diff --git a/include/crm/common/clone_internal.h b/include/crm/common/clone_internal.h index 1abda0cda7..85e325d556 100644 --- a/include/crm/common/clone_internal.h +++ b/include/crm/common/clone_internal.h @@ -1,52 +1,66 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_CLONE_INTERNAL__H # define PCMK__CRM_COMMON_CLONE_INTERNAL__H #include // pcmk_resource_t #include // pcmk_rsc_variant_clone #ifdef __cplusplus extern "C" { #endif // Clone resource flags (used in variant data) enum pcmk__clone_flags { // Whether instances should be started sequentially pcmk__clone_ordered = (1 << 0), // Whether promotion scores have been added pcmk__clone_promotion_added = (1 << 1), // Whether promotion constraints have been added pcmk__clone_promotion_constrained = (1 << 2), }; /*! * \internal * \brief Check whether a resource is a clone resource * * \param[in] rsc Resource to check * * \return true if \p rsc is a clone, otherwise false * * \note This does not return true if \p rsc has a clone ancestor. */ static inline bool pcmk__is_clone(const pcmk_resource_t *rsc) { return (rsc != NULL) && (rsc->variant == pcmk_rsc_variant_clone); } +/*! + * \internal + * \brief Check whether a resource is a globally unique clone + * + * \param[in] rsc Resource to check + * + * \return true if \p rsc is a unique clone, otherwise false + */ +static inline bool +pcmk__is_unique_clone(const pcmk_resource_t *rsc) +{ + return pcmk__is_clone(rsc) && pcmk_is_set(rsc->flags, pcmk_rsc_unique); +} + #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_CLONE_INTERNAL__H diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c index 4c4dd21019..80b4d06938 100644 --- a/lib/pacemaker/pcmk_sched_primitive.c +++ b/lib/pacemaker/pcmk_sched_primitive.c @@ -1,1679 +1,1679 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include // uint8_t, uint32_t #include #include #include "libpacemaker_private.h" static void stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void start_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); #define RSC_ROLE_MAX (pcmk_role_promoted + 1) static enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = { /* This array lists the immediate next role when transitioning from one role * to a target role. For example, when going from Stopped to Promoted, the * next role is Unpromoted, because the resource must be started before it * can be promoted. The current state then becomes Started, which is fed * into this array again, giving a next role of Promoted. * * Current role Immediate next role Final target role * ------------ ------------------- ----------------- */ /* Unknown */ { pcmk_role_unknown, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_stopped, /* Started */ pcmk_role_stopped, /* Unpromoted */ pcmk_role_stopped, /* Promoted */ }, /* Stopped */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_started, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_unpromoted, /* Promoted */ }, /* Started */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_started, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, /* Unpromoted */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_stopped, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, /* Promoted */ { pcmk_role_stopped, /* Unknown */ pcmk_role_unpromoted, /* Stopped */ pcmk_role_unpromoted, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, }; /*! * \internal * \brief Function to schedule actions needed for a role change * * \param[in,out] rsc Resource whose role is changing * \param[in,out] node Node where resource will be in its next role * \param[in] optional Whether scheduled actions should be optional */ typedef void (*rsc_transition_fn)(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static rsc_transition_fn rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = { /* This array lists the function needed to transition directly from one role * to another. NULL indicates that nothing is needed. * * Current role Transition function Next role * ------------ ------------------- ---------- */ /* Unknown */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ assert_role_error, /* Started */ assert_role_error, /* Unpromoted */ assert_role_error, /* Promoted */ }, /* Stopped */ { assert_role_error, /* Unknown */ NULL, /* Stopped */ start_resource, /* Started */ start_resource, /* Unpromoted */ assert_role_error, /* Promoted */ }, /* Started */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ NULL, /* Started */ NULL, /* Unpromoted */ promote_resource, /* Promoted */ }, /* Unpromoted */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ stop_resource, /* Started */ NULL, /* Unpromoted */ promote_resource, /* Promoted */ }, /* Promoted */ { assert_role_error, /* Unknown */ demote_resource, /* Stopped */ demote_resource, /* Started */ demote_resource, /* Unpromoted */ NULL, /* Promoted */ }, }; /*! * \internal * \brief Get a list of a resource's allowed nodes sorted by node score * * \param[in] rsc Resource to check * * \return List of allowed nodes sorted by node score */ static GList * sorted_allowed_nodes(const pcmk_resource_t *rsc) { if (rsc->allowed_nodes != NULL) { GList *nodes = g_hash_table_get_values(rsc->allowed_nodes); if (nodes != NULL) { return pcmk__sort_nodes(nodes, pcmk__current_node(rsc)); } } return NULL; } /*! * \internal * \brief Assign a resource to its best allowed node, if possible * * \param[in,out] rsc Resource to choose a node for * \param[in] prefer If not \c NULL, prefer this node when all else * equal * \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a * node, set next role to stopped and update * existing actions * * \return true if \p rsc could be assigned to a node, otherwise false * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ static bool assign_best_node(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { GList *nodes = NULL; pcmk_node_t *chosen = NULL; pcmk_node_t *best = NULL; const pcmk_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc); if (prefer == NULL) { prefer = most_free_node; } if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { // We've already finished assignment of resources to nodes return rsc->allocated_to != NULL; } // Sort allowed nodes by score nodes = sorted_allowed_nodes(rsc); if (nodes != NULL) { best = (pcmk_node_t *) nodes->data; // First node has best score } if ((prefer != NULL) && (nodes != NULL)) { // Get the allowed node version of prefer chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id); if (chosen == NULL) { pcmk__rsc_trace(rsc, "Preferred node %s for %s was unknown", pcmk__node_name(prefer), rsc->id); /* Favor the preferred node as long as its score is at least as good as * the best allowed node's. * * An alternative would be to favor the preferred node even if the best * node is better, when the best node's score is less than INFINITY. */ } else if (chosen->weight < best->weight) { pcmk__rsc_trace(rsc, "Preferred node %s for %s was unsuitable", pcmk__node_name(chosen), rsc->id); chosen = NULL; } else if (!pcmk__node_available(chosen, true, false)) { pcmk__rsc_trace(rsc, "Preferred node %s for %s was unavailable", pcmk__node_name(chosen), rsc->id); chosen = NULL; } else { pcmk__rsc_trace(rsc, "Chose preferred node %s for %s " "(ignoring %d candidates)", pcmk__node_name(chosen), rsc->id, g_list_length(nodes)); } } if ((chosen == NULL) && (best != NULL)) { /* Either there is no preferred node, or the preferred node is not * suitable, but another node is allowed to run the resource. */ chosen = best; - if (!pe_rsc_is_unique_clone(rsc->parent) + if (!pcmk__is_unique_clone(rsc->parent) && (chosen->weight > 0) // Zero not acceptable && pcmk__node_available(chosen, false, false)) { /* If the resource is already running on a node, prefer that node if * it is just as good as the chosen node. * * We don't do this for unique clone instances, because * pcmk__assign_instances() has already assigned instances to their * running nodes when appropriate, and if we get here, we don't want * remaining unassigned instances to prefer a node that's already * running another instance. */ pcmk_node_t *running = pcmk__current_node(rsc); if (running == NULL) { // Nothing to do } else if (!pcmk__node_available(running, true, false)) { pcmk__rsc_trace(rsc, "Current node for %s (%s) can't run resources", rsc->id, pcmk__node_name(running)); } else { int nodes_with_best_score = 1; for (GList *iter = nodes->next; iter; iter = iter->next) { pcmk_node_t *allowed = (pcmk_node_t *) iter->data; if (allowed->weight != chosen->weight) { // The nodes are sorted by score, so no more are equal break; } if (pcmk__same_node(allowed, running)) { // Scores are equal, so prefer the current node chosen = allowed; } nodes_with_best_score++; } if (nodes_with_best_score > 1) { uint8_t log_level = LOG_INFO; if (chosen->weight >= INFINITY) { log_level = LOG_WARNING; } do_crm_log(log_level, "Chose %s for %s from %d nodes with score %s", pcmk__node_name(chosen), rsc->id, nodes_with_best_score, pcmk_readable_score(chosen->weight)); } } } pcmk__rsc_trace(rsc, "Chose %s for %s from %d candidates", pcmk__node_name(chosen), rsc->id, g_list_length(nodes)); } pcmk__assign_resource(rsc, chosen, false, stop_if_fail); g_list_free(nodes); return rsc->allocated_to != NULL; } /*! * \internal * \brief Apply a "this with" colocation to a node's allowed node scores * * \param[in,out] colocation Colocation to apply * \param[in,out] rsc Resource being assigned */ static void apply_this_with(pcmk__colocation_t *colocation, pcmk_resource_t *rsc) { GHashTable *archive = NULL; pcmk_resource_t *other = colocation->primary; // In certain cases, we will need to revert the node scores if ((colocation->dependent_role >= pcmk_role_promoted) || ((colocation->score < 0) && (colocation->score > -INFINITY))) { archive = pcmk__copy_node_table(rsc->allowed_nodes); } if (pcmk_is_set(other->flags, pcmk_rsc_unassigned)) { pcmk__rsc_trace(rsc, "%s: Assigning colocation %s primary %s first" "(score=%d role=%s)", rsc->id, colocation->id, other->id, colocation->score, pcmk_role_text(colocation->dependent_role)); other->cmds->assign(other, NULL, true); } // Apply the colocation score to this resource's allowed node scores rsc->cmds->apply_coloc_score(rsc, other, colocation, true); if ((archive != NULL) && !pcmk__any_node_available(rsc->allowed_nodes)) { pcmk__rsc_info(rsc, "%s: Reverting scores from colocation with %s " "because no nodes allowed", rsc->id, other->id); g_hash_table_destroy(rsc->allowed_nodes); rsc->allowed_nodes = archive; archive = NULL; } if (archive != NULL) { g_hash_table_destroy(archive); } } /*! * \internal * \brief Update a Pacemaker Remote node once its connection has been assigned * * \param[in] connection Connection resource that has been assigned */ static void remote_connection_assigned(const pcmk_resource_t *connection) { pcmk_node_t *remote_node = pe_find_node(connection->cluster->nodes, connection->id); CRM_CHECK(remote_node != NULL, return); if ((connection->allocated_to != NULL) && (connection->next_role != pcmk_role_stopped)) { crm_trace("Pacemaker Remote node %s will be online", remote_node->details->id); remote_node->details->online = TRUE; if (remote_node->details->unseen) { // Avoid unnecessary fence, since we will attempt connection remote_node->details->unclean = FALSE; } } else { crm_trace("Pacemaker Remote node %s will be shut down " "(%sassigned connection's next role is %s)", remote_node->details->id, ((connection->allocated_to == NULL)? "un" : ""), pcmk_role_text(connection->next_role)); remote_node->details->shutdown = TRUE; } } /*! * \internal * \brief Assign a primitive resource to a node * * \param[in,out] rsc Resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a * node, set next role to stopped and update * existing actions * * \return Node that \p rsc is assigned to, if assigned entirely to one node * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ pcmk_node_t * pcmk__primitive_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { GList *this_with_colocations = NULL; GList *with_this_colocations = NULL; GList *iter = NULL; pcmk__colocation_t *colocation = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); // Never assign a child without parent being assigned first if ((rsc->parent != NULL) && !pcmk_is_set(rsc->parent->flags, pcmk_rsc_assigning)) { pcmk__rsc_debug(rsc, "%s: Assigning parent %s first", rsc->id, rsc->parent->id); rsc->parent->cmds->assign(rsc->parent, prefer, stop_if_fail); } if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { // Assignment has already been done const char *node_name = "no node"; if (rsc->allocated_to != NULL) { node_name = pcmk__node_name(rsc->allocated_to); } pcmk__rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name); return rsc->allocated_to; } // Ensure we detect assignment loops if (pcmk_is_set(rsc->flags, pcmk_rsc_assigning)) { pcmk__rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id); return NULL; } pcmk__set_rsc_flags(rsc, pcmk_rsc_assigning); pe__show_node_scores(true, rsc, "Pre-assignment", rsc->allowed_nodes, rsc->cluster); this_with_colocations = pcmk__this_with_colocations(rsc); with_this_colocations = pcmk__with_this_colocations(rsc); // Apply mandatory colocations first, to satisfy as many as possible for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score <= -CRM_SCORE_INFINITY) || (colocation->score >= CRM_SCORE_INFINITY)) { apply_this_with(colocation, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score <= -CRM_SCORE_INFINITY) || (colocation->score >= CRM_SCORE_INFINITY)) { pcmk__add_dependent_scores(colocation, rsc); } } pe__show_node_scores(true, rsc, "Mandatory-colocations", rsc->allowed_nodes, rsc->cluster); // Then apply optional colocations for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score > -CRM_SCORE_INFINITY) && (colocation->score < CRM_SCORE_INFINITY)) { apply_this_with(colocation, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score > -CRM_SCORE_INFINITY) && (colocation->score < CRM_SCORE_INFINITY)) { pcmk__add_dependent_scores(colocation, rsc); } } g_list_free(this_with_colocations); g_list_free(with_this_colocations); if (rsc->next_role == pcmk_role_stopped) { pcmk__rsc_trace(rsc, "Banning %s from all nodes because it will be stopped", rsc->id); resource_location(rsc, NULL, -INFINITY, PCMK_META_TARGET_ROLE, rsc->cluster); } else if ((rsc->next_role > rsc->role) && !pcmk_is_set(rsc->cluster->flags, pcmk_sched_quorate) && (rsc->cluster->no_quorum_policy == pcmk_no_quorum_freeze)) { crm_notice("Resource %s cannot be elevated from %s to %s due to " PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE, rsc->id, pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role)); pe__set_next_role(rsc, rsc->role, PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE); } pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); // Unmanage resource if fencing is enabled but no device is configured if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled) && !pcmk_is_set(rsc->cluster->flags, pcmk_sched_have_fencing)) { pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed); } if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { // Unmanaged resources stay on their current node const char *reason = NULL; pcmk_node_t *assign_to = NULL; pe__set_next_role(rsc, rsc->role, "unmanaged"); assign_to = pcmk__current_node(rsc); if (assign_to == NULL) { reason = "inactive"; } else if (rsc->role == pcmk_role_promoted) { reason = "promoted"; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { reason = "failed"; } else { reason = "active"; } pcmk__rsc_info(rsc, "Unmanaged resource %s assigned to %s: %s", rsc->id, (assign_to? assign_to->details->uname : "no node"), reason); pcmk__assign_resource(rsc, assign_to, true, stop_if_fail); } else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_stop_all)) { // Must stop at some point, but be consistent with stop_if_fail if (stop_if_fail) { pcmk__rsc_debug(rsc, "Forcing %s to stop: " PCMK_OPT_STOP_ALL_RESOURCES, rsc->id); } pcmk__assign_resource(rsc, NULL, true, stop_if_fail); } else if (!assign_best_node(rsc, prefer, stop_if_fail)) { // Assignment failed if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { pcmk__rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id); } else if ((rsc->running_on != NULL) && stop_if_fail) { pcmk__rsc_info(rsc, "Stopping orphan resource %s", rsc->id); } } pcmk__clear_rsc_flags(rsc, pcmk_rsc_assigning); if (rsc->is_remote_node) { remote_connection_assigned(rsc); } return rsc->allocated_to; } /*! * \internal * \brief Schedule actions to bring resource down and back to current role * * \param[in,out] rsc Resource to restart * \param[in,out] current Node that resource should be brought down on * \param[in] need_stop Whether the resource must be stopped * \param[in] need_promote Whether the resource must be promoted * * \return Role that resource would have after scheduled actions are taken */ static void schedule_restart_actions(pcmk_resource_t *rsc, pcmk_node_t *current, bool need_stop, bool need_promote) { enum rsc_role_e role = rsc->role; enum rsc_role_e next_role; rsc_transition_fn fn = NULL; pcmk__set_rsc_flags(rsc, pcmk_rsc_restarting); // Bring resource down to a stop on its current node while (role != pcmk_role_stopped) { next_role = rsc_state_matrix[role][pcmk_role_stopped]; pcmk__rsc_trace(rsc, "Creating %s action to take %s down from %s to %s", (need_stop? "required" : "optional"), rsc->id, pcmk_role_text(role), pcmk_role_text(next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, current, !need_stop); role = next_role; } // Bring resource up to its next role on its next node while ((rsc->role <= rsc->next_role) && (role != rsc->role) && !pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) { bool required = need_stop; next_role = rsc_state_matrix[role][rsc->role]; if ((next_role == pcmk_role_promoted) && need_promote) { required = true; } pcmk__rsc_trace(rsc, "Creating %s action to take %s up from %s to %s", (required? "required" : "optional"), rsc->id, pcmk_role_text(role), pcmk_role_text(next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, rsc->allocated_to, !required); role = next_role; } pcmk__clear_rsc_flags(rsc, pcmk_rsc_restarting); } /*! * \internal * \brief If a resource's next role is not explicitly specified, set a default * * \param[in,out] rsc Resource to set next role for * * \return "explicit" if next role was explicitly set, otherwise "implicit" */ static const char * set_default_next_role(pcmk_resource_t *rsc) { if (rsc->next_role != pcmk_role_unknown) { return "explicit"; } if (rsc->allocated_to == NULL) { pe__set_next_role(rsc, pcmk_role_stopped, "assignment"); } else { pe__set_next_role(rsc, pcmk_role_started, "assignment"); } return "implicit"; } /*! * \internal * \brief Create an action to represent an already pending start * * \param[in,out] rsc Resource to create start action for */ static void create_pending_start(pcmk_resource_t *rsc) { pcmk_action_t *start = NULL; pcmk__rsc_trace(rsc, "Creating action for %s to represent already pending start", rsc->id); start = start_action(rsc, rsc->allocated_to, TRUE); pcmk__set_action_flags(start, pcmk_action_always_in_graph); } /*! * \internal * \brief Schedule actions needed to take a resource to its next role * * \param[in,out] rsc Resource to schedule actions for */ static void schedule_role_transition_actions(pcmk_resource_t *rsc) { enum rsc_role_e role = rsc->role; while (role != rsc->next_role) { enum rsc_role_e next_role = rsc_state_matrix[role][rsc->next_role]; rsc_transition_fn fn = NULL; pcmk__rsc_trace(rsc, "Creating action to take %s from %s to %s " "(ending at %s)", rsc->id, pcmk_role_text(role), pcmk_role_text(next_role), pcmk_role_text(rsc->next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, rsc->allocated_to, false); role = next_role; } } /*! * \internal * \brief Create all actions needed for a given primitive resource * * \param[in,out] rsc Primitive resource to create actions for */ void pcmk__primitive_create_actions(pcmk_resource_t *rsc) { bool need_stop = false; bool need_promote = false; bool is_moving = false; bool allow_migrate = false; bool multiply_active = false; pcmk_node_t *current = NULL; unsigned int num_all_active = 0; unsigned int num_clean_active = 0; const char *next_role_source = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); next_role_source = set_default_next_role(rsc); pcmk__rsc_trace(rsc, "Creating all actions for %s transition from %s to %s " "(%s) on %s", rsc->id, pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role), next_role_source, pcmk__node_name(rsc->allocated_to)); current = rsc->fns->active_node(rsc, &num_all_active, &num_clean_active); g_list_foreach(rsc->dangling_migrations, pcmk__abort_dangling_migration, rsc); if ((current != NULL) && (rsc->allocated_to != NULL) && !pcmk__same_node(current, rsc->allocated_to) && (rsc->next_role >= pcmk_role_started)) { pcmk__rsc_trace(rsc, "Moving %s from %s to %s", rsc->id, pcmk__node_name(current), pcmk__node_name(rsc->allocated_to)); is_moving = true; allow_migrate = pcmk__rsc_can_migrate(rsc, current); // This is needed even if migrating (though I'm not sure why ...) need_stop = true; } // Check whether resource is partially migrated and/or multiply active if ((rsc->partial_migration_source != NULL) && (rsc->partial_migration_target != NULL) && allow_migrate && (num_all_active == 2) && pcmk__same_node(current, rsc->partial_migration_source) && pcmk__same_node(rsc->allocated_to, rsc->partial_migration_target)) { /* A partial migration is in progress, and the migration target remains * the same as when the migration began. */ pcmk__rsc_trace(rsc, "Partial migration of %s from %s to %s will continue", rsc->id, pcmk__node_name(rsc->partial_migration_source), pcmk__node_name(rsc->partial_migration_target)); } else if ((rsc->partial_migration_source != NULL) || (rsc->partial_migration_target != NULL)) { // A partial migration is in progress but can't be continued if (num_all_active > 2) { // The resource is migrating *and* multiply active! crm_notice("Forcing recovery of %s because it is migrating " "from %s to %s and possibly active elsewhere", rsc->id, pcmk__node_name(rsc->partial_migration_source), pcmk__node_name(rsc->partial_migration_target)); } else { // The migration source or target isn't available crm_notice("Forcing recovery of %s because it can no longer " "migrate from %s to %s", rsc->id, pcmk__node_name(rsc->partial_migration_source), pcmk__node_name(rsc->partial_migration_target)); } need_stop = true; rsc->partial_migration_source = rsc->partial_migration_target = NULL; allow_migrate = false; } else if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) { multiply_active = (num_all_active > 1); } else { /* If a resource has PCMK_META_REQUIRES set to PCMK_VALUE_NOTHING or * PCMK_VALUE_QUORUM, don't consider it active on unclean nodes (similar * to how all resources behave when PCMK_OPT_STONITH_ENABLED is false). * We can start such resources elsewhere before fencing completes, and * if we considered the resource active on the failed node, we would * attempt recovery for being active on multiple nodes. */ multiply_active = (num_clean_active > 1); } if (multiply_active) { const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS); // Resource was (possibly) incorrectly multiply active pcmk__sched_err("%s resource %s might be active on %u nodes (%s)", pcmk__s(class, "Untyped"), rsc->id, num_all_active, pcmk_multiply_active_text(rsc->recovery_type)); crm_notice("For more information, see \"What are multiply active " "resources?\" at " "https://projects.clusterlabs.org/w/clusterlabs/faq/"); switch (rsc->recovery_type) { case pcmk_multiply_active_restart: need_stop = true; break; case pcmk_multiply_active_unexpected: need_stop = true; // stop_resource() will skip expected node pcmk__set_rsc_flags(rsc, pcmk_rsc_stop_unexpected); break; default: break; } } else { pcmk__clear_rsc_flags(rsc, pcmk_rsc_stop_unexpected); } if (pcmk_is_set(rsc->flags, pcmk_rsc_start_pending)) { create_pending_start(rsc); } if (is_moving) { // Remaining tests are only for resources staying where they are } else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { if (pcmk_is_set(rsc->flags, pcmk_rsc_stop_if_failed)) { need_stop = true; pcmk__rsc_trace(rsc, "Recovering %s", rsc->id); } else { pcmk__rsc_trace(rsc, "Recovering %s by demotion", rsc->id); if (rsc->next_role == pcmk_role_promoted) { need_promote = true; } } } else if (pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) { pcmk__rsc_trace(rsc, "Blocking further actions on %s", rsc->id); need_stop = true; } else if ((rsc->role > pcmk_role_started) && (current != NULL) && (rsc->allocated_to != NULL)) { pcmk_action_t *start = NULL; pcmk__rsc_trace(rsc, "Creating start action for promoted resource %s", rsc->id); start = start_action(rsc, rsc->allocated_to, TRUE); if (!pcmk_is_set(start->flags, pcmk_action_optional)) { // Recovery of a promoted resource pcmk__rsc_trace(rsc, "%s restart is required for recovery", rsc->id); need_stop = true; } } // Create any actions needed to bring resource down and back up to same role schedule_restart_actions(rsc, current, need_stop, need_promote); // Create any actions needed to take resource from this role to the next schedule_role_transition_actions(rsc); pcmk__create_recurring_actions(rsc); if (allow_migrate) { pcmk__create_migration_actions(rsc, current); } } /*! * \internal * \brief Ban a resource from any allowed nodes that are Pacemaker Remote nodes * * \param[in] rsc Resource to check */ static void rsc_avoids_remote_nodes(const pcmk_resource_t *rsc) { GHashTableIter iter; pcmk_node_t *node = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (node->details->remote_rsc != NULL) { node->weight = -INFINITY; } } } /*! * \internal * \brief Return allowed nodes as (possibly sorted) list * * Convert a resource's hash table of allowed nodes to a list. If printing to * stdout, sort the list, to keep action ID numbers consistent for regression * test output (while avoiding the performance hit on a live cluster). * * \param[in] rsc Resource to check for allowed nodes * * \return List of resource's allowed nodes * \note Callers should take care not to rely on the list being sorted. */ static GList * allowed_nodes_as_list(const pcmk_resource_t *rsc) { GList *allowed_nodes = NULL; if (rsc->allowed_nodes) { allowed_nodes = g_hash_table_get_values(rsc->allowed_nodes); } if (!pcmk__is_daemon) { allowed_nodes = g_list_sort(allowed_nodes, pe__cmp_node_name); } return allowed_nodes; } /*! * \internal * \brief Create implicit constraints needed for a primitive resource * * \param[in,out] rsc Primitive resource to create implicit constraints for */ void pcmk__primitive_internal_constraints(pcmk_resource_t *rsc) { GList *allowed_nodes = NULL; bool check_unfencing = false; bool check_utilization = false; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk__rsc_trace(rsc, "Skipping implicit constraints for unmanaged resource " "%s", rsc->id); return; } // Whether resource requires unfencing check_unfencing = !pcmk_is_set(rsc->flags, pcmk_rsc_fence_device) && pcmk_is_set(rsc->cluster->flags, pcmk_sched_enable_unfencing) && pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing); // Whether a non-default placement strategy is used check_utilization = (g_hash_table_size(rsc->utilization) > 0) && !pcmk__str_eq(rsc->cluster->placement_strategy, PCMK_VALUE_DEFAULT, pcmk__str_casei); // Order stops before starts (i.e. restart) pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered |pcmk__ar_first_implies_then |pcmk__ar_intermediate_stop, rsc->cluster); // Promotable ordering: demote before stop, start before promote if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk_rsc_promotable) || (rsc->role > pcmk_role_unpromoted)) { pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_DEMOTE, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, pcmk__ar_promoted_then_implies_first, rsc->cluster); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_PROMOTE, 0), NULL, pcmk__ar_unrunnable_first_blocks, rsc->cluster); } // Don't clear resource history if probing on same node pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0), NULL, pcmk__ar_if_on_same_node|pcmk__ar_then_cancels_first, rsc->cluster); // Certain checks need allowed nodes if (check_unfencing || check_utilization || (rsc->container != NULL)) { allowed_nodes = allowed_nodes_as_list(rsc); } if (check_unfencing) { g_list_foreach(allowed_nodes, pcmk__order_restart_vs_unfence, rsc); } if (check_utilization) { pcmk__create_utilization_constraints(rsc, allowed_nodes); } if (rsc->container != NULL) { pcmk_resource_t *remote_rsc = NULL; if (rsc->is_remote_node) { // rsc is the implicit remote connection for a guest or bundle node /* Guest resources are not allowed to run on Pacemaker Remote nodes, * to avoid nesting remotes. However, bundles are allowed. */ if (!pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) { rsc_avoids_remote_nodes(rsc->container); } /* If someone cleans up a guest or bundle node's container, we will * likely schedule a (re-)probe of the container and recovery of the * connection. Order the connection stop after the container probe, * so that if we detect the container running, we will trigger a new * transition and avoid the unnecessary recovery. */ pcmk__order_resource_actions(rsc->container, PCMK_ACTION_MONITOR, rsc, PCMK_ACTION_STOP, pcmk__ar_ordered); /* A user can specify that a resource must start on a Pacemaker Remote * node by explicitly configuring it with the container=NODENAME * meta-attribute. This is of questionable merit, since location * constraints can accomplish the same thing. But we support it, so here * we check whether a resource (that is not itself a remote connection) * has container set to a remote node or guest node resource. */ } else if (rsc->container->is_remote_node) { remote_rsc = rsc->container; } else { remote_rsc = pe__resource_contains_guest_node(rsc->cluster, rsc->container); } if (remote_rsc != NULL) { /* Force the resource on the Pacemaker Remote node instead of * colocating the resource with the container resource. */ for (GList *item = allowed_nodes; item; item = item->next) { pcmk_node_t *node = item->data; if (node->details->remote_rsc != remote_rsc) { node->weight = -INFINITY; } } } else { /* This resource is either a filler for a container that does NOT * represent a Pacemaker Remote node, or a Pacemaker Remote * connection resource for a guest node or bundle. */ int score; crm_trace("Order and colocate %s relative to its container %s", rsc->id, rsc->container->id); pcmk__new_ordering(rsc->container, pcmk__op_key(rsc->container->id, PCMK_ACTION_START, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks, rsc->cluster); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, rsc->container, pcmk__op_key(rsc->container->id, PCMK_ACTION_STOP, 0), NULL, pcmk__ar_then_implies_first, rsc->cluster); if (pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) { score = 10000; /* Highly preferred but not essential */ } else { score = INFINITY; /* Force them to run on the same host */ } pcmk__new_colocation("#resource-with-container", NULL, score, rsc, rsc->container, NULL, NULL, pcmk__coloc_influence); } } if (rsc->is_remote_node || pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) { /* Remote connections and fencing devices are not allowed to run on * Pacemaker Remote nodes */ rsc_avoids_remote_nodes(rsc); } g_list_free(allowed_nodes); } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__primitive_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { enum pcmk__coloc_affects filter_results; CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if (for_dependent) { // Always process on behalf of primary resource primary->cmds->apply_coloc_score(dependent, primary, colocation, false); return; } filter_results = pcmk__colocation_affects(dependent, primary, colocation, false); pcmk__rsc_trace(dependent, "%s %s with %s (%s, score=%d, filter=%d)", ((colocation->score > 0)? "Colocating" : "Anti-colocating"), dependent->id, primary->id, colocation->id, colocation->score, filter_results); switch (filter_results) { case pcmk__coloc_affects_role: pcmk__apply_coloc_to_priority(dependent, primary, colocation); break; case pcmk__coloc_affects_location: pcmk__apply_coloc_to_scores(dependent, primary, colocation); break; default: // pcmk__coloc_affects_nothing return; } } /* Primitive implementation of * pcmk_assignment_methods_t:with_this_colocations() */ void pcmk__with_primitive_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (list != NULL)); if (rsc == orig_rsc) { /* For the resource itself, add all of its own colocations and relevant * colocations from its parent (if any). */ pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); if (rsc->parent != NULL) { rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list); } } else { // For an ancestor, add only explicitly configured constraints for (GList *iter = rsc->rsc_cons_lhs; iter != NULL; iter = iter->next) { pcmk__colocation_t *colocation = iter->data; if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) { pcmk__add_with_this(list, colocation, orig_rsc); } } } } /* Primitive implementation of * pcmk_assignment_methods_t:this_with_colocations() */ void pcmk__primitive_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (list != NULL)); if (rsc == orig_rsc) { /* For the resource itself, add all of its own colocations and relevant * colocations from its parent (if any). */ pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); if (rsc->parent != NULL) { rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list); } } else { // For an ancestor, add only explicitly configured constraints for (GList *iter = rsc->rsc_cons; iter != NULL; iter = iter->next) { pcmk__colocation_t *colocation = iter->data; if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) { pcmk__add_this_with(list, colocation, orig_rsc); } } } } /*! * \internal * \brief Return action flags for a given primitive resource action * * \param[in,out] action Action to get flags for * \param[in] node If not NULL, limit effects to this node (ignored) * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__primitive_action_flags(pcmk_action_t *action, const pcmk_node_t *node) { CRM_ASSERT(action != NULL); return (uint32_t) action->flags; } /*! * \internal * \brief Check whether a node is a multiply active resource's expected node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return \c true if \p rsc is multiply active with * \c PCMK_META_MULTIPLE_ACTIVE set to \c stop_unexpected, and \p node * is the node where it will remain active * \note This assumes that the resource's next role cannot be changed to stopped * after this is called, which should be reasonable if status has already * been unpacked and resources have been assigned to nodes. */ static bool is_expected_node(const pcmk_resource_t *rsc, const pcmk_node_t *node) { return pcmk_all_flags_set(rsc->flags, pcmk_rsc_stop_unexpected|pcmk_rsc_restarting) && (rsc->next_role > pcmk_role_stopped) && pcmk__same_node(rsc->allocated_to, node); } /*! * \internal * \brief Schedule actions needed to stop a resource wherever it is active * * \param[in,out] rsc Resource being stopped * \param[in] node Node where resource is being stopped (ignored) * \param[in] optional Whether actions should be optional */ static void stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *current = (pcmk_node_t *) iter->data; pcmk_action_t *stop = NULL; if (is_expected_node(rsc, current)) { /* We are scheduling restart actions for a multiply active resource * with PCMK_META_MULTIPLE_ACTIVE=stop_unexpected, and this is where * it should not be stopped. */ pcmk__rsc_trace(rsc, "Skipping stop of multiply active resource %s " "on expected node %s", rsc->id, pcmk__node_name(current)); continue; } if (rsc->partial_migration_target != NULL) { // Continue migration if node originally was and remains target if (pcmk__same_node(current, rsc->partial_migration_target) && pcmk__same_node(current, rsc->allocated_to)) { pcmk__rsc_trace(rsc, "Skipping stop of %s on %s " "because partial migration there will continue", rsc->id, pcmk__node_name(current)); continue; } else { pcmk__rsc_trace(rsc, "Forcing stop of %s on %s " "because migration target changed", rsc->id, pcmk__node_name(current)); optional = false; } } pcmk__rsc_trace(rsc, "Scheduling stop of %s on %s", rsc->id, pcmk__node_name(current)); stop = stop_action(rsc, current, optional); if (rsc->allocated_to == NULL) { pe_action_set_reason(stop, "node availability", true); } else if (pcmk_all_flags_set(rsc->flags, pcmk_rsc_restarting |pcmk_rsc_stop_unexpected)) { /* We are stopping a multiply active resource on a node that is * not its expected node, and we are still scheduling restart * actions, so the stop is for being multiply active. */ pe_action_set_reason(stop, "being multiply active", true); } if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { pcmk__clear_action_flags(stop, pcmk_action_runnable); } if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_remove_after_stop)) { pcmk__schedule_cleanup(rsc, current, optional); } if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing)) { pcmk_action_t *unfence = pe_fence_op(current, PCMK_ACTION_ON, true, NULL, false, rsc->cluster); order_actions(stop, unfence, pcmk__ar_then_implies_first); if (!pcmk__node_unfenced(current)) { pcmk__sched_err("Stopping %s until %s can be unfenced", rsc->id, pcmk__node_name(current)); } } } } /*! * \internal * \brief Schedule actions needed to start a resource on a node * * \param[in,out] rsc Resource being started * \param[in,out] node Node where resource should be started * \param[in] optional Whether actions should be optional */ static void start_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { pcmk_action_t *start = NULL; CRM_ASSERT(node != NULL); pcmk__rsc_trace(rsc, "Scheduling %s start of %s on %s (score %d)", (optional? "optional" : "required"), rsc->id, pcmk__node_name(node), node->weight); start = start_action(rsc, node, TRUE); pcmk__order_vs_unfence(rsc, node, start, pcmk__ar_first_implies_then); if (pcmk_is_set(start->flags, pcmk_action_runnable) && !optional) { pcmk__clear_action_flags(start, pcmk_action_optional); } if (is_expected_node(rsc, node)) { /* This could be a problem if the start becomes necessary for other * reasons later. */ pcmk__rsc_trace(rsc, "Start of multiply active resouce %s " "on expected node %s will be a pseudo-action", rsc->id, pcmk__node_name(node)); pcmk__set_action_flags(start, pcmk_action_pseudo); } } /*! * \internal * \brief Schedule actions needed to promote a resource on a node * * \param[in,out] rsc Resource being promoted * \param[in] node Node where resource should be promoted * \param[in] optional Whether actions should be optional */ static void promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { GList *iter = NULL; GList *action_list = NULL; bool runnable = true; CRM_ASSERT(node != NULL); // Any start must be runnable for promotion to be runnable action_list = pe__resource_actions(rsc, node, PCMK_ACTION_START, true); for (iter = action_list; iter != NULL; iter = iter->next) { pcmk_action_t *start = (pcmk_action_t *) iter->data; if (!pcmk_is_set(start->flags, pcmk_action_runnable)) { runnable = false; } } g_list_free(action_list); if (runnable) { pcmk_action_t *promote = promote_action(rsc, node, optional); pcmk__rsc_trace(rsc, "Scheduling %s promotion of %s on %s", (optional? "optional" : "required"), rsc->id, pcmk__node_name(node)); if (is_expected_node(rsc, node)) { /* This could be a problem if the promote becomes necessary for * other reasons later. */ pcmk__rsc_trace(rsc, "Promotion of multiply active resouce %s " "on expected node %s will be a pseudo-action", rsc->id, pcmk__node_name(node)); pcmk__set_action_flags(promote, pcmk_action_pseudo); } } else { pcmk__rsc_trace(rsc, "Not promoting %s on %s: start unrunnable", rsc->id, pcmk__node_name(node)); action_list = pe__resource_actions(rsc, node, PCMK_ACTION_PROMOTE, true); for (iter = action_list; iter != NULL; iter = iter->next) { pcmk_action_t *promote = (pcmk_action_t *) iter->data; pcmk__clear_action_flags(promote, pcmk_action_runnable); } g_list_free(action_list); } } /*! * \internal * \brief Schedule actions needed to demote a resource wherever it is active * * \param[in,out] rsc Resource being demoted * \param[in] node Node where resource should be demoted (ignored) * \param[in] optional Whether actions should be optional */ static void demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { /* Since this will only be called for a primitive (possibly as an instance * of a collective resource), the resource is multiply active if it is * running on more than one node, so we want to demote on all of them as * part of recovery, regardless of which one is the desired node. */ for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *current = (pcmk_node_t *) iter->data; if (is_expected_node(rsc, current)) { pcmk__rsc_trace(rsc, "Skipping demote of multiply active resource %s " "on expected node %s", rsc->id, pcmk__node_name(current)); } else { pcmk__rsc_trace(rsc, "Scheduling %s demotion of %s on %s", (optional? "optional" : "required"), rsc->id, pcmk__node_name(current)); demote_action(rsc, current, optional); } } } static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { CRM_ASSERT(false); } /*! * \internal * \brief Schedule cleanup of a resource * * \param[in,out] rsc Resource to clean up * \param[in] node Node to clean up on * \param[in] optional Whether clean-up should be optional */ void pcmk__schedule_cleanup(pcmk_resource_t *rsc, const pcmk_node_t *node, bool optional) { /* If the cleanup is required, its orderings are optional, because they're * relevant only if both actions are required. Conversely, if the cleanup is * optional, the orderings make the then action required if the first action * becomes required. */ uint32_t flag = optional? pcmk__ar_first_implies_then : pcmk__ar_ordered; CRM_CHECK((rsc != NULL) && (node != NULL), return); if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) { pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: resource failed", rsc->id, pcmk__node_name(node)); return; } if (node->details->unclean || !node->details->online) { pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: node unavailable", rsc->id, pcmk__node_name(node)); return; } crm_notice("Scheduling clean-up of %s on %s", rsc->id, pcmk__node_name(node)); delete_action(rsc, node, optional); // stop -> clean-up -> start pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, rsc, PCMK_ACTION_DELETE, flag); pcmk__order_resource_actions(rsc, PCMK_ACTION_DELETE, rsc, PCMK_ACTION_START, flag); } /*! * \internal * \brief Add primitive meta-attributes relevant to graph actions to XML * * \param[in] rsc Primitive resource whose meta-attributes should be added * \param[in,out] xml Transition graph action attributes XML to add to */ void pcmk__primitive_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml) { char *name = NULL; char *value = NULL; const pcmk_resource_t *parent = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (xml != NULL)); /* Clone instance numbers get set internally as meta-attributes, and are * needed in the transition graph (for example, to tell unique clone * instances apart). */ value = g_hash_table_lookup(rsc->meta, PCMK__META_CLONE); if (value != NULL) { name = crm_meta_name(PCMK__META_CLONE); crm_xml_add(xml, name, value); free(name); } // Not sure if this one is really needed ... value = g_hash_table_lookup(rsc->meta, PCMK_META_REMOTE_NODE); if (value != NULL) { name = crm_meta_name(PCMK_META_REMOTE_NODE); crm_xml_add(xml, name, value); free(name); } /* The container meta-attribute can be set on the primitive itself or one of * its parents (for example, a group inside a container resource), so check * them all, and keep the highest one found. */ for (parent = rsc; parent != NULL; parent = parent->parent) { if (parent->container != NULL) { crm_xml_add(xml, CRM_META "_" PCMK__META_CONTAINER, parent->container->id); } } /* Bundle replica children will get their external-ip set internally as a * meta-attribute. The graph action needs it, but under a different naming * convention than other meta-attributes. */ value = g_hash_table_lookup(rsc->meta, "external-ip"); if (value != NULL) { crm_xml_add(xml, "pcmk_external_ip", value); } } // Primitive implementation of pcmk_assignment_methods_t:add_utilization() void pcmk__primitive_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (orig_rsc != NULL) && (utilization != NULL)); if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) { return; } pcmk__rsc_trace(orig_rsc, "%s: Adding primitive %s as colocated utilization", orig_rsc->id, rsc->id); pcmk__release_node_capacity(utilization, rsc); } /*! * \internal * \brief Get epoch time of node's shutdown attribute (or now if none) * * \param[in,out] node Node to check * * \return Epoch time corresponding to shutdown attribute if set or now if not */ static time_t shutdown_time(pcmk_node_t *node) { const char *shutdown = pe_node_attribute_raw(node, PCMK__NODE_ATTR_SHUTDOWN); time_t result = 0; if (shutdown != NULL) { long long result_ll; if (pcmk__scan_ll(shutdown, &result_ll, 0LL) == pcmk_rc_ok) { result = (time_t) result_ll; } } return (result == 0)? get_effective_time(node->details->data_set) : result; } /*! * \internal * \brief Ban a resource from a node if it's not locked to the node * * \param[in] data Node to check * \param[in,out] user_data Resource to check */ static void ban_if_not_locked(gpointer data, gpointer user_data) { const pcmk_node_t *node = (const pcmk_node_t *) data; pcmk_resource_t *rsc = (pcmk_resource_t *) user_data; if (strcmp(node->details->uname, rsc->lock_node->details->uname) != 0) { resource_location(rsc, node, -CRM_SCORE_INFINITY, PCMK_OPT_SHUTDOWN_LOCK, rsc->cluster); } } // Primitive implementation of pcmk_assignment_methods_t:shutdown_lock() void pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc) { const char *class = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); class = crm_element_value(rsc->xml, PCMK_XA_CLASS); // Fence devices and remote connections can't be locked if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches) || pe__resource_is_remote_conn(rsc)) { return; } if (rsc->lock_node != NULL) { // The lock was obtained from resource history if (rsc->running_on != NULL) { /* The resource was started elsewhere even though it is now * considered locked. This shouldn't be possible, but as a * failsafe, we don't want to disturb the resource now. */ pcmk__rsc_info(rsc, "Cancelling shutdown lock " "because %s is already active", rsc->id); pe__clear_resource_history(rsc, rsc->lock_node); rsc->lock_node = NULL; rsc->lock_time = 0; } // Only a resource active on exactly one node can be locked } else if (pcmk__list_of_1(rsc->running_on)) { pcmk_node_t *node = rsc->running_on->data; if (node->details->shutdown) { if (node->details->unclean) { pcmk__rsc_debug(rsc, "Not locking %s to unclean %s for shutdown", rsc->id, pcmk__node_name(node)); } else { rsc->lock_node = node; rsc->lock_time = shutdown_time(node); } } } if (rsc->lock_node == NULL) { // No lock needed return; } if (rsc->cluster->shutdown_lock > 0) { time_t lock_expiration = rsc->lock_time + rsc->cluster->shutdown_lock; pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)", rsc->id, pcmk__node_name(rsc->lock_node), (long long) lock_expiration); pe__update_recheck_time(++lock_expiration, rsc->cluster, "shutdown lock expiration"); } else { pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown", rsc->id, pcmk__node_name(rsc->lock_node)); } // If resource is locked to one node, ban it from all other nodes g_list_foreach(rsc->cluster->nodes, ban_if_not_locked, rsc); } diff --git a/tools/crm_resource_runtime.c b/tools/crm_resource_runtime.c index 8d5742ef44..8c867094f8 100644 --- a/tools/crm_resource_runtime.c +++ b/tools/crm_resource_runtime.c @@ -1,2265 +1,2265 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include static GList * build_node_info_list(const pcmk_resource_t *rsc) { GList *retval = NULL; for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *child = (const pcmk_resource_t *) iter->data; for (const GList *iter2 = child->running_on; iter2 != NULL; iter2 = iter2->next) { const pcmk_node_t *node = (const pcmk_node_t *) iter2->data; node_info_t *ni = calloc(1, sizeof(node_info_t)); ni->node_name = node->details->uname; ni->promoted = pcmk_is_set(rsc->flags, pcmk_rsc_promotable) && child->fns->state(child, TRUE) == pcmk_role_promoted; retval = g_list_prepend(retval, ni); } } return retval; } GList * cli_resource_search(pcmk_resource_t *rsc, const char *requested_name, pcmk_scheduler_t *scheduler) { GList *retval = NULL; const pcmk_resource_t *parent = pe__const_top_resource(rsc, false); if (pcmk__is_clone(rsc)) { retval = build_node_info_list(rsc); /* The anonymous clone children's common ID is supplied */ } else if (pcmk__is_clone(parent) && !pcmk_is_set(rsc->flags, pcmk_rsc_unique) && rsc->clone_name && pcmk__str_eq(requested_name, rsc->clone_name, pcmk__str_casei) && !pcmk__str_eq(requested_name, rsc->id, pcmk__str_casei)) { retval = build_node_info_list(parent); } else if (rsc->running_on != NULL) { for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; node_info_t *ni = calloc(1, sizeof(node_info_t)); ni->node_name = node->details->uname; ni->promoted = (rsc->fns->state(rsc, TRUE) == pcmk_role_promoted); retval = g_list_prepend(retval, ni); } } return retval; } // \return Standard Pacemaker return code static int find_resource_attr(pcmk__output_t *out, cib_t * the_cib, const char *attr, const char *rsc, const char *attr_set_type, const char *set_name, const char *attr_id, const char *attr_name, char **value) { int rc = pcmk_rc_ok; xmlNode *xml_search = NULL; GString *xpath = NULL; const char *xpath_base = NULL; if(value) { *value = NULL; } if(the_cib == NULL) { return ENOTCONN; } xpath_base = pcmk_cib_xpath_for(PCMK_XE_RESOURCES); if (xpath_base == NULL) { crm_err(PCMK_XE_RESOURCES " CIB element not known (bug?)"); return ENOMSG; } xpath = g_string_sized_new(1024); pcmk__g_strcat(xpath, xpath_base, "//*[@" PCMK_XA_ID "=\"", rsc, "\"]", NULL); if (attr_set_type != NULL) { pcmk__g_strcat(xpath, "/", attr_set_type, NULL); if (set_name != NULL) { pcmk__g_strcat(xpath, "[@" PCMK_XA_ID "=\"", set_name, "\"]", NULL); } } g_string_append(xpath, "//" PCMK_XE_NVPAIR "["); if (attr_id != NULL) { pcmk__g_strcat(xpath, "@" PCMK_XA_ID "=\"", attr_id, "\"", NULL); } if (attr_name != NULL) { if (attr_id != NULL) { g_string_append(xpath, " and "); } pcmk__g_strcat(xpath, "@" PCMK_XA_NAME "=\"", attr_name, "\"", NULL); } g_string_append_c(xpath, ']'); rc = the_cib->cmds->query(the_cib, (const char *) xpath->str, &xml_search, cib_sync_call | cib_scope_local | cib_xpath); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { goto done; } crm_log_xml_debug(xml_search, "Match"); if (xml_search->children != NULL) { xmlNode *child = NULL; rc = ENOTUNIQ; out->info(out, "Multiple attributes match name=%s", attr_name); for (child = pcmk__xml_first_child(xml_search); child != NULL; child = pcmk__xml_next(child)) { out->info(out, " Value: %s \t(id=%s)", crm_element_value(child, PCMK_XA_VALUE), ID(child)); } out->spacer(out); } else if(value) { pcmk__str_update(value, crm_element_value(xml_search, attr)); } done: g_string_free(xpath, TRUE); free_xml(xml_search); return rc; } /* PRIVATE. Use the find_matching_attr_resources instead. */ static void find_matching_attr_resources_recursive(pcmk__output_t *out, GList /* */ **result, pcmk_resource_t *rsc, const char *rsc_id, const char * attr_set, const char * attr_set_type, const char * attr_id, const char * attr_name, cib_t * cib, const char * cmd, int depth) { int rc = pcmk_rc_ok; char *lookup_id = clone_strip(rsc->id); char *local_attr_id = NULL; /* visit the children */ for(GList *gIter = rsc->children; gIter; gIter = gIter->next) { find_matching_attr_resources_recursive(out, result, (pcmk_resource_t *) gIter->data, rsc_id, attr_set, attr_set_type, attr_id, attr_name, cib, cmd, depth+1); /* do it only once for clones */ if (rsc->variant == pcmk_rsc_variant_clone) { break; } } rc = find_resource_attr(out, cib, PCMK_XA_ID, lookup_id, attr_set_type, attr_set, attr_id, attr_name, &local_attr_id); /* Post-order traversal. * The root is always on the list and it is the last item. */ if((0 == depth) || (pcmk_rc_ok == rc)) { /* push the head */ *result = g_list_append(*result, rsc); } free(local_attr_id); free(lookup_id); } /* The result is a linearized pre-ordered tree of resources. */ static GList/**/ * find_matching_attr_resources(pcmk__output_t *out, pcmk_resource_t *rsc, const char * rsc_id, const char * attr_set, const char * attr_set_type, const char * attr_id, const char * attr_name, cib_t * cib, const char * cmd, gboolean force) { int rc = pcmk_rc_ok; char *lookup_id = NULL; char *local_attr_id = NULL; GList * result = NULL; /* If --force is used, update only the requested resource (clone or primitive). * Otherwise, if the primitive has the attribute, use that. * Otherwise use the clone. */ if(force == TRUE) { return g_list_append(result, rsc); } if ((rsc->parent != NULL) && (rsc->parent->variant == pcmk_rsc_variant_clone)) { int rc = pcmk_rc_ok; char *local_attr_id = NULL; rc = find_resource_attr(out, cib, PCMK_XA_ID, rsc_id, attr_set_type, attr_set, attr_id, attr_name, &local_attr_id); free(local_attr_id); if(rc != pcmk_rc_ok) { rsc = rsc->parent; out->info(out, "Performing %s of '%s' on '%s', the parent of '%s'", cmd, attr_name, rsc->id, rsc_id); } return g_list_append(result, rsc); } else if ((rsc->parent == NULL) && (rsc->children != NULL) && (rsc->variant == pcmk_rsc_variant_clone)) { pcmk_resource_t *child = rsc->children->data; if (child->variant == pcmk_rsc_variant_primitive) { lookup_id = clone_strip(child->id); /* Could be a cloned group! */ rc = find_resource_attr(out, cib, PCMK_XA_ID, lookup_id, attr_set_type, attr_set, attr_id, attr_name, &local_attr_id); if(rc == pcmk_rc_ok) { rsc = child; out->info(out, "A value for '%s' already exists in child '%s', performing %s on that instead of '%s'", attr_name, lookup_id, cmd, rsc_id); } free(local_attr_id); free(lookup_id); } return g_list_append(result, rsc); } /* If the resource is a group ==> children inherit the attribute if defined. */ find_matching_attr_resources_recursive(out, &result, rsc, rsc_id, attr_set, attr_set_type, attr_id, attr_name, cib, cmd, 0); return result; } // \return Standard Pacemaker return code int cli_resource_update_attribute(pcmk_resource_t *rsc, const char *requested_name, const char *attr_set, const char *attr_set_type, const char *attr_id, const char *attr_name, const char *attr_value, gboolean recursive, cib_t *cib, int cib_options, gboolean force) { pcmk__output_t *out = rsc->cluster->priv; int rc = pcmk_rc_ok; char *found_attr_id = NULL; GList/**/ *resources = NULL; const char *top_id = pe__const_top_resource(rsc, false)->id; if ((attr_id == NULL) && !force) { find_resource_attr(out, cib, PCMK_XA_ID, top_id, NULL, NULL, NULL, attr_name, NULL); } if (pcmk__str_eq(attr_set_type, PCMK_XE_INSTANCE_ATTRIBUTES, pcmk__str_casei)) { if (!force) { rc = find_resource_attr(out, cib, PCMK_XA_ID, top_id, PCMK_XE_META_ATTRIBUTES, attr_set, attr_id, attr_name, &found_attr_id); if ((rc == pcmk_rc_ok) && !out->is_quiet(out)) { out->err(out, "WARNING: There is already a meta attribute " "for '%s' called '%s' (id=%s)", top_id, attr_name, found_attr_id); out->err(out, " Delete '%s' first or use the force option " "to override", found_attr_id); } free(found_attr_id); if (rc == pcmk_rc_ok) { return ENOTUNIQ; } } resources = g_list_append(resources, rsc); } else if (pcmk__str_eq(attr_set_type, ATTR_SET_ELEMENT, pcmk__str_none)) { crm_xml_add(rsc->xml, attr_name, attr_value); CRM_ASSERT(cib != NULL); rc = cib->cmds->replace(cib, PCMK_XE_RESOURCES, rsc->xml, cib_options); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { out->info(out, "Set attribute: name=%s value=%s", attr_name, attr_value); } return rc; } else { resources = find_matching_attr_resources(out, rsc, requested_name, attr_set, attr_set_type, attr_id, attr_name, cib, "update", force); } /* If the user specified attr_set or attr_id, the intent is to modify a * single resource, which will be the last item in the list. */ if ((attr_set != NULL) || (attr_id != NULL)) { GList *last = g_list_last(resources); resources = g_list_remove_link(resources, last); g_list_free(resources); resources = last; } for (GList *iter = resources; iter != NULL; iter = iter->next) { char *lookup_id = NULL; char *local_attr_set = NULL; const char *rsc_attr_id = attr_id; const char *rsc_attr_set = attr_set; xmlNode *xml_top = NULL; xmlNode *xml_obj = NULL; found_attr_id = NULL; rsc = (pcmk_resource_t *) iter->data; lookup_id = clone_strip(rsc->id); /* Could be a cloned group! */ rc = find_resource_attr(out, cib, PCMK_XA_ID, lookup_id, attr_set_type, attr_set, attr_id, attr_name, &found_attr_id); switch (rc) { case pcmk_rc_ok: crm_debug("Found a match for name=%s: id=%s", attr_name, found_attr_id); rsc_attr_id = found_attr_id; break; case ENXIO: if (rsc_attr_set == NULL) { local_attr_set = crm_strdup_printf("%s-%s", lookup_id, attr_set_type); rsc_attr_set = local_attr_set; } if (rsc_attr_id == NULL) { found_attr_id = crm_strdup_printf("%s-%s", rsc_attr_set, attr_name); rsc_attr_id = found_attr_id; } xml_top = create_xml_node(NULL, (const char *) rsc->xml->name); crm_xml_add(xml_top, PCMK_XA_ID, lookup_id); xml_obj = create_xml_node(xml_top, attr_set_type); crm_xml_add(xml_obj, PCMK_XA_ID, rsc_attr_set); break; default: free(lookup_id); free(found_attr_id); g_list_free(resources); return rc; } xml_obj = crm_create_nvpair_xml(xml_obj, rsc_attr_id, attr_name, attr_value); if (xml_top == NULL) { xml_top = xml_obj; } crm_log_xml_debug(xml_top, "Update"); rc = cib->cmds->modify(cib, PCMK_XE_RESOURCES, xml_top, cib_options); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { out->info(out, "Set '%s' option: id=%s%s%s%s%s value=%s", lookup_id, found_attr_id, ((rsc_attr_set == NULL)? "" : " set="), pcmk__s(rsc_attr_set, ""), ((attr_name == NULL)? "" : " name="), pcmk__s(attr_name, ""), attr_value); } free_xml(xml_top); free(lookup_id); free(found_attr_id); free(local_attr_set); if (recursive && pcmk__str_eq(attr_set_type, PCMK_XE_META_ATTRIBUTES, pcmk__str_casei)) { GList *lpc = NULL; static bool need_init = true; if (need_init) { need_init = false; pcmk__unpack_constraints(rsc->cluster); pe__clear_resource_flags_on_all(rsc->cluster, pcmk_rsc_detect_loop); } /* We want to set the attribute only on resources explicitly * colocated with this one, so we use rsc->rsc_cons_lhs directly * rather than the with_this_colocations() method. */ pcmk__set_rsc_flags(rsc, pcmk_rsc_detect_loop); for (lpc = rsc->rsc_cons_lhs; lpc != NULL; lpc = lpc->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data; crm_debug("Checking %s %d", cons->id, cons->score); if (!pcmk_is_set(cons->dependent->flags, pcmk_rsc_detect_loop) && (cons->score > 0)) { crm_debug("Setting %s=%s for dependent resource %s", attr_name, attr_value, cons->dependent->id); cli_resource_update_attribute(cons->dependent, cons->dependent->id, NULL, attr_set_type, NULL, attr_name, attr_value, recursive, cib, cib_options, force); } } } } g_list_free(resources); return rc; } // \return Standard Pacemaker return code int cli_resource_delete_attribute(pcmk_resource_t *rsc, const char *requested_name, const char *attr_set, const char *attr_set_type, const char *attr_id, const char *attr_name, cib_t *cib, int cib_options, gboolean force) { pcmk__output_t *out = rsc->cluster->priv; int rc = pcmk_rc_ok; GList/**/ *resources = NULL; if ((attr_id == NULL) && !force) { find_resource_attr(out, cib, PCMK_XA_ID, pe__const_top_resource(rsc, false)->id, NULL, NULL, NULL, attr_name, NULL); } if (pcmk__str_eq(attr_set_type, PCMK_XE_META_ATTRIBUTES, pcmk__str_casei)) { resources = find_matching_attr_resources(out, rsc, requested_name, attr_set, attr_set_type, attr_id, attr_name, cib, "delete", force); } else if (pcmk__str_eq(attr_set_type, ATTR_SET_ELEMENT, pcmk__str_none)) { xml_remove_prop(rsc->xml, attr_name); CRM_ASSERT(cib != NULL); rc = cib->cmds->replace(cib, PCMK_XE_RESOURCES, rsc->xml, cib_options); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { out->info(out, "Deleted attribute: %s", attr_name); } return rc; } else { resources = g_list_append(resources, rsc); } for (GList *iter = resources; iter != NULL; iter = iter->next) { char *lookup_id = NULL; xmlNode *xml_obj = NULL; char *found_attr_id = NULL; const char *rsc_attr_id = attr_id; rsc = (pcmk_resource_t *) iter->data; lookup_id = clone_strip(rsc->id); rc = find_resource_attr(out, cib, PCMK_XA_ID, lookup_id, attr_set_type, attr_set, attr_id, attr_name, &found_attr_id); switch (rc) { case pcmk_rc_ok: break; case ENXIO: free(lookup_id); rc = pcmk_rc_ok; continue; default: free(lookup_id); g_list_free(resources); return rc; } if (rsc_attr_id == NULL) { rsc_attr_id = found_attr_id; } xml_obj = crm_create_nvpair_xml(NULL, rsc_attr_id, attr_name, NULL); crm_log_xml_debug(xml_obj, "Delete"); CRM_ASSERT(cib); rc = cib->cmds->remove(cib, PCMK_XE_RESOURCES, xml_obj, cib_options); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { out->info(out, "Deleted '%s' option: id=%s%s%s%s%s", lookup_id, found_attr_id, ((attr_set == NULL)? "" : " set="), pcmk__s(attr_set, ""), ((attr_name == NULL)? "" : " name="), pcmk__s(attr_name, "")); } free(lookup_id); free_xml(xml_obj); free(found_attr_id); } g_list_free(resources); return rc; } // \return Standard Pacemaker return code static int send_lrm_rsc_op(pcmk_ipc_api_t *controld_api, bool do_fail_resource, const char *host_uname, const char *rsc_id, pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv; const char *router_node = host_uname; const char *rsc_api_id = NULL; const char *rsc_long_id = NULL; const char *rsc_class = NULL; const char *rsc_provider = NULL; const char *rsc_type = NULL; bool cib_only = false; pcmk_resource_t *rsc = pe_find_resource(scheduler->resources, rsc_id); if (rsc == NULL) { out->err(out, "Resource %s not found", rsc_id); return ENXIO; } else if (rsc->variant != pcmk_rsc_variant_primitive) { out->err(out, "We can only process primitive resources, not %s", rsc_id); return EINVAL; } rsc_class = crm_element_value(rsc->xml, PCMK_XA_CLASS); rsc_provider = crm_element_value(rsc->xml, PCMK_XA_PROVIDER), rsc_type = crm_element_value(rsc->xml, PCMK_XA_TYPE); if ((rsc_class == NULL) || (rsc_type == NULL)) { out->err(out, "Resource %s does not have a class and type", rsc_id); return EINVAL; } { pcmk_node_t *node = pe_find_node(scheduler->nodes, host_uname); if (node == NULL) { out->err(out, "Node %s not found", host_uname); return pcmk_rc_node_unknown; } if (!(node->details->online)) { if (do_fail_resource) { out->err(out, "Node %s is not online", host_uname); return ENOTCONN; } else { cib_only = true; } } if (!cib_only && pe__is_guest_or_remote_node(node)) { node = pcmk__current_node(node->details->remote_rsc); if (node == NULL) { out->err(out, "No cluster connection to Pacemaker Remote node %s detected", host_uname); return ENOTCONN; } router_node = node->details->uname; } } if (rsc->clone_name) { rsc_api_id = rsc->clone_name; rsc_long_id = rsc->id; } else { rsc_api_id = rsc->id; } if (do_fail_resource) { return pcmk_controld_api_fail(controld_api, host_uname, router_node, rsc_api_id, rsc_long_id, rsc_class, rsc_provider, rsc_type); } else { return pcmk_controld_api_refresh(controld_api, host_uname, router_node, rsc_api_id, rsc_long_id, rsc_class, rsc_provider, rsc_type, cib_only); } } /*! * \internal * \brief Get resource name as used in failure-related node attributes * * \param[in] rsc Resource to check * * \return Newly allocated string containing resource's fail name * \note The caller is responsible for freeing the result. */ static inline char * rsc_fail_name(const pcmk_resource_t *rsc) { const char *name = (rsc->clone_name? rsc->clone_name : rsc->id); if (pcmk_is_set(rsc->flags, pcmk_rsc_unique)) { return strdup(name); } return clone_strip(name); } // \return Standard Pacemaker return code static int clear_rsc_history(pcmk_ipc_api_t *controld_api, const char *host_uname, const char *rsc_id, pcmk_scheduler_t *scheduler) { int rc = pcmk_rc_ok; /* Erase the resource's entire LRM history in the CIB, even if we're only * clearing a single operation's fail count. If we erased only entries for a * single operation, we might wind up with a wrong idea of the current * resource state, and we might not re-probe the resource. */ rc = send_lrm_rsc_op(controld_api, false, host_uname, rsc_id, scheduler); if (rc != pcmk_rc_ok) { return rc; } crm_trace("Processing %d mainloop inputs", pcmk_controld_api_replies_expected(controld_api)); while (g_main_context_iteration(NULL, FALSE)) { crm_trace("Processed mainloop input, %d still remaining", pcmk_controld_api_replies_expected(controld_api)); } return rc; } // \return Standard Pacemaker return code static int clear_rsc_failures(pcmk__output_t *out, pcmk_ipc_api_t *controld_api, const char *node_name, const char *rsc_id, const char *operation, const char *interval_spec, pcmk_scheduler_t *scheduler) { int rc = pcmk_rc_ok; const char *failed_value = NULL; const char *failed_id = NULL; char *interval_ms_s = NULL; GHashTable *rscs = NULL; GHashTableIter iter; /* Create a hash table to use as a set of resources to clean. This lets us * clean each resource only once (per node) regardless of how many failed * operations it has. */ rscs = pcmk__strkey_table(NULL, NULL); // Normalize interval to milliseconds for comparison to history entry if (operation) { guint interval_ms = 0U; pcmk_parse_interval_spec(interval_spec, &interval_ms); interval_ms_s = crm_strdup_printf("%u", interval_ms); } for (xmlNode *xml_op = pcmk__xml_first_child(scheduler->failed); xml_op != NULL; xml_op = pcmk__xml_next(xml_op)) { failed_id = crm_element_value(xml_op, PCMK__XA_RSC_ID); if (failed_id == NULL) { // Malformed history entry, should never happen continue; } // No resource specified means all resources match if (rsc_id) { pcmk_resource_t *fail_rsc = NULL; fail_rsc = pe_find_resource_with_flags(scheduler->resources, failed_id, pcmk_rsc_match_history |pcmk_rsc_match_anon_basename); if (!fail_rsc || !pcmk__str_eq(rsc_id, fail_rsc->id, pcmk__str_casei)) { continue; } } // Host name should always have been provided by this point failed_value = crm_element_value(xml_op, PCMK_XA_UNAME); if (!pcmk__str_eq(node_name, failed_value, pcmk__str_casei)) { continue; } // No operation specified means all operations match if (operation) { failed_value = crm_element_value(xml_op, PCMK_XA_OPERATION); if (!pcmk__str_eq(operation, failed_value, pcmk__str_casei)) { continue; } // Interval (if operation was specified) defaults to 0 (not all) failed_value = crm_element_value(xml_op, PCMK_META_INTERVAL); if (!pcmk__str_eq(interval_ms_s, failed_value, pcmk__str_casei)) { continue; } } g_hash_table_add(rscs, (gpointer) failed_id); } free(interval_ms_s); g_hash_table_iter_init(&iter, rscs); while (g_hash_table_iter_next(&iter, (gpointer *) &failed_id, NULL)) { crm_debug("Erasing failures of %s on %s", failed_id, node_name); rc = clear_rsc_history(controld_api, node_name, failed_id, scheduler); if (rc != pcmk_rc_ok) { return rc; } } g_hash_table_destroy(rscs); return rc; } // \return Standard Pacemaker return code static int clear_rsc_fail_attrs(const pcmk_resource_t *rsc, const char *operation, const char *interval_spec, const