diff --git a/lib/pacemaker/libpacemaker_private.h b/lib/pacemaker/libpacemaker_private.h index e32064821e..02b012a05b 100644 --- a/lib/pacemaker/libpacemaker_private.h +++ b/lib/pacemaker/libpacemaker_private.h @@ -1,1159 +1,1162 @@ /* * Copyright 2021-2025 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__PACEMAKER_LIBPACEMAKER_PRIVATE__H #define PCMK__PACEMAKER_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 // NULL #include // uint32_t #include // bool, false #include // guint, gpointer, GList, GHashTable #include // xmlNode #include // pcmk_action_t, pcmk_node_t, etc. #include // pcmk__location_t, etc. #include // cib_t #include // lrmd_event_data_t #include // pe__const_top_resource(), etc. #include // pcmk_injections_t #include // pcmk__colocation_t #ifdef __cplusplus extern "C" { #endif // 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 pcmk__assignment_methods { /*! * \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 * * \return The score added to the dependent's priority */ int (*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) G_GNUC_INTERNAL 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 int pcmk__parse_constraint_role(const char *id, const char *role_spec, enum rsc_role_e *role); 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__idref_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, }; G_GNUC_INTERNAL const char *pcmk__colocation_node_attr(const pcmk_node_t *node, const char *attr, const pcmk_resource_t *rsc); 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 int 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_spec, const char *primary_role_spec, uint32_t flags); G_GNUC_INTERNAL void pcmk__block_colocation_dependents(pcmk_action_t *action); G_GNUC_INTERNAL bool pcmk__colocation_has_influence(const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc); // 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)->priv->scheduler) #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 int 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 int 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 int 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 int 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 int 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, cib_t *cib_conn, xmlNode *cib_node, const char *resource, const char *task, guint interval_ms, int rc, bool infinity); G_GNUC_INTERNAL xmlNode *pcmk__inject_action_result(xmlNode *cib_resource, lrmd_event_data_t *op, const char *node, int target_rc); // Nodes (pcmk_sched_nodes.c) //! Options for checking node availability enum pcmk__node_availability { //! Disallow offline or unclean nodes (always implied) pcmk__node_alive = 0, //! Disallow shutting down, standby, and maintenance nodes pcmk__node_usable = (1 << 0), + //! Disallow nodes with zero score + pcmk__node_no_zero = (1 << 1), + //! Disallow nodes with negative scores pcmk__node_no_negative = (1 << 2), //! Disallow guest nodes whose guest resource is unrunnable pcmk__node_no_unrunnable_guest = (1 << 4), //! Exempt guest nodes from alive and usable checks pcmk__node_exempt_guest = (1 << 5), }; G_GNUC_INTERNAL bool pcmk__node_available(const pcmk_node_t *node, uint32_t flags); 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); // General setup functions (pcmk_setup.c) G_GNUC_INTERNAL int pcmk__setup_output_cib_sched(pcmk__output_t **out, cib_t **cib, pcmk_scheduler_t **scheduler, xmlNode **xml); G_GNUC_INTERNAL int pcmk__setup_output_fencing(pcmk__output_t **out, stonith_t **st, xmlNode **xml); #ifdef __cplusplus } #endif #endif // PCMK__PACEMAKER_LIBPACEMAKER_PRIVATE__H diff --git a/lib/pacemaker/pcmk_sched_nodes.c b/lib/pacemaker/pcmk_sched_nodes.c index cdeb13a1aa..73a845785e 100644 --- a/lib/pacemaker/pcmk_sched_nodes.c +++ b/lib/pacemaker/pcmk_sched_nodes.c @@ -1,459 +1,463 @@ /* * Copyright 2004-2025 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 "libpacemaker_private.h" /*! * \internal * \brief Check whether a node is available to run resources * * \param[in] node Node to check * \param[in] flags Group of enum pcmk__node_availability flags * * \return true if node is available per flags, otherwise false */ bool pcmk__node_available(const pcmk_node_t *node, uint32_t flags) { if ((node == NULL) || (node->details == NULL)) { return false; // A nonexistent node is not available } // Guest nodes may be exempted from alive and usable checks if (!pcmk_is_set(flags, pcmk__node_exempt_guest) || !pcmk__is_guest_or_bundle_node(node)) { // pcmk__node_alive is implicit if (!node->details->online || node->details->unclean) { return false; } if (pcmk_is_set(flags, pcmk__node_usable) && (node->details->shutdown || pcmk_is_set(node->priv->flags, pcmk__node_standby) || node->details->maintenance)) { return false; } } + if (pcmk_is_set(flags, pcmk__node_no_zero) && (node->assign->score == 0)) { + return false; + } + if (pcmk_is_set(flags, pcmk__node_no_negative) && (node->assign->score < 0)) { return false; } if (pcmk_is_set(flags, pcmk__node_no_unrunnable_guest) && pcmk__is_guest_or_bundle_node(node)) { pcmk_resource_t *guest = node->priv->remote->priv->launcher; if (guest->priv->fns->location(guest, NULL, pcmk__rsc_node_assigned) == NULL) { return false; } } return true; } /*! * \internal * \brief Create a hash table with copies of another table's nodes * * \param[in] nodes Hash table to copy * * \return New table with copies of nodes in \p nodes, or \c NULL if \p nodes is * \c NULL */ GHashTable * pcmk__copy_node_table(GHashTable *nodes) { GHashTable *new_table = NULL; GHashTableIter iter; pcmk_node_t *node = NULL; if (nodes == NULL) { return NULL; } new_table = pcmk__strkey_table(NULL, pcmk__free_node_copy); g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { pcmk_node_t *new_node = pe__copy_node(node); g_hash_table_insert(new_table, (gpointer) new_node->priv->id, new_node); } return new_table; } /*! * \internal * \brief Free a table of node tables * * \param[in,out] data Table to free * * \note This is a \c GDestroyNotify wrapper for \c g_hash_table_destroy(). */ static void destroy_node_tables(gpointer data) { g_hash_table_destroy((GHashTable *) data); } /*! * \internal * \brief Recursively copy the node tables of a resource * * Build a hash table containing copies of the allowed nodes tables of \p rsc * and its entire tree of descendants. The key is the resource ID, and the value * is a copy of the resource's node table. * * \param[in] rsc Resource whose node table to copy * \param[in,out] copy Where to store the copied node tables * * \note \p *copy should be \c NULL for the top-level call. * \note The caller is responsible for freeing \p copy using * \c g_hash_table_destroy(). */ void pcmk__copy_node_tables(const pcmk_resource_t *rsc, GHashTable **copy) { pcmk__assert((rsc != NULL) && (copy != NULL)); if (*copy == NULL) { *copy = pcmk__strkey_table(NULL, destroy_node_tables); } g_hash_table_insert(*copy, rsc->id, pcmk__copy_node_table(rsc->priv->allowed_nodes)); for (const GList *iter = rsc->priv->children; iter != NULL; iter = iter->next) { pcmk__copy_node_tables((const pcmk_resource_t *) iter->data, copy); } } /*! * \internal * \brief Recursively restore the node tables of a resource from backup * * Given a hash table containing backup copies of the allowed nodes tables of * \p rsc and its entire tree of descendants, replace the resources' current * node tables with the backed-up copies. * * \param[in,out] rsc Resource whose node tables to restore * \param[in] backup Table of backup node tables (created by * \c pcmk__copy_node_tables()) * * \note This function frees the resources' current node tables. */ void pcmk__restore_node_tables(pcmk_resource_t *rsc, GHashTable *backup) { pcmk__assert((rsc != NULL) && (backup != NULL)); g_hash_table_destroy(rsc->priv->allowed_nodes); // Copy to avoid danger with multiple restores rsc->priv->allowed_nodes = pcmk__copy_node_table(g_hash_table_lookup(backup, rsc->id)); for (GList *iter = rsc->priv->children; iter != NULL; iter = iter->next) { pcmk__restore_node_tables((pcmk_resource_t *) iter->data, backup); } } /*! * \internal * \brief Copy a list of node objects * * \param[in] list List to copy * \param[in] reset Set copies' scores to 0 * * \return New list of shallow copies of nodes in original list */ GList * pcmk__copy_node_list(const GList *list, bool reset) { GList *result = NULL; for (const GList *iter = list; iter != NULL; iter = iter->next) { pcmk_node_t *new_node = NULL; pcmk_node_t *this_node = iter->data; new_node = pe__copy_node(this_node); if (reset) { new_node->assign->score = 0; } result = g_list_prepend(result, new_node); } return result; } /*! * \internal * \brief Compare two nodes for assignment preference * * Given two nodes, check which one is more preferred by assignment criteria * such as node score and utilization. * * \param[in] a First node to compare * \param[in] b Second node to compare * \param[in] data Node to prefer if all else equal * * \return -1 if \p a is preferred, +1 if \p b is preferred, or 0 if they are * equally preferred */ static gint compare_nodes(gconstpointer a, gconstpointer b, gpointer data) { const pcmk_node_t *node1 = (const pcmk_node_t *) a; const pcmk_node_t *node2 = (const pcmk_node_t *) b; const pcmk_node_t *preferred = (const pcmk_node_t *) data; int node1_score = -PCMK_SCORE_INFINITY; int node2_score = -PCMK_SCORE_INFINITY; int result = 0; if (a == NULL) { return 1; } if (b == NULL) { return -1; } // Compare node scores if (pcmk__node_available(node1, pcmk__node_alive|pcmk__node_usable)) { node1_score = node1->assign->score; } if (pcmk__node_available(node2, pcmk__node_alive|pcmk__node_usable)) { node2_score = node2->assign->score; } if (node1_score > node2_score) { crm_trace("%s before %s (score %d > %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1_score, node2_score); return -1; } if (node1_score < node2_score) { crm_trace("%s after %s (score %d < %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1_score, node2_score); return 1; } // If appropriate, compare node utilization if (pcmk__str_eq(node1->priv->scheduler->priv->placement_strategy, PCMK_VALUE_MINIMAL, pcmk__str_casei)) { goto equal; } if (pcmk__str_eq(node1->priv->scheduler->priv->placement_strategy, PCMK_VALUE_BALANCED, pcmk__str_casei)) { result = pcmk__compare_node_capacities(node1, node2); if (result < 0) { crm_trace("%s before %s (greater capacity by %d attributes)", pcmk__node_name(node1), pcmk__node_name(node2), result * -1); return -1; } else if (result > 0) { crm_trace("%s after %s (lower capacity by %d attributes)", pcmk__node_name(node1), pcmk__node_name(node2), result); return 1; } } // Compare number of resources already assigned to node if (node1->priv->num_resources < node2->priv->num_resources) { crm_trace("%s before %s (%d resources < %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1->priv->num_resources, node2->priv->num_resources); return -1; } else if (node1->priv->num_resources > node2->priv->num_resources) { crm_trace("%s after %s (%d resources > %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1->priv->num_resources, node2->priv->num_resources); return 1; } // Check whether one node is already running desired resource if (preferred != NULL) { if (pcmk__same_node(preferred, node1)) { crm_trace("%s before %s (preferred node)", pcmk__node_name(node1), pcmk__node_name(node2)); return -1; } else if (pcmk__same_node(preferred, node2)) { crm_trace("%s after %s (not preferred node)", pcmk__node_name(node1), pcmk__node_name(node2)); return 1; } } // If all else is equal, prefer node with lowest-sorting name equal: result = strcmp(node1->priv->name, node2->priv->name); if (result < 0) { crm_trace("%s before %s (name)", pcmk__node_name(node1), pcmk__node_name(node2)); return -1; } else if (result > 0) { crm_trace("%s after %s (name)", pcmk__node_name(node1), pcmk__node_name(node2)); return 1; } crm_trace("%s == %s", pcmk__node_name(node1), pcmk__node_name(node2)); return 0; } /*! * \internal * \brief Sort a list of nodes by assigment preference * * \param[in,out] nodes Node list to sort * \param[in] active_node Node where resource being assigned is active * * \return New head of sorted list */ GList * pcmk__sort_nodes(GList *nodes, pcmk_node_t *active_node) { return g_list_sort_with_data(nodes, compare_nodes, active_node); } /*! * \internal * \brief Check whether any node is available to run resources * * \param[in] nodes Nodes to check * * \return true if any node in \p nodes is available to run resources, * otherwise false */ bool pcmk__any_node_available(GHashTable *nodes) { GHashTableIter iter; const pcmk_node_t *node = NULL; if (nodes == NULL) { return false; } g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (pcmk__node_available(node, pcmk__node_alive |pcmk__node_usable |pcmk__node_no_negative)) { return