string_append(str, "Node"); } pcmk__g_strcat(str, " ", node_name, ": ", node_text_status(node), NULL); if (health < 0) { g_string_append(str, " (health is RED)"); } else if (health == 0) { g_string_append(str, " (health is YELLOW)"); } if (pcmk_is_set(show_opts, pcmk_show_feature_set)) { const char *feature_set = get_node_feature_set(node); if (feature_set != NULL) { pcmk__g_strcat(str, ", feature set ", feature_set, NULL); } } /* If we're grouping by node, print its resources */ if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) { if (pcmk_is_set(show_opts, pcmk_show_brief)) { GList *rscs = pe__filter_rsc_list(node->details->running_rsc, only_rsc); if (rscs != NULL) { uint32_t new_show_opts = (show_opts | pcmk_show_rsc_only) & ~pcmk_show_inactive_rscs; out->begin_list(out, NULL, NULL, "%s", str->str); out->begin_list(out, NULL, NULL, "Resources"); pe__rscs_brief_output(out, rscs, new_show_opts); out->end_list(out); out->end_list(out); g_list_free(rscs); } } else { GList *gIter2 = NULL; out->begin_list(out, NULL, NULL, "%s", str->str); out->begin_list(out, NULL, NULL, "Resources"); for (gIter2 = node->details->running_rsc; gIter2 != NULL; gIter2 = gIter2->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter2->data; show_opts |= pcmk_show_rsc_only; out->message(out, crm_map_element_name(rsc->xml), show_opts, rsc, only_node, only_rsc); } out->end_list(out); out->end_list(out); } } else { out->list_item(out, NULL, "%s", str->str); } g_string_free(str, TRUE); free(node_name); } else { char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id)); out->begin_list(out, NULL, NULL, "Node: %s", node_name); free(node_name); } return pcmk_rc_ok; } /*! * \internal * \brief Convert an integer health value to a string representation * * \param[in] health Integer health value * * \retval \c PCMK_VALUE_RED if \p health is less than 0 * \retval \c PCMK_VALUE_YELLOW if \p health is equal to 0 * \retval \c PCMK_VALUE_GREEN if \p health is greater than 0 */ static const char * health_text(int health) { if (health < 0) { return PCMK_VALUE_RED; } else if (health == 0) { return PCMK_VALUE_YELLOW; } else { return PCMK_VALUE_GREEN; } } /*! * \internal * \brief Convert a node type to a string representation * * \param[in] type Node type * * \retval \c PCMK_VALUE_MEMBER if \p node_type is \c pcmk_node_variant_cluster * \retval \c PCMK_VALUE_REMOTE if \p node_type is \c pcmk_node_variant_remote * \retval \c PCMK__VALUE_PING if \p node_type is \c node_ping * \retval \c PCMK_VALUE_UNKNOWN otherwise */ static const char * node_type_str(enum node_type type) { switch (type) { case pcmk_node_variant_cluster: return PCMK_VALUE_MEMBER; case pcmk_node_variant_remote: return PCMK_VALUE_REMOTE; case node_ping: return PCMK__VALUE_PING; default: return PCMK_VALUE_UNKNOWN; } } PCMK__OUTPUT_ARGS("node", "pcmk_node_t *", "uint32_t", "bool", "GList *", "GList *") static int node_xml(pcmk__output_t *out, va_list args) { pcmk_node_t *node = va_arg(args, pcmk_node_t *); uint32_t show_opts G_GNUC_UNUSED = va_arg(args, uint32_t); bool full = va_arg(args, int); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); if (full) { const char *online = pcmk__btoa(node->details->online); const char *standby = pcmk__btoa(node->details->standby); const char *standby_onfail = pcmk__btoa(node->details->standby_onfail); const char *maintenance = pcmk__btoa(node->details->maintenance); const char *pending = pcmk__btoa(node->details->pending); const char *unclean = pcmk__btoa(node->details->unclean); const char *health = health_text(pe__node_health(node)); const char *feature_set = get_node_feature_set(node); const char *shutdown = pcmk__btoa(node->details->shutdown); const char *expected_up = pcmk__btoa(node->details->expected_up); const char *is_dc = pcmk__btoa(node->details->is_dc); int length = g_list_length(node->details->running_rsc); char *resources_running = pcmk__itoa(length); const char *node_type = node_type_str(node->details->type); pe__name_and_nvpairs_xml(out, true, PCMK_XE_NODE, 15, PCMK_XA_NAME, node->details->uname, PCMK_XA_ID, node->details->id, PCMK_XA_ONLINE, online, PCMK_XA_STANDBY, standby, PCMK_XA_STANDBY_ONFAIL, standby_onfail, PCMK_XA_MAINTENANCE, maintenance, PCMK_XA_PENDING, pending, PCMK_XA_UNCLEAN, unclean, PCMK_XA_HEALTH, health, PCMK_XA_FEATURE_SET, feature_set, PCMK_XA_SHUTDOWN, shutdown, PCMK_XA_EXPECTED_UP, expected_up, PCMK_XA_IS_DC, is_dc, PCMK_XA_RESOURCES_RUNNING, resources_running, PCMK_XA_TYPE, node_type); if (pe__is_guest_node(node)) { xmlNodePtr xml_node = pcmk__output_xml_peek_parent(out); crm_xml_add(xml_node, PCMK_XA_ID_AS_RESOURCE, node->details->remote_rsc->container->id); } if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) { GList *lpc = NULL; for (lpc = node->details->running_rsc; lpc != NULL; lpc = lpc->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) lpc->data; show_opts |= pcmk_show_rsc_only; out->message(out, crm_map_element_name(rsc->xml), show_opts, rsc, only_node, only_rsc); } } free(resources_running); out->end_list(out); } else { pcmk__output_xml_create_parent(out, PCMK_XE_NODE, PCMK_XA_NAME, node->details->uname, NULL); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int") static int node_attribute_text(pcmk__output_t *out, va_list args) { const char *name = va_arg(args, const char *); const char *value = va_arg(args, const char *); bool add_extra = va_arg(args, int); int expected_score = va_arg(args, int); if (add_extra) { int v; if (value == NULL) { v = 0; } else { pcmk__scan_min_int(value, &v, INT_MIN); } if (v <= 0) { out->list_item(out, NULL, "%-32s\t: %-10s\t: Connectivity is lost", name, value); } else if (v < expected_score) { out->list_item(out, NULL, "%-32s\t: %-10s\t: Connectivity is degraded (Expected=%d)", name, value, expected_score); } else { out->list_item(out, NULL, "%-32s\t: %-10s", name, value); } } else { out->list_item(out, NULL, "%-32s\t: %-10s", name, value); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int") static int node_attribute_html(pcmk__output_t *out, va_list args) { const char *name = va_arg(args, const char *); const char *value = va_arg(args, const char *); bool add_extra = va_arg(args, int); int expected_score = va_arg(args, int); if (add_extra) { int v; char *s = crm_strdup_printf("%s: %s", name, value); xmlNodePtr item_node = pcmk__output_create_xml_node(out, "li", NULL); if (value == NULL) { v = 0; } else { pcmk__scan_min_int(value, &v, INT_MIN); } pcmk_create_html_node(item_node, "span", NULL, NULL, s); free(s); if (v <= 0) { pcmk_create_html_node(item_node, "span", NULL, "bold", "(connectivity is lost)"); } else if (v < expected_score) { char *buf = crm_strdup_printf("(connectivity is degraded -- expected %d", expected_score); pcmk_create_html_node(item_node, "span", NULL, "bold", buf); free(buf); } } else { out->list_item(out, NULL, "%s: %s", name, value); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-and-op", "pcmk_scheduler_t *", "xmlNodePtr") static int node_and_op(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); xmlNodePtr xml_op = va_arg(args, xmlNodePtr); pcmk_resource_t *rsc = NULL; gchar *node_str = NULL; char *last_change_str = NULL; const char *op_rsc = crm_element_value(xml_op, PCMK_XA_RESOURCE); int status; time_t last_change = 0; pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS), &status, PCMK_EXEC_UNKNOWN); rsc = pe_find_resource(scheduler->resources, op_rsc); if (rsc) { const pcmk_node_t *node = pcmk__current_node(rsc); const char *target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); uint32_t show_opts = pcmk_show_rsc_only | pcmk_show_pending; if (node == NULL) { node = rsc->pending_node; } node_str = pcmk__native_output_string(rsc, rsc_printable_id(rsc), node, show_opts, target_role, false); } else { node_str = crm_strdup_printf("Unknown resource %s", op_rsc); } if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &last_change) == pcmk_ok) { const char *exec_time = crm_element_value(xml_op, PCMK_XA_EXEC_TIME); last_change_str = crm_strdup_printf(", %s='%s', exec=%sms", PCMK_XA_LAST_RC_CHANGE, pcmk__trim(ctime(&last_change)), exec_time); } out->list_item(out, NULL, "%s: %s (node=%s, call=%s, rc=%s%s): %s", node_str, pcmk__xe_history_key(xml_op), crm_element_value(xml_op, PCMK_XA_UNAME), crm_element_value(xml_op, PCMK__XA_CALL_ID), crm_element_value(xml_op, PCMK__XA_RC_CODE), last_change_str ? last_change_str : "", pcmk_exec_status_str(status)); g_free(node_str); free(last_change_str); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-and-op", "pcmk_scheduler_t *", "xmlNodePtr") static int node_and_op_xml(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); xmlNodePtr xml_op = va_arg(args, xmlNodePtr); pcmk_resource_t *rsc = NULL; const char *uname = crm_element_value(xml_op, PCMK_XA_UNAME); const char *call_id = crm_element_value(xml_op, PCMK__XA_CALL_ID); const char *rc_s = crm_element_value(xml_op, PCMK__XA_RC_CODE); const char *status_s = NULL; const char *op_rsc = crm_element_value(xml_op, PCMK_XA_RESOURCE); int status; time_t last_change = 0; xmlNode *node = NULL; pcmk__scan_min_int(crm_element_value(xml_op, PCMK__XA_OP_STATUS), &status, PCMK_EXEC_UNKNOWN); status_s = pcmk_exec_status_str(status); node = pcmk__output_create_xml_node(out, PCMK_XE_OPERATION, PCMK_XA_OP, pcmk__xe_history_key(xml_op), PCMK_XA_NODE, uname, PCMK_XA_CALL, call_id, "rc", rc_s, PCMK_XA_STATUS, status_s, NULL); rsc = pe_find_resource(scheduler->resources, op_rsc); if (rsc) { const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS); const char *provider = crm_element_value(rsc->xml, PCMK_XA_PROVIDER); const char *kind = crm_element_value(rsc->xml, PCMK_XA_TYPE); bool has_provider = pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider); char *agent_tuple = crm_strdup_printf("%s:%s:%s", class, (has_provider? provider : ""), kind); pcmk__xe_set_props(node, PCMK_XA_RSC, rsc_printable_id(rsc), PCMK_XA_AGENT, agent_tuple, NULL); free(agent_tuple); } if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &last_change) == pcmk_ok) { const char *last_rc_change = pcmk__trim(ctime(&last_change)); const char *exec_time = crm_element_value(xml_op, PCMK_XA_EXEC_TIME); pcmk__xe_set_props(node, PCMK_XA_LAST_RC_CHANGE, last_rc_change, PCMK_XA_EXEC_TIME, exec_time, NULL); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-attribute", "const char *", "const char *", "bool", "int") static int node_attribute_xml(pcmk__output_t *out, va_list args) { const char *name = va_arg(args, const char *); const char *value = va_arg(args, const char *); bool add_extra = va_arg(args, int); int expected_score = va_arg(args, int); xmlNodePtr node = pcmk__output_create_xml_node(out, PCMK_XE_ATTRIBUTE, PCMK_XA_NAME, name, PCMK_XA_VALUE, value, NULL); if (add_extra) { char *buf = pcmk__itoa(expected_score); crm_xml_add(node, PCMK_XA_EXPECTED, buf); free(buf); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-attribute-list", "pcmk_scheduler_t *", "uint32_t", "bool", "GList *", "GList *") static int node_attribute_list(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); uint32_t show_opts = va_arg(args, uint32_t); bool print_spacer = va_arg(args, int); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); int rc = pcmk_rc_no_output; /* Display each node's attributes */ for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = gIter->data; GList *attr_list = NULL; GHashTableIter iter; gpointer key; if (!node || !node->details || !node->details->online) { continue; } g_hash_table_iter_init(&iter, node->details->attrs); while (g_hash_table_iter_next (&iter, &key, NULL)) { attr_list = filter_attr_list(attr_list, key); } if (attr_list == NULL) { continue; } if (!pcmk__str_in_list(node->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { g_list_free(attr_list); continue; } PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Node Attributes"); out->message(out, "node", node, show_opts, false, only_node, only_rsc); for (GList *aIter = attr_list; aIter != NULL; aIter = aIter->next) { const char *name = aIter->data; const char *value = NULL; int expected_score = 0; bool add_extra = false; value = pe_node_attribute_raw(node, name); add_extra = add_extra_info(node, node->details->running_rsc, scheduler, name, &expected_score); /* Print attribute name and value */ out->message(out, "node-attribute", name, value, add_extra, expected_score); } g_list_free(attr_list); out->end_list(out); } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("node-capacity", "const pcmk_node_t *", "const char *") static int node_capacity(pcmk__output_t *out, va_list args) { const pcmk_node_t *node = va_arg(args, pcmk_node_t *); const char *comment = va_arg(args, const char *); char *dump_text = crm_strdup_printf("%s: %s capacity:", comment, pcmk__node_name(node)); g_hash_table_foreach(node->details->utilization, append_dump_text, &dump_text); out->list_item(out, NULL, "%s", dump_text); free(dump_text); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-capacity", "const pcmk_node_t *", "const char *") static int node_capacity_xml(pcmk__output_t *out, va_list args) { const pcmk_node_t *node = va_arg(args, pcmk_node_t *); const char *uname = node->details->uname; const char *comment = va_arg(args, const char *); xmlNodePtr xml_node = pcmk__output_create_xml_node(out, PCMK_XE_CAPACITY, PCMK_XA_NODE, uname, PCMK_XA_COMMENT, comment, NULL); g_hash_table_foreach(node->details->utilization, add_dump_node, xml_node); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-history-list", "pcmk_scheduler_t *", "pcmk_node_t *", "xmlNodePtr", "GList *", "GList *", "uint32_t", "uint32_t") static int node_history_list(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); pcmk_node_t *node = va_arg(args, pcmk_node_t *); xmlNode *node_state = va_arg(args, xmlNode *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); uint32_t section_opts = va_arg(args, uint32_t); uint32_t show_opts = va_arg(args, uint32_t); xmlNode *lrm_rsc = NULL; xmlNode *rsc_entry = NULL; int rc = pcmk_rc_no_output; lrm_rsc = find_xml_node(node_state, PCMK__XE_LRM, FALSE); lrm_rsc = find_xml_node(lrm_rsc, PCMK__XE_LRM_RESOURCES, FALSE); /* Print history of each of the node's resources */ for (rsc_entry = first_named_child(lrm_rsc, PCMK__XE_LRM_RESOURCE); rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) { const char *rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID); pcmk_resource_t *rsc = pe_find_resource(scheduler->resources, rsc_id); const pcmk_resource_t *parent = pe__const_top_resource(rsc, false); /* We can't use is_filtered here to filter group resources. For is_filtered, * we have to decide whether to check the parent or not. If we check the * parent, all elements of a group will always be printed because that's how * is_filtered works for groups. If we do not check the parent, sometimes * this will filter everything out. * * For other resource types, is_filtered is okay. */ if (parent->variant == pcmk_rsc_variant_group) { if (!pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) && !pcmk__str_in_list(rsc_printable_id(parent), only_rsc, pcmk__str_star_matches)) { continue; } } else { if (rsc->fns->is_filtered(rsc, only_rsc, TRUE)) { continue; } } if (!pcmk_is_set(section_opts, pcmk_section_operations)) { time_t last_failure = 0; int failcount = pe_get_failcount(node, rsc, &last_failure, pcmk__fc_default, NULL); if (failcount <= 0) { continue; } if (rc == pcmk_rc_no_output) { rc = pcmk_rc_ok; out->message(out, "node", node, show_opts, false, only_node, only_rsc); } out->message(out, "resource-history", rsc, rsc_id, false, failcount, last_failure, false); } else { GList *op_list = get_operation_list(rsc_entry); pcmk_resource_t *rsc = NULL; if (op_list == NULL) { continue; } rsc = pe_find_resource(scheduler->resources, crm_element_value(rsc_entry, PCMK_XA_ID)); if (rc == pcmk_rc_no_output) { rc = pcmk_rc_ok; out->message(out, "node", node, show_opts, false, only_node, only_rsc); } out->message(out, "resource-operation-list", scheduler, rsc, node, op_list, show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool") static int node_list_html(pcmk__output_t *out, va_list args) { GList *nodes = va_arg(args, GList *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); uint32_t show_opts = va_arg(args, uint32_t); bool print_spacer G_GNUC_UNUSED = va_arg(args, int); int rc = pcmk_rc_no_output; for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = (pcmk_node_t *) gIter->data; if (!pcmk__str_in_list(node->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { continue; } PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Node List"); out->message(out, "node", node, show_opts, true, only_node, only_rsc); } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool") static int node_list_text(pcmk__output_t *out, va_list args) { GList *nodes = va_arg(args, GList *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); uint32_t show_opts = va_arg(args, uint32_t); bool print_spacer = va_arg(args, int); /* space-separated lists of node names */ GString *online_nodes = NULL; GString *online_remote_nodes = NULL; GString *online_guest_nodes = NULL; GString *offline_nodes = NULL; GString *offline_remote_nodes = NULL; int rc = pcmk_rc_no_output; for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = (pcmk_node_t *) gIter->data; char *node_name = pe__node_display_name(node, pcmk_is_set(show_opts, pcmk_show_node_id)); if (!pcmk__str_in_list(node->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { free(node_name); continue; } PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, "Node List"); // Determine whether to display node individually or in a list if (node->details->unclean || node->details->pending || (node->details->standby_onfail && node->details->online) || node->details->standby || node->details->maintenance || pcmk_is_set(show_opts, pcmk_show_rscs_by_node) || pcmk_is_set(show_opts, pcmk_show_feature_set) || (pe__node_health(node) <= 0)) { // Display node individually } else if (node->details->online) { // Display online node in a list if (pe__is_guest_node(node)) { pcmk__add_word(&online_guest_nodes, 1024, node_name); } else if (pe__is_remote_node(node)) { pcmk__add_word(&online_remote_nodes, 1024, node_name); } else { pcmk__add_word(&online_nodes, 1024, node_name); } free(node_name); continue; } else { // Display offline node in a list if (pe__is_remote_node(node)) { pcmk__add_word(&offline_remote_nodes, 1024, node_name); } else if (pe__is_guest_node(node)) { /* ignore offline guest nodes */ } else { pcmk__add_word(&offline_nodes, 1024, node_name); } free(node_name); continue; } /* If we get here, node is in bad state, or we're grouping by node */ out->message(out, "node", node, show_opts, true, only_node, only_rsc); free(node_name); } /* If we're not grouping by node, summarize nodes by status */ if (online_nodes != NULL) { out->list_item(out, "Online", "[ %s ]", (const char *) online_nodes->str); g_string_free(online_nodes, TRUE); } if (offline_nodes != NULL) { out->list_item(out, "OFFLINE", "[ %s ]", (const char *) offline_nodes->str); g_string_free(offline_nodes, TRUE); } if (online_remote_nodes) { out->list_item(out, "RemoteOnline", "[ %s ]", (const char *) online_remote_nodes->str); g_string_free(online_remote_nodes, TRUE); } if (offline_remote_nodes) { out->list_item(out, "RemoteOFFLINE", "[ %s ]", (const char *) offline_remote_nodes->str); g_string_free(offline_remote_nodes, TRUE); } if (online_guest_nodes != NULL) { out->list_item(out, "GuestOnline", "[ %s ]", (const char *) online_guest_nodes->str); g_string_free(online_guest_nodes, TRUE); } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("node-list", "GList *", "GList *", "GList *", "uint32_t", "bool") static int node_list_xml(pcmk__output_t *out, va_list args) { GList *nodes = va_arg(args, GList *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); uint32_t show_opts = va_arg(args, uint32_t); bool print_spacer G_GNUC_UNUSED = va_arg(args, int); /* PCMK_XE_NODES acts as the list's element name for CLI tools that force * --xml-simple-list. Otherwise PCMK_XE_NODES is the value of the list's * PCMK_XA_NAME attribute. */ out->begin_list(out, NULL, NULL, PCMK_XE_NODES); for (GList *gIter = nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = (pcmk_node_t *) gIter->data; if (!pcmk__str_in_list(node->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { continue; } out->message(out, "node", node, show_opts, true, only_node, only_rsc); } out->end_list(out); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-summary", "pcmk_scheduler_t *", "GList *", "GList *", "uint32_t", "uint32_t", "bool") static int node_summary(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); uint32_t section_opts = va_arg(args, uint32_t); uint32_t show_opts = va_arg(args, uint32_t); bool print_spacer = va_arg(args, int); xmlNode *node_state = NULL; xmlNode *cib_status = pcmk_find_cib_element(scheduler->input, PCMK_XE_STATUS); int rc = pcmk_rc_no_output; if (xmlChildElementCount(cib_status) == 0) { return rc; } for (node_state = first_named_child(cib_status, PCMK__XE_NODE_STATE); node_state != NULL; node_state = crm_next_same_xml(node_state)) { pcmk_node_t *node = pe_find_node_id(scheduler->nodes, ID(node_state)); if (!node || !node->details || !node->details->online) { continue; } if (!pcmk__str_in_list(node->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { continue; } PCMK__OUTPUT_LIST_HEADER(out, print_spacer, rc, pcmk_is_set(section_opts, pcmk_section_operations) ? "Operations" : "Migration Summary"); out->message(out, "node-history-list", scheduler, node, node_state, only_node, only_rsc, section_opts, show_opts); } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("node-weight", "const pcmk_resource_t *", "const char *", "const char *", "const char *") static int node_weight(pcmk__output_t *out, va_list args) { const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *); const char *prefix = va_arg(args, const char *); const char *uname = va_arg(args, const char *); const char *score = va_arg(args, const char *); if (rsc) { out->list_item(out, NULL, "%s: %s allocation score on %s: %s", prefix, rsc->id, uname, score); } else { out->list_item(out, NULL, "%s: %s = %s", prefix, uname, score); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-weight", "const pcmk_resource_t *", "const char *", "const char *", "const char *") static int node_weight_xml(pcmk__output_t *out, va_list args) { const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *); const char *prefix = va_arg(args, const char *); const char *uname = va_arg(args, const char *); const char *score = va_arg(args, const char *); xmlNodePtr node = pcmk__output_create_xml_node(out, PCMK_XE_NODE_WEIGHT, PCMK_XA_FUNCTION, prefix, PCMK_XA_NODE, uname, PCMK_XA_SCORE, score, NULL); if (rsc) { crm_xml_add(node, PCMK_XA_ID, rsc->id); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("op-history", "xmlNodePtr", "const char *", "const char *", "int", "uint32_t") static int op_history_text(pcmk__output_t *out, va_list args) { xmlNodePtr xml_op = va_arg(args, xmlNodePtr); const char *task = va_arg(args, const char *); const char *interval_ms_s = va_arg(args, const char *); int rc = va_arg(args, int); uint32_t show_opts = va_arg(args, uint32_t); char *buf = op_history_string(xml_op, task, interval_ms_s, rc, pcmk_is_set(show_opts, pcmk_show_timing)); out->list_item(out, NULL, "%s", buf); free(buf); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("op-history", "xmlNodePtr", "const char *", "const char *", "int", "uint32_t") static int op_history_xml(pcmk__output_t *out, va_list args) { xmlNodePtr xml_op = va_arg(args, xmlNodePtr); const char *task = va_arg(args, const char *); const char *interval_ms_s = va_arg(args, const char *); int rc = va_arg(args, int); uint32_t show_opts = va_arg(args, uint32_t); const char *call_id = crm_element_value(xml_op, PCMK__XA_CALL_ID); char *rc_s = pcmk__itoa(rc); xmlNodePtr node = pcmk__output_create_xml_node(out, "operation_history", PCMK_XA_CALL, call_id, "task", task, "rc", rc_s, "rc_text", services_ocf_exitcode_str(rc), NULL); free(rc_s); if (interval_ms_s && !pcmk__str_eq(interval_ms_s, "0", pcmk__str_casei)) { char *s = crm_strdup_printf("%sms", interval_ms_s); crm_xml_add(node, PCMK_XA_INTERVAL, s); free(s); } if (pcmk_is_set(show_opts, pcmk_show_timing)) { const char *value = NULL; time_t epoch = 0; if ((crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &epoch) == pcmk_ok) && (epoch > 0)) { char *s = pcmk__epoch2str(&epoch, 0); crm_xml_add(node, PCMK_XA_LAST_RC_CHANGE, s); free(s); } value = crm_element_value(xml_op, PCMK_XA_EXEC_TIME); if (value) { char *s = crm_strdup_printf("%sms", value); crm_xml_add(node, PCMK_XA_EXEC_TIME, s); free(s); } value = crm_element_value(xml_op, PCMK_XA_QUEUE_TIME); if (value) { char *s = crm_strdup_printf("%sms", value); crm_xml_add(node, PCMK_XA_QUEUE_TIME, s); free(s); } } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("promotion-score", "pcmk_resource_t *", "pcmk_node_t *", "const char *") static int promotion_score(pcmk__output_t *out, va_list args) { pcmk_resource_t *child_rsc = va_arg(args, pcmk_resource_t *); pcmk_node_t *chosen = va_arg(args, pcmk_node_t *); const char *score = va_arg(args, const char *); out->list_item(out, NULL, "%s promotion score on %s: %s", child_rsc->id, chosen? chosen->details->uname : "none", score); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("promotion-score", "pcmk_resource_t *", "pcmk_node_t *", "const char *") static int promotion_score_xml(pcmk__output_t *out, va_list args) { pcmk_resource_t *child_rsc = va_arg(args, pcmk_resource_t *); pcmk_node_t *chosen = va_arg(args, pcmk_node_t *); const char *score = va_arg(args, const char *); xmlNodePtr node = pcmk__output_create_xml_node(out, "promotion_score", PCMK_XA_ID, child_rsc->id, PCMK_XA_SCORE, score, NULL); if (chosen) { crm_xml_add(node, PCMK_XA_NODE, chosen->details->uname); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("resource-config", "const pcmk_resource_t *", "bool") static int resource_config(pcmk__output_t *out, va_list args) { const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *); bool raw = va_arg(args, int); char *rsc_xml = formatted_xml_buf(rsc, raw); out->output_xml(out, "xml", rsc_xml); free(rsc_xml); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("resource-config", "const pcmk_resource_t *", "bool") static int resource_config_text(pcmk__output_t *out, va_list args) { const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *); bool raw = va_arg(args, int); char *rsc_xml = formatted_xml_buf(rsc, raw); pcmk__formatted_printf(out, "Resource XML:\n"); out->output_xml(out, "xml", rsc_xml); free(rsc_xml); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("resource-history", "pcmk_resource_t *", "const char *", "bool", "int", "time_t", "bool") static int resource_history_text(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); const char *rsc_id = va_arg(args, const char *); bool all = va_arg(args, int); int failcount = va_arg(args, int); time_t last_failure = va_arg(args, time_t); bool as_header = va_arg(args, int); char *buf = resource_history_string(rsc, rsc_id, all, failcount, last_failure); if (as_header) { out->begin_list(out, NULL, NULL, "%s", buf); } else { out->list_item(out, NULL, "%s", buf); } free(buf); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("resource-history", "pcmk_resource_t *", "const char *", "bool", "int", "time_t", "bool") static int resource_history_xml(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); const char *rsc_id = va_arg(args, const char *); bool all = va_arg(args, int); int failcount = va_arg(args, int); time_t last_failure = va_arg(args, time_t); bool as_header = va_arg(args, int); xmlNodePtr node = pcmk__output_xml_create_parent(out, "resource_history", PCMK_XA_ID, rsc_id, NULL); if (rsc == NULL) { pcmk__xe_set_bool_attr(node, "orphan", true); } else if (all || failcount || last_failure > 0) { char *migration_s = pcmk__itoa(rsc->migration_threshold); pcmk__xe_set_props(node, "orphan", PCMK_VALUE_FALSE, PCMK_META_MIGRATION_THRESHOLD, migration_s, NULL); free(migration_s); if (failcount > 0) { char *s = pcmk__itoa(failcount); crm_xml_add(node, PCMK_XA_FAIL_COUNT, s); free(s); } if (last_failure > 0) { char *s = pcmk__epoch2str(&last_failure, 0); crm_xml_add(node, PCMK_XA_LAST_FAILURE, s); free(s); } } if (!as_header) { pcmk__output_xml_pop_parent(out); } return pcmk_rc_ok; } static void print_resource_header(pcmk__output_t *out, uint32_t show_opts) { if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) { /* Active resources have already been printed by node */ out->begin_list(out, NULL, NULL, "Inactive Resources"); } else if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { out->begin_list(out, NULL, NULL, "Full List of Resources"); } else { out->begin_list(out, NULL, NULL, "Active Resources"); } } PCMK__OUTPUT_ARGS("resource-list", "pcmk_scheduler_t *", "uint32_t", "bool", "GList *", "GList *", "bool") static int resource_list(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); uint32_t show_opts = va_arg(args, uint32_t); bool print_summary = va_arg(args, int); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); bool print_spacer = va_arg(args, int); GList *rsc_iter; int rc = pcmk_rc_no_output; bool printed_header = false; /* If we already showed active resources by node, and * we're not showing inactive resources, we have nothing to do */ if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node) && !pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { return rc; } /* If we haven't already printed resources grouped by node, * and brief output was requested, print resource summary */ if (pcmk_is_set(show_opts, pcmk_show_brief) && !pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) { GList *rscs = pe__filter_rsc_list(scheduler->resources, only_rsc); PCMK__OUTPUT_SPACER_IF(out, print_spacer); print_resource_header(out, show_opts); printed_header = true; rc = pe__rscs_brief_output(out, rscs, show_opts); g_list_free(rscs); } /* For each resource, display it if appropriate */ for (rsc_iter = scheduler->resources; rsc_iter != NULL; rsc_iter = rsc_iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) rsc_iter->data; int x; /* Complex resources may have some sub-resources active and some inactive */ gboolean is_active = rsc->fns->active(rsc, TRUE); gboolean partially_active = rsc->fns->active(rsc, FALSE); /* Skip inactive orphans (deleted but still in CIB) */ if (pcmk_is_set(rsc->flags, pcmk_rsc_removed) && !is_active) { continue; /* Skip active resources if we already displayed them by node */ } else if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) { if (is_active) { continue; } /* Skip primitives already counted in a brief summary */ } else if (pcmk_is_set(show_opts, pcmk_show_brief) && (rsc->variant == pcmk_rsc_variant_primitive)) { continue; /* Skip resources that aren't at least partially active, * unless we're displaying inactive resources */ } else if (!partially_active && !pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { continue; } else if (partially_active && !pe__rsc_running_on_any(rsc, only_node)) { continue; } if (!printed_header) { PCMK__OUTPUT_SPACER_IF(out, print_spacer); print_resource_header(out, show_opts); printed_header = true; } /* Print this resource */ x = out->message(out, crm_map_element_name(rsc->xml), show_opts, rsc, only_node, only_rsc); if (x == pcmk_rc_ok) { rc = pcmk_rc_ok; } } if (print_summary && rc != pcmk_rc_ok) { if (!printed_header) { PCMK__OUTPUT_SPACER_IF(out, print_spacer); print_resource_header(out, show_opts); printed_header = true; } if (pcmk_is_set(show_opts, pcmk_show_rscs_by_node)) { out->list_item(out, NULL, "No inactive resources"); } else if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { out->list_item(out, NULL, "No resources"); } else { out->list_item(out, NULL, "No active resources"); } } if (printed_header) { out->end_list(out); } return rc; } PCMK__OUTPUT_ARGS("resource-operation-list", "pcmk_scheduler_t *", "pcmk_resource_t *", "pcmk_node_t *", "GList *", "uint32_t") static int resource_operation_list(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler G_GNUC_UNUSED = va_arg(args, pcmk_scheduler_t *); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); pcmk_node_t *node = va_arg(args, pcmk_node_t *); GList *op_list = va_arg(args, GList *); uint32_t show_opts = va_arg(args, uint32_t); GList *gIter = NULL; int rc = pcmk_rc_no_output; /* Print each operation */ for (gIter = op_list; gIter != NULL; gIter = gIter->next) { xmlNode *xml_op = (xmlNode *) gIter->data; const char *task = crm_element_value(xml_op, PCMK_XA_OPERATION); const char *interval_ms_s = crm_element_value(xml_op, PCMK_META_INTERVAL); const char *op_rc = crm_element_value(xml_op, PCMK__XA_RC_CODE); int op_rc_i; pcmk__scan_min_int(op_rc, &op_rc_i, 0); /* Display 0-interval monitors as "probe" */ if (pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei) && pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches | pcmk__str_casei)) { task = "probe"; } /* If this is the first printed operation, print heading for resource */ if (rc == pcmk_rc_no_output) { time_t last_failure = 0; int failcount = pe_get_failcount(node, rsc, &last_failure, pcmk__fc_default, NULL); out->message(out, "resource-history", rsc, rsc_printable_id(rsc), true, failcount, last_failure, true); rc = pcmk_rc_ok; } /* Print the operation */ out->message(out, "op-history", xml_op, task, interval_ms_s, op_rc_i, show_opts); } /* Free the list we created (no need to free the individual items) */ g_list_free(op_list); PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("resource-util", "pcmk_resource_t *", "pcmk_node_t *", "const char *") static int resource_util(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); pcmk_node_t *node = va_arg(args, pcmk_node_t *); const char *fn = va_arg(args, const char *); char *dump_text = crm_strdup_printf("%s: %s utilization on %s:", fn, rsc->id, pcmk__node_name(node)); g_hash_table_foreach(rsc->utilization, append_dump_text, &dump_text); out->list_item(out, NULL, "%s", dump_text); free(dump_text); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("resource-util", "pcmk_resource_t *", "pcmk_node_t *", "const char *") static int resource_util_xml(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); pcmk_node_t *node = va_arg(args, pcmk_node_t *); const char *uname = node->details->uname; const char *fn = va_arg(args, const char *); xmlNodePtr xml_node = NULL; xml_node = pcmk__output_create_xml_node(out, PCMK_XE_UTILIZATION, PCMK_XA_RESOURCE, rsc->id, PCMK_XA_NODE, uname, PCMK_XA_FUNCTION, fn, NULL); g_hash_table_foreach(rsc->utilization, add_dump_node, xml_node); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ticket", "pcmk_ticket_t *") static int ticket_html(pcmk__output_t *out, va_list args) { pcmk_ticket_t *ticket = va_arg(args, pcmk_ticket_t *); if (ticket->last_granted > -1) { char *epoch_str = pcmk__epoch2str(&(ticket->last_granted), 0); out->list_item(out, NULL, "%s:\t%s%s last-granted=\"%s\"", ticket->id, (ticket->granted? "granted" : "revoked"), (ticket->standby? " [standby]" : ""), pcmk__s(epoch_str, "")); free(epoch_str); } else { out->list_item(out, NULL, "%s:\t%s%s", ticket->id, ticket->granted ? "granted" : "revoked", ticket->standby ? " [standby]" : ""); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ticket", "pcmk_ticket_t *") static int ticket_text(pcmk__output_t *out, va_list args) { pcmk_ticket_t *ticket = va_arg(args, pcmk_ticket_t *); if (ticket->last_granted > -1) { char *epoch_str = pcmk__epoch2str(&(ticket->last_granted), 0); out->list_item(out, ticket->id, "%s%s last-granted=\"%s\"", (ticket->granted? "granted" : "revoked"), (ticket->standby? " [standby]" : ""), pcmk__s(epoch_str, "")); free(epoch_str); } else { out->list_item(out, ticket->id, "%s%s", ticket->granted ? "granted" : "revoked", ticket->standby ? " [standby]" : ""); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ticket", "pcmk_ticket_t *") static int ticket_xml(pcmk__output_t *out, va_list args) { pcmk_ticket_t *ticket = va_arg(args, pcmk_ticket_t *); const char *status = NULL; const char *standby = pcmk__btoa(ticket->standby); xmlNodePtr node = NULL; status = ticket->granted? PCMK_VALUE_GRANTED : PCMK_VALUE_REVOKED; node = pcmk__output_create_xml_node(out, PCMK_XE_TICKET, PCMK_XA_ID, ticket->id, PCMK_XA_STATUS, status, PCMK_XA_STANDBY, standby, NULL); if (ticket->last_granted > -1) { char *buf = pcmk__epoch2str(&ticket->last_granted, 0); crm_xml_add(node, PCMK_XA_LAST_GRANTED, buf); free(buf); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ticket-list", "pcmk_scheduler_t *", "bool") static int ticket_list(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); bool print_spacer = va_arg(args, int); GHashTableIter iter; gpointer key, value; if (g_hash_table_size(scheduler->tickets) == 0) { return pcmk_rc_no_output; } PCMK__OUTPUT_SPACER_IF(out, print_spacer); /* Print section heading */ out->begin_list(out, NULL, NULL, "Tickets"); /* Print each ticket */ g_hash_table_iter_init(&iter, scheduler->tickets); while (g_hash_table_iter_next(&iter, &key, &value)) { pcmk_ticket_t *ticket = (pcmk_ticket_t *) value; out->message(out, "ticket", ticket); } /* Close section */ out->end_list(out); return pcmk_rc_ok; } static pcmk__message_entry_t fmt_functions[] = { { "ban", "default", ban_text }, { "ban", "html", ban_html }, { "ban", "xml", ban_xml }, { "ban-list", "default", ban_list }, { "bundle", "default", pe__bundle_text }, { "bundle", "xml", pe__bundle_xml }, { "bundle", "html", pe__bundle_html }, { "clone", "default", pe__clone_default }, { "clone", "xml", pe__clone_xml }, { "cluster-counts", "default", cluster_counts_text }, { "cluster-counts", "html", cluster_counts_html }, { "cluster-counts", "xml", cluster_counts_xml }, { "cluster-dc", "default", cluster_dc_text }, { "cluster-dc", "html", cluster_dc_html }, { "cluster-dc", "xml", cluster_dc_xml }, { "cluster-options", "default", cluster_options_text }, { "cluster-options", "html", cluster_options_html }, { "cluster-options", "log", cluster_options_log }, { "cluster-options", "xml", cluster_options_xml }, { "cluster-summary", "default", cluster_summary }, { "cluster-summary", "html", cluster_summary_html }, { "cluster-stack", "default", cluster_stack_text }, { "cluster-stack", "html", cluster_stack_html }, { "cluster-stack", "xml", cluster_stack_xml }, { "cluster-times", "default", cluster_times_text }, { "cluster-times", "html", cluster_times_html }, { "cluster-times", "xml", cluster_times_xml }, { "failed-action", "default", failed_action_default }, { "failed-action", "xml", failed_action_xml }, { "failed-action-list", "default", failed_action_list }, { "group", "default", pe__group_default}, { "group", "xml", pe__group_xml }, { "maint-mode", "text", cluster_maint_mode_text }, { "node", "default", node_text }, { "node", "html", node_html }, { "node", "xml", node_xml }, { "node-and-op", "default", node_and_op }, { "node-and-op", "xml", node_and_op_xml }, { "node-capacity", "default", node_capacity }, { "node-capacity", "xml", node_capacity_xml }, { "node-history-list", "default", node_history_list }, { "node-list", "default", node_list_text }, { "node-list", "html", node_list_html }, { "node-list", "xml", node_list_xml }, { "node-weight", "default", node_weight }, { "node-weight", "xml", node_weight_xml }, { "node-attribute", "default", node_attribute_text }, { "node-attribute", "html", node_attribute_html }, { "node-attribute", "xml", node_attribute_xml }, { "node-attribute-list", "default", node_attribute_list }, { "node-summary", "default", node_summary }, { "op-history", "default", op_history_text }, { "op-history", "xml", op_history_xml }, { "primitive", "default", pe__resource_text }, { "primitive", "xml", pe__resource_xml }, { "primitive", "html", pe__resource_html }, { "promotion-score", "default", promotion_score }, { "promotion-score", "xml", promotion_score_xml }, { "resource-config", "default", resource_config }, { "resource-config", "text", resource_config_text }, { "resource-history", "default", resource_history_text }, { "resource-history", "xml", resource_history_xml }, { "resource-list", "default", resource_list }, { "resource-operation-list", "default", resource_operation_list }, { "resource-util", "default", resource_util }, { "resource-util", "xml", resource_util_xml }, { "ticket", "default", ticket_text }, { "ticket", "html", ticket_html }, { "ticket", "xml", ticket_xml }, { "ticket-list", "default", ticket_list }, { NULL, NULL, NULL } }; void pe__register_messages(pcmk__output_t *out) { pcmk__register_messages(out, fmt_functions); } diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c index 0cd760eedb..5b71f506bc 100644 --- a/lib/pengine/unpack.c +++ b/lib/pengine/unpack.c @@ -1,5173 +1,5173 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include CRM_TRACE_INIT_DATA(pe_status); // A (parsed) resource action history entry struct action_history { pcmk_resource_t *rsc; // Resource that history is for pcmk_node_t *node; // Node that history is for xmlNode *xml; // History entry XML // Parsed from entry XML const char *id; // XML ID of history entry const char *key; // Operation key of action const char *task; // Action name const char *exit_reason; // Exit reason given for result guint interval_ms; // Action interval int call_id; // Call ID of action int expected_exit_status; // Expected exit status of action int exit_status; // Actual exit status of action int execution_status; // Execution status of action }; /* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than * use pcmk__set_scheduler_flags()/pcmk__clear_scheduler_flags() so that the * flag is stringified more readably in log messages. */ #define set_config_flag(scheduler, option, flag) do { \ GHashTable *config_hash = (scheduler)->config_hash; \ const char *scf_value = pcmk__cluster_option(config_hash, (option)); \ \ if (scf_value != NULL) { \ if (crm_is_true(scf_value)) { \ (scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", \ crm_system_name, (scheduler)->flags, \ (flag), #flag); \ } else { \ (scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", \ crm_system_name, (scheduler)->flags, \ (flag), #flag); \ } \ } \ } while(0) static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *failed); static void determine_remote_online_status(pcmk_scheduler_t *scheduler, pcmk_node_t *this_node); static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite, pcmk_scheduler_t *scheduler); static void determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node, pcmk_scheduler_t *scheduler); static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml, pcmk_scheduler_t *scheduler); static gboolean is_dangling_guest_node(pcmk_node_t *node) { /* we are looking for a remote-node that was supposed to be mapped to a * container resource, but all traces of that container have disappeared * from both the config and the status section. */ if (pe__is_guest_or_remote_node(node) && node->details->remote_rsc && node->details->remote_rsc->container == NULL && pcmk_is_set(node->details->remote_rsc->flags, pcmk_rsc_removed_filler)) { return TRUE; } return FALSE; } /*! * \brief Schedule a fence action for a node * * \param[in,out] scheduler Scheduler data * \param[in,out] node Node to fence * \param[in] reason Text description of why fencing is needed * \param[in] priority_delay Whether to consider * \c PCMK_OPT_PRIORITY_FENCING_DELAY */ void pe_fence_node(pcmk_scheduler_t *scheduler, pcmk_node_t *node, const char *reason, bool priority_delay) { CRM_CHECK(node, return); /* A guest node is fenced by marking its container as failed */ if (pe__is_guest_node(node)) { pcmk_resource_t *rsc = node->details->remote_rsc->container; if (!pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { crm_notice("Not fencing guest node %s " "(otherwise would because %s): " "its guest resource %s is unmanaged", pcmk__node_name(node), reason, rsc->id); } else { pcmk__sched_warn("Guest node %s will be fenced " "(by recovering its guest resource %s): %s", pcmk__node_name(node), rsc->id, reason); /* We don't mark the node as unclean because that would prevent the * node from running resources. We want to allow it to run resources * in this transition if the recovery succeeds. */ node->details->remote_requires_reset = TRUE; pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); } } } else if (is_dangling_guest_node(node)) { crm_info("Cleaning up dangling connection for guest node %s: " "fencing was already done because %s, " "and guest resource no longer exists", pcmk__node_name(node), reason); pcmk__set_rsc_flags(node->details->remote_rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); } else if (pe__is_remote_node(node)) { pcmk_resource_t *rsc = node->details->remote_rsc; if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { crm_notice("Not fencing remote node %s " "(otherwise would because %s): connection is unmanaged", pcmk__node_name(node), reason); } else if(node->details->remote_requires_reset == FALSE) { node->details->remote_requires_reset = TRUE; pcmk__sched_warn("Remote node %s %s: %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "will be fenced" : "is unclean", reason); } node->details->unclean = TRUE; // No need to apply PCMK_OPT_PRIORITY_FENCING_DELAY for remote nodes pe_fence_op(node, NULL, TRUE, reason, FALSE, scheduler); } else if (node->details->unclean) { crm_trace("Cluster node %s %s because %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean", reason); } else { pcmk__sched_warn("Cluster node %s %s: %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "will be fenced" : "is unclean", reason); node->details->unclean = TRUE; pe_fence_op(node, NULL, TRUE, reason, priority_delay, scheduler); } } // @TODO xpaths can't handle templates, rules, or id-refs // nvpair with provides or requires set to unfencing #define XPATH_UNFENCING_NVPAIR PCMK_XE_NVPAIR \ "[(@" PCMK_XA_NAME "='" PCMK_STONITH_PROVIDES "'" \ "or @" PCMK_XA_NAME "='" PCMK_META_REQUIRES "') " \ "and @" PCMK_XA_VALUE "='" PCMK__VALUE_UNFENCING "']" // unfencing in rsc_defaults or any resource #define XPATH_ENABLE_UNFENCING \ "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \ "//" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR \ "|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RSC_DEFAULTS \ "/" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR static void set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler) { xmlXPathObjectPtr result = NULL; if (!pcmk_is_set(scheduler->flags, flag)) { result = xpath_search(scheduler->input, xpath); if (result && (numXpathResults(result) > 0)) { pcmk__set_scheduler_flags(scheduler, flag); } freeXpathObject(result); } } gboolean unpack_config(xmlNode *config, pcmk_scheduler_t *scheduler) { const char *value = NULL; guint interval_ms = 0U; GHashTable *config_hash = pcmk__strkey_table(free, free); pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = pcmk_role_unknown, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; scheduler->config_hash = config_hash; pe__unpack_dataset_nvpairs(config, PCMK_XE_CLUSTER_PROPERTY_SET, &rule_data, config_hash, PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, FALSE, scheduler); pcmk__validate_cluster_options(config_hash); set_config_flag(scheduler, PCMK_OPT_ENABLE_STARTUP_PROBES, pcmk_sched_probe_resources); if (!pcmk_is_set(scheduler->flags, pcmk_sched_probe_resources)) { crm_info("Startup probes: disabled (dangerous)"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_HAVE_WATCHDOG); if (value && crm_is_true(value)) { crm_info("Watchdog-based self-fencing will be performed via SBD if " "fencing is required and " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " is nonzero"); pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_fencing); } /* Set certain flags via xpath here, so they can be used before the relevant * configuration sections are unpacked. */ set_if_xpath(pcmk_sched_enable_unfencing, XPATH_ENABLE_UNFENCING, scheduler); value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT); pcmk_parse_interval_spec(value, &interval_ms); if (interval_ms >= INT_MAX) { scheduler->stonith_timeout = INT_MAX; } else { scheduler->stonith_timeout = (int) interval_ms; } crm_debug("STONITH timeout: %d", scheduler->stonith_timeout); set_config_flag(scheduler, PCMK_OPT_STONITH_ENABLED, pcmk_sched_fencing_enabled); if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { crm_debug("STONITH of failed nodes is enabled"); } else { crm_debug("STONITH of failed nodes is disabled"); } scheduler->stonith_action = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_ACTION); if (!strcmp(scheduler->stonith_action, "poweroff")) { pcmk__warn_once(pcmk__wo_poweroff, "Support for " PCMK_OPT_STONITH_ACTION " of " "'poweroff' is deprecated and will be removed in a " "future release (use 'off' instead)"); scheduler->stonith_action = PCMK_ACTION_OFF; } crm_trace("STONITH will %s nodes", scheduler->stonith_action); set_config_flag(scheduler, PCMK_OPT_CONCURRENT_FENCING, pcmk_sched_concurrent_fencing); if (pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing)) { crm_debug("Concurrent fencing is enabled"); } else { crm_debug("Concurrent fencing is disabled"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_PRIORITY_FENCING_DELAY); if (value) { pcmk_parse_interval_spec(value, &interval_ms); scheduler->priority_fencing_delay = (int) (interval_ms / 1000); crm_trace("Priority fencing delay is %ds", scheduler->priority_fencing_delay); } set_config_flag(scheduler, PCMK_OPT_STOP_ALL_RESOURCES, pcmk_sched_stop_all); crm_debug("Stop all active resources: %s", pcmk__flag_text(scheduler->flags, pcmk_sched_stop_all)); set_config_flag(scheduler, PCMK_OPT_SYMMETRIC_CLUSTER, pcmk_sched_symmetric_cluster); if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) { crm_debug("Cluster is symmetric" " - resources can run anywhere by default"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_NO_QUORUM_POLICY); if (pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_ignore; } else if (pcmk__str_eq(value, PCMK_VALUE_FREEZE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_freeze; } else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_demote; } else if (pcmk__str_eq(value, PCMK_VALUE_FENCE_LEGACY, pcmk__str_casei)) { if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { int do_panic = 0; crm_element_value_int(scheduler->input, PCMK_XA_NO_QUORUM_PANIC, &do_panic); if (do_panic || pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) { scheduler->no_quorum_policy = pcmk_no_quorum_fence; } else { crm_notice("Resetting " PCMK_OPT_NO_QUORUM_POLICY " to 'stop': cluster has never had quorum"); scheduler->no_quorum_policy = pcmk_no_quorum_stop; } } else { pcmk__config_err("Resetting " PCMK_OPT_NO_QUORUM_POLICY " to 'stop' because fencing is disabled"); scheduler->no_quorum_policy = pcmk_no_quorum_stop; } } else { scheduler->no_quorum_policy = pcmk_no_quorum_stop; } switch (scheduler->no_quorum_policy) { case pcmk_no_quorum_freeze: crm_debug("On loss of quorum: Freeze resources"); break; case pcmk_no_quorum_stop: crm_debug("On loss of quorum: Stop ALL resources"); break; case pcmk_no_quorum_demote: crm_debug("On loss of quorum: " "Demote promotable resources and stop other resources"); break; case pcmk_no_quorum_fence: crm_notice("On loss of quorum: Fence all remaining nodes"); break; case pcmk_no_quorum_ignore: crm_notice("On loss of quorum: Ignore"); break; } set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_RESOURCES, pcmk_sched_stop_removed_resources); if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { crm_trace("Orphan resources are stopped"); } else { crm_trace("Orphan resources are ignored"); } set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_ACTIONS, pcmk_sched_cancel_removed_actions); if (pcmk_is_set(scheduler->flags, pcmk_sched_cancel_removed_actions)) { crm_trace("Orphan resource actions are stopped"); } else { crm_trace("Orphan resource actions are ignored"); } value = pcmk__cluster_option(config_hash, PCMK__OPT_REMOVE_AFTER_STOP); if (value != NULL) { if (crm_is_true(value)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_remove_after_stop); #ifndef PCMK__COMPAT_2_0 pcmk__warn_once(pcmk__wo_remove_after, "Support for the " PCMK__OPT_REMOVE_AFTER_STOP " cluster property is deprecated and will be " "removed in a future release"); #endif } else { pcmk__clear_scheduler_flags(scheduler, pcmk_sched_remove_after_stop); } } set_config_flag(scheduler, PCMK_OPT_MAINTENANCE_MODE, pcmk_sched_in_maintenance); crm_trace("Maintenance mode: %s", pcmk__flag_text(scheduler->flags, pcmk_sched_in_maintenance)); set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL, pcmk_sched_start_failure_fatal); if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) { crm_trace("Start failures are always fatal"); } else { crm_trace("Start failures are handled by failcount"); } if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING, pcmk_sched_startup_fencing); } if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) { crm_trace("Unseen nodes will be fenced"); } else { pcmk__warn_once(pcmk__wo_blind, "Blind faith: not fencing unseen nodes"); } pe__unpack_node_health_scores(scheduler); scheduler->placement_strategy = pcmk__cluster_option(config_hash, PCMK_OPT_PLACEMENT_STRATEGY); crm_trace("Placement strategy: %s", scheduler->placement_strategy); set_config_flag(scheduler, PCMK_OPT_SHUTDOWN_LOCK, pcmk_sched_shutdown_lock); if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { value = pcmk__cluster_option(config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT); pcmk_parse_interval_spec(value, &(scheduler->shutdown_lock)); scheduler->shutdown_lock /= 1000; crm_trace("Resources will be locked to nodes that were cleanly " "shut down (locks expire after %s)", pcmk__readable_interval(scheduler->shutdown_lock)); } else { crm_trace("Resources will not be locked to nodes that were cleanly " "shut down"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT); pcmk_parse_interval_spec(value, &(scheduler->node_pending_timeout)); scheduler->node_pending_timeout /= 1000; if (scheduler->node_pending_timeout == 0) { crm_trace("Do not fence pending nodes"); } else { crm_trace("Fence pending nodes after %s", pcmk__readable_interval(scheduler->node_pending_timeout * 1000)); } return TRUE; } pcmk_node_t * pe_create_node(const char *id, const char *uname, const char *type, const char *score, pcmk_scheduler_t *scheduler) { pcmk_node_t *new_node = NULL; if (pe_find_node(scheduler->nodes, uname) != NULL) { pcmk__config_warn("More than one node entry has name '%s'", uname); } new_node = calloc(1, sizeof(pcmk_node_t)); if (new_node == NULL) { pcmk__sched_err("Could not allocate memory for node %s", uname); return NULL; } new_node->weight = char2score(score); new_node->details = calloc(1, sizeof(struct pe_node_shared_s)); if (new_node->details == NULL) { free(new_node); pcmk__sched_err("Could not allocate memory for node %s", uname); return NULL; } crm_trace("Creating node for entry %s/%s", uname, id); new_node->details->id = id; new_node->details->uname = uname; new_node->details->online = FALSE; new_node->details->shutdown = FALSE; new_node->details->rsc_discovery_enabled = TRUE; new_node->details->running_rsc = NULL; new_node->details->data_set = scheduler; if (pcmk__str_eq(type, PCMK_VALUE_MEMBER, pcmk__str_null_matches|pcmk__str_casei)) { new_node->details->type = pcmk_node_variant_cluster; } else if (pcmk__str_eq(type, PCMK_VALUE_REMOTE, pcmk__str_casei)) { new_node->details->type = pcmk_node_variant_remote; pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_remote_nodes); } else { /* @COMPAT 'ping' is the default for backward compatibility, but it * should be changed to 'member' at a compatibility break */ if (!pcmk__str_eq(type, PCMK__VALUE_PING, pcmk__str_casei)) { pcmk__config_warn("Node %s has unrecognized type '%s', " "assuming '" PCMK__VALUE_PING "'", pcmk__s(uname, "without name"), type); } pcmk__warn_once(pcmk__wo_ping_node, "Support for nodes of type '" PCMK__VALUE_PING "' " "(such as %s) is deprecated and will be removed in a " "future release", pcmk__s(uname, "unnamed node")); new_node->details->type = node_ping; } new_node->details->attrs = pcmk__strkey_table(free, free); if (pe__is_guest_or_remote_node(new_node)) { g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("remote")); } else { g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("cluster")); } new_node->details->utilization = pcmk__strkey_table(free, free); new_node->details->digest_cache = pcmk__strkey_table(free, pe__free_digests); scheduler->nodes = g_list_insert_sorted(scheduler->nodes, new_node, pe__cmp_node_name); return new_node; } static const char * expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pcmk_scheduler_t *data) { xmlNode *attr_set = NULL; xmlNode *attr = NULL; const char *container_id = ID(xml_obj); const char *remote_name = NULL; const char *remote_server = NULL; const char *remote_port = NULL; const char *connect_timeout = "60s"; const char *remote_allow_migrate=NULL; const char *is_managed = NULL; for (attr_set = pcmk__xe_first_child(xml_obj); attr_set != NULL; attr_set = pcmk__xe_next(attr_set)) { if (!pcmk__str_eq((const char *) attr_set->name, PCMK_XE_META_ATTRIBUTES, pcmk__str_casei)) { continue; } for (attr = pcmk__xe_first_child(attr_set); attr != NULL; attr = pcmk__xe_next(attr)) { const char *value = crm_element_value(attr, PCMK_XA_VALUE); const char *name = crm_element_value(attr, PCMK_XA_NAME); if (name == NULL) { // Sanity continue; } if (strcmp(name, PCMK_META_REMOTE_NODE) == 0) { remote_name = value; } else if (strcmp(name, PCMK_META_REMOTE_ADDR) == 0) { remote_server = value; } else if (strcmp(name, PCMK_META_REMOTE_PORT) == 0) { remote_port = value; } else if (strcmp(name, PCMK_META_REMOTE_CONNECT_TIMEOUT) == 0) { connect_timeout = value; } else if (strcmp(name, PCMK_META_REMOTE_ALLOW_MIGRATE) == 0) { remote_allow_migrate = value; } else if (strcmp(name, PCMK_META_IS_MANAGED) == 0) { is_managed = value; } } } if (remote_name == NULL) { return NULL; } if (pe_find_resource(data->resources, remote_name) != NULL) { return NULL; } pe_create_remote_xml(parent, remote_name, container_id, remote_allow_migrate, is_managed, connect_timeout, remote_server, remote_port); return remote_name; } static void handle_startup_fencing(pcmk_scheduler_t *scheduler, pcmk_node_t *new_node) { if ((new_node->details->type == pcmk_node_variant_remote) && (new_node->details->remote_rsc == NULL)) { /* Ignore fencing for remote nodes that don't have a connection resource * associated with them. This happens when remote node entries get left * in the nodes section after the connection resource is removed. */ return; } if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) { // All nodes are unclean until we've seen their status entry new_node->details->unclean = TRUE; } else { // Blind faith ... new_node->details->unclean = FALSE; } /* We need to be able to determine if a node's status section * exists or not separate from whether the node is unclean. */ new_node->details->unseen = TRUE; } gboolean unpack_nodes(xmlNode *xml_nodes, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; pcmk_node_t *new_node = NULL; const char *id = NULL; const char *uname = NULL; const char *type = NULL; const char *score = NULL; for (xml_obj = pcmk__xe_first_child(xml_nodes); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (pcmk__str_eq((const char *) xml_obj->name, PCMK_XE_NODE, pcmk__str_none)) { new_node = NULL; id = crm_element_value(xml_obj, PCMK_XA_ID); uname = crm_element_value(xml_obj, PCMK_XA_UNAME); type = crm_element_value(xml_obj, PCMK_XA_TYPE); score = crm_element_value(xml_obj, PCMK_XA_SCORE); crm_trace("Processing node %s/%s", uname, id); if (id == NULL) { pcmk__config_err("Ignoring <" PCMK_XE_NODE "> entry in configuration without id"); continue; } new_node = pe_create_node(id, uname, type, score, scheduler); if (new_node == NULL) { return FALSE; } handle_startup_fencing(scheduler, new_node); add_node_attrs(xml_obj, new_node, FALSE, scheduler); crm_trace("Done with node %s", crm_element_value(xml_obj, PCMK_XA_UNAME)); } } if (scheduler->localhost && (pe_find_node(scheduler->nodes, scheduler->localhost) == NULL)) { crm_info("Creating a fake local node"); pe_create_node(scheduler->localhost, scheduler->localhost, NULL, 0, scheduler); } return TRUE; } static void setup_container(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { const char *container_id = NULL; if (rsc->children) { g_list_foreach(rsc->children, (GFunc) setup_container, scheduler); return; } container_id = g_hash_table_lookup(rsc->meta, PCMK__META_CONTAINER); if (container_id && !pcmk__str_eq(container_id, rsc->id, pcmk__str_casei)) { pcmk_resource_t *container = pe_find_resource(scheduler->resources, container_id); if (container) { rsc->container = container; pcmk__set_rsc_flags(container, pcmk_rsc_has_filler); container->fillers = g_list_append(container->fillers, rsc); pcmk__rsc_trace(rsc, "Resource %s's container is %s", rsc->id, container_id); } else { pcmk__config_err("Resource %s: Unknown resource container (%s)", rsc->id, container_id); } } } gboolean unpack_remote_nodes(xmlNode *xml_resources, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; /* Create remote nodes and guest nodes from the resource configuration * before unpacking resources. */ for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { const char *new_node_id = NULL; /* Check for remote nodes, which are defined by ocf:pacemaker:remote * primitives. */ if (xml_contains_remote_node(xml_obj)) { new_node_id = ID(xml_obj); /* The "pe_find_node" check is here to make sure we don't iterate over * an expanded node that has already been added to the node list. */ if (new_node_id && (pe_find_node(scheduler->nodes, new_node_id) == NULL)) { crm_trace("Found remote node %s defined by resource %s", new_node_id, ID(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } continue; } /* Check for guest nodes, which are defined by special meta-attributes * of a primitive of any type (for example, VirtualDomain or Xen). */ if (pcmk__str_eq((const char *) xml_obj->name, PCMK_XE_PRIMITIVE, pcmk__str_none)) { /* This will add an ocf:pacemaker:remote primitive to the * configuration for the guest node's connection, to be unpacked * later. */ new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources, scheduler); if (new_node_id && (pe_find_node(scheduler->nodes, new_node_id) == NULL)) { crm_trace("Found guest node %s in resource %s", new_node_id, ID(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } continue; } /* Check for guest nodes inside a group. Clones are currently not * supported as guest nodes. */ if (pcmk__str_eq((const char *) xml_obj->name, PCMK_XE_GROUP, pcmk__str_none)) { xmlNode *xml_obj2 = NULL; for (xml_obj2 = pcmk__xe_first_child(xml_obj); xml_obj2 != NULL; xml_obj2 = pcmk__xe_next(xml_obj2)) { new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources, scheduler); if (new_node_id && (pe_find_node(scheduler->nodes, new_node_id) == NULL)) { crm_trace("Found guest node %s in resource %s inside group %s", new_node_id, ID(xml_obj2), ID(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } } } } return TRUE; } /* Call this after all the nodes and resources have been * unpacked, but before the status section is read. * * A remote node's online status is reflected by the state * of the remote node's connection resource. We need to link * the remote node to this connection resource so we can have * easy access to the connection resource during the scheduler calculations. */ static void link_rsc2remotenode(pcmk_scheduler_t *scheduler, pcmk_resource_t *new_rsc) { pcmk_node_t *remote_node = NULL; if (new_rsc->is_remote_node == FALSE) { return; } if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { /* remote_nodes and remote_resources are not linked in quick location calculations */ return; } remote_node = pe_find_node(scheduler->nodes, new_rsc->id); CRM_CHECK(remote_node != NULL, return); pcmk__rsc_trace(new_rsc, "Linking remote connection resource %s to %s", new_rsc->id, pcmk__node_name(remote_node)); remote_node->details->remote_rsc = new_rsc; if (new_rsc->container == NULL) { /* Handle start-up fencing for remote nodes (as opposed to guest nodes) * the same as is done for cluster nodes. */ handle_startup_fencing(scheduler, remote_node); } else { /* pe_create_node() marks the new node as "remote" or "cluster"; now * that we know the node is a guest node, update it correctly. */ g_hash_table_replace(remote_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("container")); } } static void destroy_tag(gpointer data) { pcmk_tag_t *tag = data; if (tag) { free(tag->id); g_list_free_full(tag->refs, free); free(tag); } } /*! * \internal * \brief Parse configuration XML for resource information * * \param[in] xml_resources Top of resource configuration XML * \param[in,out] scheduler Scheduler data * * \return TRUE * * \note unpack_remote_nodes() MUST be called before this, so that the nodes can * be used when pe__unpack_resource() calls resource_location() */ gboolean unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; GList *gIter = NULL; scheduler->template_rsc_sets = pcmk__strkey_table(free, destroy_tag); for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { pcmk_resource_t *new_rsc = NULL; const char *id = ID(xml_obj); if (pcmk__str_empty(id)) { pcmk__config_err("Ignoring <%s> resource without ID", xml_obj->name); continue; } if (pcmk__str_eq((const char *) xml_obj->name, PCMK_XE_TEMPLATE, pcmk__str_none)) { if (g_hash_table_lookup_extended(scheduler->template_rsc_sets, id, NULL, NULL) == FALSE) { /* Record the template's ID for the knowledge of its existence anyway. */ g_hash_table_insert(scheduler->template_rsc_sets, strdup(id), NULL); } continue; } crm_trace("Unpacking <%s " PCMK_XA_ID "='%s'>", xml_obj->name, id); if (pe__unpack_resource(xml_obj, &new_rsc, NULL, scheduler) == pcmk_rc_ok) { scheduler->resources = g_list_append(scheduler->resources, new_rsc); pcmk__rsc_trace(new_rsc, "Added resource %s", new_rsc->id); } else { pcmk__config_err("Ignoring <%s> resource '%s' " "because configuration is invalid", xml_obj->name, id); } } for (gIter = scheduler->resources; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; setup_container(rsc, scheduler); link_rsc2remotenode(scheduler, rsc); } scheduler->resources = g_list_sort(scheduler->resources, pe__cmp_rsc_priority); if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { /* Ignore */ } else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled) && !pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) { pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined"); pcmk__config_err("Either configure some or disable STONITH with the " PCMK_OPT_STONITH_ENABLED " option"); pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity"); } return TRUE; } gboolean unpack_tags(xmlNode *xml_tags, pcmk_scheduler_t *scheduler) { xmlNode *xml_tag = NULL; scheduler->tags = pcmk__strkey_table(free, destroy_tag); for (xml_tag = pcmk__xe_first_child(xml_tags); xml_tag != NULL; xml_tag = pcmk__xe_next(xml_tag)) { xmlNode *xml_obj_ref = NULL; const char *tag_id = ID(xml_tag); if (!pcmk__str_eq((const char *) xml_tag->name, PCMK_XE_TAG, pcmk__str_none)) { continue; } if (tag_id == NULL) { pcmk__config_err("Ignoring <%s> without " PCMK_XA_ID, (const char *) xml_tag->name); continue; } for (xml_obj_ref = pcmk__xe_first_child(xml_tag); xml_obj_ref != NULL; xml_obj_ref = pcmk__xe_next(xml_obj_ref)) { const char *obj_ref = ID(xml_obj_ref); if (!pcmk__str_eq((const char *) xml_obj_ref->name, PCMK_XE_OBJ_REF, pcmk__str_none)) { continue; } if (obj_ref == NULL) { pcmk__config_err("Ignoring <%s> for tag '%s' without " PCMK_XA_ID, xml_obj_ref->name, tag_id); continue; } if (add_tag_ref(scheduler->tags, tag_id, obj_ref) == FALSE) { return FALSE; } } } return TRUE; } /* The ticket state section: * "/cib/status/tickets/ticket_state" */ static gboolean unpack_ticket_state(xmlNode *xml_ticket, pcmk_scheduler_t *scheduler) { const char *ticket_id = NULL; const char *granted = NULL; const char *last_granted = NULL; const char *standby = NULL; xmlAttrPtr xIter = NULL; pcmk_ticket_t *ticket = NULL; ticket_id = ID(xml_ticket); if (pcmk__str_empty(ticket_id)) { return FALSE; } crm_trace("Processing ticket state for %s", ticket_id); ticket = g_hash_table_lookup(scheduler->tickets, ticket_id); if (ticket == NULL) { ticket = ticket_new(ticket_id, scheduler); if (ticket == NULL) { return FALSE; } } for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = pcmk__xml_attr_value(xIter); if (pcmk__str_eq(prop_name, PCMK_XA_ID, pcmk__str_none)) { continue; } g_hash_table_replace(ticket->state, strdup(prop_name), strdup(prop_value)); } granted = g_hash_table_lookup(ticket->state, PCMK__XA_GRANTED); if (granted && crm_is_true(granted)) { ticket->granted = TRUE; crm_info("We have ticket '%s'", ticket->id); } else { ticket->granted = FALSE; crm_info("We do not have ticket '%s'", ticket->id); } last_granted = g_hash_table_lookup(ticket->state, PCMK_XA_LAST_GRANTED); if (last_granted) { long long last_granted_ll; pcmk__scan_ll(last_granted, &last_granted_ll, 0LL); ticket->last_granted = (time_t) last_granted_ll; } standby = g_hash_table_lookup(ticket->state, PCMK_XA_STANDBY); if (standby && crm_is_true(standby)) { ticket->standby = TRUE; if (ticket->granted) { crm_info("Granted ticket '%s' is in standby-mode", ticket->id); } } else { ticket->standby = FALSE; } crm_trace("Done with ticket state for %s", ticket_id); return TRUE; } static gboolean unpack_tickets_state(xmlNode *xml_tickets, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; for (xml_obj = pcmk__xe_first_child(xml_tickets); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (!pcmk__str_eq((const char *) xml_obj->name, PCMK__XE_TICKET_STATE, pcmk__str_none)) { continue; } unpack_ticket_state(xml_obj, scheduler); } return TRUE; } static void unpack_handle_remote_attrs(pcmk_node_t *this_node, const xmlNode *state, pcmk_scheduler_t *scheduler) { const char *discovery = NULL; const xmlNode *attrs = NULL; pcmk_resource_t *rsc = NULL; if (!pcmk__str_eq((const char *) state->name, PCMK__XE_NODE_STATE, pcmk__str_none)) { return; } if ((this_node == NULL) || !pe__is_guest_or_remote_node(this_node)) { return; } crm_trace("Processing Pacemaker Remote node %s", pcmk__node_name(this_node)); pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_IN_MAINTENANCE), &(this_node->details->remote_maintenance), 0); rsc = this_node->details->remote_rsc; if (this_node->details->remote_requires_reset == FALSE) { this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; } attrs = find_xml_node(state, PCMK__XE_TRANSIENT_ATTRIBUTES, FALSE); add_node_attrs(attrs, this_node, TRUE, scheduler); if (pe__shutdown_requested(this_node)) { crm_info("%s is shutting down", pcmk__node_name(this_node)); this_node->details->shutdown = TRUE; } if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) { crm_info("%s is in standby mode", pcmk__node_name(this_node)); this_node->details->standby = TRUE; } if (crm_is_true(pe_node_attribute_raw(this_node, PCMK_NODE_ATTR_MAINTENANCE)) || ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_managed))) { crm_info("%s is in maintenance mode", pcmk__node_name(this_node)); this_node->details->maintenance = TRUE; } discovery = pe_node_attribute_raw(this_node, PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED); if ((discovery != NULL) && !crm_is_true(discovery)) { pcmk__warn_once(pcmk__wo_rdisc_enabled, "Support for the " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " node attribute is deprecated and will be removed" " (and behave as 'true') in a future release."); if (pe__is_remote_node(this_node) && !pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("Ignoring " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " attribute on Pacemaker Remote node %s" " because fencing is disabled", pcmk__node_name(this_node)); } else { /* This is either a remote node with fencing enabled, or a guest * node. We don't care whether fencing is enabled when fencing guest * nodes, because they are "fenced" by recovering their containing * resource. */ crm_info("%s has resource discovery disabled", pcmk__node_name(this_node)); this_node->details->rsc_discovery_enabled = FALSE; } } } /*! * \internal * \brief Unpack a cluster node's transient attributes * * \param[in] state CIB node state XML * \param[in,out] node Cluster node whose attributes are being unpacked * \param[in,out] scheduler Scheduler data */ static void unpack_transient_attributes(const xmlNode *state, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { const char *discovery = NULL; const xmlNode *attrs = find_xml_node(state, PCMK__XE_TRANSIENT_ATTRIBUTES, FALSE); add_node_attrs(attrs, node, TRUE, scheduler); if (crm_is_true(pe_node_attribute_raw(node, "standby"))) { crm_info("%s is in standby mode", pcmk__node_name(node)); node->details->standby = TRUE; } if (crm_is_true(pe_node_attribute_raw(node, PCMK_NODE_ATTR_MAINTENANCE))) { crm_info("%s is in maintenance mode", pcmk__node_name(node)); node->details->maintenance = TRUE; } discovery = pe_node_attribute_raw(node, PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED); if ((discovery != NULL) && !crm_is_true(discovery)) { pcmk__config_warn("Ignoring " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " attribute for %s because disabling resource" " discovery is not allowed for cluster nodes", pcmk__node_name(node)); } } /*! * \internal * \brief Unpack a node state entry (first pass) * * Unpack one node state entry from status. This unpacks information from the * \C PCMK__XE_NODE_STATE element itself and node attributes inside it, but not * the resource history inside it. Multiple passes through the status are needed * to fully unpack everything. * * \param[in] state CIB node state XML * \param[in,out] scheduler Scheduler data */ static void unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *uname = NULL; pcmk_node_t *this_node = NULL; id = crm_element_value(state, PCMK_XA_ID); if (id == NULL) { pcmk__config_err("Ignoring invalid " PCMK__XE_NODE_STATE " entry without " PCMK_XA_ID); crm_log_xml_info(state, "missing-id"); return; } uname = crm_element_value(state, PCMK_XA_UNAME); if (uname == NULL) { /* If a joining peer makes the cluster acquire the quorum from corosync * meanwhile it has not joined CPG membership of pacemaker-controld yet, * it's possible that the created PCMK__XE_NODE_STATE entry doesn't have * a PCMK_XA_UNAME yet. We should recognize the node as `pending` and * wait for it to join CPG. */ crm_trace("Handling " PCMK__XE_NODE_STATE " entry with id=\"%s\" " "without " PCMK_XA_UNAME, id); } this_node = pe_find_node_any(scheduler->nodes, id, uname); if (this_node == NULL) { pcmk__config_warn("Ignoring recorded node state for id=\"%s\" (%s) " "because it is no longer in the configuration", id, pcmk__s(uname, "uname unknown")); return; } if (pe__is_guest_or_remote_node(this_node)) { /* We can't determine the online status of Pacemaker Remote nodes until * after all resource history has been unpacked. In this first pass, we * do need to mark whether the node has been fenced, as this plays a * role during unpacking cluster node resource state. */ pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_FENCED), &(this_node->details->remote_was_fenced), 0); return; } unpack_transient_attributes(state, this_node, scheduler); /* Provisionally mark this cluster node as clean. We have at least seen it * in the current cluster's lifetime. */ this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; crm_trace("Determining online status of cluster node %s (id %s)", pcmk__node_name(this_node), id); determine_online_status(state, this_node, scheduler); if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate) && this_node->details->online && (scheduler->no_quorum_policy == pcmk_no_quorum_fence)) { /* Everything else should flow from this automatically * (at least until the scheduler becomes able to migrate off * healthy resources) */ pe_fence_node(scheduler, this_node, "cluster does not have quorum", FALSE); } } /*! * \internal * \brief Unpack nodes' resource history as much as possible * * Unpack as many nodes' resource history as possible in one pass through the * status. We need to process Pacemaker Remote nodes' connections/containers * before unpacking their history; the connection/container history will be * in another node's history, so it might take multiple passes to unpack * everything. * * \param[in] status CIB XML status section * \param[in] fence If true, treat any not-yet-unpacked nodes as unseen * \param[in,out] scheduler Scheduler data * * \return Standard Pacemaker return code (specifically pcmk_rc_ok if done, * or EAGAIN if more unpacking remains to be done) */ static int unpack_node_history(const xmlNode *status, bool fence, pcmk_scheduler_t *scheduler) { int rc = pcmk_rc_ok; // Loop through all PCMK__XE_NODE_STATE entries in CIB status for (const xmlNode *state = first_named_child(status, PCMK__XE_NODE_STATE); state != NULL; state = crm_next_same_xml(state)) { const char *id = ID(state); const char *uname = crm_element_value(state, PCMK_XA_UNAME); pcmk_node_t *this_node = NULL; if ((id == NULL) || (uname == NULL)) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history from malformed " PCMK__XE_NODE_STATE " without id and/or uname"); continue; } this_node = pe_find_node_any(scheduler->nodes, id, uname); if (this_node == NULL) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history for node %s because " "no longer in configuration", id); continue; } if (this_node->details->unpacked) { crm_trace("Not unpacking resource history for node %s because " "already unpacked", id); continue; } if (fence) { // We're processing all remaining nodes } else if (pe__is_guest_node(this_node)) { /* We can unpack a guest node's history only after we've unpacked * other resource history to the point that we know that the node's * connection and containing resource are both up. */ pcmk_resource_t *rsc = this_node->details->remote_rsc; if ((rsc == NULL) || (rsc->role != pcmk_role_started) || (rsc->container->role != pcmk_role_started)) { crm_trace("Not unpacking resource history for guest node %s " "because container and connection are not known to " "be up", id); continue; } } else if (pe__is_remote_node(this_node)) { /* We can unpack a remote node's history only after we've unpacked * other resource history to the point that we know that the node's * connection is up, with the exception of when shutdown locks are * in use. */ pcmk_resource_t *rsc = this_node->details->remote_rsc; if ((rsc == NULL) || (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock) && (rsc->role != pcmk_role_started))) { crm_trace("Not unpacking resource history for remote node %s " "because connection is not known to be up", id); continue; } /* If fencing and shutdown locks are disabled and we're not processing * unseen nodes, then we don't want to unpack offline nodes until online * nodes have been unpacked. This allows us to number active clone * instances first. */ } else if (!pcmk_any_flags_set(scheduler->flags, pcmk_sched_fencing_enabled |pcmk_sched_shutdown_lock) && !this_node->details->online) { crm_trace("Not unpacking resource history for offline " "cluster node %s", id); continue; } if (pe__is_guest_or_remote_node(this_node)) { determine_remote_online_status(scheduler, this_node); unpack_handle_remote_attrs(this_node, state, scheduler); } crm_trace("Unpacking resource history for %snode %s", (fence? "unseen " : ""), id); this_node->details->unpacked = TRUE; unpack_node_lrm(this_node, state, scheduler); rc = EAGAIN; // Other node histories might depend on this one } return rc; } /* remove nodes that are down, stopping */ /* create positive rsc_to_node constraints between resources and the nodes they are running on */ /* anything else? */ gboolean unpack_status(xmlNode *status, pcmk_scheduler_t *scheduler) { xmlNode *state = NULL; crm_trace("Beginning unpack"); if (scheduler->tickets == NULL) { scheduler->tickets = pcmk__strkey_table(free, destroy_ticket); } for (state = pcmk__xe_first_child(status); state != NULL; state = pcmk__xe_next(state)) { if (pcmk__str_eq((const char *) state->name, PCMK_XE_TICKETS, pcmk__str_none)) { unpack_tickets_state((xmlNode *) state, scheduler); } else if (pcmk__str_eq((const char *) state->name, PCMK__XE_NODE_STATE, pcmk__str_none)) { unpack_node_state(state, scheduler); } } while (unpack_node_history(status, FALSE, scheduler) == EAGAIN) { crm_trace("Another pass through node resource histories is needed"); } // Now catch any nodes we didn't see unpack_node_history(status, pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled), scheduler); /* Now that we know where resources are, we can schedule stops of containers * with failed bundle connections */ if (scheduler->stop_needed != NULL) { for (GList *item = scheduler->stop_needed; item; item = item->next) { pcmk_resource_t *container = item->data; pcmk_node_t *node = pcmk__current_node(container); if (node) { stop_action(container, node, FALSE); } } g_list_free(scheduler->stop_needed); scheduler->stop_needed = NULL; } /* Now that we know status of all Pacemaker Remote connections and nodes, * we can stop connections for node shutdowns, and check the online status * of remote/guest nodes that didn't have any node history to unpack. */ for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *this_node = gIter->data; if (!pe__is_guest_or_remote_node(this_node)) { continue; } if (this_node->details->shutdown && (this_node->details->remote_rsc != NULL)) { pe__set_next_role(this_node->details->remote_rsc, pcmk_role_stopped, "remote shutdown"); } if (!this_node->details->unpacked) { determine_remote_online_status(scheduler, this_node); } } return TRUE; } /*! * \internal * \brief Unpack node's time when it became a member at the cluster layer * * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry * \param[in,out] scheduler Scheduler data * * \return Epoch time when node became a cluster member * (or scheduler effective time for legacy entries) if a member, * 0 if not a member, or -1 if no valid information available */ static long long unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler) { const char *member_time = crm_element_value(node_state, PCMK__XA_IN_CCM); int member = 0; if (member_time == NULL) { return -1LL; } else if (crm_str_to_boolean(member_time, &member) == 1) { /* If in_ccm=0, we'll return 0 here. If in_ccm=1, either the entry was * recorded as a boolean for a DC < 2.1.7, or the node is pending * shutdown and has left the CPG, in which case it was set to 1 to avoid * fencing for PCMK_OPT_NODE_PENDING_TIMEOUT. * * We return the effective time for in_ccm=1 because what's important to * avoid fencing is that effective time minus this value is less than * the pending node timeout. */ return member? (long long) get_effective_time(scheduler) : 0LL; } else { long long when_member = 0LL; if ((pcmk__scan_ll(member_time, &when_member, 0LL) != pcmk_rc_ok) || (when_member < 0LL)) { crm_warn("Unrecognized value '%s' for " PCMK__XA_IN_CCM " in " PCMK__XE_NODE_STATE " entry", member_time); return -1LL; } return when_member; } } /*! * \internal * \brief Unpack node's time when it became online in process group * * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry * * \return Epoch time when node became online in process group (or 0 if not * online, or 1 for legacy online entries) */ static long long unpack_node_online(const xmlNode *node_state) { const char *peer_time = crm_element_value(node_state, PCMK_XA_CRMD); // @COMPAT Entries recorded for DCs < 2.1.7 have "online" or "offline" - if (pcmk__str_eq(peer_time, OFFLINESTATUS, + if (pcmk__str_eq(peer_time, PCMK_VALUE_OFFLINE, pcmk__str_casei|pcmk__str_null_matches)) { return 0LL; } else if (pcmk__str_eq(peer_time, PCMK_VALUE_ONLINE, pcmk__str_casei)) { return 1LL; } else { long long when_online = 0LL; if ((pcmk__scan_ll(peer_time, &when_online, 0LL) != pcmk_rc_ok) || (when_online < 0)) { crm_warn("Unrecognized value '%s' for " PCMK_XA_CRMD " in " PCMK__XE_NODE_STATE " entry, assuming offline", peer_time); return 0LL; } return when_online; } } /*! * \internal * \brief Unpack node attribute for user-requested fencing * * \param[in] node Node to check * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry in CIB status * * \return \c true if fencing has been requested for \p node, otherwise \c false */ static bool unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state) { long long value = 0LL; int value_i = 0; const char *value_s = pe_node_attribute_raw(node, PCMK_NODE_ATTR_TERMINATE); // Value may be boolean or an epoch time if (crm_str_to_boolean(value_s, &value_i) == 1) { return (value_i != 0); } if (pcmk__scan_ll(value_s, &value, 0LL) == pcmk_rc_ok) { return (value > 0); } crm_warn("Ignoring unrecognized value '%s' for " PCMK_NODE_ATTR_TERMINATE "node attribute for %s", value_s, pcmk__node_name(node)); return false; } static gboolean determine_online_status_no_fencing(pcmk_scheduler_t *scheduler, const xmlNode *node_state, pcmk_node_t *this_node) { gboolean online = FALSE; const char *join = crm_element_value(node_state, PCMK__XA_JOIN); const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); long long when_member = unpack_node_member(node_state, scheduler); long long when_online = unpack_node_online(node_state); if (when_member <= 0) { crm_trace("Node %s is %sdown", pcmk__node_name(this_node), ((when_member < 0)? "presumed " : "")); } else if (when_online > 0) { if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { online = TRUE; } else { crm_debug("Node %s is not ready to run resources: %s", pcmk__node_name(this_node), join); } } else if (this_node->details->expected_up == FALSE) { crm_trace("Node %s controller is down: " "member@%lld online@%lld join=%s expected=%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, "")); } else { /* mark it unclean */ pe_fence_node(scheduler, this_node, "peer is unexpectedly down", FALSE); crm_info("Node %s member@%lld online@%lld join=%s expected=%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, "")); } return online; } /*! * \internal * \brief Check whether a node has taken too long to join controller group * * \param[in,out] scheduler Scheduler data * \param[in] node Node to check * \param[in] when_member Epoch time when node became a cluster member * \param[in] when_online Epoch time when node joined controller group * * \return true if node has been pending (on the way up) longer than * \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false * \note This will also update the cluster's recheck time if appropriate. */ static inline bool pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, long long when_member, long long when_online) { if ((scheduler->node_pending_timeout > 0) && (when_member > 0) && (when_online <= 0)) { // There is a timeout on pending nodes, and node is pending time_t timeout = when_member + scheduler->node_pending_timeout; if (get_effective_time(node->details->data_set) >= timeout) { return true; // Node has timed out } // Node is pending, but still has time pe__update_recheck_time(timeout, scheduler, "pending node timeout"); } return false; } static bool determine_online_status_fencing(pcmk_scheduler_t *scheduler, const xmlNode *node_state, pcmk_node_t *this_node) { bool termination_requested = unpack_node_terminate(this_node, node_state); const char *join = crm_element_value(node_state, PCMK__XA_JOIN); const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); long long when_member = unpack_node_member(node_state, scheduler); long long when_online = unpack_node_online(node_state); /* - PCMK__XA_JOIN ::= member|down|pending|banned - PCMK_XA_EXPECTED ::= member|down @COMPAT with entries recorded for DCs < 2.1.7 - PCMK__XA_IN_CCM ::= true|false - PCMK_XA_CRMD ::= online|offline Since crm_feature_set 3.18.0 (pacemaker-2.1.7): - PCMK__XA_IN_CCM ::= |0 Since when node has been a cluster member. A value 0 of means the node is not a cluster member. - PCMK_XA_CRMD ::= |0 Since when peer has been online in CPG. A value 0 means the peer is offline in CPG. */ crm_trace("Node %s member@%lld online@%lld join=%s expected=%s%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, ""), (termination_requested? " (termination requested)" : "")); if (this_node->details->shutdown) { crm_debug("%s is shutting down", pcmk__node_name(this_node)); /* Slightly different criteria since we can't shut down a dead peer */ return (when_online > 0); } if (when_member < 0) { pe_fence_node(scheduler, this_node, "peer has not been seen by the cluster", FALSE); return false; } if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_none)) { pe_fence_node(scheduler, this_node, "peer failed Pacemaker membership criteria", FALSE); } else if (termination_requested) { if ((when_member <= 0) && (when_online <= 0) && pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_none)) { crm_info("%s was fenced as requested", pcmk__node_name(this_node)); return false; } pe_fence_node(scheduler, this_node, "fencing was requested", false); } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN, pcmk__str_null_matches)) { if (pending_too_long(scheduler, this_node, when_member, when_online)) { pe_fence_node(scheduler, this_node, "peer pending timed out on joining the process group", FALSE); } else if ((when_member > 0) || (when_online > 0)) { crm_info("- %s is not ready to run resources", pcmk__node_name(this_node)); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { crm_trace("%s is down or still coming up", pcmk__node_name(this_node)); } } else if (when_member <= 0) { // Consider PCMK_OPT_PRIORITY_FENCING_DELAY for lost nodes pe_fence_node(scheduler, this_node, "peer is no longer part of the cluster", TRUE); } else if (when_online <= 0) { pe_fence_node(scheduler, this_node, "peer process is no longer available", FALSE); /* Everything is running at this point, now check join state */ } else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_none)) { crm_info("%s is active", pcmk__node_name(this_node)); } else if (pcmk__str_any_of(join, CRMD_JOINSTATE_PENDING, CRMD_JOINSTATE_DOWN, NULL)) { crm_info("%s is not ready to run resources", pcmk__node_name(this_node)); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { pe_fence_node(scheduler, this_node, "peer was in an unknown state", FALSE); } return (when_member > 0); } static void determine_remote_online_status(pcmk_scheduler_t *scheduler, pcmk_node_t *this_node) { pcmk_resource_t *rsc = this_node->details->remote_rsc; pcmk_resource_t *container = NULL; pcmk_node_t *host = NULL; /* If there is a node state entry for a (former) Pacemaker Remote node * but no resource creating that node, the node's connection resource will * be NULL. Consider it an offline remote node in that case. */ if (rsc == NULL) { this_node->details->online = FALSE; goto remote_online_done; } container = rsc->container; if (container && pcmk__list_of_1(rsc->running_on)) { host = rsc->running_on->data; } /* If the resource is currently started, mark it online. */ if (rsc->role == pcmk_role_started) { crm_trace("%s node %s presumed ONLINE because connection resource is started", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = TRUE; } /* consider this node shutting down if transitioning start->stop */ if ((rsc->role == pcmk_role_started) && (rsc->next_role == pcmk_role_stopped)) { crm_trace("%s node %s shutting down because connection resource is stopping", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->shutdown = TRUE; } /* Now check all the failure conditions. */ if(container && pcmk_is_set(container->flags, pcmk_rsc_failed)) { crm_trace("Guest node %s UNCLEAN because guest resource failed", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { crm_trace("%s node %s OFFLINE because connection resource failed", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; } else if ((rsc->role == pcmk_role_stopped) || ((container != NULL) && (container->role == pcmk_role_stopped))) { crm_trace("%s node %s OFFLINE because its resource is stopped", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = FALSE; } else if (host && (host->details->online == FALSE) && host->details->unclean) { crm_trace("Guest node %s UNCLEAN because host is unclean", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } remote_online_done: crm_trace("Remote node %s online=%s", this_node->details->id, this_node->details->online ? "TRUE" : "FALSE"); } static void determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node, pcmk_scheduler_t *scheduler) { gboolean online = FALSE; const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); CRM_CHECK(this_node != NULL, return); this_node->details->shutdown = FALSE; this_node->details->expected_up = FALSE; if (pe__shutdown_requested(this_node)) { this_node->details->shutdown = TRUE; } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { this_node->details->expected_up = TRUE; } if (this_node->details->type == node_ping) { this_node->details->unclean = FALSE; online = FALSE; /* As far as resource management is concerned, * the node is safely offline. * Anyone caught abusing this logic will be shot */ } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { online = determine_online_status_no_fencing(scheduler, node_state, this_node); } else { online = determine_online_status_fencing(scheduler, node_state, this_node); } if (online) { this_node->details->online = TRUE; } else { /* remove node from contention */ this_node->fixed = TRUE; // @COMPAT deprecated and unused this_node->weight = -INFINITY; } if (online && this_node->details->shutdown) { /* don't run resources here */ this_node->fixed = TRUE; // @COMPAT deprecated and unused this_node->weight = -INFINITY; } if (this_node->details->type == node_ping) { crm_info("%s is not a Pacemaker node", pcmk__node_name(this_node)); } else if (this_node->details->unclean) { pcmk__sched_warn("%s is unclean", pcmk__node_name(this_node)); } else if (this_node->details->online) { crm_info("%s is %s", pcmk__node_name(this_node), this_node->details->shutdown ? "shutting down" : this_node->details->pending ? "pending" : this_node->details->standby ? "standby" : this_node->details->maintenance ? "maintenance" : "online"); } else { crm_trace("%s is offline", pcmk__node_name(this_node)); } } /*! * \internal * \brief Find the end of a resource's name, excluding any clone suffix * * \param[in] id Resource ID to check * * \return Pointer to last character of resource's base name */ const char * pe_base_name_end(const char *id) { if (!pcmk__str_empty(id)) { const char *end = id + strlen(id) - 1; for (const char *s = end; s > id; --s) { switch (*s) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': break; case ':': return (s == end)? s : (s - 1); default: return end; } } return end; } return NULL; } /*! * \internal * \brief Get a resource name excluding any clone suffix * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_strip(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); char *basename = NULL; CRM_ASSERT(end); basename = strndup(last_rsc_id, end - last_rsc_id + 1); CRM_ASSERT(basename); return basename; } /*! * \internal * \brief Get the name of the first instance of a cloned resource * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name plus :0 * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_zero(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); size_t base_name_len = end - last_rsc_id + 1; char *zero = NULL; CRM_ASSERT(end); zero = calloc(base_name_len + 3, sizeof(char)); CRM_ASSERT(zero); memcpy(zero, last_rsc_id, base_name_len); zero[base_name_len] = ':'; zero[base_name_len + 1] = '0'; return zero; } static pcmk_resource_t * create_fake_resource(const char *rsc_id, const xmlNode *rsc_entry, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = NULL; xmlNode *xml_rsc = create_xml_node(NULL, PCMK_XE_PRIMITIVE); copy_in_properties(xml_rsc, rsc_entry); crm_xml_add(xml_rsc, PCMK_XA_ID, rsc_id); crm_log_xml_debug(xml_rsc, "Orphan resource"); if (pe__unpack_resource(xml_rsc, &rsc, NULL, scheduler) != pcmk_rc_ok) { return NULL; } if (xml_contains_remote_node(xml_rsc)) { pcmk_node_t *node; crm_debug("Detected orphaned remote node %s", rsc_id); node = pe_find_node(scheduler->nodes, rsc_id); if (node == NULL) { node = pe_create_node(rsc_id, rsc_id, PCMK_VALUE_REMOTE, NULL, scheduler); } link_rsc2remotenode(scheduler, rsc); if (node) { crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id); node->details->shutdown = TRUE; } } if (crm_element_value(rsc_entry, PCMK__META_CONTAINER)) { /* This orphaned rsc needs to be mapped to a container. */ crm_trace("Detected orphaned container filler %s", rsc_id); pcmk__set_rsc_flags(rsc, pcmk_rsc_removed_filler); } pcmk__set_rsc_flags(rsc, pcmk_rsc_removed); scheduler->resources = g_list_append(scheduler->resources, rsc); return rsc; } /*! * \internal * \brief Create orphan instance for anonymous clone resource history * * \param[in,out] parent Clone resource that orphan will be added to * \param[in] rsc_id Orphan's resource ID * \param[in] node Where orphan is active (for logging only) * \param[in,out] scheduler Scheduler data * * \return Newly added orphaned instance of \p parent */ static pcmk_resource_t * create_anonymous_orphan(pcmk_resource_t *parent, const char *rsc_id, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk_resource_t *top = pe__create_clone_child(parent, scheduler); // find_rsc() because we might be a cloned group pcmk_resource_t *orphan = top->fns->find_rsc(top, rsc_id, NULL, pcmk_rsc_match_clone_only); pcmk__rsc_debug(parent, "Created orphan %s for %s: %s on %s", top->id, parent->id, rsc_id, pcmk__node_name(node)); return orphan; } /*! * \internal * \brief Check a node for an instance of an anonymous clone * * Return a child instance of the specified anonymous clone, in order of * preference: (1) the instance running on the specified node, if any; * (2) an inactive instance (i.e. within the total of \c PCMK_META_CLONE_MAX * instances); (3) a newly created orphan (that is, \c PCMK_META_CLONE_MAX * instances are already active). * * \param[in,out] scheduler Scheduler data * \param[in] node Node on which to check for instance * \param[in,out] parent Clone to check * \param[in] rsc_id Name of cloned resource in history (no instance) */ static pcmk_resource_t * find_anonymous_clone(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, pcmk_resource_t *parent, const char *rsc_id) { GList *rIter = NULL; pcmk_resource_t *rsc = NULL; pcmk_resource_t *inactive_instance = NULL; gboolean skip_inactive = FALSE; CRM_ASSERT(parent != NULL); CRM_ASSERT(pe_rsc_is_clone(parent)); CRM_ASSERT(!pcmk_is_set(parent->flags, pcmk_rsc_unique)); // Check for active (or partially active, for cloned groups) instance pcmk__rsc_trace(parent, "Looking for %s on %s in %s", rsc_id, pcmk__node_name(node), parent->id); for (rIter = parent->children; rsc == NULL && rIter; rIter = rIter->next) { GList *locations = NULL; pcmk_resource_t *child = rIter->data; /* Check whether this instance is already known to be active or pending * anywhere, at this stage of unpacking. Because this function is called * for a resource before the resource's individual operation history * entries are unpacked, locations will generally not contain the * desired node. * * However, there are three exceptions: * (1) when child is a cloned group and we have already unpacked the * history of another member of the group on the same node; * (2) when we've already unpacked the history of another numbered * instance on the same node (which can happen if * PCMK_META_GLOBALLY_UNIQUE was flipped from true to false); and * (3) when we re-run calculations on the same scheduler data as part of * a simulation. */ child->fns->location(child, &locations, 2); if (locations) { /* We should never associate the same numbered anonymous clone * instance with multiple nodes, and clone instances can't migrate, * so there must be only one location, regardless of history. */ CRM_LOG_ASSERT(locations->next == NULL); if (pcmk__same_node((pcmk_node_t *) locations->data, node)) { /* This child instance is active on the requested node, so check * for a corresponding configured resource. We use find_rsc() * instead of child because child may be a cloned group, and we * need the particular member corresponding to rsc_id. * * If the history entry is orphaned, rsc will be NULL. */ rsc = parent->fns->find_rsc(child, rsc_id, NULL, pcmk_rsc_match_clone_only); if (rsc) { /* If there are multiple instance history entries for an * anonymous clone in a single node's history (which can * happen if PCMK_META_GLOBALLY_UNIQUE is switched from true * to false), we want to consider the instances beyond the * first as orphans, even if there are inactive instance * numbers available. */ if (rsc->running_on) { crm_notice("Active (now-)anonymous clone %s has " "multiple (orphan) instance histories on %s", parent->id, pcmk__node_name(node)); skip_inactive = TRUE; rsc = NULL; } else { pcmk__rsc_trace(parent, "Resource %s, active", rsc->id); } } } g_list_free(locations); } else { pcmk__rsc_trace(parent, "Resource %s, skip inactive", child->id); if (!skip_inactive && !inactive_instance && !pcmk_is_set(child->flags, pcmk_rsc_blocked)) { // Remember one inactive instance in case we don't find active inactive_instance = parent->fns->find_rsc(child, rsc_id, NULL, pcmk_rsc_match_clone_only); /* ... but don't use it if it was already associated with a * pending action on another node */ if ((inactive_instance != NULL) && (inactive_instance->pending_node != NULL) && !pcmk__same_node(inactive_instance->pending_node, node)) { inactive_instance = NULL; } } } } if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) { pcmk__rsc_trace(parent, "Resource %s, empty slot", inactive_instance->id); rsc = inactive_instance; } /* If the resource has "requires" set to "quorum" or "nothing", and we don't * have a clone instance for every node, we don't want to consume a valid * instance number for unclean nodes. Such instances may appear to be active * according to the history, but should be considered inactive, so we can * start an instance elsewhere. Treat such instances as orphans. * * An exception is instances running on guest nodes -- since guest node * "fencing" is actually just a resource stop, requires shouldn't apply. * * @TODO Ideally, we'd use an inactive instance number if it is not needed * for any clean instances. However, we don't know that at this point. */ if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing) && (!node->details->online || node->details->unclean) && !pe__is_guest_node(node) && !pe__is_universal_clone(parent, scheduler)) { rsc = NULL; } if (rsc == NULL) { rsc = create_anonymous_orphan(parent, rsc_id, node, scheduler); pcmk__rsc_trace(parent, "Resource %s, orphan", rsc->id); } return rsc; } static pcmk_resource_t * unpack_find_resource(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, const char *rsc_id) { pcmk_resource_t *rsc = NULL; pcmk_resource_t *parent = NULL; crm_trace("looking for %s", rsc_id); rsc = pe_find_resource(scheduler->resources, rsc_id); if (rsc == NULL) { /* If we didn't find the resource by its name in the operation history, * check it again as a clone instance. Even when PCMK_META_CLONE_MAX=0, * we create a single :0 orphan to match against here. */ char *clone0_id = clone_zero(rsc_id); pcmk_resource_t *clone0 = pe_find_resource(scheduler->resources, clone0_id); if (clone0 && !pcmk_is_set(clone0->flags, pcmk_rsc_unique)) { rsc = clone0; parent = uber_parent(clone0); crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id); } else { crm_trace("%s is not known as %s either (orphan)", rsc_id, clone0_id); } free(clone0_id); } else if (rsc->variant > pcmk_rsc_variant_primitive) { crm_trace("Resource history for %s is orphaned because it is no longer primitive", rsc_id); return NULL; } else { parent = uber_parent(rsc); } if (pe_rsc_is_anon_clone(parent)) { if (pe_rsc_is_bundled(parent)) { rsc = pe__find_bundle_replica(parent->parent, node); } else { char *base = clone_strip(rsc_id); rsc = find_anonymous_clone(scheduler, node, parent, base); free(base); CRM_ASSERT(rsc != NULL); } } if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_casei) && !pcmk__str_eq(rsc_id, rsc->clone_name, pcmk__str_casei)) { pcmk__str_update(&rsc->clone_name, rsc_id); pcmk__rsc_debug(rsc, "Internally renamed %s on %s to %s%s", rsc_id, pcmk__node_name(node), rsc->id, pcmk_is_set(rsc->flags, pcmk_rsc_removed)? " (ORPHAN)" : ""); } return rsc; } static pcmk_resource_t * process_orphan_resource(const xmlNode *rsc_entry, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = NULL; const char *rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID); crm_debug("Detected orphan resource %s on %s", rsc_id, pcmk__node_name(node)); rsc = create_fake_resource(rsc_id, rsc_entry, scheduler); if (rsc == NULL) { return NULL; } if (!pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed); } else { CRM_CHECK(rsc != NULL, return NULL); pcmk__rsc_trace(rsc, "Added orphan %s", rsc->id); resource_location(rsc, NULL, -INFINITY, "__orphan_do_not_run__", scheduler); } return rsc; } static void process_rsc_state(pcmk_resource_t *rsc, pcmk_node_t *node, enum action_fail_response on_fail) { pcmk_node_t *tmpnode = NULL; char *reason = NULL; enum action_fail_response save_on_fail = pcmk_on_fail_ignore; CRM_ASSERT(rsc); pcmk__rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s", rsc->id, pcmk_role_text(rsc->role), pcmk__node_name(node), fail2text(on_fail)); /* process current state */ if (rsc->role != pcmk_role_unknown) { pcmk_resource_t *iter = rsc; while (iter) { if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) { pcmk_node_t *n = pe__copy_node(node); pcmk__rsc_trace(rsc, "%s%s%s known on %s", rsc->id, ((rsc->clone_name == NULL)? "" : " also known as "), ((rsc->clone_name == NULL)? "" : rsc->clone_name), pcmk__node_name(n)); g_hash_table_insert(iter->known_on, (gpointer) n->details->id, n); } if (pcmk_is_set(iter->flags, pcmk_rsc_unique)) { break; } iter = iter->parent; } } /* If a managed resource is believed to be running, but node is down ... */ if ((rsc->role > pcmk_role_stopped) && node->details->online == FALSE && node->details->maintenance == FALSE && pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { gboolean should_fence = FALSE; /* If this is a guest node, fence it (regardless of whether fencing is * enabled, because guest node fencing is done by recovery of the * container resource rather than by the fencer). Mark the resource * we're processing as failed. When the guest comes back up, its * operation history in the CIB will be cleared, freeing the affected * resource to run again once we are sure we know its state. */ if (pe__is_guest_node(node)) { pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); should_fence = TRUE; } else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { if (pe__is_remote_node(node) && node->details->remote_rsc && !pcmk_is_set(node->details->remote_rsc->flags, pcmk_rsc_failed)) { /* Setting unseen means that fencing of the remote node will * occur only if the connection resource is not going to start * somewhere. This allows connection resources on a failed * cluster node to move to another node without requiring the * remote nodes to be fenced as well. */ node->details->unseen = TRUE; reason = crm_strdup_printf("%s is active there (fencing will be" " revoked if remote connection can " "be re-established elsewhere)", rsc->id); } should_fence = TRUE; } if (should_fence) { if (reason == NULL) { reason = crm_strdup_printf("%s is thought to be active there", rsc->id); } pe_fence_node(rsc->cluster, node, reason, FALSE); } free(reason); } /* In order to calculate priority_fencing_delay correctly, save the failure information and pass it to native_add_running(). */ save_on_fail = on_fail; if (node->details->unclean) { /* No extra processing needed * Also allows resources to be started again after a node is shot */ on_fail = pcmk_on_fail_ignore; } switch (on_fail) { case pcmk_on_fail_ignore: /* nothing to do */ break; case pcmk_on_fail_demote: pcmk__set_rsc_flags(rsc, pcmk_rsc_failed); demote_action(rsc, node, FALSE); break; case pcmk_on_fail_fence_node: /* treat it as if it is still running * but also mark the node as unclean */ reason = crm_strdup_printf("%s failed there", rsc->id); pe_fence_node(rsc->cluster, node, reason, FALSE); free(reason); break; case pcmk_on_fail_standby_node: node->details->standby = TRUE; node->details->standby_onfail = TRUE; break; case pcmk_on_fail_block: /* is_managed == FALSE will prevent any * actions being sent for the resource */ pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed); pcmk__set_rsc_flags(rsc, pcmk_rsc_blocked); break; case pcmk_on_fail_ban: /* make sure it comes up somewhere else * or not at all */ resource_location(rsc, node, -INFINITY, "__action_migration_auto__", rsc->cluster); break; case pcmk_on_fail_stop: pe__set_next_role(rsc, pcmk_role_stopped, PCMK_META_ON_FAIL "=" PCMK_VALUE_STOP); break; case pcmk_on_fail_restart: if ((rsc->role != pcmk_role_stopped) && (rsc->role != pcmk_role_unknown)) { pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); stop_action(rsc, node, FALSE); } break; case pcmk_on_fail_restart_container: pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); if (rsc->container && pe_rsc_is_bundled(rsc)) { /* A bundle's remote connection can run on a different node than * the bundle's container. We don't necessarily know where the * container is running yet, so remember it and add a stop * action for it later. */ rsc->cluster->stop_needed = g_list_prepend(rsc->cluster->stop_needed, rsc->container); } else if (rsc->container) { stop_action(rsc->container, node, FALSE); } else if ((rsc->role != pcmk_role_stopped) && (rsc->role != pcmk_role_unknown)) { stop_action(rsc, node, FALSE); } break; case pcmk_on_fail_reset_remote: pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)) { tmpnode = NULL; if (rsc->is_remote_node) { tmpnode = pe_find_node(rsc->cluster->nodes, rsc->id); } if (tmpnode && pe__is_remote_node(tmpnode) && tmpnode->details->remote_was_fenced == 0) { /* The remote connection resource failed in a way that * should result in fencing the remote node. */ pe_fence_node(rsc->cluster, tmpnode, "remote connection is unrecoverable", FALSE); } } /* require the stop action regardless if fencing is occurring or not. */ if (rsc->role > pcmk_role_stopped) { stop_action(rsc, node, FALSE); } /* if reconnect delay is in use, prevent the connection from exiting the * "STOPPED" role until the failure is cleared by the delay timeout. */ if (rsc->remote_reconnect_ms) { pe__set_next_role(rsc, pcmk_role_stopped, "remote reset"); } break; } /* ensure a remote-node connection failure forces an unclean remote-node * to be fenced. By setting unseen = FALSE, the remote-node failure will * result in a fencing operation regardless if we're going to attempt to * reconnect to the remote-node in this transition or not. */ if (pcmk_is_set(rsc->flags, pcmk_rsc_failed) && rsc->is_remote_node) { tmpnode = pe_find_node(rsc->cluster->nodes, rsc->id); if (tmpnode && tmpnode->details->unclean) { tmpnode->details->unseen = FALSE; } } if ((rsc->role != pcmk_role_stopped) && (rsc->role != pcmk_role_unknown)) { if (pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { if (pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk__config_warn("Detected active orphan %s running on %s", rsc->id, pcmk__node_name(node)); } else { pcmk__config_warn("Resource '%s' must be stopped manually on " "%s because cluster is configured not to " "stop active orphans", rsc->id, pcmk__node_name(node)); } } native_add_running(rsc, node, rsc->cluster, (save_on_fail != pcmk_on_fail_ignore)); switch (on_fail) { case pcmk_on_fail_ignore: break; case pcmk_on_fail_demote: case pcmk_on_fail_block: pcmk__set_rsc_flags(rsc, pcmk_rsc_failed); break; default: pcmk__set_rsc_flags(rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); break; } } else if (rsc->clone_name && strchr(rsc->clone_name, ':') != NULL) { /* Only do this for older status sections that included instance numbers * Otherwise stopped instances will appear as orphans */ pcmk__rsc_trace(rsc, "Resetting clone_name %s for %s (stopped)", rsc->clone_name, rsc->id); free(rsc->clone_name); rsc->clone_name = NULL; } else { GList *possible_matches = pe__resource_actions(rsc, node, PCMK_ACTION_STOP, FALSE); GList *gIter = possible_matches; for (; gIter != NULL; gIter = gIter->next) { pcmk_action_t *stop = (pcmk_action_t *) gIter->data; pcmk__set_action_flags(stop, pcmk_action_optional); } g_list_free(possible_matches); } /* A successful stop after migrate_to on the migration source doesn't make * the partially migrated resource stopped on the migration target. */ if ((rsc->role == pcmk_role_stopped) && rsc->partial_migration_source && rsc->partial_migration_source->details == node->details && rsc->partial_migration_target && rsc->running_on) { rsc->role = pcmk_role_started; } } /* create active recurring operations as optional */ static void process_recurring(pcmk_node_t *node, pcmk_resource_t *rsc, int start_index, int stop_index, GList *sorted_op_list, pcmk_scheduler_t *scheduler) { int counter = -1; const char *task = NULL; const char *status = NULL; GList *gIter = sorted_op_list; CRM_ASSERT(rsc); pcmk__rsc_trace(rsc, "%s: Start index %d, stop index = %d", rsc->id, start_index, stop_index); for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; guint interval_ms = 0; char *key = NULL; const char *id = ID(rsc_op); counter++; if (node->details->online == FALSE) { pcmk__rsc_trace(rsc, "Skipping %s on %s: node is offline", rsc->id, pcmk__node_name(node)); break; /* Need to check if there's a monitor for role="Stopped" */ } else if (start_index < stop_index && counter <= stop_index) { pcmk__rsc_trace(rsc, "Skipping %s on %s: resource is not active", id, pcmk__node_name(node)); continue; } else if (counter < start_index) { pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d", id, pcmk__node_name(node), counter); continue; } crm_element_value_ms(rsc_op, PCMK_META_INTERVAL, &interval_ms); if (interval_ms == 0) { pcmk__rsc_trace(rsc, "Skipping %s on %s: non-recurring", id, pcmk__node_name(node)); continue; } status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS); if (pcmk__str_eq(status, "-1", pcmk__str_casei)) { pcmk__rsc_trace(rsc, "Skipping %s on %s: status", id, pcmk__node_name(node)); continue; } task = crm_element_value(rsc_op, PCMK_XA_OPERATION); /* create the action */ key = pcmk__op_key(rsc->id, task, interval_ms); pcmk__rsc_trace(rsc, "Creating %s on %s", key, pcmk__node_name(node)); custom_action(rsc, key, task, node, TRUE, scheduler); } } void calculate_active_ops(const GList *sorted_op_list, int *start_index, int *stop_index) { int counter = -1; int implied_monitor_start = -1; int implied_clone_start = -1; const char *task = NULL; const char *status = NULL; *stop_index = -1; *start_index = -1; for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) { const xmlNode *rsc_op = (const xmlNode *) iter->data; counter++; task = crm_element_value(rsc_op, PCMK_XA_OPERATION); status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS); if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_casei) && pcmk__str_eq(status, "0", pcmk__str_casei)) { *stop_index = counter; } else if (pcmk__strcase_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { *start_index = counter; } else if ((implied_monitor_start <= *stop_index) && pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { const char *rc = crm_element_value(rsc_op, PCMK__XA_RC_CODE); if (pcmk__strcase_any_of(rc, "0", "8", NULL)) { implied_monitor_start = counter; } } else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, NULL)) { implied_clone_start = counter; } } if (*start_index == -1) { if (implied_clone_start != -1) { *start_index = implied_clone_start; } else if (implied_monitor_start != -1) { *start_index = implied_monitor_start; } } } // If resource history entry has shutdown lock, remember lock node and time static void unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { time_t lock_time = 0; // When lock started (i.e. node shutdown time) if ((crm_element_value_epoch(rsc_entry, PCMK_OPT_SHUTDOWN_LOCK, &lock_time) == pcmk_ok) && (lock_time != 0)) { if ((scheduler->shutdown_lock > 0) && (get_effective_time(scheduler) > (lock_time + scheduler->shutdown_lock))) { pcmk__rsc_info(rsc, "Shutdown lock for %s on %s expired", rsc->id, pcmk__node_name(node)); pe__clear_resource_history(rsc, node); } else { /* @COMPAT I don't like breaking const signatures, but * rsc->lock_node should really be const -- we just can't change it * until the next API compatibility break. */ rsc->lock_node = (pcmk_node_t *) node; rsc->lock_time = lock_time; } } } /*! * \internal * \brief Unpack one \c PCMK__XE_LRM_RESOURCE entry from a node's CIB status * * \param[in,out] node Node whose status is being unpacked * \param[in] rsc_entry \c PCMK__XE_LRM_RESOURCE XML being unpacked * \param[in,out] scheduler Scheduler data * * \return Resource corresponding to the entry, or NULL if no operation history */ static pcmk_resource_t * unpack_lrm_resource(pcmk_node_t *node, const