diff --git a/lib/pacemaker/libpacemaker_private.h b/lib/pacemaker/libpacemaker_private.h index b18cb00a7e..ccdc623d91 100644 --- a/lib/pacemaker/libpacemaker_private.h +++ b/lib/pacemaker/libpacemaker_private.h @@ -1,1167 +1,1174 @@ /* * Copyright 2021-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__LIBPACEMAKER_PRIVATE__H # define PCMK__LIBPACEMAKER_PRIVATE__H /* This header is for the sole use of libpacemaker, so that functions can be * declared with G_GNUC_INTERNAL for efficiency. */ #include // lrmd_event_data_t #include // pcmk_action_t, pcmk_node_t, etc. #include // pcmk__location_t // Colocation flags enum pcmk__coloc_flags { pcmk__coloc_none = 0U, // Primary is affected even if already active pcmk__coloc_influence = (1U << 0), // Colocation was explicitly configured in CIB pcmk__coloc_explicit = (1U << 1), }; // Flags to modify the behavior of add_colocated_node_scores() enum pcmk__coloc_select { // With no other flags, apply all "with this" colocations pcmk__coloc_select_default = 0, // Apply "this with" colocations instead of "with this" colocations pcmk__coloc_select_this_with = (1 << 0), // Apply only colocations with non-negative scores pcmk__coloc_select_nonnegative = (1 << 1), // Apply only colocations with at least one matching node pcmk__coloc_select_active = (1 << 2), }; // Flags the update_ordered_actions() method can return enum pcmk__updated { pcmk__updated_none = 0, // Nothing changed pcmk__updated_first = (1 << 0), // First action was updated pcmk__updated_then = (1 << 1), // Then action was updated }; #define pcmk__set_updated_flags(au_flags, action, flags_to_set) do { \ au_flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Action update", \ (action)->uuid, au_flags, \ (flags_to_set), #flags_to_set); \ } while (0) #define pcmk__clear_updated_flags(au_flags, action, flags_to_clear) do { \ au_flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Action update", \ (action)->uuid, au_flags, \ (flags_to_clear), #flags_to_clear); \ } while (0) // Resource assignment methods struct resource_alloc_functions_s { /*! * \internal * \brief Assign a 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 (if \p rsc is not a * primitive, this applies to its primitive * descendants instead) * * \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 *(*assign)(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail); /*! * \internal * \brief Create all actions needed for a given resource * * \param[in,out] rsc Resource to create actions for */ void (*create_actions)(pcmk_resource_t *rsc); /*! * \internal * \brief Schedule any probes needed for a resource on a node * * \param[in,out] rsc Resource to create probe for * \param[in,out] node Node to create probe on * * \return true if any probe was created, otherwise false */ bool (*create_probe)(pcmk_resource_t *rsc, pcmk_node_t *node); /*! * \internal * \brief Create implicit constraints needed for a resource * * \param[in,out] rsc Resource to create implicit constraints for */ void (*internal_constraints)(pcmk_resource_t *rsc); /*! * \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 (*apply_coloc_score)(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); /*! * \internal * \brief Create list of all resources in colocations with a given resource * * Given a resource, create a list of all resources involved in mandatory * colocations with it, whether directly or via chained colocations. * * \param[in] rsc Resource to add to colocated list * \param[in] orig_rsc Resource originally requested * \param[in,out] colocated_rscs Existing list * * \return List of given resource and all resources involved in colocations * * \note This function is recursive; top-level callers should pass NULL as * \p colocated_rscs and \p orig_rsc, and the desired resource as * \p rsc. The recursive calls will use other values. */ GList *(*colocated_resources)(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs); /*! * \internal * \brief Add colocations affecting a resource as primary to a list * * Given a resource being assigned (\p orig_rsc) and a resource somewhere in * its chain of ancestors (\p rsc, which may be \p orig_rsc), get * colocations that affect the ancestor as primary and should affect the * resource, and add them to a given list. * * \param[in] rsc Resource whose colocations should be added * \param[in] orig_rsc Affected resource (\p rsc or a descendant) * \param[in,out] list List of colocations to add to * * \note All arguments should be non-NULL. * \note The pcmk__with_this_colocations() wrapper should usually be used * instead of using this method directly. */ void (*with_this_colocations)(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); /*! * \internal * \brief Add colocations affecting a resource as dependent to a list * * Given a resource being assigned (\p orig_rsc) and a resource somewhere in * its chain of ancestors (\p rsc, which may be \p orig_rsc), get * colocations that affect the ancestor as dependent and should affect the * resource, and add them to a given list. * * * \param[in] rsc Resource whose colocations should be added * \param[in] orig_rsc Affected resource (\p rsc or a descendant) * \param[in,out] list List of colocations to add to * * \note All arguments should be non-NULL. * \note The pcmk__this_with_colocations() wrapper should usually be used * instead of using this method directly. */ void (*this_with_colocations)(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); /*! * \internal * \brief Update nodes with scores of colocated resources' nodes * * Given a table of nodes and a resource, update the nodes' scores with the * scores of the best nodes matching the attribute used for each of the * resource's relevant colocations. * * \param[in,out] source_rsc Resource whose node scores to add * \param[in] target_rsc Resource on whose behalf to update \p *nodes * \param[in] log_id Resource ID for logs (if \c NULL, use * \p source_rsc ID) * \param[in,out] nodes Nodes to update (set initial contents to * \c NULL to copy allowed nodes from * \p source_rsc) * \param[in] colocation Original colocation constraint (used to get * configured primary resource's stickiness, and * to get colocation node attribute; if \c NULL, * source_rsc's own matching node scores * will not be added, and \p *nodes must be * \c NULL as well) * \param[in] factor Incorporate scores multiplied by this factor * \param[in] flags Bitmask of enum pcmk__coloc_select values * * \note \c NULL \p target_rsc, \c NULL \p *nodes, \c NULL \p colocation, * and the \c pcmk__coloc_select_this_with flag are used together (and * only by \c cmp_resources()). * \note The caller remains responsible for freeing \p *nodes. */ void (*add_colocated_node_scores)(pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags); /*! * \internal * \brief Apply a location constraint to a resource's allowed node scores * * \param[in,out] rsc Resource to apply constraint to * \param[in,out] location Location constraint to apply */ void (*apply_location)(pcmk_resource_t *rsc, pcmk__location_t *location); /*! * \internal * \brief Return action flags for a given resource action * * \param[in,out] action Action to get flags for * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node * \note For primitives, this will be the same as action->flags regardless * of node. For collective resources, the flags can differ due to * multiple instances possibly being involved. */ uint32_t (*action_flags)(pcmk_action_t *action, const pcmk_node_t *node); /*! * \internal * \brief Update two actions according to an ordering between