pcmk_node_t *node) { int rc = pcmk_rc_ok; int attr_options = pcmk__node_attr_none; char *rsc_name = rsc_fail_name(rsc); if (pe__is_guest_or_remote_node(node)) { attr_options |= pcmk__node_attr_remote; } rc = pcmk__attrd_api_clear_failures(NULL, node->details->uname, rsc_name, operation, interval_spec, NULL, attr_options); free(rsc_name); return rc; } // \return Standard Pacemaker return code int cli_resource_delete(pcmk_ipc_api_t *controld_api, const char *host_uname, const pcmk_resource_t *rsc, const char *operation, const char *interval_spec, bool just_failures, pcmk_scheduler_t *scheduler, gboolean force) { pcmk__output_t *out = scheduler->priv; int rc = pcmk_rc_ok; pcmk_node_t *node = NULL; if (rsc == NULL) { return ENXIO; } else if (rsc->children) { for (const GList *lpc = rsc->children; lpc != NULL; lpc = lpc->next) { const pcmk_resource_t *child = (const pcmk_resource_t *) lpc->data; rc = cli_resource_delete(controld_api, host_uname, child, operation, interval_spec, just_failures, scheduler, force); if (rc != pcmk_rc_ok) { return rc; } } return pcmk_rc_ok; } else if (host_uname == NULL) { GList *lpc = NULL; GList *nodes = g_hash_table_get_values(rsc->known_on); if(nodes == NULL && force) { nodes = pcmk__copy_node_list(scheduler->nodes, false); } else if(nodes == NULL && rsc->exclusive_discover) { GHashTableIter iter; pcmk_node_t *node = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void**)&node)) { if(node->weight >= 0) { nodes = g_list_prepend(nodes, node); } } } else if(nodes == NULL) { nodes = g_hash_table_get_values(rsc->allowed_nodes); } for (lpc = nodes; lpc != NULL; lpc = lpc->next) { node = (pcmk_node_t *) lpc->data; if (node->details->online) { rc = cli_resource_delete(controld_api, node->details->uname, rsc, operation, interval_spec, just_failures, scheduler, force); } if (rc != pcmk_rc_ok) { g_list_free(nodes); return rc; } } g_list_free(nodes); return pcmk_rc_ok; } node = pe_find_node(scheduler->nodes, host_uname); if (node == NULL) { out->err(out, "Unable to clean up %s because node %s not found", rsc->id, host_uname); return ENODEV; } if (!node->details->rsc_discovery_enabled) { out->err(out, "Unable to clean up %s because resource discovery disabled on %s", rsc->id, host_uname); return EOPNOTSUPP; } if (controld_api == NULL) { out->err(out, "Dry run: skipping clean-up of %s on %s due to CIB_file", rsc->id, host_uname); return pcmk_rc_ok; } rc = clear_rsc_fail_attrs(rsc, operation, interval_spec, node); if (rc != pcmk_rc_ok) { out->err(out, "Unable to clean up %s failures on %s: %s", rsc->id, host_uname, pcmk_rc_str(rc)); return rc; } if (just_failures) { rc = clear_rsc_failures(out, controld_api, host_uname, rsc->id, operation, interval_spec, scheduler); } else { rc = clear_rsc_history(controld_api, host_uname, rsc->id, scheduler); } if (rc != pcmk_rc_ok) { out->err(out, "Cleaned %s failures on %s, but unable to clean history: %s", rsc->id, host_uname, pcmk_rc_str(rc)); } else { out->info(out, "Cleaned up %s on %s", rsc->id, host_uname); } return rc; } // \return Standard Pacemaker return code int cli_cleanup_all(pcmk_ipc_api_t *controld_api, const char *node_name, const char *operation, const char *interval_spec, pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv; int rc = pcmk_rc_ok; int attr_options = pcmk__node_attr_none; const char *display_name = node_name? node_name : "all nodes"; if (controld_api == NULL) { out->info(out, "Dry run: skipping clean-up of %s due to CIB_file", display_name); return rc; } if (node_name) { pcmk_node_t *node = pe_find_node(scheduler->nodes, node_name); if (node == NULL) { out->err(out, "Unknown node: %s", node_name); return ENXIO; } if (pe__is_guest_or_remote_node(node)) { attr_options |= pcmk__node_attr_remote; } } rc = pcmk__attrd_api_clear_failures(NULL, node_name, NULL, operation, interval_spec, NULL, attr_options); if (rc != pcmk_rc_ok) { out->err(out, "Unable to clean up all failures on %s: %s", display_name, pcmk_rc_str(rc)); return rc; } if (node_name) { rc = clear_rsc_failures(out, controld_api, node_name, NULL, operation, interval_spec, scheduler); if (rc != pcmk_rc_ok) { out->err(out, "Cleaned all resource failures on %s, but unable to clean history: %s", node_name, pcmk_rc_str(rc)); return rc; } } else { for (GList *iter = scheduler->nodes; iter; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; rc = clear_rsc_failures(out, controld_api, node->details->uname, NULL, operation, interval_spec, scheduler); if (rc != pcmk_rc_ok) { out->err(out, "Cleaned all resource failures on all nodes, but unable to clean history: %s", pcmk_rc_str(rc)); return rc; } } } out->info(out, "Cleaned up all resources on %s", display_name); return rc; } static void check_role(resource_checks_t *checks) { const char *role_s = g_hash_table_lookup(checks->rsc->meta, PCMK_META_TARGET_ROLE); if (role_s == NULL) { return; } switch (pcmk_parse_role(role_s)) { case pcmk_role_stopped: checks->flags |= rsc_remain_stopped; break; case pcmk_role_unpromoted: if (pcmk_is_set(pe__const_top_resource(checks->rsc, false)->flags, pcmk_rsc_promotable)) { checks->flags |= rsc_unpromotable; } break; default: break; } } static void check_managed(resource_checks_t *checks) { const char *managed_s = g_hash_table_lookup(checks->rsc->meta, PCMK_META_IS_MANAGED); if ((managed_s != NULL) && !crm_is_true(managed_s)) { checks->flags |= rsc_unmanaged; } } static void check_locked(resource_checks_t *checks) { if (checks->rsc->lock_node != NULL) { checks->flags |= rsc_locked; checks->lock_node = checks->rsc->lock_node->details->uname; } } static bool node_is_unhealthy(pcmk_node_t *node) { switch (pe__health_strategy(node->details->data_set)) { case pcmk__health_strategy_none: break; case pcmk__health_strategy_no_red: if (pe__node_health(node) < 0) { return true; } break; case pcmk__health_strategy_only_green: if (pe__node_health(node) <= 0) { return true; } break; case pcmk__health_strategy_progressive: case pcmk__health_strategy_custom: /* @TODO These are finite scores, possibly with rules, and possibly * combining with other scores, so attributing these as a cause is * nontrivial. */ break; } return false; } static void check_node_health(resource_checks_t *checks, pcmk_node_t *node) { if (node == NULL) { GHashTableIter iter; bool allowed = false; bool all_nodes_unhealthy = true; g_hash_table_iter_init(&iter, checks->rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { allowed = true; if (!node_is_unhealthy(node)) { all_nodes_unhealthy = false; break; } } if (allowed && all_nodes_unhealthy) { checks->flags |= rsc_node_health; } } else if (node_is_unhealthy(node)) { checks->flags |= rsc_node_health; } } int cli_resource_check(pcmk__output_t *out, pcmk_resource_t *rsc, pcmk_node_t *node) { resource_checks_t checks = { .rsc = rsc }; check_role(&checks); check_managed(&checks); check_locked(&checks); check_node_health(&checks, node); return out->message(out, "resource-check-list", &checks); } // \return Standard Pacemaker return code int cli_resource_fail(pcmk_ipc_api_t *controld_api, const char *host_uname, const char *rsc_id, pcmk_scheduler_t *scheduler) { crm_notice("Failing %s on %s", rsc_id, host_uname); return send_lrm_rsc_op(controld_api, true, host_uname, rsc_id, scheduler); } static GHashTable * generate_resource_params(pcmk_resource_t *rsc, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { GHashTable *params = NULL; GHashTable *meta = NULL; GHashTable *combined = NULL; GHashTableIter iter; char *key = NULL; char *value = NULL; combined = pcmk__strkey_table(free, free); params = pe_rsc_params(rsc, node, scheduler); if (params != NULL) { g_hash_table_iter_init(&iter, params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { g_hash_table_insert(combined, strdup(key), strdup(value)); } } meta = pcmk__strkey_table(free, free); get_meta_attributes(meta, rsc, node, scheduler); if (meta != NULL) { g_hash_table_iter_init(&iter, meta); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { char *crm_name = crm_meta_name(key); g_hash_table_insert(combined, crm_name, strdup(value)); } g_hash_table_destroy(meta); } return combined; } bool resource_is_running_on(pcmk_resource_t *rsc, const char *host) { bool found = true; GList *hIter = NULL; GList *hosts = NULL; if (rsc == NULL) { return false; } rsc->fns->location(rsc, &hosts, TRUE); for (hIter = hosts; host != NULL && hIter != NULL; hIter = hIter->next) { pcmk_node_t *node = (pcmk_node_t *) hIter->data; if (pcmk__strcase_any_of(host, node->details->uname, node->details->id, NULL)) { crm_trace("Resource %s is running on %s\n", rsc->id, host); goto done; } } if (host != NULL) { crm_trace("Resource %s is not running on: %s\n", rsc->id, host); found = false; } else if(host == NULL && hosts == NULL) { crm_trace("Resource %s is not running\n", rsc->id); found = false; } done: g_list_free(hosts); return found; } /*! * \internal * \brief Create a list of all resources active on host from a given list * * \param[in] host Name of host to check whether resources are active * \param[in] rsc_list List of resources to check * * \return New list of resources from list that are active on host */ static GList * get_active_resources(const char *host, GList *rsc_list) { GList *rIter = NULL; GList *active = NULL; for (rIter = rsc_list; rIter != NULL; rIter = rIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) rIter->data; /* Expand groups to their members, because if we're restarting a member * other than the first, we can't otherwise tell which resources are * stopping and starting. */ if (rsc->variant == pcmk_rsc_variant_group) { active = g_list_concat(active, get_active_resources(host, rsc->children)); } else if (resource_is_running_on(rsc, host)) { active = g_list_append(active, strdup(rsc->id)); } } return active; } static void dump_list(GList *items, const char *tag) { int lpc = 0; GList *item = NULL; for (item = items; item != NULL; item = item->next) { crm_trace("%s[%d]: %s", tag, lpc, (char*)item->data); lpc++; } } static void display_list(pcmk__output_t *out, GList *items, const char *tag) { GList *item = NULL; for (item = items; item != NULL; item = item->next) { out->info(out, "%s%s", tag, (const char *)item->data); } } /*! * \internal * \brief Upgrade XML to latest schema version and use it as scheduler input * * This also updates the scheduler timestamp to the current time. * * \param[in,out] scheduler Scheduler data to update * \param[in,out] xml XML to use as input * * \return Standard Pacemaker return code * \note On success, caller is responsible for freeing memory allocated for * scheduler->now. * \todo This follows the example of other callers of cli_config_update() * and returns ENOKEY ("Required key not available") if that fails, * but perhaps pcmk_rc_schema_validation would be better in that case. */ int update_scheduler_input(pcmk_scheduler_t *scheduler, xmlNode **xml) { if (cli_config_update(xml, NULL, FALSE) == FALSE) { return ENOKEY; } scheduler->input = *xml; scheduler->now = crm_time_new(NULL); return pcmk_rc_ok; } /*! * \internal * \brief Update scheduler XML input based on a CIB query * * \param[in] scheduler Scheduler data to initialize * \param[in] cib Connection to the CIB manager * * \return Standard Pacemaker return code * \note On success, caller is responsible for freeing memory allocated for * scheduler->input and scheduler->now. */ static int update_scheduler_input_to_cib(pcmk__output_t *out, pcmk_scheduler_t *scheduler, cib_t *cib) { xmlNode *cib_xml_copy = NULL; int rc = pcmk_rc_ok; rc = cib->cmds->query(cib, NULL, &cib_xml_copy, cib_scope_local | cib_sync_call); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { out->err(out, "Could not obtain the current CIB: %s (%d)", pcmk_rc_str(rc), rc); return rc; } rc = update_scheduler_input(scheduler, &cib_xml_copy); if (rc != pcmk_rc_ok) { out->err(out, "Could not upgrade the current CIB XML"); free_xml(cib_xml_copy); return rc; } return rc; } // \return Standard Pacemaker return code static int update_dataset(cib_t *cib, pcmk_scheduler_t *scheduler, bool simulate) { char *pid = NULL; char *shadow_file = NULL; cib_t *shadow_cib = NULL; int rc = pcmk_rc_ok; pcmk__output_t *out = scheduler->priv; pe_reset_working_set(scheduler); pcmk__set_scheduler_flags(scheduler, pcmk_sched_no_counts|pcmk_sched_no_compat); rc = update_scheduler_input_to_cib(out, scheduler, cib); if (rc != pcmk_rc_ok) { return rc; } if(simulate) { bool prev_quiet = false; pid = pcmk__getpid_s(); shadow_cib = cib_shadow_new(pid); shadow_file = get_shadow_file(pid); if (shadow_cib == NULL) { out->err(out, "Could not create shadow cib: '%s'", pid); rc = ENXIO; goto done; } rc = write_xml_file(scheduler->input, shadow_file, FALSE); if (rc < 0) { out->err(out, "Could not populate shadow cib: %s (%d)", pcmk_strerror(rc), rc); goto done; } rc = shadow_cib->cmds->signon(shadow_cib, crm_system_name, cib_command); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { out->err(out, "Could not connect to shadow cib: %s (%d)", pcmk_rc_str(rc), rc); goto done; } pcmk__schedule_actions(scheduler->input, pcmk_sched_no_counts|pcmk_sched_no_compat, scheduler); prev_quiet = out->is_quiet(out); out->quiet = true; pcmk__simulate_transition(scheduler, shadow_cib, NULL); out->quiet = prev_quiet; rc = update_dataset(shadow_cib, scheduler, false); } else { cluster_status(scheduler); } done: // Do not free scheduler->input here, we need rsc->xml to be valid later on cib_delete(shadow_cib); free(pid); if(shadow_file) { unlink(shadow_file); free(shadow_file); } return rc; } /*! * \internal * \brief Find