true; } } return false; } /*! * \internal * \brief Apply node health values for all nodes in cluster * * \param[in,out] scheduler Scheduler data */ void pcmk__apply_node_health(pcmk_scheduler_t *scheduler) { int base_health = 0; enum pcmk__health_strategy strategy; const char *strategy_str = pcmk__cluster_option(scheduler->priv->options, PCMK_OPT_NODE_HEALTH_STRATEGY); strategy = pcmk__parse_health_strategy(strategy_str); if (strategy == pcmk__health_strategy_none) { return; } crm_info("Applying node health strategy '%s'", strategy_str); // The progressive strategy can use a base health score if (strategy == pcmk__health_strategy_progressive) { base_health = pcmk__health_score(PCMK_OPT_NODE_HEALTH_BASE, scheduler); } for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; int health = pe__sum_node_health_scores(node, base_health); // An overall health score of 0 has no effect if (health == 0) { continue; } crm_info("Overall system health of %s is %d", pcmk__node_name(node), health); // Use node health as a location score for each resource on the node for (GList *r = scheduler->priv->resources; r != NULL; r = r->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) r->data; bool constrain = true; if (health < 0) { /* Negative health scores do not apply to resources with * PCMK_META_ALLOW_UNHEALTHY_NODES=true. */ constrain = !crm_is_true(g_hash_table_lookup(rsc->priv->meta, PCMK_META_ALLOW_UNHEALTHY_NODES)); } if (constrain) { pcmk__new_location(strategy_str, rsc, health, NULL, node); } else { pcmk__rsc_trace(rsc, "%s is immune from health ban on %s", rsc->id, pcmk__node_name(node)); } } } } /*! * \internal * \brief Check for a node in a resource's parent's allowed nodes * * \param[in] rsc Resource whose parent should be checked * \param[in] node Node to check for * * \return Equivalent of \p node from \p rsc's parent's allowed nodes if any, * otherwise NULL */ pcmk_node_t * pcmk__top_allowed_node(const pcmk_resource_t *rsc, const pcmk_node_t *node) { GHashTable *allowed_nodes = NULL; if ((rsc == NULL) || (node == NULL)) { return NULL; } if (rsc->priv->parent == NULL) { allowed_nodes = rsc->priv->allowed_nodes; } else { allowed_nodes = rsc->priv->parent->priv->allowed_nodes; } return g_hash_table_lookup(allowed_nodes, node->priv->id); } diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c index 4b7d11746e..a5b1a1b384 100644 --- a/lib/pacemaker/pcmk_sched_primitive.c +++ b/lib/pacemaker/pcmk_sched_primitive.c @@ -1,1725 +1,1726 @@ /* * Copyright 2004-2025 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->priv->allowed_nodes != NULL) { GList *nodes = g_hash_table_get_values(rsc->priv->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->priv->assigned_node != 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->priv->allowed_nodes, prefer->priv->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->assign->score < best->assign->score) { 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, pcmk__node_alive |pcmk__node_usable |pcmk__node_no_negative)) { 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 (!pcmk__is_unique_clone(rsc->priv->parent) - && (chosen->assign->score > 0) // Zero not acceptable - && pcmk__node_available(chosen, - pcmk__node_alive|pcmk__node_usable)) { + && pcmk__node_available(chosen, pcmk__node_alive + |pcmk__node_usable + |pcmk__node_no_zero + |pcmk__node_no_negative)) { /* 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, pcmk__node_alive |pcmk__node_usable |pcmk__node_no_negative)) { 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->assign->score != chosen->assign->score) { // 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->assign->score >= PCMK_SCORE_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->assign->score)); } } } 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->priv->assigned_node != 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 > -PCMK_SCORE_INFINITY))) { archive = pcmk__copy_node_table(rsc->priv->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->priv->cmds->assign(other, NULL, true); } // Apply the colocation score to this resource's allowed node scores rsc->priv->cmds->apply_coloc_score(rsc, other, colocation, true); if ((archive != NULL) && !pcmk__any_node_available(rsc->priv->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->priv->allowed_nodes); rsc->priv->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 = pcmk_find_node(connection->priv->scheduler, connection->id); CRM_CHECK(remote_node != NULL, return); if ((connection->priv->assigned_node != NULL) && (connection->priv->next_role != pcmk_role_stopped)) { crm_trace("Pacemaker Remote node %s will be online", remote_node->priv->id); remote_node->details->online = TRUE; if (!pcmk_is_set(remote_node->priv->flags, pcmk__node_seen)) { // 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->priv->id, ((connection->priv->assigned_node == NULL)? "un" : ""), pcmk_role_text(connection->priv->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_resource_t *parent = NULL; pcmk__colocation_t *colocation = NULL; pcmk_scheduler_t *scheduler = NULL; pcmk__assert(pcmk__is_primitive(rsc)); scheduler = rsc->priv->scheduler; parent = rsc->priv->parent; // Never assign a child without parent being assigned first if ((parent != NULL) && !pcmk_is_set(parent->flags, pcmk__rsc_assigning)) { pcmk__rsc_debug(rsc, "%s: Assigning parent %s first", rsc->id, parent->id); parent->priv->cmds->assign(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->priv->assigned_node != NULL) { node_name = pcmk__node_name(rsc->priv->assigned_node); } pcmk__rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name); return rsc->priv->assigned_node; } // 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->priv->allowed_nodes, scheduler); 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 <= -PCMK_SCORE_INFINITY) || (colocation->score >= PCMK_SCORE_INFINITY)) { apply_this_with(colocation, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score <= -PCMK_SCORE_INFINITY) || (colocation->score >= PCMK_SCORE_INFINITY)) { pcmk__add_dependent_scores(colocation, rsc); } } pe__show_node_scores(true, rsc, "Mandatory-colocations", rsc->priv->allowed_nodes, scheduler); // Then apply optional colocations for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score > -PCMK_SCORE_INFINITY) && (colocation->score < PCMK_SCORE_INFINITY)) { apply_this_with(colocation, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score > -PCMK_SCORE_INFINITY) && (colocation->score < PCMK_SCORE_INFINITY)) { pcmk__add_dependent_scores(colocation, rsc); } } g_list_free(this_with_colocations); g_list_free(with_this_colocations); if (rsc->priv->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, -PCMK_SCORE_INFINITY, PCMK_META_TARGET_ROLE, scheduler); } else if ((rsc->priv->next_role > rsc->priv->orig_role) && !pcmk_is_set(scheduler->flags, pcmk__sched_quorate) && (scheduler->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->priv->orig_role), pcmk_role_text(rsc->priv->next_role)); pe__set_next_role(rsc, rsc->priv->orig_role, PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE); } pe__show_node_scores(!pcmk_is_set(scheduler->flags, pcmk__sched_output_scores), rsc, __func__, rsc->priv->allowed_nodes, scheduler); // Unmanage resource if fencing is enabled but no device is configured if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled) && !pcmk_is_set(scheduler->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->priv->orig_role, "unmanaged"); assign_to = pcmk__current_node(rsc); if (assign_to == NULL) { reason = "inactive"; } else if (rsc->priv->orig_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->priv->name : "no node"), reason); pcmk__assign_resource(rsc, assign_to, true, stop_if_fail); } else if (pcmk_is_set(scheduler->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->priv->active_nodes != NULL) && stop_if_fail) { pcmk__rsc_info(rsc, "Stopping removed resource %s", rsc->id); } } pcmk__clear_rsc_flags(rsc, pcmk__rsc_assigning); if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { remote_connection_assigned(rsc); } return rsc->priv->assigned_node; } /*! * \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->priv->orig_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->priv->orig_role <= rsc->priv->next_role) && (role != rsc->priv->orig_role) && !pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) { bool required = need_stop; next_role = rsc_state_matrix[role][rsc->priv->orig_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->priv->assigned_node, !