xmlNode *lrm_resource, pcmk_scheduler_t *scheduler) { GList *gIter = NULL; int stop_index = -1; int start_index = -1; enum rsc_role_e req_role = pcmk_role_unknown; const char *rsc_id = ID(lrm_resource); pcmk_resource_t *rsc = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; xmlNode *rsc_op = NULL; xmlNode *last_failure = NULL; enum action_fail_response on_fail = pcmk_on_fail_ignore; enum rsc_role_e saved_role = pcmk_role_unknown; if (rsc_id == NULL) { pcmk__config_err("Ignoring invalid " PCMK__XE_LRM_RESOURCE " entry: No " PCMK_XA_ID); crm_log_xml_info(lrm_resource, "missing-id"); return NULL; } crm_trace("Unpacking " PCMK__XE_LRM_RESOURCE " for %s on %s", rsc_id, pcmk__node_name(node)); /* Build a list of individual PCMK__XE_LRM_RSC_OP entries, so we can sort * them */ for (rsc_op = first_named_child(lrm_resource, PCMK__XE_LRM_RSC_OP); rsc_op != NULL; rsc_op = crm_next_same_xml(rsc_op)) { op_list = g_list_prepend(op_list, rsc_op); } if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } } /* find the resource */ rsc = unpack_find_resource(scheduler, node, rsc_id); if (rsc == NULL) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } else { rsc = process_orphan_resource(lrm_resource, node, scheduler); } } CRM_ASSERT(rsc != NULL); // Check whether the resource is "shutdown-locked" to this node if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { unpack_shutdown_lock(lrm_resource, rsc, node, scheduler); } /* process operations */ saved_role = rsc->role; rsc->role = pcmk_role_unknown; sorted_op_list = g_list_sort(op_list, sort_op_by_callid); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail); } /* create active recurring operations as optional */ calculate_active_ops(sorted_op_list, &start_index, &stop_index); process_recurring(node, rsc, start_index, stop_index, sorted_op_list, scheduler); /* no need to free the contents */ g_list_free(sorted_op_list); process_rsc_state(rsc, node, on_fail); if (get_target_role(rsc, &req_role)) { if ((rsc->next_role == pcmk_role_unknown) || (req_role < rsc->next_role)) { pe__set_next_role(rsc, req_role, PCMK_META_TARGET_ROLE); } else if (req_role > rsc->next_role) { pcmk__rsc_info(rsc, "%s: Not overwriting calculated next role %s" " with requested next role %s", rsc->id, pcmk_role_text(rsc->next_role), pcmk_role_text(req_role)); } } if (saved_role > rsc->role) { rsc->role = saved_role; } return rsc; } static void handle_orphaned_container_fillers(const xmlNode *lrm_rsc_list, pcmk_scheduler_t *scheduler) { for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) { pcmk_resource_t *rsc; pcmk_resource_t *container; const char *rsc_id; const char *container_id; if (!pcmk__str_eq((const char *) rsc_entry->name, PCMK__XE_LRM_RESOURCE, pcmk__str_casei)) { continue; } container_id = crm_element_value(rsc_entry, PCMK__META_CONTAINER); rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID); if (container_id == NULL || rsc_id == NULL) { continue; } container = pe_find_resource(scheduler->resources, container_id); if (container == NULL) { continue; } rsc = pe_find_resource(scheduler->resources, rsc_id); if ((rsc == NULL) || (rsc->container != NULL) || !pcmk_is_set(rsc->flags, pcmk_rsc_removed_filler)) { continue; } pcmk__rsc_trace(rsc, "Mapped container of orphaned resource %s to %s", rsc->id, container_id); rsc->container = container; container->fillers = g_list_append(container->fillers, rsc); } } /*! * \internal * \brief Unpack one node's lrm status section * * \param[in,out] node Node whose status is being unpacked * \param[in] xml CIB node state XML * \param[in,out] scheduler Scheduler data */ static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml, pcmk_scheduler_t *scheduler) { bool found_orphaned_container_filler = false; // Drill down to PCMK__XE_LRM_RESOURCES section xml = find_xml_node(xml, PCMK__XE_LRM, FALSE); if (xml == NULL) { return; } xml = find_xml_node(xml, PCMK__XE_LRM_RESOURCES, FALSE); if (xml == NULL) { return; } // Unpack each PCMK__XE_LRM_RESOURCE entry for (const xmlNode *rsc_entry = first_named_child(xml, PCMK__XE_LRM_RESOURCE); rsc_entry != NULL; rsc_entry = crm_next_same_xml(rsc_entry)) { pcmk_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, scheduler); if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk_rsc_removed_filler)) { found_orphaned_container_filler = true; } } /* Now that all resource state has been unpacked for this node, map any * orphaned container fillers to their container resource. */ if (found_orphaned_container_filler) { handle_orphaned_container_fillers(xml, scheduler); } } static void set_active(pcmk_resource_t *rsc) { const pcmk_resource_t *top = pe__const_top_resource(rsc, false); if (top && pcmk_is_set(top->flags, pcmk_rsc_promotable)) { rsc->role = pcmk_role_unpromoted; } else { rsc->role = pcmk_role_started; } } static void set_node_score(gpointer key, gpointer value, gpointer user_data) { pcmk_node_t *node = value; int *score = user_data; node->weight = *score; } #define XPATH_NODE_STATE "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE #define SUB_XPATH_LRM_RESOURCE "/" PCMK__XE_LRM \ "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE #define SUB_XPATH_LRM_RSC_OP "/" PCMK__XE_LRM_RSC_OP static xmlNode * find_lrm_op(const char *resource, const char *op, const char *node, const char *source, int target_rc, pcmk_scheduler_t *scheduler) { GString *xpath = NULL; xmlNode *xml = NULL; CRM_CHECK((resource != NULL) && (op != NULL) && (node != NULL), return NULL); xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node, "']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", resource, "']" SUB_XPATH_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='", op, "'", NULL); /* Need to check against transition_magic too? */ if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_TO) == 0)) { pcmk__g_strcat(xpath, " and @" PCMK__META_MIGRATE_TARGET "='", source, "']", NULL); } else if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) { pcmk__g_strcat(xpath, " and @" PCMK__META_MIGRATE_SOURCE "='", source, "']", NULL); } else { g_string_append_c(xpath, ']'); } xml = get_xpath_object((const char *) xpath->str, scheduler->input, LOG_DEBUG); g_string_free(xpath, TRUE); if (xml && target_rc >= 0) { int rc = PCMK_OCF_UNKNOWN_ERROR; int status = PCMK_EXEC_ERROR; crm_element_value_int(xml, PCMK__XA_RC_CODE, &rc); crm_element_value_int(xml, PCMK__XA_OP_STATUS, &status); if ((rc != target_rc) || (status != PCMK_EXEC_DONE)) { return NULL; } } return xml; } static xmlNode * find_lrm_resource(const char *rsc_id, const char *node_name, pcmk_scheduler_t *scheduler) { GString *xpath = NULL; xmlNode *xml = NULL; CRM_CHECK((rsc_id != NULL) && (node_name != NULL), return NULL); xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node_name, "']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", rsc_id, "']", NULL); xml = get_xpath_object((const char *) xpath->str, scheduler->input, LOG_DEBUG); g_string_free(xpath, TRUE); return xml; } /*! * \internal * \brief Check whether a resource has no completed action history on a node * * \param[in,out] rsc Resource to check * \param[in] node_name Node to check * * \return true if \p rsc_id is unknown on \p node_name, otherwise false */ static bool unknown_on_node(pcmk_resource_t *rsc, const char *node_name) { bool result = false; xmlXPathObjectPtr search; char *xpath = NULL; xpath = crm_strdup_printf(XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" SUB_XPATH_LRM_RSC_OP "[@" PCMK__XA_RC_CODE "!='%d']", node_name, rsc->id, PCMK_OCF_UNKNOWN); search = xpath_search(rsc->cluster->input, xpath); result = (numXpathResults(search) == 0); freeXpathObject(search); free(xpath); return result; } /*! * \brief Check whether a probe/monitor indicating the resource was not running * on a node happened after some event * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] xml_op Event that monitor is being compared to * \param[in] same_node Whether the operations are on the same node * \param[in,out] scheduler Scheduler data * * \return true if such a monitor happened after event, false otherwise */ static bool monitor_not_running_after(const char *rsc_id, const char *node_name, const xmlNode *xml_op, bool same_node, pcmk_scheduler_t *scheduler) { /* Any probe/monitor operation on the node indicating it was not running * there */ xmlNode *monitor = find_lrm_op(rsc_id, PCMK_ACTION_MONITOR, node_name, NULL, PCMK_OCF_NOT_RUNNING, scheduler); return (monitor && pe__is_newer_op(monitor, xml_op, same_node) > 0); } /*! * \brief Check whether any non-monitor operation on a node happened after some * event * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] xml_op Event that non-monitor is being compared to * \param[in] same_node Whether the operations are on the same node * \param[in,out] scheduler Scheduler data * * \return true if such a operation happened after event, false otherwise */ static bool non_monitor_after(const char *rsc_id, const char *node_name, const xmlNode *xml_op, bool same_node, pcmk_scheduler_t *scheduler) { xmlNode *lrm_resource = NULL; lrm_resource = find_lrm_resource(rsc_id, node_name, scheduler); if (lrm_resource == NULL) { return false; } for (xmlNode *op = first_named_child(lrm_resource, PCMK__XE_LRM_RSC_OP); op != NULL; op = crm_next_same_xml(op)) { const char * task = NULL; if (op == xml_op) { continue; } task = crm_element_value(op, PCMK_XA_OPERATION); if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL) && pe__is_newer_op(op, xml_op, same_node) > 0) { return true; } } return false; } /*! * \brief Check whether the resource has newer state on a node after a migration * attempt * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] migrate_to Any migrate_to event that is being compared to * \param[in] migrate_from Any migrate_from event that is being compared to * \param[in,out] scheduler Scheduler data * * \return true if such a operation happened after event, false otherwise */ static bool newer_state_after_migrate(const char *rsc_id, const char *node_name, const xmlNode *migrate_to, const xmlNode *migrate_from, pcmk_scheduler_t *scheduler) { const xmlNode *xml_op = migrate_to; const char *source = NULL; const char *target = NULL; bool same_node = false; if (migrate_from) { xml_op = migrate_from; } source = crm_element_value(xml_op, PCMK__META_MIGRATE_SOURCE); target = crm_element_value(xml_op, PCMK__META_MIGRATE_TARGET); /* It's preferred to compare to the migrate event on the same node if * existing, since call ids are more reliable. */ if (pcmk__str_eq(node_name, target, pcmk__str_casei)) { if (migrate_from) { xml_op = migrate_from; same_node = true; } else { xml_op = migrate_to; } } else if (pcmk__str_eq(node_name, source, pcmk__str_casei)) { if (migrate_to) { xml_op = migrate_to; same_node = true; } else { xml_op = migrate_from; } } /* If there's any newer non-monitor operation on the node, or any newer * probe/monitor operation on the node indicating it was not running there, * the migration events potentially no longer matter for the node. */ return non_monitor_after(rsc_id, node_name, xml_op, same_node, scheduler) || monitor_not_running_after(rsc_id, node_name, xml_op, same_node, scheduler); } /*! * \internal * \brief Parse migration source and target node names from history entry * * \param[in] entry Resource history entry for a migration action * \param[in] source_node If not NULL, source must match this node * \param[in] target_node If not NULL, target must match this node * \param[out] source_name Where to store migration source node name * \param[out] target_name Where to store migration target node name * * \return Standard Pacemaker return code */ static int get_migration_node_names(const xmlNode *entry, const pcmk_node_t *source_node, const pcmk_node_t *target_node, const char **source_name, const char **target_name) { *source_name = crm_element_value(entry, PCMK__META_MIGRATE_SOURCE); *target_name = crm_element_value(entry, PCMK__META_MIGRATE_TARGET); if ((*source_name == NULL) || (*target_name == NULL)) { pcmk__config_err("Ignoring resource history entry %s without " PCMK__META_MIGRATE_SOURCE " and " PCMK__META_MIGRATE_TARGET, ID(entry)); return pcmk_rc_unpack_error; } if ((source_node != NULL) && !pcmk__str_eq(*source_name, source_node->details->uname, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_err("Ignoring resource history entry %s because " PCMK__META_MIGRATE_SOURCE "='%s' does not match %s", ID(entry), *source_name, pcmk__node_name(source_node)); return pcmk_rc_unpack_error; } if ((target_node != NULL) && !pcmk__str_eq(*target_name, target_node->details->uname, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_err("Ignoring resource history entry %s because " PCMK__META_MIGRATE_TARGET "='%s' does not match %s", ID(entry), *target_name, pcmk__node_name(target_node)); return pcmk_rc_unpack_error; } return pcmk_rc_ok; } /* * \internal * \brief Add a migration source to a resource's list of dangling migrations * * If the migrate_to and migrate_from actions in a live migration both * succeeded, but there is no stop on the source, the migration is considered * "dangling." Add the source to the resource's dangling migration list, which * will be used to schedule a stop on the source without affecting the target. * * \param[in,out] rsc Resource involved in migration * \param[in] node Migration source */ static void add_dangling_migration(pcmk_resource_t *rsc, const pcmk_node_t *node) { pcmk__rsc_trace(rsc, "Dangling migration of %s requires stop on %s", rsc->id, pcmk__node_name(node)); rsc->role = pcmk_role_stopped; rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, (gpointer) node); } /*! * \internal * \brief Update resource role etc. after a successful migrate_to action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_to_success(struct action_history *history) { /* A complete migration sequence is: * 1. migrate_to on source node (which succeeded if we get to this function) * 2. migrate_from on target node * 3. stop on source node * * If no migrate_from has happened, the migration is considered to be * "partial". If the migrate_from succeeded but no stop has happened, the * migration is considered to be "dangling". * * If a successful migrate_to and stop have happened on the source node, we * still need to check for a partial migration, due to scenarios (easier to * produce with batch-limit=1) like: * * - A resource is migrating from node1 to node2, and a migrate_to is * initiated for it on node1. * * - node2 goes into standby mode while the migrate_to is pending, which * aborts the transition. * * - Upon completion of the migrate_to, a new transition schedules a stop * on both nodes and a start on node1. * * - If the new transition is aborted for any reason while the resource is * stopping on node1, the transition after that stop completes will see * the migrate_to and stop on the source, but it's still a partial * migration, and the resource must be stopped on node2 because it is * potentially active there due to the migrate_to. * * We also need to take into account that either node's history may be * cleared at any point in the migration process. */ int from_rc = PCMK_OCF_OK; int from_status = PCMK_EXEC_PENDING; pcmk_node_t *target_node = NULL; xmlNode *migrate_from = NULL; const char *source = NULL; const char *target = NULL; bool source_newer_op = false; bool target_newer_state = false; bool active_on_target = false; // Get source and target node names from XML if (get_migration_node_names(history->xml, history->node, NULL, &source, &target) != pcmk_rc_ok) { return; } // Check for newer state on the source source_newer_op = non_monitor_after(history->rsc->id, source, history->xml, true, history->rsc->cluster); // Check for a migrate_from action from this source on the target migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM, target, source, -1, history->rsc->cluster); if (migrate_from != NULL) { if (source_newer_op) { /* There's a newer non-monitor operation on the source and a * migrate_from on the target, so this migrate_to is irrelevant to * the resource's state. */ return; } crm_element_value_int(migrate_from, PCMK__XA_RC_CODE, &from_rc); crm_element_value_int(migrate_from, PCMK__XA_OP_STATUS, &from_status); } /* If the resource has newer state on both the source and target after the * migration events, this migrate_to is irrelevant to the resource's state. */ target_newer_state = newer_state_after_migrate(history->rsc->id, target, history->xml, migrate_from, history->rsc->cluster); if (source_newer_op && target_newer_state) { return; } /* Check for dangling migration (migrate_from succeeded but stop not done). * We know there's no stop because we already returned if the target has a * migrate_from and the source has any newer non-monitor operation. */ if ((from_rc == PCMK_OCF_OK) && (from_status == PCMK_EXEC_DONE)) { add_dangling_migration(history->rsc, history->node); return; } /* Without newer state, this migrate_to implies the resource is active. * (Clones are not allowed to migrate, so role can't be promoted.) */ history->rsc->role = pcmk_role_started; target_node = pe_find_node(history->rsc->cluster->nodes, target); active_on_target = !target_newer_state && (target_node != NULL) && target_node->details->online; if (from_status != PCMK_EXEC_PENDING) { // migrate_from failed on target if (active_on_target) { native_add_running(history->rsc, target_node, history->rsc->cluster, TRUE); } else { // Mark resource as failed, require recovery, and prevent migration pcmk__set_rsc_flags(history->rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); pcmk__clear_rsc_flags(history->rsc, pcmk_rsc_migratable); } return; } // The migrate_from is pending, complete but erased, or to be scheduled /* If there is no history at all for the resource on an online target, then * it was likely cleaned. Just return, and we'll schedule a probe. Once we * have the probe result, it will be reflected in target_newer_state. */ if ((target_node != NULL) && target_node->details->online && unknown_on_node(history->rsc, target)) { return; } if (active_on_target) { pcmk_node_t *source_node = pe_find_node(history->rsc->cluster->nodes, source); native_add_running(history->rsc, target_node, history->rsc->cluster, FALSE); if ((source_node != NULL) && source_node->details->online) { /* This is a partial migration: the migrate_to completed * successfully on the source, but the migrate_from has not * completed. Remember the source and target; if the newly * chosen target remains the same when we schedule actions * later, we may continue with the migration. */ history->rsc->partial_migration_target = target_node; history->rsc->partial_migration_source = source_node; } } else if (!source_newer_op) { // Mark resource as failed, require recovery, and prevent migration pcmk__set_rsc_flags(history->rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); pcmk__clear_rsc_flags(history->rsc, pcmk_rsc_migratable); } } /*! * \internal * \brief Update resource role etc. after a failed migrate_to action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_to_failure(struct action_history *history) { xmlNode *target_migrate_from = NULL; const char *source = NULL; const char *target = NULL; // Get source and target node names from XML if (get_migration_node_names(history->xml, history->node, NULL, &source, &target) != pcmk_rc_ok) { return; } /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ history->rsc->role = pcmk_role_started; // Check for migrate_from on the target target_migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM, target, source, PCMK_OCF_OK, history->rsc->cluster); if (/* If the resource state is unknown on the target, it will likely be * probed there. * Don't just consider it running there. We will get back here anyway in * case the probe detects it's running there. */ !unknown_on_node(history->rsc, target) /* If the resource has newer state on the target after the migration * events, this migrate_to no longer matters for the target. */ && !newer_state_after_migrate(history->rsc->id, target, history->xml, target_migrate_from, history->rsc->cluster)) { /* The resource has no newer state on the target, so assume it's still * active there. * (if it is up). */ pcmk_node_t *target_node = pe_find_node(history->rsc->cluster->nodes, target); if (target_node && target_node->details->online) { native_add_running(history->rsc, target_node, history->rsc->cluster, FALSE); } } else if (!non_monitor_after(history->rsc->id, source, history->xml, true, history->rsc->cluster)) { /* We know the resource has newer state on the target, but this * migrate_to still matters for the source as long as there's no newer * non-monitor operation there. */ // Mark node as having dangling migration so we can force a stop later history->rsc->dangling_migrations = g_list_prepend(history->rsc->dangling_migrations, (gpointer) history->node); } } /*! * \internal * \brief Update resource role etc. after a failed migrate_from action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_from_failure(struct action_history *history) { xmlNode *source_migrate_to = NULL; const char *source = NULL; const char *target = NULL; // Get source and target node names from XML if (get_migration_node_names(history->xml, NULL, history->node, &source, &target) != pcmk_rc_ok) { return; } /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ history->rsc->role = pcmk_role_started; // Check for a migrate_to on the source source_migrate_to = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_TO, source, target, PCMK_OCF_OK, history->rsc->cluster); if (/* If the resource state is unknown on the source, it will likely be * probed there. * Don't just consider it running there. We will get back here anyway in * case the probe detects it's running there. */ !unknown_on_node(history->rsc, source) /* If the resource has newer state on the source after the migration * events, this migrate_from no longer matters for the source. */ && !newer_state_after_migrate(history->rsc->id, source, source_migrate_to, history->xml, history->rsc->cluster)) { /* The resource has no newer state on the source, so assume it's still * active there (if it is up). */ pcmk_node_t *source_node = pe_find_node(history->rsc->cluster->nodes, source); if (source_node && source_node->details->online) { native_add_running(history->rsc, source_node, history->rsc->cluster, TRUE); } } } /*! * \internal * \brief Add an action to cluster's list of failed actions * * \param[in,out] history Parsed action result history */ static void record_failed_op(struct action_history *history) { if (!