them * * Given information about an ordering of two actions, update the actions' * flags (and runnable_before members if appropriate) as appropriate for the * ordering. Effects may cascade to other orderings involving the actions as * well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this * node (only used when interleaving instances) * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates * (may include pcmk_action_optional to affect * only mandatory actions and pcmk_action_runnable * to affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t (*update_ordered_actions)(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler); /*! * \internal * \brief Output a summary of scheduled actions for a resource * * \param[in,out] rsc Resource to output actions for */ void (*output_actions)(pcmk_resource_t *rsc); /*! * \internal * \brief Add a resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void (*add_actions_to_graph)(pcmk_resource_t *rsc); /*! * \internal * \brief Add meta-attributes relevant to transition graph actions to XML * * If a given resource supports variant-specific meta-attributes that are * needed for transition graph actions, add them to a given XML element. * * \param[in] rsc Resource whose meta-attributes should be added * \param[in,out] xml Transition graph action attributes XML to add to */ void (*add_graph_meta)(const pcmk_resource_t *rsc, xmlNode *xml); /*! * \internal * \brief Add a resource's utilization to a table of utilization values * * This function is used when summing the utilization of a resource and all * resources colocated with it, to determine whether a node has sufficient * capacity. Given a resource and a table of utilization values, it will add * the resource's utilization to the existing values, if the resource has * not yet been assigned to a node. * * \param[in] rsc Resource with utilization to add * \param[in] orig_rsc Resource being assigned (for logging only) * \param[in] all_rscs List of all resources that will be summed * \param[in,out] utilization Table of utilization values to add to */ void (*add_utilization)(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); /*! * \internal * \brief Apply a shutdown lock for a resource, if appropriate * * \param[in,out] rsc Resource to check for shutdown lock */ void (*shutdown_lock)(pcmk_resource_t *rsc); }; // Actions (pcmk_sched_actions.c) G_GNUC_INTERNAL void pcmk__update_action_for_orderings(pcmk_action_t *action, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL uint32_t pcmk__update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__log_action(const char *pre_text, const pcmk_action_t *action, bool details); G_GNUC_INTERNAL pcmk_action_t *pcmk__new_cancel_action(pcmk_resource_t *rsc, const char *name, guint interval_ms, const pcmk_node_t *node); G_GNUC_INTERNAL pcmk_action_t *pcmk__new_shutdown_action(pcmk_node_t *node); G_GNUC_INTERNAL bool pcmk__action_locks_rsc_to_node(const pcmk_action_t *action); G_GNUC_INTERNAL void pcmk__deduplicate_action_inputs(pcmk_action_t *action); G_GNUC_INTERNAL void pcmk__output_actions(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL bool pcmk__check_action_config(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *xml_op); G_GNUC_INTERNAL void pcmk__handle_rsc_config_changes(pcmk_scheduler_t *scheduler); // Recurring actions (pcmk_sched_recurring.c) G_GNUC_INTERNAL void pcmk__create_recurring_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__schedule_cancel(pcmk_resource_t *rsc, const char *call_id, const char *task, guint interval_ms, const pcmk_node_t *node, const char *reason); G_GNUC_INTERNAL void pcmk__reschedule_recurring(pcmk_resource_t *rsc, const char *task, guint interval_ms, pcmk_node_t *node); G_GNUC_INTERNAL bool pcmk__action_is_recurring(const pcmk_action_t *action); // Producing transition graphs (pcmk_graph_producer.c) G_GNUC_INTERNAL bool pcmk__graph_has_loop(const pcmk_action_t *init_action, const pcmk_action_t *action, pcmk__related_action_t *input); G_GNUC_INTERNAL void pcmk__add_rsc_actions_to_graph(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_graph(pcmk_scheduler_t *scheduler); // Fencing (pcmk_sched_fencing.c) G_GNUC_INTERNAL void pcmk__order_vs_fence(pcmk_action_t *stonith_op, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__order_vs_unfence(const pcmk_resource_t *rsc, pcmk_node_t *node, pcmk_action_t *action, enum pcmk__action_relation_flags order); G_GNUC_INTERNAL void pcmk__fence_guest(pcmk_node_t *node); G_GNUC_INTERNAL bool pcmk__node_unfenced(const pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__order_restart_vs_unfence(gpointer data, gpointer user_data); // Injected scheduler inputs (pcmk_sched_injections.c) void pcmk__inject_scheduler_input(pcmk_scheduler_t *scheduler, cib_t *cib, const pcmk_injections_t *injections); // Constraints of any type (pcmk_sched_constraints.c) G_GNUC_INTERNAL pcmk_resource_t *pcmk__find_constraint_resource(GList *rsc_list, const char *id); G_GNUC_INTERNAL xmlNode *pcmk__expand_tags_in_sets(xmlNode *xml_obj, const pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL bool pcmk__valid_resource_or_tag(const pcmk_scheduler_t *scheduler, const char *id, pcmk_resource_t **rsc, pcmk_tag_t **tag); G_GNUC_INTERNAL bool pcmk__tag_to_set(xmlNode *xml_obj, xmlNode **rsc_set, const char *attr, bool convert_rsc, const pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__create_internal_constraints(pcmk_scheduler_t *scheduler); // Location constraints G_GNUC_INTERNAL void pcmk__unpack_location(xmlNode *xml_obj, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL pcmk__location_t *pcmk__new_location(const char *id, pcmk_resource_t *rsc, int node_score, const char *discover_mode, pcmk_node_t *foo_node); G_GNUC_INTERNAL void pcmk__apply_locations(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__apply_location(pcmk_resource_t *rsc, pcmk__location_t *constraint); // Colocation constraints (pcmk_sched_colocation.c) enum pcmk__coloc_affects { pcmk__coloc_affects_nothing = 0, pcmk__coloc_affects_location, pcmk__coloc_affects_role, }; /*! * \internal * \brief Get the value of a colocation's node attribute * * \param[in] node Node on which to look up the attribute * \param[in] attr Name of attribute to look up * \param[in] rsc Resource on whose behalf to look up the attribute * * \return Value of \p attr on \p node or on the host of \p node, as appropriate */ static inline const char * pcmk__colocation_node_attr(const pcmk_node_t *node, const char *attr, const pcmk_resource_t *rsc) { const char *target = NULL; /* A resource colocated with a bundle or its primitive can't run on the * bundle node itself (where only the primitive, if any, can run). Instead, * we treat it as a colocation with the bundle's containers, so always look * up colocation node attributes on the container host. */ if (pcmk__is_bundle_node(node) && pcmk__is_bundled(rsc) && (pe__const_top_resource(rsc, false) == pe__bundled_resource(rsc))) { target = PCMK_VALUE_HOST; } else if (rsc != NULL) { target = g_hash_table_lookup(rsc->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); } return pcmk__node_attr(node, attr, target, pcmk__rsc_node_assigned); } G_GNUC_INTERNAL enum pcmk__coloc_affects pcmk__colocation_affects(const pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool preview); G_GNUC_INTERNAL void pcmk__apply_coloc_to_scores(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__apply_coloc_to_priority(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__add_colocated_node_scores(pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags); G_GNUC_INTERNAL