the maximum stop timeout of a resource and its children (if any) * * \param[in,out] rsc Resource to get timeout for * * \return Maximum stop timeout for \p rsc (in milliseconds) */ static int max_rsc_stop_timeout(pcmk_resource_t *rsc) { long long result_ll; int max_delay = 0; xmlNode *config = NULL; GHashTable *meta = NULL; if (rsc == NULL) { return 0; } // If resource is collective, use maximum of its children's stop timeouts if (rsc->children != NULL) { for (GList *iter = rsc->children; iter; iter = iter->next) { pcmk_resource_t *child = iter->data; int delay = max_rsc_stop_timeout(child); if (delay > max_delay) { pcmk__rsc_trace(rsc, "Maximum stop timeout for %s is now %s " "due to %s", rsc->id, pcmk__readable_interval(delay), child->id); max_delay = delay; } } return max_delay; } // Get resource's stop action configuration from CIB config = pcmk__find_action_config(rsc, PCMK_ACTION_STOP, 0, true); /* Get configured timeout for stop action (fully evaluated for rules, * defaults, etc.). * * @TODO This currently ignores node (which might matter for rules) */ meta = pcmk__unpack_action_meta(rsc, NULL, PCMK_ACTION_STOP, 0, config); if ((pcmk__scan_ll(g_hash_table_lookup(meta, PCMK_META_TIMEOUT), &result_ll, -1LL) == pcmk_rc_ok) && (result_ll >= 0) && (result_ll <= INT_MAX)) { max_delay = (int) result_ll; } g_hash_table_destroy(meta); return max_delay; } /*! * \internal * \brief Find a reasonable waiting time for stopping any one resource in a list * * \param[in,out] scheduler Scheduler data * \param[in] resources List of names of resources that will be stopped * * \return Rough estimate of a reasonable time to wait (in seconds) to stop any * one resource in \p resources * \note This estimate is very rough, simply the maximum stop timeout of all * given resources and their children, plus a small fudge factor. It does * not account for children that must be stopped in sequence, action * throttling, or any demotions needed. It checks the stop timeout, even * if the resources in question are actually being started. */ static int wait_time_estimate(pcmk_scheduler_t *scheduler, const GList *resources) { int max_delay = 0; // Find maximum stop timeout in milliseconds for (const GList *item = resources; item != NULL; item = item->next) { pcmk_resource_t *rsc = pe_find_resource(scheduler->resources, (const char *) item->data); int delay = max_rsc_stop_timeout(rsc); if (delay > max_delay) { pcmk__rsc_trace(rsc, "Wait time is now %s due to %s", pcmk__readable_interval(delay), rsc->id); max_delay = delay; } } return (max_delay / 1000) + 5; } #define waiting_for_starts(d, r, h) ((d != NULL) || \ (!resource_is_running_on((r), (h)))) /*! * \internal * \brief Restart a resource (on a particular host if requested). * * \param[in,out] out Output object * \param[in,out] rsc The resource to restart * \param[in] node Node to restart resource on (NULL for all) * \param[in] move_lifetime If not NULL, how long constraint should * remain in effect (as ISO 8601 string) * \param[in] timeout_ms Consider failed if actions do not complete * in this time (specified in milliseconds, * but a two-second granularity is actually * used; if 0, it will be calculated based on * the resource timeout) * \param[in,out] cib Connection to the CIB manager * \param[in] cib_options Group of enum cib_call_options flags to * use with CIB calls * \param[in] promoted_role_only If true, limit to promoted instances * \param[in] force If true, apply only to requested instance * if part of a collective resource * * \return Standard Pacemaker return code (exits on certain failures) */ int cli_resource_restart(pcmk__output_t *out, pcmk_resource_t *rsc, const pcmk_node_t *node, const char *move_lifetime, int timeout_ms, cib_t *cib, int cib_options, gboolean promoted_role_only, gboolean force) { int rc = pcmk_rc_ok; int lpc = 0; int before = 0; int step_timeout_s = 0; int sleep_interval = 2; int timeout = timeout_ms / 1000; bool stop_via_ban = false; char *rsc_id = NULL; char *lookup_id = NULL; char *orig_target_role = NULL; GList *list_delta = NULL; GList *target_active = NULL; GList *current_active = NULL; GList *restart_target_active = NULL; pcmk_scheduler_t *scheduler = NULL; pcmk_resource_t *parent = uber_parent(rsc); bool running = false; const char *id = rsc->clone_name ? rsc->clone_name : rsc->id; const char *host = node ? node->details->uname : NULL; /* If the implicit resource or primitive resource of a bundle is given, operate on the * bundle itself instead. */ if (pe_rsc_is_bundled(rsc)) { rsc = parent->parent; } running = resource_is_running_on(rsc, host); if (pcmk__is_clone(parent) && !running) { - if (pe_rsc_is_unique_clone(parent)) { + if (pcmk__is_unique_clone(parent)) { lookup_id = strdup(rsc->id); } else { lookup_id = clone_strip(rsc->id); } rsc = parent->fns->find_rsc(parent, lookup_id, node, pcmk_rsc_match_basename |pcmk_rsc_match_current_node); free(lookup_id); running = resource_is_running_on(rsc, host); } if (!running) { if (host) { out->err(out, "%s is not running on %s and so cannot be restarted", id, host); } else { out->err(out, "%s is not running anywhere and so cannot be restarted", id); } return ENXIO; } if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { out->err(out, "Unmanaged resources cannot be restarted."); return EAGAIN; } rsc_id = strdup(rsc->id); - if (pe_rsc_is_unique_clone(parent)) { + if (pcmk__is_unique_clone(parent)) { lookup_id = strdup(rsc->id); } else { lookup_id = clone_strip(rsc->id); } if (host) { if (pcmk__is_clone(rsc) || pe_bundle_replicas(rsc)) { stop_via_ban = true; } else if (pcmk__is_clone(parent)) { stop_via_ban = true; free(lookup_id); lookup_id = strdup(parent->id); } } /* grab full cib determine originally active resources disable or ban poll cib and watch for affected resources to get stopped without --timeout, calculate the stop timeout for each step and wait for that if we hit --timeout or the service timeout, re-enable or un-ban, report failure and indicate which resources we couldn't take down if everything stopped, re-enable or un-ban poll cib and watch for affected resources to get started without --timeout, calculate the start timeout for each step and wait for that if we hit --timeout or the service timeout, report (different) failure and indicate which resources we couldn't bring back up report success Optimizations: - use constraints to determine ordered list of affected resources - Allow a --no-deps option (aka. --force-restart) */ scheduler = pe_new_working_set(); if (scheduler == NULL) { rc = errno; out->err(out, "Could not allocate scheduler data: %s", pcmk_rc_str(rc)); goto done; } scheduler->priv = out; rc = update_dataset(cib, scheduler, false); if(rc != pcmk_rc_ok) { out->err(out, "Could not get new resource list: %s (%d)", pcmk_rc_str(rc), rc); goto done; } restart_target_active = get_active_resources(host, scheduler->resources); current_active = get_active_resources(host, scheduler->resources); dump_list(current_active, "Origin"); if (stop_via_ban) { /* Stop the clone or bundle instance by banning it from the host */ out->quiet = true; rc = cli_resource_ban(out, lookup_id, host, move_lifetime, cib, cib_options, promoted_role_only, PCMK__ROLE_PROMOTED); } else { /* Stop the resource by setting PCMK_META_TARGET_ROLE to Stopped. * Remember any existing PCMK_META_TARGET_ROLE so we can restore it * later (though it only makes any difference if it's Unpromoted). */ find_resource_attr(out, cib, PCMK_XA_VALUE, lookup_id, NULL, NULL, NULL, PCMK_META_TARGET_ROLE, &orig_target_role); rc = cli_resource_update_attribute(rsc, rsc_id, NULL, PCMK_XE_META_ATTRIBUTES, NULL, PCMK_META_TARGET_ROLE, PCMK_ACTION_STOPPED, FALSE, cib, cib_options, force); } if(rc != pcmk_rc_ok) { out->err(out, "Could not set " PCMK_META_TARGET_ROLE " for %s: %s (%d)", rsc_id, pcmk_rc_str(rc), rc); if (current_active != NULL) { g_list_free_full(current_active, free); current_active = NULL; } if (restart_target_active != NULL) { g_list_free_full(restart_target_active, free); restart_target_active = NULL; } goto done; } rc = update_dataset(cib, scheduler, true); if(rc != pcmk_rc_ok) { out->err(out, "Could not determine which resources would be stopped"); goto failure; } target_active = get_active_resources(host, scheduler->resources); dump_list(target_active, "Target"); list_delta = pcmk__subtract_lists(current_active, target_active, (GCompareFunc) strcmp); out->info(out, "Waiting for %d resources to stop:", g_list_length(list_delta)); display_list(out, list_delta, " * "); step_timeout_s = timeout / sleep_interval; while (list_delta != NULL) { before = g_list_length(list_delta); if(timeout_ms == 0) { step_timeout_s = wait_time_estimate(scheduler, list_delta) / sleep_interval; } /* We probably don't need the entire step timeout */ for(lpc = 0; (lpc < step_timeout_s) && (list_delta != NULL); lpc++) { sleep(sleep_interval); if(timeout) { timeout -= sleep_interval; crm_trace("%ds remaining", timeout); } rc = update_dataset(cib, scheduler, FALSE); if(rc != pcmk_rc_ok) { out->err(out, "Could not determine which resources were stopped"); goto failure; } if (current_active != NULL) { g_list_free_full(current_active, free); } current_active = get_active_resources(host, scheduler->resources); g_list_free(list_delta); list_delta = pcmk__subtract_lists(current_active, target_active, (GCompareFunc) strcmp); dump_list(current_active, "Current"); dump_list(list_delta, "Delta"); } crm_trace("%d (was %d) resources remaining", g_list_length(list_delta), before); if(before == g_list_length(list_delta)) { /* aborted during stop phase, print the contents of list_delta */ out->err(out, "Could not complete shutdown of %s, %d resources remaining", rsc_id, g_list_length(list_delta)); display_list(out, list_delta, " * "); rc = ETIME; goto failure; } } if (stop_via_ban) { rc = cli_resource_clear(lookup_id, host, NULL, cib, cib_options, true, force); } else if (orig_target_role) { rc = cli_resource_update_attribute(rsc, rsc_id, NULL, PCMK_XE_META_ATTRIBUTES, NULL, PCMK_META_TARGET_ROLE, orig_target_role, FALSE, cib, cib_options, force); free(orig_target_role); orig_target_role = NULL; } else { rc = cli_resource_delete_attribute(rsc, rsc_id, NULL, PCMK_XE_META_ATTRIBUTES, NULL, PCMK_META_TARGET_ROLE, cib, cib_options, force); } if(rc != pcmk_rc_ok) { out->err(out, "Could not unset " PCMK_META_TARGET_ROLE " for %s: %s (%d)", rsc_id, pcmk_rc_str(rc), rc); goto done; } if (target_active != NULL) { g_list_free_full(target_active, free); } target_active = restart_target_active; list_delta = pcmk__subtract_lists(target_active, current_active, (GCompareFunc) strcmp); out->info(out, "Waiting for %d resources to start again:", g_list_length(list_delta)); display_list(out, list_delta, " * "); step_timeout_s = timeout / sleep_interval; while (waiting_for_starts(list_delta, rsc, host)) { before = g_list_length(list_delta); if(timeout_ms == 0) { step_timeout_s = wait_time_estimate(scheduler, list_delta) / sleep_interval; } /* We probably don't need the entire step timeout */ for (lpc = 0; (lpc < step_timeout_s) && waiting_for_starts(list_delta, rsc, host); lpc++) { sleep(sleep_interval); if(timeout) { timeout -= sleep_interval; crm_trace("%ds remaining", timeout); } rc = update_dataset(cib, scheduler, false); if(rc != pcmk_rc_ok) { out->err(out, "Could not determine which resources were started"); goto failure; } /* It's OK if dependent resources moved to a different node, * so we check active resources on all nodes. */ if (current_active != NULL) { g_list_free_full(current_active, free); } current_active = get_active_resources(NULL, scheduler->resources); g_list_free(list_delta); list_delta = pcmk__subtract_lists(target_active, current_active, (GCompareFunc) strcmp); dump_list(current_active, "Current"); dump_list(list_delta, "Delta"); } if(before == g_list_length(list_delta)) { /* aborted during start phase, print the contents of list_delta */ out->err(out, "Could not complete restart of %s, %d resources remaining", rsc_id, g_list_length(list_delta)); display_list(out, list_delta, " * "); rc = ETIME; goto failure; } } rc = pcmk_rc_ok; goto done; failure: if (stop_via_ban) { cli_resource_clear(lookup_id, host, NULL, cib, cib_options, true, force); } else if (orig_target_role) { cli_resource_update_attribute(rsc, rsc_id, NULL, PCMK_XE_META_ATTRIBUTES, NULL, PCMK_META_TARGET_ROLE, orig_target_role, FALSE, cib, cib_options, force); free(orig_target_role); } else { cli_resource_delete_attribute(rsc, rsc_id, NULL, PCMK_XE_META_ATTRIBUTES, NULL, PCMK_META_TARGET_ROLE, cib, cib_options, force); } done: if (list_delta != NULL) { g_list_free(list_delta); } if (current_active != NULL) { g_list_free_full(current_active, free); } if (target_active != NULL && (target_active != restart_target_active)) { g_list_free_full(target_active, free); } if (restart_target_active != NULL) { g_list_free_full(restart_target_active, free); } free(rsc_id); free(lookup_id); pe_free_working_set(scheduler); return rc; } static inline bool action_is_pending(const pcmk_action_t *action) { if (pcmk_any_flags_set(action->flags, pcmk_action_optional|pcmk_action_pseudo) || !pcmk_is_set(action->flags, pcmk_action_runnable) || pcmk__str_eq(PCMK_ACTION_NOTIFY, action->task, pcmk__str_casei)) { return false; } return true; } /*! * \internal * \brief Check whether any actions in a list are pending * * \param[in] actions List of actions to check * * \return true if any actions in the list are pending, otherwise false */ static bool actions_are_pending(const GList *actions) { for (const