(history->node->details->online)) { return; } for (const xmlNode *xIter = history->rsc->cluster->failed->children; xIter != NULL; xIter = xIter->next) { const char *key = pcmk__xe_history_key(xIter); const char *uname = crm_element_value(xIter, PCMK_XA_UNAME); if (pcmk__str_eq(history->key, key, pcmk__str_none) && pcmk__str_eq(uname, history->node->details->uname, pcmk__str_casei)) { crm_trace("Skipping duplicate entry %s on %s", history->key, pcmk__node_name(history->node)); return; } } crm_trace("Adding entry for %s on %s to failed action list", history->key, pcmk__node_name(history->node)); crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->details->uname); crm_xml_add(history->xml, PCMK__XA_RSC_ID, history->rsc->id); add_node_copy(history->rsc->cluster->failed, history->xml); } static char * last_change_str(const xmlNode *xml_op) { time_t when; char *result = NULL; if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &when) == pcmk_ok) { char *when_s = pcmk__epoch2str(&when, 0); const char *p = strchr(when_s, ' '); // Skip day of week to make message shorter if ((p != NULL) && (*(++p) != '\0')) { result = strdup(p); CRM_ASSERT(result != NULL); } free(when_s); } if (result == NULL) { result = strdup("unknown time"); CRM_ASSERT(result != NULL); } return result; } /*! * \internal * \brief Compare two on-fail values * * \param[in] first One on-fail value to compare * \param[in] second The other on-fail value to compare * * \return A negative number if second is more severe than first, zero if they * are equal, or a positive number if first is more severe than second. * \note This is only needed until the action_fail_response values can be * renumbered at the next API compatibility break. */ static int cmp_on_fail(enum action_fail_response first, enum action_fail_response second) { switch (first) { case pcmk_on_fail_demote: switch (second) { case pcmk_on_fail_ignore: return 1; case pcmk_on_fail_demote: return 0; default: return -1; } break; case pcmk_on_fail_reset_remote: switch (second) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: return 1; case pcmk_on_fail_reset_remote: return 0; default: return -1; } break; case pcmk_on_fail_restart_container: switch (second) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: case pcmk_on_fail_reset_remote: return 1; case pcmk_on_fail_restart_container: return 0; default: return -1; } break; default: break; } switch (second) { case pcmk_on_fail_demote: return (first == pcmk_on_fail_ignore)? -1 : 1; case pcmk_on_fail_reset_remote: switch (first) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: return -1; default: return 1; } break; case pcmk_on_fail_restart_container: switch (first) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: case pcmk_on_fail_reset_remote: return -1; default: return 1; } break; default: break; } return first - second; } /*! * \internal * \brief Ban a resource (or its clone if an anonymous instance) from all nodes * * \param[in,out] rsc Resource to ban */ static void ban_from_all_nodes(pcmk_resource_t *rsc) { int score = -INFINITY; pcmk_resource_t *fail_rsc = rsc; if (fail_rsc->parent != NULL) { pcmk_resource_t *parent = uber_parent(fail_rsc); if (pe_rsc_is_anon_clone(parent)) { /* For anonymous clones, if an operation with * PCMK_META_ON_FAIL=PCMK_VALUE_STOP fails for any instance, the * entire clone must stop. */ fail_rsc = parent; } } // Ban the resource from all nodes crm_notice("%s will not be started under current conditions", fail_rsc->id); if (fail_rsc->allowed_nodes != NULL) { g_hash_table_destroy(fail_rsc->allowed_nodes); } fail_rsc->allowed_nodes = pe__node_list2table(rsc->cluster->nodes); g_hash_table_foreach(fail_rsc->allowed_nodes, set_node_score, &score); } /*! * \internal * \brief Get configured failure handling and role after failure for an action * * \param[in,out] history Unpacked action history entry * \param[out] on_fail Where to set configured failure handling * \param[out] fail_role Where to set to role after failure */ static void unpack_failure_handling(struct action_history *history, enum action_fail_response *on_fail, enum rsc_role_e *fail_role) { xmlNode *config = pcmk__find_action_config(history->rsc, history->task, history->interval_ms, true); GHashTable *meta = pcmk__unpack_action_meta(history->rsc, history->node, history->task, history->interval_ms, config); const char *on_fail_str = g_hash_table_lookup(meta, PCMK_META_ON_FAIL); *on_fail = pcmk__parse_on_fail(history->rsc, history->task, history->interval_ms, on_fail_str); *fail_role = pcmk__role_after_failure(history->rsc, history->task, *on_fail, meta); g_hash_table_destroy(meta); } /*! * \internal * \brief Update resource role, failure handling, etc., after a failed action * * \param[in,out] history Parsed action result history * \param[in] config_on_fail Action failure handling from configuration * \param[in] fail_role Resource's role after failure of this action * \param[out] last_failure This will be set to the history XML * \param[in,out] on_fail Actual handling of action result */ static void unpack_rsc_op_failure(struct action_history *history, enum action_fail_response config_on_fail, enum rsc_role_e fail_role, xmlNode **last_failure, enum action_fail_response *on_fail) { bool is_probe = false; char *last_change_s = NULL; *last_failure = history->xml; is_probe = pcmk_xe_is_probe(history->xml); last_change_s = last_change_str(history->xml); if (!pcmk_is_set(history->rsc->cluster->flags, pcmk_sched_symmetric_cluster) && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) { crm_trace("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " CRM_XS " exit-status=%d id=%s", services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), (is_probe? "probe" : history->task), history->rsc->id, pcmk__node_name(history->node), last_change_s, history->exit_status, history->id); } else { pcmk__sched_warn("Unexpected result (%s%s%s) was recorded for %s of " "%s on %s at %s " CRM_XS " exit-status=%d id=%s", services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), (is_probe? "probe" : history->task), history->rsc->id, pcmk__node_name(history->node), last_change_s, history->exit_status, history->id); if (is_probe && (history->exit_status != PCMK_OCF_OK) && (history->exit_status != PCMK_OCF_NOT_RUNNING) && (history->exit_status != PCMK_OCF_RUNNING_PROMOTED)) { /* A failed (not just unexpected) probe result could mean the user * didn't know resources will be probed even where they can't run. */ crm_notice("If it is not possible for %s to run on %s, see " "the " PCMK_XA_RESOURCE_DISCOVERY " option for location " "constraints", history->rsc->id, pcmk__node_name(history->node)); } record_failed_op(history); } free(last_change_s); if (cmp_on_fail(*on_fail, config_on_fail) < 0) { pcmk__rsc_trace(history->rsc, "on-fail %s -> %s for %s", fail2text(*on_fail), fail2text(config_on_fail), history->key); *on_fail = config_on_fail; } if (strcmp(history->task, PCMK_ACTION_STOP) == 0) { resource_location(history->rsc, history->node, -INFINITY, "__stop_fail__", history->rsc->cluster); } else if (strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) { unpack_migrate_to_failure(history); } else if (strcmp(history->task, PCMK_ACTION_MIGRATE_FROM) == 0) { unpack_migrate_from_failure(history); } else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) { history->rsc->role = pcmk_role_promoted; } else if (strcmp(history->task, PCMK_ACTION_DEMOTE) == 0) { if (config_on_fail == pcmk_on_fail_block) { history->rsc->role = pcmk_role_promoted; pe__set_next_role(history->rsc, pcmk_role_stopped, "demote with on-fail=block"); } else if (history->exit_status == PCMK_OCF_NOT_RUNNING) { history->rsc->role = pcmk_role_stopped; } else { /* Staying in the promoted role would put the scheduler and * controller into a loop. Setting the role to unpromoted is not * dangerous because the resource will be stopped as part of * recovery, and any promotion will be ordered after that stop. */ history->rsc->role = pcmk_role_unpromoted; } } if (is_probe && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) { /* leave stopped */ pcmk__rsc_trace(history->rsc, "Leaving %s stopped", history->rsc->id); history->rsc->role = pcmk_role_stopped; } else if (history->rsc->role < pcmk_role_started) { pcmk__rsc_trace(history->rsc, "Setting %s active", history->rsc->id); set_active(history->rsc); } pcmk__rsc_trace(history->rsc, "Resource %s: role=%s unclean=%s on_fail=%s fail_role=%s", history->rsc->id, pcmk_role_text(history->rsc->role), pcmk__btoa(history->node->details->unclean), fail2text(config_on_fail), pcmk_role_text(fail_role)); if ((fail_role != pcmk_role_started) && (history->rsc->next_role < fail_role)) { pe__set_next_role(history->rsc, fail_role, "failure"); } if (fail_role == pcmk_role_stopped) { ban_from_all_nodes(history->rsc); } } /*! * \internal * \brief Block a resource with a failed action if it cannot be recovered * * If resource action is a failed stop and fencing is not possible, mark the * resource as unmanaged and blocked, since recovery cannot be done. * * \param[in,out] history Parsed action history entry */ static void block_if_unrecoverable(struct action_history *history) { char *last_change_s = NULL; if (strcmp(history->task, PCMK_ACTION_STOP) != 0) { return; // All actions besides stop are always recoverable } if (pe_can_fence(history->node->details->data_set, history->node)) { return; // Failed stops are recoverable via fencing } last_change_s = last_change_str(history->xml); pcmk__sched_err("No further recovery can be attempted for %s " "because %s on %s failed (%s%s%s) at %s " CRM_XS " rc=%d id=%s", history->rsc->id, history->task, pcmk__node_name(history->node), services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), last_change_s, history->exit_status, history->id); free(last_change_s); pcmk__clear_rsc_flags(history->rsc, pcmk_rsc_managed); pcmk__set_rsc_flags(history->rsc, pcmk_rsc_blocked); } /*! * \internal * \brief Update action history's execution status and why * * \param[in,out] history Parsed action history entry * \param[out] why Where to store reason for update * \param[in] value New value * \param[in] reason Description of why value was changed */ static inline void remap_because(struct action_history *history, const char **why, int value, const char *reason) { if (history->execution_status != value) { history->execution_status = value; *why = reason; } } /*! * \internal * \brief Remap informational monitor results and operation status * * For the monitor results, certain OCF codes are for providing extended information * to the user about services that aren't yet failed but not entirely healthy either. * These must be treated as the "normal" result by Pacemaker. * * For operation status, the action result can be used to determine an appropriate * status for the purposes of responding to the action. The status provided by the * executor is not directly usable since the executor does not know what was expected. * * \param[in,out] history Parsed action history entry * \param[in,out] on_fail What should be done about the result * \param[in] expired Whether result is expired * * \note If the result is remapped and the node is not shutting down or failed, * the operation will be recorded in the scheduler data's list of failed * operations to highlight it for the user. * * \note This may update the resource's current and next role. */ static void remap_operation(struct action_history *history, enum action_fail_response *on_fail, bool expired) { bool is_probe = false; int orig_exit_status = history->exit_status; int orig_exec_status = history->execution_status; const char *why = NULL; const char *task = history->task; // Remap degraded results to their successful counterparts history->exit_status = pcmk__effective_rc(history->exit_status); if (history->exit_status != orig_exit_status) { why = "degraded result"; if (!expired && (!history->node->details->shutdown || history->node->details->online)) { record_failed_op(history); } } if (!pe_rsc_is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml) && ((history->execution_status != PCMK_EXEC_DONE) || (history->exit_status != PCMK_OCF_NOT_RUNNING))) { history->execution_status = PCMK_EXEC_DONE; history->exit_status = PCMK_OCF_NOT_RUNNING; why = "equivalent probe result"; } /* If the executor reported an execution status of anything but done or * error, consider that final. But for done or error, we know better whether * it should be treated as a failure or not, because we know the expected * result. */ switch (history->execution_status) { case PCMK_EXEC_DONE: case PCMK_EXEC_ERROR: break; // These should be treated as node-fatal case PCMK_EXEC_NO_FENCE_DEVICE: case PCMK_EXEC_NO_SECRETS: remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "node-fatal error"); goto remap_done; default: goto remap_done; } is_probe = pcmk_xe_is_probe(history->xml); if (is_probe) { task = "probe"; } if (history->expected_exit_status < 0) { /* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with * Heartbeat 2.0.7 or earlier as the cluster layer, did not include the * expected exit status in the transition key, which (along with the * similar case of a corrupted transition key in the CIB) will be * reported to this function as -1. Pacemaker 2.0+ does not support * rolling upgrades from those versions or processing of saved CIB files * from those versions, so we do not need to care much about this case. */ remap_because(history, &why, PCMK_EXEC_ERROR, "obsolete history format"); pcmk__config_warn("Expected result not found for %s on %s " "(corrupt or obsolete CIB?)", history->key, pcmk__node_name(history->node)); } else if (history->exit_status == history->expected_exit_status) { remap_because(history, &why, PCMK_EXEC_DONE, "expected result"); } else { remap_because(history, &why, PCMK_EXEC_ERROR, "unexpected result"); pcmk__rsc_debug(history->rsc, "%s on %s: expected %d (%s), got %d (%s%s%s)", history->key, pcmk__node_name(history->node), history->expected_exit_status, services_ocf_exitcode_str(history->expected_exit_status), history->exit_status, services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, "")); } switch (history->exit_status) { case PCMK_OCF_OK: if (is_probe && (history->expected_exit_status == PCMK_OCF_NOT_RUNNING)) { char *last_change_s = last_change_str(history->xml); remap_because(history, &why, PCMK_EXEC_DONE, "probe"); pcmk__rsc_info(history->rsc, "Probe found %s active on %s at %s", history->rsc->id, pcmk__node_name(history->node), last_change_s); free(last_change_s); } break; case PCMK_OCF_NOT_RUNNING: if (is_probe || (history->expected_exit_status == history->exit_status) || !pcmk_is_set(history->rsc->flags, pcmk_rsc_managed)) { /* For probes, recurring monitors for the Stopped role, and * unmanaged resources, "not running" is not considered a * failure. */ remap_because(history, &why, PCMK_EXEC_DONE, "exit status"); history->rsc->role = pcmk_role_stopped; *on_fail = pcmk_on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "not running"); } break; case PCMK_OCF_RUNNING_PROMOTED: if (is_probe && (history->exit_status != history->expected_exit_status)) { char *last_change_s = last_change_str(history->xml); remap_because(history, &why, PCMK_EXEC_DONE, "probe"); pcmk__rsc_info(history->rsc, "Probe found %s active and promoted on %s at %s", history->rsc->id, pcmk__node_name(history->node), last_change_s); free(last_change_s); } if (!expired || (history->exit_status == history->expected_exit_status)) { history->rsc->role = pcmk_role_promoted; } break; case PCMK_OCF_FAILED_PROMOTED: if (!expired) { history->rsc->role = pcmk_role_promoted; } remap_because(history, &why, PCMK_EXEC_ERROR, "exit status"); break; case PCMK_OCF_NOT_CONFIGURED: remap_because(history, &why, PCMK_EXEC_ERROR_FATAL, "exit status"); break; case PCMK_OCF_UNIMPLEMENT_FEATURE: { guint interval_ms = 0; crm_element_value_ms(history->xml, PCMK_META_INTERVAL, &interval_ms); if (interval_ms == 0) { if (!expired) { block_if_unrecoverable(history); } remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status"); } else { remap_because(history, &why, PCMK_EXEC_NOT_SUPPORTED, "exit status"); } } break; case PCMK_OCF_NOT_INSTALLED: case PCMK_OCF_INVALID_PARAM: case PCMK_OCF_INSUFFICIENT_PRIV: if (!expired) { block_if_unrecoverable(history); } remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status"); break; default: if (history->execution_status == PCMK_EXEC_DONE) { char *last_change_s = last_change_str(history->xml); crm_info("Treating unknown exit status %d from %s of %s " "on %s at %s as failure", history->exit_status, task, history->rsc->id, pcmk__node_name(history->node), last_change_s); remap_because(history, &why, PCMK_EXEC_ERROR, "unknown exit status"); free(last_change_s); } break; } remap_done: if (why != NULL) { pcmk__rsc_trace(history->rsc, "Remapped %s result from [%s: %s] to [%s: %s] " "because of %s", history->key, pcmk_exec_status_str(orig_exec_status), crm_exit_str(orig_exit_status), pcmk_exec_status_str(history->execution_status), crm_exit_str(history->exit_status), why); } } // return TRUE if start or monitor last failure but parameters changed static bool should_clear_for_param_change(const xmlNode *xml_op, const char *task, pcmk_resource_t *rsc, pcmk_node_t *node) { if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) { if (pe__bundle_needs_remote_name(rsc)) { /* We haven't allocated resources yet, so we can't reliably * substitute addr parameters for the REMOTE_CONTAINER_HACK. * When that's needed, defer the check until later. */ pe__add_param_check(xml_op, rsc, node, pcmk__check_last_failure, rsc->cluster); } else { pcmk__op_digest_t *digest_data = NULL; digest_data = rsc_action_digest_cmp(rsc, xml_op, node, rsc->cluster); switch (digest_data->rc) { case pcmk__digest_unknown: crm_trace("Resource %s history entry %s on %s" " has no digest to compare", rsc->id, pcmk__xe_history_key(xml_op), node->details->id); break; case pcmk__digest_match: break; default: return TRUE; } } } return FALSE; } // Order action after fencing of remote node, given connection rsc static void order_after_remote_fencing(pcmk_action_t *action, pcmk_resource_t *remote_conn, pcmk_scheduler_t *scheduler) { pcmk_node_t *remote_node = pe_find_node(scheduler->nodes, remote_conn->id); if (remote_node) { pcmk_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL, FALSE, scheduler); order_actions(fence, action, pcmk__ar_first_implies_then); } } static bool should_ignore_failure_timeout(const pcmk_resource_t *rsc, const char *task, guint interval_ms, bool is_last_failure) { /* Clearing failures of recurring monitors has special concerns. The * executor reports only changes in the monitor result, so if the * monitor is still active and still getting the same failure result, * that will go undetected after the failure is cleared. * * Also, the operation history will have the time when the recurring * monitor result changed to the given code, not the time when the * result last happened. * * @TODO We probably should clear such failures only when the failure * timeout has passed since the last occurrence of the failed result. * However we don't record that information. We could maybe approximate * that by clearing only if there is a more recent successful monitor or * stop result, but we don't even have that information at this point * since we are still unpacking the resource's operation history. * * This is especially important for remote connection resources with a * reconnect interval, so in that case, we skip clearing failures * if the remote node hasn't been fenced. */ if (rsc->remote_reconnect_ms && pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled) && (interval_ms != 0) && pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { pcmk_node_t *remote_node = pe_find_node(rsc->cluster->nodes, rsc->id); if (remote_node && !remote_node->details->remote_was_fenced) { if (is_last_failure) { crm_info("Waiting to clear monitor failure for remote node %s" " until fencing has occurred", rsc->id); } return TRUE; } } return FALSE; } /*! * \internal * \brief Check operation age and schedule failure clearing when appropriate * * This function has two distinct purposes. The first is to check whether an * operation history entry is expired (i.e. the resource has a failure timeout, * the entry is older than the timeout, and the resource either has no fail * count or its fail count is entirely older than the timeout). The second is to * schedule fail count clearing when appropriate (i.e. the operation is expired * and either the resource has an expired fail count or the operation is a * last_failure for a remote connection resource with a reconnect interval, * or the operation is a last_failure for a start or monitor operation and the * resource's parameters have changed since the operation). * * \param[in,out] history Parsed action result history * * \return true if operation history entry is expired, otherwise false */ static bool check_operation_expiry(struct action_history *history) { bool expired = false; bool is_last_failure = pcmk__ends_with(history->id, "_last_failure_0"); time_t last_run = 0; int unexpired_fail_count = 0; const char *clear_reason = NULL; if (history->execution_status == PCMK_EXEC_NOT_INSTALLED) { pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not expired: " "Not Installed does not expire", history->id, pcmk__node_name(history->node)); return false; // "Not installed" must always be cleared manually } if ((history->rsc->failure_timeout > 0) && (crm_element_value_epoch(history->xml, PCMK_XA_LAST_RC_CHANGE, &last_run) == 0)) { /* Resource has a PCMK_META_FAILURE_TIMEOUT and history entry has a * timestamp */ time_t now = get_effective_time(history->rsc->cluster); time_t last_failure = 0; // Is this particular operation history older than the failure timeout? if ((now >= (last_run + history->rsc->failure_timeout)) && !should_ignore_failure_timeout(history->rsc, history->task, history->interval_ms, is_last_failure)) { expired = true; } // Does the resource as a whole have an unexpired fail count? unexpired_fail_count = pe_get_failcount(history->node, history->rsc, &last_failure, pcmk__fc_effective, history->xml); // Update scheduler recheck time according to *last* failure crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d timeout=%ds" " last-failure@%lld", history->id, (long long) last_run, (expired? "" : "not "), (long long) now, unexpired_fail_count, history->rsc->failure_timeout, (long long) last_failure); last_failure += history->rsc->failure_timeout + 1; if (unexpired_fail_count && (now < last_failure)) { pe__update_recheck_time(last_failure, history->rsc->cluster, "fail count expiration"); } } if (expired) { if (pe_get_failcount(history->node, history->rsc, NULL, pcmk__fc_default, history->xml)) { // There is a fail count ignoring timeout if (unexpired_fail_count == 0) { // There is no fail count considering timeout clear_reason = "it expired"; } else { /* This operation is old, but there is an unexpired fail count. * In a properly functioning cluster, this should only be * possible if this operation is not a failure (otherwise the * fail count should be expired too), so this is really just a * failsafe. */ pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not " "expired: Unexpired fail count", history->id, pcmk__node_name(history->node)); expired = false; } } else if (is_last_failure && (history->rsc->remote_reconnect_ms != 0)) { /* Clear any expired last failure when reconnect interval is set, * even if there is no fail count. */ clear_reason = "reconnect interval is set"; } } if (!expired && is_last_failure && should_clear_for_param_change(history->xml, history->task, history->rsc, history->node)) { clear_reason = "resource parameters have changed"; } if (clear_reason != NULL) { pcmk_action_t *clear_op = NULL; // Schedule clearing of the fail count clear_op = pe__clear_failcount(history->rsc, history->node, clear_reason, history->rsc->cluster); if (pcmk_is_set(history->rsc->cluster->flags, pcmk_sched_fencing_enabled) && (history->rsc->remote_reconnect_ms != 0)) { /* If we're clearing a remote connection due to a reconnect * interval, we want to wait until any scheduled fencing * completes. * * We could limit this to remote_node->details->unclean, but at * this point, that's always true (it won't be reliable until * after