void pcmk__add_dependent_scores(gpointer data, gpointer user_data); G_GNUC_INTERNAL void pcmk__colocation_intersect_nodes(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, const GList *primary_nodes, bool merge_scores); G_GNUC_INTERNAL void pcmk__unpack_colocation(xmlNode *xml_obj, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__add_this_with(GList **list, const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__add_this_with_list(GList **list, GList *addition, const pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__add_with_this(GList **list, const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__add_with_this_list(GList **list, GList *addition, const pcmk_resource_t *rsc); G_GNUC_INTERNAL GList *pcmk__with_this_colocations(const pcmk_resource_t *rsc); G_GNUC_INTERNAL GList *pcmk__this_with_colocations(const pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__new_colocation(const char *id, const char *node_attr, int score, pcmk_resource_t *dependent, pcmk_resource_t *primary, const char *dependent_role, const char *primary_role, uint32_t flags); G_GNUC_INTERNAL void pcmk__block_colocation_dependents(pcmk_action_t *action); /*! * \internal * \brief Check whether colocation's dependent preferences should be considered * * \param[in] colocation Colocation constraint * \param[in] rsc Primary instance (normally this will be * colocation->primary, which NULL will be treated as, * but for clones or bundles with multiple instances * this can be a particular instance) * * \return true if colocation influence should be effective, otherwise false */ static inline bool pcmk__colocation_has_influence(const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc) { if (rsc == NULL) { rsc = colocation->primary; } /* A bundle replica colocates its remote connection with its container, * using a finite score so that the container can run on Pacemaker Remote * nodes. * * Moving a connection is lightweight and does not interrupt the service, * while moving a container is heavyweight and does interrupt the service, * so don't move a clean, active container based solely on the preferences * of its connection. * * This also avoids problematic scenarios where two containers want to * perpetually swap places. */ if (pcmk_is_set(colocation->dependent->flags, pcmk_rsc_remote_nesting_allowed) && !pcmk_is_set(rsc->flags, pcmk_rsc_failed) && pcmk__list_of_1(rsc->running_on)) { return false; } /* The dependent in a colocation influences the primary's location * if the PCMK_XA_INFLUENCE option is true or the primary is not yet active. */ return pcmk_is_set(colocation->flags, pcmk__coloc_influence) || (rsc->running_on == NULL); } // Ordering constraints (pcmk_sched_ordering.c) G_GNUC_INTERNAL void pcmk__new_ordering(pcmk_resource_t *first_rsc, char *first_task, pcmk_action_t *first_action, pcmk_resource_t *then_rsc, char *then_task, pcmk_action_t *then_action, uint32_t flags, pcmk_scheduler_t *sched); G_GNUC_INTERNAL void pcmk__unpack_ordering(xmlNode *xml_obj, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__disable_invalid_orderings(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__order_stops_before_shutdown(pcmk_node_t *node, pcmk_action_t *shutdown_op); G_GNUC_INTERNAL void pcmk__apply_orderings(pcmk_scheduler_t *sched); G_GNUC_INTERNAL void pcmk__order_after_each(pcmk_action_t *after, GList *list); /*! * \internal * \brief Create a new ordering between two resource actions * * \param[in,out] first_rsc Resource for 'first' action * \param[in,out] first_task Action key for 'first' action * \param[in] then_rsc Resource for 'then' action * \param[in,out] then_task Action key for 'then' action * \param[in] flags Group of enum pcmk__action_relation_flags */ #define pcmk__order_resource_actions(first_rsc, first_task, \ then_rsc, then_task, flags) \ pcmk__new_ordering((first_rsc), \ pcmk__op_key((first_rsc)->id, (first_task), 0), \ NULL, \ (then_rsc), \ pcmk__op_key((then_rsc)->id, (then_task), 0), \ NULL, (flags), (first_rsc)->cluster) #define pcmk__order_starts(rsc1, rsc2, flags) \ pcmk__order_resource_actions((rsc1), PCMK_ACTION_START, \ (rsc2), PCMK_ACTION_START, (flags)) #define pcmk__order_stops(rsc1, rsc2, flags) \ pcmk__order_resource_actions((rsc1), PCMK_ACTION_STOP, \ (rsc2), PCMK_ACTION_STOP, (flags)) // Ticket constraints (pcmk_sched_tickets.c) G_GNUC_INTERNAL void pcmk__unpack_rsc_ticket(xmlNode *xml_obj, pcmk_scheduler_t *scheduler); // Promotable clone resources (pcmk_sched_promotable.c) G_GNUC_INTERNAL void pcmk__add_promotion_scores(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__require_promotion_tickets(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__set_instance_roles(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_promotable_actions(pcmk_resource_t *clone); G_GNUC_INTERNAL void pcmk__promotable_restart_ordering(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__order_promotable_instances(pcmk_resource_t *clone); G_GNUC_INTERNAL void pcmk__update_dependent_with_promotable(const pcmk_resource_t *primary, pcmk_resource_t *dependent, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__update_promotable_dependent_priority(const pcmk_resource_t *primary, pcmk_resource_t *dependent, const pcmk__colocation_t *colocation); // Pacemaker Remote nodes (pcmk_sched_remote.c) G_GNUC_INTERNAL bool pcmk__is_failed_remote_node(const pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__order_remote_connection_actions(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL bool pcmk__rsc_corresponds_to_guest(const pcmk_resource_t *rsc, const pcmk_node_t *node); G_GNUC_INTERNAL pcmk_node_t *pcmk__connection_host_for_action(const pcmk_action_t *action); G_GNUC_INTERNAL void pcmk__substitute_remote_addr(pcmk_resource_t *rsc, GHashTable *params); G_GNUC_INTERNAL void pcmk__add_guest_meta_to_xml(xmlNode *args_xml, const pcmk_action_t *action); // Primitives (pcmk_sched_primitive.c) G_GNUC_INTERNAL pcmk_node_t *pcmk__primitive_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail); G_GNUC_INTERNAL void pcmk__primitive_create_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__primitive_internal_constraints(pcmk_resource_t *rsc); G_GNUC_INTERNAL uint32_t pcmk__primitive_action_flags(pcmk_action_t *action, const pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__primitive_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_primitive_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__primitive_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__schedule_cleanup(pcmk_resource_t *rsc, const pcmk_node_t *node, bool optional); G_GNUC_INTERNAL void pcmk__primitive_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__primitive_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc); // Groups (pcmk_sched_group.c) G_GNUC_INTERNAL pcmk_node_t *pcmk__group_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail); G_GNUC_INTERNAL void pcmk__group_create_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__group_internal_constraints(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__group_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_group_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__group_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__group_add_colocated_node_scores(pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags); G_GNUC_INTERNAL void pcmk__group_apply_location(pcmk_resource_t *rsc, pcmk__location_t *location); G_GNUC_INTERNAL uint32_t pcmk__group_action_flags(pcmk_action_t *action, const pcmk_node_t *node); G_GNUC_INTERNAL uint32_t