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->priv->next_role != pcmk_role_unknown) { return "explicit"; } if (rsc->priv->assigned_node == 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->priv->assigned_node, 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->priv->orig_role; while (role != rsc->priv->next_role) { enum rsc_role_e next_role = rsc_state_matrix[role][rsc->priv->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->priv->next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, rsc->priv->assigned_node, 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; pcmk_node_t *migration_target = NULL; unsigned int num_all_active = 0; unsigned int num_clean_active = 0; const char *next_role_source = NULL; pcmk__assert(pcmk__is_primitive(rsc)); 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->priv->orig_role), pcmk_role_text(rsc->priv->next_role), next_role_source, pcmk__node_name(rsc->priv->assigned_node)); current = rsc->priv->fns->active_node(rsc, &num_all_active, &num_clean_active); g_list_foreach(rsc->priv->dangling_migration_sources, pcmk__abort_dangling_migration, rsc); if ((current != NULL) && (rsc->priv->assigned_node != NULL) && !pcmk__same_node(current, rsc->priv->assigned_node) && (rsc->priv->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->priv->assigned_node)); 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 migration_target = rsc->priv->partial_migration_target; if ((rsc->priv->partial_migration_source != NULL) && (migration_target != NULL) && allow_migrate && (num_all_active == 2) && pcmk__same_node(current, rsc->priv->partial_migration_source) && pcmk__same_node(rsc->priv->assigned_node, 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->priv->partial_migration_source), pcmk__node_name(migration_target)); } else if ((rsc->priv->partial_migration_source != NULL) || (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->priv->partial_migration_source), pcmk__node_name(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->priv->partial_migration_source), pcmk__node_name(migration_target)); } need_stop = true; rsc->priv->partial_migration_source = NULL; rsc->priv->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->priv->xml, PCMK_XA_CLASS); // Resource was (possibly) incorrectly multiply active pcmk__sched_err(rsc->priv->scheduler, "%s resource %s might be active on %u nodes (%s)", pcmk__s(class, "Untyped"), rsc->id, num_all_active, pcmk__multiply_active_text(rsc)); crm_notice("For more information, see \"What are multiply active " "resources?\" at " "https://projects.clusterlabs.org/w/clusterlabs/faq/"); switch (rsc->priv->multiply_active_policy) { 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->priv->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->priv->orig_role > pcmk_role_started) && (current != NULL) && (rsc->priv->assigned_node != NULL)) { pcmk_action_t *start = NULL; pcmk__rsc_trace(rsc, "Creating start action for promoted resource %s", rsc->id); start = start_action(rsc, rsc->priv->assigned_node, 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->priv->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (node->priv->remote != NULL) { node->assign->score = -PCMK_SCORE_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->priv->allowed_nodes != NULL) { allowed_nodes = g_hash_table_get_values(rsc->priv->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; pcmk_scheduler_t *scheduler = NULL; pcmk__assert(pcmk__is_primitive(rsc)); scheduler = rsc->priv->scheduler; 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(scheduler->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->priv->utilization) > 0) && !pcmk__str_eq(scheduler->priv->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, scheduler); // Promotable ordering: demote before stop, start before promote if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_promotable) || (rsc->priv->orig_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, scheduler); 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, scheduler); } // 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, scheduler); // Certain checks need allowed nodes if (check_unfencing || check_utilization || (rsc->priv->launcher != 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->priv->launcher != NULL) { pcmk_resource_t *remote_rsc = NULL; if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { // 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->priv->launcher); } /* If someone cleans up a guest or bundle node's launcher, we will * likely schedule a (re-)probe of the launcher and recovery of the * connection. Order the connection stop after the launcher probe, * so that if we detect the launcher running, we will trigger a new * transition and avoid the unnecessary recovery. */ pcmk__order_resource_actions(rsc->priv->launcher, 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 PCMK__META_CONTAINER * 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 PCMK__META_CONTAINER set to a remote node or guest node resource. */ } else if (pcmk_is_set(rsc->priv->launcher->flags, pcmk__rsc_is_remote_connection)) { remote_rsc = rsc->priv->launcher; } else { remote_rsc = pe__resource_contains_guest_node(scheduler, rsc->priv->launcher); } if (remote_rsc != NULL) { /* Force the resource on the Pacemaker Remote node instead of * colocating the resource with the launcher. */ for (GList *item = allowed_nodes; item; item = item->next) { pcmk_node_t *node = item->data; if (node->priv->remote != remote_rsc) { node->assign->score = -PCMK_SCORE_INFINITY; } } } else { /* This resource is either launched by a resource 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 launcher %s", rsc->id, rsc->priv->launcher->id); pcmk__new_ordering(rsc->priv->launcher, pcmk__op_key(rsc->priv->launcher->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, scheduler); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, rsc->priv->launcher, pcmk__op_key(rsc->priv->launcher->id, PCMK_ACTION_STOP, 0), NULL, pcmk__ar_then_implies_first, scheduler); if (pcmk_is_set(rsc->flags, pcmk__rsc_remote_nesting_allowed) /* @TODO: && non-bundle Pacemaker Remote nodes exist */) { score = 10000; /* Highly preferred but not essential */ } else { score = PCMK_SCORE_INFINITY; // Force to run on same host } pcmk__new_colocation("#resource-with-container", NULL, score, rsc, rsc->priv->launcher, NULL, NULL, pcmk__coloc_influence); } } if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection) || 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 * * \return The score added to the dependent's priority */ int 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; pcmk__assert((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if (for_dependent) { // Always process on behalf of primary resource return primary->priv->cmds->apply_coloc_score(dependent, primary, colocation, false); } 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: return pcmk__apply_coloc_to_priority(dependent, primary, colocation); case pcmk__coloc_affects_location: pcmk__apply_coloc_to_scores(dependent, primary, colocation); return 0; default: // pcmk__coloc_affects_nothing return 0; } } /* 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) { const pcmk_resource_t *parent = NULL; pcmk__assert(pcmk__is_primitive(rsc) && (list != NULL)); parent = rsc->priv->parent; 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->priv->with_this_colocations, orig_rsc); if (parent != NULL) { parent->priv->cmds->with_this_colocations(parent, orig_rsc, list); } } else { // For an ancestor, add only explicitly configured constraints for (GList *iter = rsc->priv->with_this_colocations; 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) { const pcmk_resource_t *parent = NULL; pcmk__assert(pcmk__is_primitive(rsc) && (list != NULL)); parent = rsc->priv->parent; 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->priv->this_with_colocations, orig_rsc); if (parent != NULL) { parent->priv->cmds->this_with_colocations(parent, orig_rsc, list); } } else { // For an ancestor, add only explicitly configured constraints for (GList *iter = rsc->priv->this_with_colocations; 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) { pcmk__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 PCMK_VALUE_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->priv->next_role > pcmk_role_stopped) && pcmk__same_node(rsc->priv->assigned_node, 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->priv->active_nodes; 