unpack_node_history() is done). */ crm_info("Clearing %s failure will wait until any scheduled " "fencing of %s completes", history->task, history->rsc->id); order_after_remote_fencing(clear_op, history->rsc, history->rsc->cluster); } } if (expired && (history->interval_ms == 0) && pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { switch (history->exit_status) { case PCMK_OCF_OK: case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_PROMOTED: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: // Don't expire probes that return these values pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not " "expired: Probe result", history->id, pcmk__node_name(history->node)); expired = false; break; } } return expired; } int pe__target_rc_from_xml(const xmlNode *xml_op) { int target_rc = 0; const char *key = crm_element_value(xml_op, PCMK__XA_TRANSITION_KEY); if (key == NULL) { return -1; } decode_transition_key(key, NULL, NULL, NULL, &target_rc); return target_rc; } /*! * \internal * \brief Update a resource's state for an action result * * \param[in,out] history Parsed action history entry * \param[in] exit_status Exit status to base new state on * \param[in] last_failure Resource's last_failure entry, if known * \param[in,out] on_fail Resource's current failure handling */ static void update_resource_state(struct action_history *history, int exit_status, const xmlNode *last_failure, enum action_fail_response *on_fail) { bool clear_past_failure = false; if ((exit_status == PCMK_OCF_NOT_INSTALLED) || (!pe_rsc_is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml))) { history->rsc->role = pcmk_role_stopped; } else if (exit_status == PCMK_OCF_NOT_RUNNING) { clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { if ((last_failure != NULL) && pcmk__str_eq(history->key, pcmk__xe_history_key(last_failure), pcmk__str_none)) { clear_past_failure = true; } if (history->rsc->role < pcmk_role_started) { set_active(history->rsc); } } else if (pcmk__str_eq(history->task, PCMK_ACTION_START, pcmk__str_none)) { history->rsc->role = pcmk_role_started; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_STOP, pcmk__str_none)) { history->rsc->role = pcmk_role_stopped; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_PROMOTE, pcmk__str_none)) { history->rsc->role = pcmk_role_promoted; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_DEMOTE, pcmk__str_none)) { if (*on_fail == pcmk_on_fail_demote) { /* Demote clears an error only if * PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE */ clear_past_failure = true; } history->rsc->role = pcmk_role_unpromoted; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_FROM, pcmk__str_none)) { history->rsc->role = pcmk_role_started; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)) { unpack_migrate_to_success(history); } else if (history->rsc->role < pcmk_role_started) { pcmk__rsc_trace(history->rsc, "%s active on %s", history->rsc->id, pcmk__node_name(history->node)); set_active(history->rsc); } if (!clear_past_failure) { return; } switch (*on_fail) { case pcmk_on_fail_stop: case pcmk_on_fail_ban: case pcmk_on_fail_standby_node: case pcmk_on_fail_fence_node: pcmk__rsc_trace(history->rsc, "%s (%s) is not cleared by a completed %s", history->rsc->id, fail2text(*on_fail), history->task); break; case pcmk_on_fail_block: case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: case pcmk_on_fail_restart_container: *on_fail = pcmk_on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "clear past failures"); break; case pcmk_on_fail_reset_remote: if (history->rsc->remote_reconnect_ms == 0) { /* With no reconnect interval, the connection is allowed to * start again after the remote node is fenced and * completely stopped. (With a reconnect interval, we wait * for the failure to be cleared entirely before attempting * to reconnect.) */ *on_fail = pcmk_on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "clear past failures and reset remote"); } break; } } /*! * \internal * \brief Check whether a given history entry matters for resource state * * \param[in] history Parsed action history entry * * \return true if action can affect resource state, otherwise false */ static inline bool can_affect_state(struct action_history *history) { #if 0 /* @COMPAT It might be better to parse only actions we know we're interested * in, rather than exclude a couple we don't. However that would be a * behavioral change that should be done at a major or minor series release. * Currently, unknown operations can affect whether a resource is considered * active and/or failed. */ return pcmk__str_any_of(history->task, PCMK_ACTION_MONITOR, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, "asyncmon", NULL); #else return !pcmk__str_any_of(history->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_META_DATA, NULL); #endif } /*! * \internal * \brief Unpack execution/exit status and exit reason from a history entry * * \param[in,out] history Action history entry to unpack * * \return Standard Pacemaker return code */ static int unpack_action_result(struct action_history *history) { if ((crm_element_value_int(history->xml, PCMK__XA_OP_STATUS, &(history->execution_status)) < 0) || (history->execution_status < PCMK_EXEC_PENDING) || (history->execution_status > PCMK_EXEC_MAX) || (history->execution_status == PCMK_EXEC_CANCELLED)) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "with invalid " PCMK__XA_OP_STATUS " '%s'", history->id, history->rsc->id, pcmk__node_name(history->node), pcmk__s(crm_element_value(history->xml, PCMK__XA_OP_STATUS), "")); return pcmk_rc_unpack_error; } if ((crm_element_value_int(history->xml, PCMK__XA_RC_CODE, &(history->exit_status)) < 0) || (history->exit_status < 0) || (history->exit_status > CRM_EX_MAX)) { #if 0 /* @COMPAT We should ignore malformed entries, but since that would * change behavior, it should be done at a major or minor series * release. */ pcmk__config_err("Ignoring resource history entry %s for %s on %s " "with invalid " PCMK__XA_RC_CODE " '%s'", history->id, history->rsc->id, pcmk__node_name(history->node), pcmk__s(crm_element_value(history->xml, PCMK__XA_RC_CODE), "")); return pcmk_rc_unpack_error; #else history->exit_status = CRM_EX_ERROR; #endif } history->exit_reason = crm_element_value(history->xml, PCMK_XA_EXIT_REASON); return pcmk_rc_ok; } /*! * \internal * \brief Process an action history entry whose result expired * * \param[in,out] history Parsed action history entry * \param[in] orig_exit_status Action exit status before remapping * * \return Standard Pacemaker return code (in particular, pcmk_rc_ok means the * entry needs no further processing) */ static int process_expired_result(struct action_history *history, int orig_exit_status) { if (!pe_rsc_is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml) && (orig_exit_status != history->expected_exit_status)) { if (history->rsc->role <= pcmk_role_stopped) { history->rsc->role = pcmk_role_unknown; } crm_trace("Ignoring resource history entry %s for probe of %s on %s: " "Masked failure expired", history->id, history->rsc->id, pcmk__node_name(history->node)); return pcmk_rc_ok; } if (history->exit_status == history->expected_exit_status) { return pcmk_rc_undetermined; // Only failures expire } if (history->interval_ms == 0) { crm_notice("Ignoring resource history entry %s for %s of %s on %s: " "Expired failure", history->id, history->task, history->rsc->id, pcmk__node_name(history->node)); return pcmk_rc_ok; } if (history->node->details->online && !history->node->details->unclean) { /* Reschedule the recurring action. schedule_cancel() won't work at * this stage, so as a hacky workaround, forcibly change the restart * digest so pcmk__check_action_config() does what we want later. * * @TODO We should skip this if there is a newer successful monitor. * Also, this causes rescheduling only if the history entry * has a PCMK__XA_OP_DIGEST (which the expire-non-blocked-failure * scheduler regression test doesn't, but that may not be a * realistic scenario in production). */ crm_notice("Rescheduling %s-interval %s of %s on %s " "after failure expired", pcmk__readable_interval(history->interval_ms), history->task, history->rsc->id, pcmk__node_name(history->node)); crm_xml_add(history->xml, PCMK__XA_OP_RESTART_DIGEST, "calculated-failure-timeout"); return pcmk_rc_ok; } return pcmk_rc_undetermined; } /*! * \internal * \brief Process a masked probe failure * * \param[in,out] history Parsed action history entry * \param[in] orig_exit_status Action exit status before remapping * \param[in] last_failure Resource's last_failure entry, if known * \param[in,out] on_fail Resource's current failure handling */ static void mask_probe_failure(struct action_history *history, int orig_exit_status, const xmlNode *last_failure, enum action_fail_response *on_fail) { pcmk_resource_t *ban_rsc = history->rsc; if (!pcmk_is_set(history->rsc->flags, pcmk_rsc_unique)) { ban_rsc = uber_parent(history->rsc); } crm_notice("Treating probe result '%s' for %s on %s as 'not running'", services_ocf_exitcode_str(orig_exit_status), history->rsc->id, pcmk__node_name(history->node)); update_resource_state(history, history->expected_exit_status, last_failure, on_fail); crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->details->uname); record_failed_op(history); resource_location(ban_rsc, history->node, -INFINITY, "masked-probe-failure", history->rsc->cluster); } /*! * \internal Check whether a given failure is for a given pending action * * \param[in] history Parsed history entry for pending action * \param[in] last_failure Resource's last_failure entry, if known * * \return true if \p last_failure is failure of pending action in \p history, * otherwise false * \note Both \p history and \p last_failure must come from the same * \c PCMK__XE_LRM_RESOURCE block, as node and resource are assumed to be * the same. */ static bool failure_is_newer(const struct action_history *history, const xmlNode *last_failure) { guint failure_interval_ms = 0U; long long failure_change = 0LL; long long this_change = 0LL; if (last_failure == NULL) { return false; // Resource has no last_failure entry } if (!pcmk__str_eq(history->task, crm_element_value(last_failure, PCMK_XA_OPERATION), pcmk__str_none)) { return false; // last_failure is for different action } if ((crm_element_value_ms(last_failure, PCMK_META_INTERVAL, &failure_interval_ms) != pcmk_ok) || (history->interval_ms != failure_interval_ms)) { return false; // last_failure is for action with different interval } if ((pcmk__scan_ll(crm_element_value(history->xml, PCMK_XA_LAST_RC_CHANGE), &this_change, 0LL) != pcmk_rc_ok) || (pcmk__scan_ll(crm_element_value(last_failure, PCMK_XA_LAST_RC_CHANGE), &failure_change, 0LL) != pcmk_rc_ok) || (failure_change < this_change)) { return false; // Failure is not known to be newer } return true; } /*! * \internal * \brief Update a resource's role etc. for a pending action * * \param[in,out] history Parsed history entry for pending action * \param[in] last_failure Resource's last_failure entry, if known */ static void process_pending_action(struct action_history *history, const xmlNode *last_failure) { /* For recurring monitors, a failure is recorded only in RSC_last_failure_0, * and there might be a RSC_monitor_INTERVAL entry with the last successful * or pending result. * * If last_failure contains the failure of the pending recurring monitor * we're processing here, and is newer, the action is no longer pending. * (Pending results have call ID -1, which sorts last, so the last failure * if any should be known.) */ if (failure_is_newer(history, last_failure)) { return; } if (strcmp(history->task, PCMK_ACTION_START) == 0) { pcmk__set_rsc_flags(history->rsc, pcmk_rsc_start_pending); set_active(history->rsc); } else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) { history->rsc->role = pcmk_role_promoted; } else if ((strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) && history->node->details->unclean) { /* A migrate_to action is pending on a unclean source, so force a stop * on the target. */ const char *migrate_target = NULL; pcmk_node_t *target = NULL; migrate_target = crm_element_value(history->xml, PCMK__META_MIGRATE_TARGET); target = pe_find_node(history->rsc->cluster->nodes, migrate_target); if (target != NULL) { stop_action(history->rsc, target, FALSE); } } if (history->rsc->pending_task != NULL) { /* There should never be multiple pending actions, but as a failsafe, * just remember the first one processed for display purposes. */ return; } if (pcmk_is_probe(history->task, history->interval_ms)) { /* Pending probes are currently never displayed, even if pending * operations are requested. If we ever want to change that, * enable the below and the corresponding part of * native.c:native_pending_task(). */ #if 0 history->rsc->pending_task = strdup("probe"); history->rsc->pending_node = history->node; #endif } else { history->rsc->pending_task = strdup(history->task); history->rsc->pending_node = history->node; } } static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *on_fail) { int old_rc = 0; bool expired = false; pcmk_resource_t *parent = rsc; enum rsc_role_e fail_role = pcmk_role_unknown; enum action_fail_response failure_strategy = pcmk_on_fail_restart; struct action_history history = { .rsc = rsc, .node = node, .xml = xml_op, .execution_status = PCMK_EXEC_UNKNOWN, }; CRM_CHECK(rsc && node && xml_op, return); history.id = ID(xml_op); if (history.id == NULL) { pcmk__config_err("Ignoring resource history entry for %s on %s " "without ID", rsc->id, pcmk__node_name(node)); return; } // Task and interval history.task = crm_element_value(xml_op, PCMK_XA_OPERATION); if (history.task == NULL) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "without " PCMK_XA_OPERATION, history.id, rsc->id, pcmk__node_name(node)); return; } crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &(history.interval_ms)); if (!can_affect_state(&history)) { pcmk__rsc_trace(rsc, "Ignoring resource history entry %s for %s on %s " "with irrelevant action '%s'", history.id, rsc->id, pcmk__node_name(node), history.task); return; } if (unpack_action_result(&history) != pcmk_rc_ok) { return; // Error already logged } history.expected_exit_status = pe__target_rc_from_xml(xml_op); history.key = pcmk__xe_history_key(xml_op); crm_element_value_int(xml_op, PCMK__XA_CALL_ID, &(history.call_id)); pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)", history.id, history.task, history.call_id, pcmk__node_name(node), pcmk_exec_status_str(history.execution_status), crm_exit_str(history.exit_status)); if (node->details->unclean) { pcmk__rsc_trace(rsc, "%s is running on %s, which is unclean (further action " "depends on value of stop's on-fail attribute)", rsc->id, pcmk__node_name(node)); } expired = check_operation_expiry(&history); old_rc = history.exit_status; remap_operation(&history, on_fail, expired); if (expired && (process_expired_result(&history, old_rc) == pcmk_rc_ok)) { goto done; } if (!pe_rsc_is_bundled(rsc) && pcmk_xe_mask_probe_failure(xml_op)) { mask_probe_failure(&history, old_rc, *last_failure, on_fail); goto done; } if (!pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { parent = uber_parent(rsc); } switch (history.execution_status) { case PCMK_EXEC_PENDING: process_pending_action(&history, *last_failure); goto done; case PCMK_EXEC_DONE: update_resource_state(&history, history.exit_status, *last_failure, on_fail); goto done; case PCMK_EXEC_NOT_INSTALLED: unpack_failure_handling(&history, &failure_strategy, &fail_role); if (failure_strategy == pcmk_on_fail_ignore) { crm_warn("Cannot ignore failed %s of %s on %s: " "Resource agent doesn't exist " CRM_XS " status=%d rc=%d id=%s", history.task, rsc->id, pcmk__node_name(node), history.execution_status, history.exit_status, history.id); /* Also for printing it as "FAILED" by marking it as * pcmk_rsc_failed later */ *on_fail = pcmk_on_fail_ban; } resource_location(parent, node, -INFINITY, "hard-error", rsc->cluster); unpack_rsc_op_failure(&history, failure_strategy, fail_role, last_failure, on_fail); goto done; case PCMK_EXEC_NOT_CONNECTED: if (pe__is_guest_or_remote_node(node) && pcmk_is_set(node->details->remote_rsc->flags, pcmk_rsc_managed)) { /* We should never get into a situation where a managed remote * connection resource is considered OK but a resource action * behind the connection gets a "not connected" status. But as a * fail-safe in case a bug or unusual circumstances do lead to * that, ensure the remote connection is considered failed. */ pcmk__set_rsc_flags(node->details->remote_rsc, pcmk_rsc_failed|pcmk_rsc_stop_if_failed); } break; // Not done, do error handling case PCMK_EXEC_ERROR: case PCMK_EXEC_ERROR_HARD: case PCMK_EXEC_ERROR_FATAL: case PCMK_EXEC_TIMEOUT: case PCMK_EXEC_NOT_SUPPORTED: case PCMK_EXEC_INVALID: break; // Not done, do error handling default: // No other value should be possible at this point break; } unpack_failure_handling(&history, &failure_strategy, &fail_role); if ((failure_strategy == pcmk_on_fail_ignore) || ((failure_strategy == pcmk_on_fail_restart_container) && (strcmp(history.task, PCMK_ACTION_STOP) == 0))) { char *last_change_s = last_change_str(xml_op); crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s succeeded " CRM_XS " %s", history.task, services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), rsc->id, pcmk__node_name(node), last_change_s, history.id); free(last_change_s); update_resource_state(&history, history.expected_exit_status, *last_failure, on_fail); crm_xml_add(xml_op, PCMK_XA_UNAME, node->details->uname); pcmk__set_rsc_flags(rsc, pcmk_rsc_ignore_failure); record_failed_op(&history); if ((failure_strategy == pcmk_on_fail_restart_container) && cmp_on_fail(*on_fail, pcmk_on_fail_restart) <= 0) { *on_fail = failure_strategy; } } else { unpack_rsc_op_failure(&history, failure_strategy, fail_role, last_failure, on_fail); if (history.execution_status == PCMK_EXEC_ERROR_HARD) { uint8_t log_level = LOG_ERR; if (history.exit_status == PCMK_OCF_NOT_INSTALLED) { log_level = LOG_NOTICE; } do_crm_log(log_level, "Preventing %s from restarting on %s because " "of hard failure (%s%s%s) " CRM_XS " %s", parent->id, pcmk__node_name(node), services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), history.id); resource_location(parent, node, -INFINITY, "hard-error", rsc->cluster); } else if (history.execution_status == PCMK_EXEC_ERROR_FATAL) { pcmk__sched_err("Preventing %s from restarting anywhere because " "of fatal failure (%s%s%s) " CRM_XS " %s", parent->id, services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), history.id); resource_location(parent, NULL, -INFINITY, "fatal-error", rsc->cluster); } } done: pcmk__rsc_trace(rsc, "%s role on %s after %s is %s (next %s)", rsc->id, pcmk__node_name(node), history.id, pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role)); } static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite, pcmk_scheduler_t *scheduler) { const char *cluster_name = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = pcmk_role_unknown, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_UNAME), strdup(node->details->uname)); g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_ID), strdup(node->details->id)); if (pcmk__str_eq(node->details->id, scheduler->dc_uuid, pcmk__str_casei)) { scheduler->dc_node = node; node->details->is_dc = TRUE; g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_IS_DC), strdup(PCMK_VALUE_TRUE)); } else { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_IS_DC), strdup(PCMK_VALUE_FALSE)); } cluster_name = g_hash_table_lookup(scheduler->config_hash, PCMK_OPT_CLUSTER_NAME); if (cluster_name) { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_CLUSTER_NAME), strdup(cluster_name)); } pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, node->details->attrs, NULL, overwrite, scheduler); pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_UTILIZATION, &rule_data, node->details->utilization, NULL, FALSE, scheduler); if (pe_node_attribute_raw(node, CRM_ATTR_SITE_NAME) == NULL) { const char *site_name = pe_node_attribute_raw(node, "site-name"); if (site_name) { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_SITE_NAME), strdup(site_name)); } else if (cluster_name) { /* Default to cluster-name if unset */ g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_SITE_NAME), strdup(cluster_name)); } } } static GList * extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter) { int counter = -1; int stop_index = -1; int start_index = -1; xmlNode *rsc_op = NULL; GList *gIter = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; /* extract operations */ op_list = NULL; sorted_op_list = NULL; for (rsc_op = pcmk__xe_first_child(rsc_entry); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) { if (pcmk__str_eq((const char *)rsc_op->name, PCMK__XE_LRM_RSC_OP, pcmk__str_none)) { crm_xml_add(rsc_op, PCMK_XA_RESOURCE, rsc); crm_xml_add(rsc_op, PCMK_XA_UNAME, node); op_list = g_list_prepend(op_list, rsc_op); } } if (op_list == NULL) { /* if there are no operations, there is nothing to do */ return NULL; } sorted_op_list = g_list_sort(op_list, sort_op_by_callid); /* create active recurring operations as optional */ if (active_filter == FALSE) { return sorted_op_list; } op_list = NULL; calculate_active_ops(sorted_op_list, &start_index, &stop_index); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; if (start_index < stop_index) { crm_trace("Skipping %s: not active", ID(rsc_entry)); break; } else if (counter < start_index) { crm_trace("Skipping %s: old", ID(rsc_op)); continue; } op_list = g_list_append(op_list, rsc_op); } g_list_free(sorted_op_list); return op_list; } GList * find_operations(const char *rsc, const char *node, gboolean active_filter, pcmk_scheduler_t *scheduler) { GList *output = NULL; GList *intermediate = NULL; xmlNode *tmp = NULL; xmlNode *status = find_xml_node(scheduler->input, PCMK_XE_STATUS, TRUE); pcmk_node_t *this_node = NULL; xmlNode *node_state = NULL; for (node_state = pcmk__xe_first_child(status); node_state != NULL; node_state = pcmk__xe_next(node_state)) { if (pcmk__str_eq((const char *) node_state->name, PCMK__XE_NODE_STATE, pcmk__str_none)) { const char *uname = crm_element_value(node_state, PCMK_XA_UNAME); if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) { continue; } this_node = pe_find_node(scheduler->nodes, uname); if(this_node == NULL) { CRM_LOG_ASSERT(this_node != NULL); continue; } else if (pe__is_guest_or_remote_node(this_node)) { determine_remote_online_status(scheduler, this_node); } else { determine_online_status(node_state, this_node, scheduler); } if (this_node->details->online || pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { /* offline nodes run no resources... * unless stonith is enabled in which case we need to * make sure rsc start events happen after the stonith */ xmlNode *lrm_rsc = NULL; tmp = find_xml_node(node_state, PCMK__XE_LRM, FALSE); tmp = find_xml_node(tmp, PCMK__XE_LRM_RESOURCES, FALSE); for (lrm_rsc = pcmk__xe_first_child(tmp); lrm_rsc != NULL; lrm_rsc = pcmk__xe_next(lrm_rsc)) { if (pcmk__str_eq((const char *) lrm_rsc->name, PCMK__XE_LRM_RESOURCE, pcmk__str_none)) { const char *rsc_id = crm_element_value(lrm_rsc, PCMK_XA_ID); if (rsc != NULL && !pcmk__str_eq(rsc_id, rsc, pcmk__str_casei)) { continue; } intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter); output = g_list_concat(output, intermediate); } } } } } return output; }