GList *action = actions; action != NULL; action = action->next) { const pcmk_action_t *a = (const pcmk_action_t *) action->data; if (action_is_pending(a)) { crm_notice("Waiting for %s (flags=%#.8x)", a->uuid, a->flags); return true; } } return false; } static void print_pending_actions(pcmk__output_t *out, GList *actions) { GList *action; out->info(out, "Pending actions:"); for (action = actions; action != NULL; action = action->next) { pcmk_action_t *a = (pcmk_action_t *) action->data; if (!action_is_pending(a)) { continue; } if (a->node) { out->info(out, "\tAction %d: %s\ton %s", a->id, a->uuid, pcmk__node_name(a->node)); } else { out->info(out, "\tAction %d: %s", a->id, a->uuid); } } } /* For --wait, timeout (in seconds) to use if caller doesn't specify one */ #define WAIT_DEFAULT_TIMEOUT_S (60 * 60) /* For --wait, how long to sleep between cluster state checks */ #define WAIT_SLEEP_S (2) /*! * \internal * \brief Wait until all pending cluster actions are complete * * This waits until either the CIB's transition graph is idle or a timeout is * reached. * * \param[in,out] out Output object * \param[in] timeout_ms Consider failed if actions do not complete in * this time (specified in milliseconds, but * one-second granularity is actually used; if 0, a * default will be used) * \param[in,out] cib Connection to the CIB manager * * \return Standard Pacemaker return code */ int wait_till_stable(pcmk__output_t *out, int timeout_ms, cib_t * cib) { pcmk_scheduler_t *scheduler = NULL; xmlXPathObjectPtr search; int rc = pcmk_rc_ok; bool pending_unknown_state_resources; int timeout_s = timeout_ms? ((timeout_ms + 999) / 1000) : WAIT_DEFAULT_TIMEOUT_S; time_t expire_time = time(NULL) + timeout_s; time_t time_diff; bool printed_version_warning = out->is_quiet(out); // i.e. don't print if quiet char *xpath = NULL; scheduler = pe_new_working_set(); if (scheduler == NULL) { return ENOMEM; } xpath = crm_strdup_printf("/" PCMK_XE_CIB "/" PCMK_XE_STATUS "/" PCMK__XE_NODE_STATE "/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES "/" PCMK__XE_LRM_RESOURCE "/" PCMK__XE_LRM_RSC_OP "[@" PCMK__XA_RC_CODE "='%d']", PCMK_OCF_UNKNOWN); do { /* Abort if timeout is reached */ time_diff = expire_time - time(NULL); if (time_diff <= 0) { print_pending_actions(out, scheduler->actions); rc = ETIME; break; } crm_info("Waiting up to %lld seconds for cluster actions to complete", (long long) time_diff); if (rc == pcmk_rc_ok) { /* this avoids sleep on first loop iteration */ sleep(WAIT_SLEEP_S); } /* Get latest transition graph */ pe_reset_working_set(scheduler); rc = update_scheduler_input_to_cib(out, scheduler, cib); if (rc != pcmk_rc_ok) { break; } pcmk__schedule_actions(scheduler->input, pcmk_sched_no_counts|pcmk_sched_no_compat, scheduler); if (!printed_version_warning) { /* If the DC has a different version than the local node, the two * could come to different conclusions about what actions need to be * done. Warn the user in this case. * * @TODO A possible long-term solution would be to reimplement the * wait as a new controller operation that would be forwarded to the * DC. However, that would have potential problems of its own. */ const char *dc_version = g_hash_table_lookup(scheduler->config_hash, PCMK_OPT_DC_VERSION); if (!pcmk__str_eq(dc_version, PACEMAKER_VERSION "-" BUILD_VERSION, pcmk__str_casei)) { out->info(out, "warning: wait option may not work properly in " "mixed-version cluster"); printed_version_warning = true; } } search = xpath_search(scheduler->input, xpath); pending_unknown_state_resources = (numXpathResults(search) > 0); freeXpathObject(search); } while (actions_are_pending(scheduler->actions) || pending_unknown_state_resources); pe_free_working_set(scheduler); free(xpath); return rc; } static const char * get_action(const char *rsc_action) { const char *action = NULL; if (pcmk__str_eq(rsc_action, "validate", pcmk__str_casei)) { action = PCMK_ACTION_VALIDATE_ALL; } else if (pcmk__str_eq(rsc_action, "force-check", pcmk__str_casei)) { action = PCMK_ACTION_MONITOR; } else if (pcmk__strcase_any_of(rsc_action, "force-start", "force-stop", "force-demote", "force-promote", NULL)) { action = rsc_action+6; } else { action = rsc_action; } return action; } /*! * \brief Set up environment variables as expected by resource agents * * When the cluster executes resource agents, it adds certain environment * variables (directly or via resource meta-attributes) expected by some * resource agents. Add the essential ones that many resource agents expect, so * the behavior is the same for command-line execution. * * \param[in,out] params Resource parameters that will be passed to agent * \param[in] timeout_ms Action timeout (in milliseconds) * \param[in] check_level OCF check level * \param[in] verbosity Verbosity level */ static void set_agent_environment(GHashTable *params, int timeout_ms, int check_level, int verbosity) { g_hash_table_insert(params, crm_meta_name(PCMK_META_TIMEOUT), crm_strdup_printf("%d", timeout_ms)); g_hash_table_insert(params, strdup(PCMK_XA_CRM_FEATURE_SET), strdup(CRM_FEATURE_SET)); if (check_level >= 0) { char *level = crm_strdup_printf("%d", check_level); setenv("OCF_CHECK_LEVEL", level, 1); free(level); } pcmk__set_env_option(PCMK__ENV_DEBUG, ((verbosity > 0)? "1" : "0"), true); if (verbosity > 1) { setenv("OCF_TRACE_RA", "1", 1); } /* A resource agent using the standard ocf-shellfuncs library will not print * messages to stderr if it doesn't have a controlling terminal (e.g. if * crm_resource is called via script or ssh). This forces it to do so. */ setenv("OCF_TRACE_FILE", "/dev/stderr", 0); } /*! * \internal * \brief Apply command-line overrides to resource parameters * * \param[in,out] params Parameters to be passed to agent * \param[in] overrides Parameters to override (or NULL if none) */ static void apply_overrides(GHashTable *params, GHashTable *overrides) { if (overrides != NULL) { GHashTableIter iter; char *name = NULL; char *value = NULL; g_hash_table_iter_init(&iter, overrides); while (g_hash_table_iter_next(&iter, (gpointer *) &name, (gpointer *) &value)) { g_hash_table_replace(params, strdup(name), strdup(value)); } } } crm_exit_t cli_resource_execute_from_params(pcmk__output_t *out, const char *rsc_name, const char *rsc_class, const char *rsc_prov, const char *rsc_type, const char *rsc_action, GHashTable *params, GHashTable *override_hash, int timeout_ms, int resource_verbose, gboolean force, int check_level) { const char *class = rsc_class; const char *action = get_action(rsc_action); crm_exit_t exit_code = CRM_EX_OK; svc_action_t *op = NULL; // If no timeout was provided, use the same default as the cluster if (timeout_ms == 0) { timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS; } set_agent_environment(params, timeout_ms, check_level, resource_verbose); apply_overrides(params, override_hash); op = services__create_resource_action(rsc_name? rsc_name : "test", rsc_class, rsc_prov, rsc_type, action, 0, timeout_ms, params, 0); if (op == NULL) { out->err(out, "Could not execute %s using %s%s%s:%s: %s", action, rsc_class, (rsc_prov? ":" : ""), (rsc_prov? rsc_prov : ""), rsc_type, strerror(ENOMEM)); g_hash_table_destroy(params); return CRM_EX_OSERR; } if (pcmk__str_eq(rsc_class, PCMK_RESOURCE_CLASS_SERVICE, pcmk__str_casei)) { class = resources_find_service_class(rsc_type); } if (!pcmk__strcase_any_of(class, PCMK_RESOURCE_CLASS_OCF, PCMK_RESOURCE_CLASS_LSB, NULL)) { services__format_result(op, CRM_EX_UNIMPLEMENT_FEATURE, PCMK_EXEC_ERROR, "Manual execution of the %s standard is " "unsupported", pcmk__s(class, "unspecified")); } if (op->rc != PCMK_OCF_UNKNOWN) { exit_code = op->rc; goto done; } services_action_sync(op); // Map results to OCF codes for consistent reporting to user { enum ocf_exitcode ocf_code = services_result2ocf(class, action, op->rc); // Cast variable instead of function return to keep compilers happy exit_code = (crm_exit_t) ocf_code; } done: out->message(out, "resource-agent-action", resource_verbose, rsc_class, rsc_prov, rsc_type, rsc_name, rsc_action, override_hash, exit_code, op->status, services__exit_reason(op), op->stdout_data, op->stderr_data); services_action_free(op); return exit_code; } crm_exit_t cli_resource_execute(pcmk_resource_t *rsc, const char *requested_name, const char *rsc_action, GHashTable *override_hash, int timeout_ms, cib_t *cib, pcmk_scheduler_t *scheduler, int resource_verbose, gboolean force, int check_level) { pcmk__output_t *out = scheduler->priv; crm_exit_t exit_code = CRM_EX_OK; const char *rid = NULL; const char *rtype = NULL; const char *rprov = NULL; const char *rclass = NULL; GHashTable *params = NULL; if (pcmk__strcase_any_of(rsc_action, "force-start", "force-demote", "force-promote", NULL)) { if (pcmk__is_clone(rsc)) { GList *nodes = cli_resource_search(rsc, requested_name, scheduler); if(nodes != NULL && force == FALSE) { out->err(out, "It is not safe to %s %s here: the cluster claims it is already active", rsc_action, rsc->id); out->err(out, "Try setting " PCMK_META_TARGET_ROLE "=" PCMK__ROLE_STOPPED " first or specifying the force option"); return CRM_EX_UNSAFE; } g_list_free_full(nodes, free); } } if (pcmk__is_clone(rsc)) { /* Grab the first child resource in the hope it's not a group */ rsc = rsc->children->data; } if (rsc->variant == pcmk_rsc_variant_group) { out->err(out, "Sorry, the %s option doesn't support group resources", rsc_action); return CRM_EX_UNIMPLEMENT_FEATURE; } else if (pe_rsc_is_bundled(rsc)) { out->err(out, "Sorry, the %s option doesn't support bundled resources", rsc_action); return CRM_EX_UNIMPLEMENT_FEATURE; } rclass = crm_element_value(rsc->xml, PCMK_XA_CLASS); rprov = crm_element_value(rsc->xml, PCMK_XA_PROVIDER); rtype = crm_element_value(rsc->xml, PCMK_XA_TYPE); params = generate_resource_params(rsc, NULL /* @TODO use local node */, scheduler); if (timeout_ms == 0) { timeout_ms = pe_get_configured_timeout(rsc, get_action(rsc_action), scheduler); } rid = pe_rsc_is_anon_clone(rsc->parent)? requested_name : rsc->id; exit_code = cli_resource_execute_from_params(out, rid, rclass, rprov, rtype, rsc_action, params, override_hash, timeout_ms, resource_verbose, force, check_level); return exit_code; } // \return Standard Pacemaker return code int cli_resource_move(const pcmk_resource_t *rsc, const char *rsc_id, const char *host_name, const char *move_lifetime, cib_t *cib, int cib_options, pcmk_scheduler_t *scheduler, gboolean promoted_role_only, gboolean force) { pcmk__output_t *out = scheduler->priv; int rc = pcmk_rc_ok; unsigned int count = 0; pcmk_node_t *current = NULL; pcmk_node_t *dest = pe_find_node(scheduler->nodes, host_name); bool cur_is_dest = false; if (dest == NULL) { return pcmk_rc_node_unknown; } if (promoted_role_only && !pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { const pcmk_resource_t *p = pe__const_top_resource(rsc, false); if (pcmk_is_set(p->flags, pcmk_rsc_promotable)) { out->info(out, "Using parent '%s' for move instead of '%s'.", rsc->id, rsc_id); rsc_id = p->id; rsc = p; } else { out->info(out, "Ignoring --promoted option: %s is not promotable", rsc_id); promoted_role_only = FALSE; } } current = pe__find_active_requires(rsc, &count); if (pcmk_is_set(rsc->flags, pcmk_rsc_promotable)) { unsigned int promoted_count = 0; pcmk_node_t *promoted_node = NULL; for (const GList *iter = rsc->children; iter; iter = iter->next) { const pcmk_resource_t *child = (const pcmk_resource_t *) iter->data; enum rsc_role_e child_role = child->fns->state(child, TRUE); if (child_role == pcmk_role_promoted) { rsc = child; promoted_node = pcmk__current_node(child); promoted_count++; } } if (promoted_role_only || (promoted_count != 0)) { count = promoted_count; current = promoted_node; } } if (count > 1) { if (pcmk__is_clone(rsc)) { current = NULL; } else { return pcmk_rc_multiple; } } if (pcmk__same_node(current, dest)) { cur_is_dest = true; if (force) { crm_info("%s is already %s on %s, reinforcing placement with location constraint.", rsc_id, promoted_role_only?"promoted":"active", pcmk__node_name(dest)); } else { return pcmk_rc_already; } } /* Clear any previous prefer constraints across all nodes. */ cli_resource_clear(rsc_id, NULL, scheduler->nodes, cib, cib_options, false, force); /* Clear any previous ban constraints on 'dest'. */ cli_resource_clear(rsc_id, dest->details->uname, scheduler->nodes, cib, cib_options, TRUE, force); /* Record an explicit preference for 'dest' */ rc = cli_resource_prefer(out, rsc_id, dest->details->uname, move_lifetime, cib, cib_options, promoted_role_only, PCMK__ROLE_PROMOTED); crm_trace("%s%s now prefers %s%s", rsc->id, (promoted_role_only? " (promoted)" : ""), pcmk__node_name(dest), force?"(forced)":""); /* only ban the previous location if current location != destination location. * it is possible to use -M to enforce a location without regard of where the * resource is currently located */ if (force && !cur_is_dest) { /* Ban the original location if possible */ if(current) { (void)cli_resource_ban(out, rsc_id, current->details->uname, move_lifetime, cib, cib_options, promoted_role_only, PCMK__ROLE_PROMOTED); } else if(count > 1) { out->info(out, "Resource '%s' is currently %s in %d locations. " "One may now move to %s", rsc_id, (promoted_role_only? "promoted" : "active"), count, pcmk__node_name(dest)); out->info(out, "To prevent '%s' from being %s at a specific location, " "specify a node.", rsc_id, (promoted_role_only? "promoted" : "active")); } else { crm_trace("Not banning %s from its current location: not active", rsc_id); } } return rc; }