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=PCMK_VALUE_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->priv->partial_migration_target != NULL) { // Continue migration if node originally was and remains target if (pcmk__same_node(current, rsc->priv->partial_migration_target) && pcmk__same_node(current, rsc->priv->assigned_node)) { 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->priv->assigned_node == 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->flags, pcmk__rsc_needs_unfencing)) { pcmk_action_t *unfence = pe_fence_op(current, PCMK_ACTION_ON, true, NULL, false, rsc->priv->scheduler); order_actions(stop, unfence, pcmk__ar_then_implies_first); if (!pcmk__node_unfenced(current)) { pcmk__sched_err(rsc->priv->scheduler, "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; pcmk__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->assign->score); 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; pcmk__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->priv->active_nodes; 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) { pcmk__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 (!pcmk__node_available(node, pcmk__node_alive)) { 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; pcmk__assert(pcmk__is_primitive(rsc) && (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->priv->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->priv->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 PCMK__META_CONTAINER meta-attribute can be set on the primitive * itself or one of its ancestors, so check them all and keep the highest. */ for (parent = rsc; parent != NULL; parent = parent->priv->parent) { if (parent->priv->launcher != NULL) { crm_xml_add(xml, CRM_META "_" PCMK__META_CONTAINER, parent->priv->launcher->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->priv->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) { pcmk__assert(pcmk__is_primitive(rsc) && (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 = pcmk__node_attr(node, PCMK__NODE_ATTR_SHUTDOWN, NULL, pcmk__rsc_node_current); time_t result = 0; if (shutdown != NULL) { long long result_ll; int rc = pcmk__scan_ll(shutdown, &result_ll, 0LL); if (rc == pcmk_rc_ok) { result = (time_t) result_ll; } else { crm_warn("Ignoring invalid value '%s' for %s " PCMK__NODE_ATTR_SHUTDOWN " attribute: %s", shutdown, pcmk__node_name(node), pcmk_rc_str(rc)); } } if (result == 0) { result = pcmk__scheduler_epoch_time(node->priv->scheduler); } return 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 (!pcmk__same_node(node, rsc->priv->lock_node)) { resource_location(rsc, node, -PCMK_SCORE_INFINITY, PCMK_OPT_SHUTDOWN_LOCK, rsc->priv->scheduler); } } // Primitive implementation of pcmk__assignment_methods_t:shutdown_lock() void pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc) { pcmk_scheduler_t *scheduler = NULL; pcmk__assert(pcmk__is_primitive(rsc)); scheduler = rsc->priv->scheduler; // Fence devices and remote connections can't be locked if (pcmk_any_flags_set(rsc->flags, pcmk__rsc_fence_device |pcmk__rsc_is_remote_connection)) { return; } if (rsc->priv->lock_node != NULL) { // The lock was obtained from resource history if (rsc->priv->active_nodes != 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->priv->lock_node); rsc->priv->lock_node = NULL; rsc->priv->lock_time = 0; } // Only a resource active on exactly one node can be locked } else if (pcmk__list_of_1(rsc->priv->active_nodes)) { pcmk_node_t *node = rsc->priv->active_nodes->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->priv->lock_node = node; rsc->priv->lock_time = shutdown_time(node); } } } if (rsc->priv->lock_node == NULL) { // No lock needed return; } if (scheduler->priv->shutdown_lock_ms > 0U) { time_t lock_expiration = rsc->priv->lock_time + pcmk__timeout_ms2s(scheduler->priv->shutdown_lock_ms); pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)", rsc->id, pcmk__node_name(rsc->priv->lock_node), (long long) lock_expiration); pcmk__update_recheck_time(++lock_expiration, scheduler, "shutdown lock expiration"); } else { pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown", rsc->id, pcmk__node_name(rsc->priv->lock_node)); } // If resource is locked to one node, ban it from all other nodes g_list_foreach(scheduler->nodes, ban_if_not_locked, rsc); }