pcmk__group_update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL GList *pcmk__group_colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs); G_GNUC_INTERNAL void pcmk__group_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__group_shutdown_lock(pcmk_resource_t *rsc); // Clones (pcmk_sched_clone.c) G_GNUC_INTERNAL pcmk_node_t *pcmk__clone_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail); G_GNUC_INTERNAL void pcmk__clone_create_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__clone_create_probe(pcmk_resource_t *rsc, pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__clone_internal_constraints(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__clone_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_clone_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__clone_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__clone_apply_location(pcmk_resource_t *rsc, pcmk__location_t *constraint); G_GNUC_INTERNAL uint32_t pcmk__clone_action_flags(pcmk_action_t *action, const pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__clone_add_actions_to_graph(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__clone_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__clone_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__clone_shutdown_lock(pcmk_resource_t *rsc); // Bundles (pcmk_sched_bundle.c) G_GNUC_INTERNAL pcmk_node_t *pcmk__bundle_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail); G_GNUC_INTERNAL void pcmk__bundle_create_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__bundle_create_probe(pcmk_resource_t *rsc, pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__bundle_internal_constraints(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__bundle_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_bundle_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__bundle_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__bundle_apply_location(pcmk_resource_t *rsc, pcmk__location_t *constraint); G_GNUC_INTERNAL uint32_t pcmk__bundle_action_flags(pcmk_action_t *action, const pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__output_bundle_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__bundle_add_actions_to_graph(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__bundle_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__bundle_shutdown_lock(pcmk_resource_t *rsc); // Clone instances or bundle replica containers (pcmk_sched_instances.c) G_GNUC_INTERNAL void pcmk__assign_instances(pcmk_resource_t *collective, GList *instances, int max_total, int max_per_node); G_GNUC_INTERNAL void pcmk__create_instance_actions(pcmk_resource_t *rsc, GList *instances); G_GNUC_INTERNAL bool pcmk__instance_matches(const pcmk_resource_t *instance, const pcmk_node_t *node, enum rsc_role_e role, bool current); G_GNUC_INTERNAL pcmk_resource_t *pcmk__find_compatible_instance(const pcmk_resource_t *match_rsc, const pcmk_resource_t *rsc, enum rsc_role_e role, bool current); G_GNUC_INTERNAL uint32_t pcmk__instance_update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL uint32_t pcmk__collective_action_flags(pcmk_action_t *action, const GList *instances, const pcmk_node_t *node); // Injections (pcmk_injections.c) G_GNUC_INTERNAL xmlNode *pcmk__inject_node(cib_t *cib_conn, const char *node, const char *uuid); G_GNUC_INTERNAL xmlNode *pcmk__inject_node_state_change(cib_t *cib_conn, const char *node, bool up); G_GNUC_INTERNAL xmlNode *pcmk__inject_resource_history(pcmk__output_t *out, xmlNode *cib_node, const char *resource, const char *lrm_name, const char *rclass, const char *rtype, const char *rprovider); G_GNUC_INTERNAL void pcmk__inject_failcount(pcmk__output_t *out, xmlNode *cib_node, const char *resource, const char *task, guint interval_ms, int rc); G_GNUC_INTERNAL xmlNode *pcmk__inject_action_result(xmlNode *cib_resource, lrmd_event_data_t *op, int target_rc); // Nodes (pcmk_sched_nodes.c) G_GNUC_INTERNAL bool pcmk__node_available(const pcmk_node_t *node, bool consider_score, bool consider_guest); G_GNUC_INTERNAL bool pcmk__any_node_available(GHashTable *nodes); G_GNUC_INTERNAL GHashTable *pcmk__copy_node_table(GHashTable *nodes); G_GNUC_INTERNAL void pcmk__copy_node_tables(const pcmk_resource_t *rsc, GHashTable **copy); G_GNUC_INTERNAL void pcmk__restore_node_tables(pcmk_resource_t *rsc, GHashTable *backup); G_GNUC_INTERNAL GList *pcmk__sort_nodes(GList *nodes, pcmk_node_t *active_node); G_GNUC_INTERNAL void pcmk__apply_node_health(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL pcmk_node_t *pcmk__top_allowed_node(const pcmk_resource_t *rsc, const pcmk_node_t *node); // Functions applying to more than one variant (pcmk_sched_resource.c) G_GNUC_INTERNAL void pcmk__set_assignment_methods(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL bool pcmk__rsc_agent_changed(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *rsc_entry, bool active_on_node); G_GNUC_INTERNAL GList *pcmk__rscs_matching_id(const char *id, const pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL GList *pcmk__colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs); G_GNUC_INTERNAL void pcmk__noop_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__output_resource_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__assign_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool force, bool stop_if_fail); G_GNUC_INTERNAL void pcmk__unassign_resource(pcmk_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__threshold_reached(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_resource_t **failed); G_GNUC_INTERNAL void pcmk__sort_resources(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL gint pcmk__cmp_instance(gconstpointer a, gconstpointer b); G_GNUC_INTERNAL gint pcmk__cmp_instance_number(gconstpointer a, gconstpointer b); // Functions related to probes (pcmk_sched_probes.c) G_GNUC_INTERNAL bool pcmk__probe_rsc_on_node(pcmk_resource_t *rsc, pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__order_probes(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL bool pcmk__probe_resource_list(GList *rscs, pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__schedule_probes(pcmk_scheduler_t *scheduler); // Functions related to live migration (pcmk_sched_migration.c) void pcmk__create_migration_actions(pcmk_resource_t *rsc, const pcmk_node_t *current); void pcmk__abort_dangling_migration(void *data, void *user_data); bool pcmk__rsc_can_migrate(const pcmk_resource_t *rsc, const pcmk_node_t *current); void pcmk__order_migration_equivalents(pcmk__action_relation_t *order); // Functions related to node utilization (pcmk_sched_utilization.c) G_GNUC_INTERNAL int pcmk__compare_node_capacities(const pcmk_node_t *node1, const pcmk_node_t *node2); G_GNUC_INTERNAL void pcmk__consume_node_capacity(GHashTable *current_utilization, const pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__release_node_capacity(GHashTable *current_utilization, const pcmk_resource_t *rsc); G_GNUC_INTERNAL const pcmk_node_t *pcmk__ban_insufficient_capacity(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_utilization_constraints(pcmk_resource_t *rsc, const GList *allowed_nodes); G_GNUC_INTERNAL void pcmk__show_node_capacities(const char *desc, pcmk_scheduler_t *scheduler); + +// Functions related to the scheduler (pcmk_scheduler.c) + +G_GNUC_INTERNAL +int pcmk__init_scheduler(pcmk__output_t *out, xmlNodePtr input, const crm_time_t *date, + pcmk_scheduler_t **scheduler); + #endif // PCMK__LIBPACEMAKER_PRIVATE__H diff --git a/lib/pacemaker/pcmk_rule.c b/lib/pacemaker/pcmk_rule.c index cb7796b60f..0795d81e41 100644 --- a/lib/pacemaker/pcmk_rule.c +++ b/lib/pacemaker/pcmk_rule.c @@ -1,277 +1,214 @@ /* * Copyright 2022-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include -/*! - * \internal - * \brief Initialize scheduler data for checking rules - * - * Make our own copies of the CIB XML and date/time object, if they're not - * \c NULL. This way we don't have to take ownership of the objects passed via - * the API. - * - * \param[in,out] out Output object - * \param[in] input The CIB XML to check (if \c NULL, use current CIB) - * \param[in] date Check whether the rule is in effect at this date - * and time (if \c NULL, use current date and time) - * \param[out] scheduler Where to store initialized scheduler data - * - * \return Standard Pacemaker return code - */ -static int -init_rule_check(pcmk__output_t *out, xmlNodePtr input, const crm_time_t *date, - pcmk_scheduler_t **scheduler) -{ - // Allows for cleaner syntax than dereferencing the scheduler argument - pcmk_scheduler_t *new_scheduler = NULL; - - new_scheduler = pe_new_working_set(); - if (new_scheduler == NULL) { - return ENOMEM; - } - - pcmk__set_scheduler_flags(new_scheduler, - pcmk_sched_no_counts|pcmk_sched_no_compat); - - // Populate the scheduler data - - // Make our own copy of the given input or fetch the CIB and use that - if (input != NULL) { - new_scheduler->input = pcmk__xml_copy(NULL, input); - if (new_scheduler->input == NULL) { - out->err(out, "Failed to copy input XML"); - pe_free_working_set(new_scheduler); - return ENOMEM; - } - - } else { - int rc = cib__signon_query(out, NULL, &(new_scheduler->input)); - - if (rc != pcmk_rc_ok) { - pe_free_working_set(new_scheduler); - return rc; - } - } - - // Make our own copy of the given crm_time_t object; otherwise - // cluster_status() populates with the current time - if (date != NULL) { - // pcmk_copy_time() guarantees non-NULL - new_scheduler->now = pcmk_copy_time(date); - } - - // Unpack everything - cluster_status(new_scheduler); - *scheduler = new_scheduler; - - return pcmk_rc_ok; -} +#include "libpacemaker_private.h" #define XPATH_NODE_RULE "//" PCMK_XE_RULE "[@" PCMK_XA_ID "='%s']" /*! * \internal * \brief Check whether a given rule is in effect * * \param[in] scheduler Scheduler data * \param[in] rule_id The ID of the rule to check * \param[out] error Where to store a rule evaluation error message * * \return Standard Pacemaker return code */ static int eval_rule(pcmk_scheduler_t *scheduler, const char *rule_id, const char **error) { xmlNodePtr cib_constraints = NULL; xmlNodePtr match = NULL; xmlXPathObjectPtr xpath_obj = NULL; char *xpath = NULL; int rc = pcmk_rc_ok; int num_results = 0; *error = NULL; /* Rules are under the constraints node in the XML, so first find that. */ cib_constraints = pcmk_find_cib_element(scheduler->input, PCMK_XE_CONSTRAINTS); /* Get all rules matching the given ID that are also simple enough for us * to check. For the moment, these rules must only have a single * date_expression child and: * - Do not have a date_spec operation, or * - Have a date_spec operation that contains years= but does not contain * moon=. * * We do this in steps to provide better error messages. First, check that * there's any rule with the given ID. */ xpath = crm_strdup_printf(XPATH_NODE_RULE, rule_id); xpath_obj = xpath_search(cib_constraints, xpath); num_results = numXpathResults(xpath_obj); free(xpath); freeXpathObject(xpath_obj); if (num_results == 0) { *error = "Rule not found"; return ENXIO; } if (num_results > 1) { // Should not be possible; schema prevents this *error = "Found more than one rule with matching ID"; return pcmk_rc_duplicate_id; } /* Next, make sure it has exactly one date_expression. */ xpath = crm_strdup_printf(XPATH_NODE_RULE "//date_expression", rule_id); xpath_obj = xpath_search(cib_constraints, xpath); num_results = numXpathResults(xpath_obj); free(xpath); freeXpathObject(xpath_obj); if (num_results != 1) { if (num_results == 0) { *error = "Rule does not have a date expression"; } else { *error = "Rule has more than one date expression"; } return EOPNOTSUPP; } /* Then, check that it's something we actually support. */ xpath = crm_strdup_printf(XPATH_NODE_RULE "//" PCMK_XE_DATE_EXPRESSION "[@" PCMK_XA_OPERATION "!='" PCMK_VALUE_DATE_SPEC "']", rule_id); xpath_obj = xpath_search(cib_constraints, xpath); num_results = numXpathResults(xpath_obj); free(xpath); if (num_results == 0) { freeXpathObject(xpath_obj); xpath = crm_strdup_printf(XPATH_NODE_RULE "//" PCMK_XE_DATE_EXPRESSION "[@" PCMK_XA_OPERATION "='" PCMK_VALUE_DATE_SPEC "' " "and " PCMK_XE_DATE_SPEC "/@" PCMK_XA_YEARS " " "and not(" PCMK_XE_DATE_SPEC "/@" PCMK__XA_MOON ")]", rule_id); xpath_obj = xpath_search(cib_constraints, xpath); num_results = numXpathResults(xpath_obj); free(xpath); if (num_results == 0) { freeXpathObject(xpath_obj); *error = "Rule must either not use " PCMK_XE_DATE_SPEC ", or use " PCMK_XE_DATE_SPEC " with " PCMK_XA_YEARS "= but not " PCMK__XA_MOON "="; return EOPNOTSUPP; } } match = getXpathResult(xpath_obj, 0); /* We should have ensured this with the xpath query above, but double- * checking can't hurt. */ CRM_ASSERT(match != NULL); CRM_ASSERT(pcmk__expression_type(match) == pcmk__subexpr_datetime); rc = pcmk__evaluate_date_expression(match, scheduler->now, NULL); if (rc == pcmk_rc_undetermined) { // Malformed or missing *error = "Error parsing rule"; } freeXpathObject(xpath_obj); return rc; } /*! * \internal * \brief Check whether each rule in a list is in effect * * \param[in,out] out Output object * \param[in] input The CIB XML to check (if \c NULL, use current CIB) * \param[in] date Check whether the rule is in effect at this date and * time (if \c NULL, use current date and time) * \param[in] rule_ids The IDs of the rules to check, as a NULL- * terminated list. * * \return Standard Pacemaker return code */ int pcmk__check_rules(pcmk__output_t *out, xmlNodePtr input, const crm_time_t *date, const char **rule_ids) { pcmk_scheduler_t *scheduler = NULL; int rc = pcmk_rc_ok; CRM_ASSERT(out != NULL); if (rule_ids == NULL) { // Trivial case; every rule specified is in effect return pcmk_rc_ok; } - rc = init_rule_check(out, input, date, &scheduler); + rc = pcmk__init_scheduler(out, input, date, &scheduler); if (rc != pcmk_rc_ok) { return rc; } for (const char **rule_id = rule_ids; *rule_id != NULL; rule_id++) { const char *error = NULL; int last_rc = eval_rule(scheduler, *rule_id, &error); out->message(out, "rule-check", *rule_id, last_rc, error); if (last_rc != pcmk_rc_ok) { rc = last_rc; } } pe_free_working_set(scheduler); return rc; } // Documented in pacemaker.h int pcmk_check_rules(xmlNodePtr *xml, xmlNodePtr input, const crm_time_t *date, const char **rule_ids) { pcmk__output_t *out = NULL; int rc = pcmk_rc_ok; rc = pcmk__xml_output_new(&out, xml); if (rc != pcmk_rc_ok) { return rc; } pcmk__register_lib_messages(out); rc = pcmk__check_rules(out, input, date, rule_ids); pcmk__xml_output_finish(out, pcmk_rc2exitc(rc), xml); return rc; } diff --git a/lib/pacemaker/pcmk_scheduler.c b/lib/pacemaker/pcmk_scheduler.c index 5674b98a3c..caa95616e8 100644 --- a/lib/pacemaker/pcmk_scheduler.c +++ b/lib/pacemaker/pcmk_scheduler.c @@ -1,821 +1,892 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include +#include #include #include #include #include #include #include #include "libpacemaker_private.h" CRM_TRACE_INIT_DATA(pacemaker); /*! * \internal * \brief Do deferred action checks after assignment * * When unpacking the resource history, the scheduler checks for resource * configurations that have changed since an action was run. However, at that * time, bundles using the REMOTE_CONTAINER_HACK don't have their final * parameter information, so instead they add a deferred check to a list. This * function processes one entry in that list. * * \param[in,out] rsc Resource that action history is for * \param[in,out] node Node that action history is for * \param[in] rsc_op Action history entry * \param[in] check Type of deferred check to do */ static void check_params(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *rsc_op, enum pcmk__check_parameters check) { const char *reason = NULL; pcmk__op_digest_t *digest_data = NULL; switch (check) { case pcmk__check_active: if (pcmk__check_action_config(rsc, node, rsc_op) && pe_get_failcount(node, rsc, NULL, pcmk__fc_effective, NULL)) { reason = "action definition changed"; } break; case pcmk__check_last_failure: digest_data = rsc_action_digest_cmp(rsc, rsc_op, node, rsc->cluster); switch (digest_data->rc) { case pcmk__digest_unknown: crm_trace("Resource %s history entry %s on %s has " "no digest to compare", rsc->id, pcmk__xe_id(rsc_op), node->details->id); break; case pcmk__digest_match: break; default: reason = "resource parameters have changed"; break; } break; } if (reason != NULL) { pe__clear_failcount(rsc, node, reason, rsc->cluster); } } /*! * \internal * \brief Check whether a resource has failcount clearing scheduled on a node * * \param[in] node Node to check * \param[in] rsc Resource to check * * \return true if \p rsc has failcount clearing scheduled on \p node, * otherwise false */ static bool failcount_clear_action_exists(const pcmk_node_t *node, const pcmk_resource_t *rsc) { GList *list = pe__resource_actions(rsc, node, PCMK_ACTION_CLEAR_FAILCOUNT, TRUE); if (list != NULL) { g_list_free(list); return true; } return false; } /*! * \internal * \brief Ban a resource from a node if it reached its failure threshold there * * \param[in,out] data Resource to check failure threshold for * \param[in] user_data Node to check resource on */ static void check_failure_threshold(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; const pcmk_node_t *node = user_data; // If this is a collective resource, apply recursively to children instead if (rsc->children != NULL) { g_list_foreach(rsc->children, check_failure_threshold, user_data); return; } if (!failcount_clear_action_exists(node, rsc)) { /* Don't force the resource away from this node due to a failcount * that's going to be cleared. * * @TODO Failcount clearing can be scheduled in * pcmk__handle_rsc_config_changes() via process_rsc_history(), or in * schedule_resource_actions() via check_params(). This runs well before * then, so it cannot detect those, meaning we might check the migration * threshold when we shouldn't. Worst case, we stop or move the * resource, then move it back in the next transition. */ pcmk_resource_t *failed = NULL; if (pcmk__threshold_reached(rsc, node, &failed)) { resource_location(failed, node, -PCMK_SCORE_INFINITY, "__fail_limit__", rsc->cluster); } } } /*! * \internal * \brief If resource has exclusive discovery, ban node if not allowed * * Location constraints have a PCMK_XA_RESOURCE_DISCOVERY option that allows * users to specify where probes are done for the affected resource. If this is * set to \c exclusive, probes will only be done on nodes listed in exclusive * constraints. This function bans the resource from the node if the node is not * listed. * * \param[in,out] data Resource to check * \param[in] user_data Node to check resource on */ static void apply_exclusive_discovery(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; const pcmk_node_t *node = user_data; if (rsc->exclusive_discover || pe__const_top_resource(rsc, false)->exclusive_discover) { pcmk_node_t *match = NULL; // If this is a collective resource, apply recursively to children g_list_foreach(rsc->children, apply_exclusive_discovery, user_data); match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if ((match != NULL) && (match->rsc_discover_mode != pcmk_probe_exclusive)) { match->weight = -PCMK_SCORE_INFINITY; } } } /*! * \internal * \brief Apply stickiness to a resource if appropriate * * \param[in,out] data Resource to check for stickiness * \param[in] user_data Ignored */ static void apply_stickiness(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; pcmk_node_t *node = NULL; // If this is a collective resource, apply recursively to children instead if (rsc->children != NULL) { g_list_foreach(rsc->children, apply_stickiness, NULL); return; } /* A resource is sticky if it is managed, has stickiness configured, and is * active on a single node. */ if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed) || (rsc->stickiness < 1) || !pcmk__list_of_1(rsc->running_on)) { return; } node = rsc->running_on->data; /* In a symmetric cluster, stickiness can always be used. In an * asymmetric cluster, we have to check whether the resource is still * allowed on the node, so we don't keep the resource somewhere it is no * longer explicitly enabled. */ if (!pcmk_is_set(rsc->cluster->flags, pcmk_sched_symmetric_cluster) && (g_hash_table_lookup(rsc->allowed_nodes, node->details->id) == NULL)) { pcmk__rsc_debug(rsc, "Ignoring %s stickiness because the cluster is " "asymmetric and %s is not explicitly allowed", rsc->id, pcmk__node_name(node)); return; } pcmk__rsc_debug(rsc, "Resource %s has %d stickiness on %s", rsc->id, rsc->stickiness, pcmk__node_name(node)); resource_location(rsc, node, rsc->stickiness, "stickiness", rsc->cluster); } /*! * \internal * \brief Apply shutdown locks for all resources as appropriate * * \param[in,out] scheduler Scheduler data */ static void apply_shutdown_locks(pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { return; } for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; rsc->cmds->shutdown_lock(rsc); } } /*! * \internal * \brief Calculate the number of available nodes in the cluster * * \param[in,out] scheduler Scheduler data */ static void count_available_nodes(pcmk_scheduler_t *scheduler) { if (pcmk_is_set(scheduler->flags, pcmk_sched_no_compat)) { return; } // @COMPAT for API backward compatibility only (cluster does not use value) for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; if ((node != NULL) && (node->weight >= 0) && node->details->online && (node->details->type != node_ping)) { scheduler->max_valid_nodes++; } } crm_trace("Online node count: %d", scheduler->max_valid_nodes); } /* * \internal * \brief Apply node-specific scheduling criteria * * After the CIB has been unpacked, process node-specific scheduling criteria * including shutdown locks, location constraints, resource stickiness, * migration thresholds, and exclusive resource discovery. */ static void apply_node_criteria(pcmk_scheduler_t *scheduler) { crm_trace("Applying node-specific scheduling criteria"); apply_shutdown_locks(scheduler); count_available_nodes(scheduler); pcmk__apply_locations(scheduler); g_list_foreach(scheduler->resources, apply_stickiness, NULL); for (GList *node_iter = scheduler->nodes; node_iter != NULL; node_iter = node_iter->next) { for (GList *rsc_iter = scheduler->resources; rsc_iter != NULL; rsc_iter = rsc_iter->next) { check_failure_threshold(rsc_iter->data, node_iter->data); apply_exclusive_discovery(rsc_iter->data, node_iter->data); } } } /*! * \internal * \brief Assign resources to nodes * * \param[in,out] scheduler Scheduler data */ static void assign_resources(pcmk_scheduler_t *scheduler) { GList *iter = NULL; crm_trace("Assigning resources to nodes"); if (!pcmk__str_eq(scheduler->placement_strategy, PCMK_VALUE_DEFAULT, pcmk__str_casei)) { pcmk__sort_resources(scheduler); } pcmk__show_node_capacities("Original", scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_have_remote_nodes)) { /* Assign remote connection resources first (which will also assign any * colocation dependencies). If the connection is migrating, always * prefer the partial migration target. */ for (iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; if (rsc->is_remote_node) { pcmk__rsc_trace(rsc, "Assigning remote connection resource '%s'", rsc->id); rsc->cmds->assign(rsc, rsc->partial_migration_target, true); } } } /* now do the rest of the resources */ for (iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; if (!rsc->is_remote_node) { pcmk__rsc_trace(rsc, "Assigning %s resource '%s'", rsc->xml->name, rsc->id); rsc->cmds->assign(rsc, NULL, true); } } pcmk__show_node_capacities("Remaining", scheduler); } /*! * \internal * \brief Schedule fail count clearing on online nodes if resource is orphaned * * \param[in,out] data Resource to check * \param[in] user_data Ignored */ static void clear_failcounts_if_orphaned(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)) { return; } crm_trace("Clear fail counts for orphaned resource %s", rsc->id); /* There's no need to recurse into rsc->children because those * should just be unassigned clone instances. */ for (GList *iter = rsc->cluster->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; pcmk_action_t *clear_op = NULL; if (!node->details->online) { continue; } if (pe_get_failcount(node, rsc, NULL, pcmk__fc_effective, NULL) == 0) { continue; } clear_op = pe__clear_failcount(rsc, node, "it is orphaned", rsc->cluster); /* We can't use order_action_then_stop() here because its * pcmk__ar_guest_allowed breaks things */ pcmk__new_ordering(clear_op->rsc, NULL, clear_op, rsc, stop_key(rsc), NULL, pcmk__ar_ordered, rsc->cluster); } } /*! * \internal * \brief Schedule any resource actions needed * * \param[in,out] scheduler Scheduler data */ static void schedule_resource_actions(pcmk_scheduler_t *scheduler) { // Process deferred action checks pe__foreach_param_check(scheduler, check_params); pe__free_param_checks(scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_probe_resources)) { crm_trace("Scheduling probes"); pcmk__schedule_probes(scheduler); } if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { g_list_foreach(scheduler->resources, clear_failcounts_if_orphaned, NULL); } crm_trace("Scheduling resource actions"); for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; rsc->cmds->create_actions(rsc); } } /*! * \internal * \brief Check whether a resource or any of its descendants are managed * * \param[in] rsc Resource to check * * \return true if resource or any descendant is managed, otherwise false */ static bool is_managed(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk_rsc_managed)) { return true; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { if (is_managed((pcmk_resource_t *) iter->data)) { return true; } } return false; } /*! * \internal * \brief Check whether any resources in the cluster are managed * * \param[in] scheduler Scheduler data * * \return true if any resource is managed, otherwise false */ static bool any_managed_resources(const pcmk_scheduler_t *scheduler) { for (const GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { if (is_managed((const pcmk_resource_t *) iter->data)) { return true; } } return false; } /*! * \internal * \brief Check whether a node requires fencing * * \param[in] node Node to check * \param[in] have_managed Whether any resource in cluster is managed * * \return true if \p node should be fenced, otherwise false */ static bool needs_fencing(const pcmk_node_t *node, bool have_managed) { return have_managed && node->details->unclean && pe_can_fence(node->details->data_set, node); } /*! * \internal * \brief Check whether a node requires shutdown * * \param[in] node Node to check * * \return true if \p node should be shut down, otherwise false */ static bool needs_shutdown(const pcmk_node_t *node) { if (pcmk__is_pacemaker_remote_node(node)) { /* Do not send shutdown actions for Pacemaker Remote nodes. * @TODO We might come up with a good use for this in the future. */ return false; } return node->details->online && node->details->shutdown; } /*! * \internal * \brief Track and order non-DC fencing * * \param[in,out] list List of existing non-DC fencing actions * \param[in,out] action Fencing action to prepend to \p list * \param[in] scheduler Scheduler data * * \return (Possibly new) head of \p list */ static GList * add_nondc_fencing(GList *list, pcmk_action_t *action, const pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing) && (list != NULL)) { /* Concurrent fencing is disabled, so order each non-DC * fencing in a chain. If there is any DC fencing or * shutdown, it will be ordered after the last action in the * chain later. */ order_actions((pcmk_action_t *) list->data, action, pcmk__ar_ordered); } return g_list_prepend(list, action); } /*! * \internal * \brief Schedule a node for fencing * * \param[in,out] node Node that requires fencing */ static pcmk_action_t * schedule_fencing(pcmk_node_t *node) { pcmk_action_t *fencing = pe_fence_op(node, NULL, FALSE, "node is unclean", FALSE, node->details->data_set); pcmk__sched_warn("Scheduling node %s for fencing", pcmk__node_name(node)); pcmk__order_vs_fence(fencing, node->details->data_set); return fencing; } /*! * \internal * \brief Create and order node fencing and shutdown actions * * \param[in,out] scheduler Scheduler data */ static void schedule_fencing_and_shutdowns(pcmk_scheduler_t *scheduler) { pcmk_action_t *dc_down = NULL; bool integrity_lost = false; bool have_managed = any_managed_resources(scheduler); GList *fencing_ops = NULL; GList *shutdown_ops = NULL; crm_trace("Scheduling fencing and shutdowns as needed"); if (!have_managed) { crm_notice("No fencing will be done until there are resources " "to manage"); } // Check each node for whether it needs fencing or shutdown for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; pcmk_action_t *fencing = NULL; /* Guest nodes are "fenced" by recovering their container resource, * so handle them separately. */ if (pcmk__is_guest_or_bundle_node(node)) { if (node->details->remote_requires_reset && have_managed && pe_can_fence(scheduler, node)) { pcmk__fence_guest(node); } continue; } if (needs_fencing(node, have_managed)) { fencing = schedule_fencing(node); // Track DC and non-DC fence actions separately if (node->details->is_dc) { dc_down = fencing; } else { fencing_ops = add_nondc_fencing(fencing_ops, fencing, scheduler); } } else if (needs_shutdown(node)) { pcmk_action_t *down_op = pcmk__new_shutdown_action(node); // Track DC and non-DC shutdown actions separately if (node->details->is_dc) { dc_down = down_op; } else { shutdown_ops = g_list_prepend(shutdown_ops, down_op); } } if ((fencing == NULL) && node->details->unclean) { integrity_lost = true; pcmk__config_warn("Node %s is unclean but cannot be fenced", pcmk__node_name(node)); } } if (integrity_lost) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("Resource functionality and data integrity " "cannot be guaranteed (configure, enable, " "and test fencing to correct this)"); } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) { crm_notice("Unclean nodes will not be fenced until quorum is " "attained or " PCMK_OPT_NO_QUORUM_POLICY " is set to " PCMK_VALUE_IGNORE); } } if (dc_down != NULL) { /* Order any non-DC shutdowns before any DC shutdown, to avoid repeated * DC elections. However, we don't want to order non-DC shutdowns before * a DC *fencing*, because even though we don't want a node that's * shutting down to become DC, the DC fencing could be ordered before a * clone stop that's also ordered before the shutdowns, thus leading to * a graph loop. */ if (pcmk__str_eq(dc_down->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { pcmk__order_after_each(dc_down, shutdown_ops); } // Order any non-DC fencing before any DC fencing or shutdown if (pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing)) { /* With concurrent fencing, order each non-DC fencing action * separately before any DC fencing or shutdown. */ pcmk__order_after_each(dc_down, fencing_ops); } else if (fencing_ops != NULL) { /* Without concurrent fencing, the non-DC fencing actions are * already ordered relative to each other, so we just need to order * the DC fencing after the last action in the chain (which is the * first item in the list). */ order_actions((pcmk_action_t *) fencing_ops->data, dc_down, pcmk__ar_ordered); } } g_list_free(fencing_ops); g_list_free(shutdown_ops); } static void log_resource_details(pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv; GList *all = NULL; /* Due to the `crm_mon --node=` feature, out->message() for all the * resource-related messages expects a list of nodes that we are allowed to * output information for. Here, we create a wildcard to match all nodes. */ all = g_list_prepend(all, (gpointer) "*"); for (GList *item = scheduler->resources; item != NULL; item = item->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) item->data; // Log all resources except inactive orphans if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed) || (rsc->role != pcmk_role_stopped)) { out->message(out, crm_map_element_name(rsc->xml), 0UL, rsc, all, all); } } g_list_free(all); } static void log_all_actions(pcmk_scheduler_t *scheduler) { /* This only ever outputs to the log, so ignore whatever output object was * previously set and just log instead. */ pcmk__output_t *prev_out = scheduler->priv; pcmk__output_t *out = NULL; if (pcmk__log_output_new(&out) != pcmk_rc_ok) { return; } pe__register_messages(out); pcmk__register_lib_messages(out); pcmk__output_set_log_level(out, LOG_NOTICE); scheduler->priv = out; out->begin_list(out, NULL, NULL, "Actions"); pcmk__output_actions(scheduler); out->end_list(out); out->finish(out, CRM_EX_OK, true, NULL); pcmk__output_free(out); scheduler->priv = prev_out; } /*! * \internal * \brief Log all required but unrunnable actions at trace level * * \param[in] scheduler Scheduler data */ static void log_unrunnable_actions(const pcmk_scheduler_t *scheduler) { const uint64_t flags = pcmk_action_optional |pcmk_action_runnable |pcmk_action_pseudo; crm_trace("Required but unrunnable actions:"); for (const GList *iter = scheduler->actions; iter != NULL; iter = iter->next) { const pcmk_action_t *action = (const pcmk_action_t *) iter->data; if (!pcmk_any_flags_set(action->flags, flags)) { pcmk__log_action("\t", action, true); } } } /*! * \internal * \brief Unpack the CIB for scheduling * * \param[in,out] cib CIB XML to unpack (may be NULL if already unpacked) * \param[in] flags Scheduler flags to set in addition to defaults * \param[in,out] scheduler Scheduler data */ static void unpack_cib(xmlNode *cib, unsigned long long flags, pcmk_scheduler_t *scheduler) { const char* localhost_save = NULL; if (pcmk_is_set(scheduler->flags, pcmk_sched_have_status)) { crm_trace("Reusing previously calculated cluster status"); pcmk__set_scheduler_flags(scheduler, flags); return; } if (scheduler->localhost) { localhost_save = scheduler->localhost; } CRM_ASSERT(cib != NULL); crm_trace("Calculating cluster status"); /* This will zero the entire struct without freeing anything first, so * callers should never call pcmk__schedule_actions() with a populated data * set unless pcmk_sched_have_status is set (i.e. cluster_status() was * previously called, whether directly or via pcmk__schedule_actions()). */ set_working_set_defaults(scheduler); if (localhost_save) { scheduler->localhost = localhost_save; } pcmk__set_scheduler_flags(scheduler, flags); scheduler->input = cib; cluster_status(scheduler); // Sets pcmk_sched_have_status } /*! * \internal * \brief Run the scheduler for a given CIB * * \param[in,out] cib CIB XML to use as scheduler input * \param[in] flags Scheduler flags to set in addition to defaults * \param[in,out] scheduler Scheduler data */ void pcmk__schedule_actions(xmlNode *cib, unsigned long long flags, pcmk_scheduler_t *scheduler) { unpack_cib(cib, flags, scheduler); pcmk__set_assignment_methods(scheduler); pcmk__apply_node_health(scheduler); pcmk__unpack_constraints(scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_validate_only)) { return; } if (!pcmk_is_set(scheduler->flags, pcmk_sched_location_only) && pcmk__is_daemon) { log_resource_details(scheduler); } apply_node_criteria(scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { return; } pcmk__create_internal_constraints(scheduler); pcmk__handle_rsc_config_changes(scheduler); assign_resources(scheduler); schedule_resource_actions(scheduler); /* Remote ordering constraints need to happen prior to calculating fencing * because it is one more place we can mark nodes as needing fencing. */ pcmk__order_remote_connection_actions(scheduler); schedule_fencing_and_shutdowns(scheduler); pcmk__apply_orderings(scheduler); log_all_actions(scheduler); pcmk__create_graph(scheduler); if (get_crm_log_level() == LOG_TRACE) { log_unrunnable_actions(scheduler); } } + +/*! + * \internal + * \brief Initialize scheduler data + * + * Make our own copies of the CIB XML and date/time object, if they're not + * \c NULL. This way we don't have to take ownership of the objects passed via + * the API. + * + * This function is most useful for public API functions that want the caller + * to retain ownership of the CIB object + * + * \param[in,out] out Output object + * \param[in] input The CIB XML to check (if \c NULL, use current CIB) + * \param[in] date Date and time to use in the scheduler (if \c NULL, + * use current date and time). This can be used for + * checking whether a rule is in effect at a certa + * date and time. + * \param[out] scheduler Where to store initialized scheduler data + * + * \return Standard Pacemaker return code + */ +int +pcmk__init_scheduler(pcmk__output_t *out, xmlNodePtr input, const crm_time_t *date, + pcmk_scheduler_t **scheduler) +{ + // Allows for cleaner syntax than dereferencing the scheduler argument + pcmk_scheduler_t *new_scheduler = NULL; + + new_scheduler = pe_new_working_set(); + if (new_scheduler == NULL) { + return ENOMEM; + } + + pcmk__set_scheduler_flags(new_scheduler, + pcmk_sched_no_counts|pcmk_sched_no_compat); + + // Populate the scheduler data + + // Make our own copy of the given input or fetch the CIB and use that + if (input != NULL) { + new_scheduler->input = pcmk__xml_copy(NULL, input); + if (new_scheduler->input == NULL) { + out->err(out, "Failed to copy input XML"); + pe_free_working_set(new_scheduler); + return ENOMEM; + } + + } else { + int rc = cib__signon_query(out, NULL, &(new_scheduler->input)); + + if (rc != pcmk_rc_ok) { + pe_free_working_set(new_scheduler); + return rc; + } + } + + // Make our own copy of the given crm_time_t object; otherwise + // cluster_status() populates with the current time + if (date != NULL) { + // pcmk_copy_time() guarantees non-NULL + new_scheduler->now = pcmk_copy_time(date); + } + + // Unpack everything + cluster_status(new_scheduler); + *scheduler = new_scheduler; + + return pcmk_rc_ok; +}