diff --git a/include/pcmki/pcmki_sched_allocate.h b/include/pcmki/pcmki_sched_allocate.h index 5f5dc2d19e..98bf8dffcd 100644 --- a/include/pcmki/pcmki_sched_allocate.h +++ b/include/pcmki/pcmki_sched_allocate.h @@ -1,59 +1,57 @@ /* * Copyright 2004-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__PCMKI_PCMKI_SCHED_ALLOCATE__H # define PCMK__PCMKI_PCMKI_SCHED_ALLOCATE__H # include # include # include # include # include # include # include # include pe_node_t *pcmk__bundle_allocate(pe_resource_t *rsc, const pe_node_t *prefer); void pcmk__bundle_create_actions(pe_resource_t *rsc); bool pcmk__bundle_create_probe(pe_resource_t *rsc, pe_node_t *node); void pcmk__bundle_internal_constraints(pe_resource_t *rsc); void pcmk__bundle_rsc_location(pe_resource_t *rsc, pe__location_t *constraint); enum pe_action_flags pcmk__bundle_action_flags(pe_action_t *action, const pe_node_t *node); void pcmk__bundle_expand(pe_resource_t *rsc); void pcmk__bundle_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); void pcmk__bundle_shutdown_lock(pe_resource_t *rsc); pe_node_t *pcmk__clone_allocate(pe_resource_t *rsc, const pe_node_t *prefer); void clone_create_actions(pe_resource_t *rsc); void clone_internal_constraints(pe_resource_t *rsc); void clone_rsc_location(pe_resource_t *rsc, pe__location_t *constraint); enum pe_action_flags clone_action_flags(pe_action_t *action, const pe_node_t *node); void clone_expand(pe_resource_t *rsc); bool clone_create_probe(pe_resource_t *rsc, pe_node_t *node); void clone_append_meta(const pe_resource_t *rsc, xmlNode *xml); void pcmk__clone_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); void pcmk__clone_shutdown_lock(pe_resource_t *rsc); uint32_t pcmk__multi_update_actions(pe_action_t *first, pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pe_working_set_t *data_set); void pcmk__log_transition_summary(const char *filename); -void clone_create_pseudo_actions(pe_resource_t *rsc, GList *children, - notify_data_t **start_notify, - notify_data_t **stop_notify); + #endif diff --git a/include/pcmki/pcmki_sched_utils.h b/include/pcmki/pcmki_sched_utils.h index fb9df7373a..acff7bf86b 100644 --- a/include/pcmki/pcmki_sched_utils.h +++ b/include/pcmki/pcmki_sched_utils.h @@ -1,47 +1,42 @@ /* * Copyright 2004-2022 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__PCMKI_PCMKI_SCHED_UTILS__H # define PCMK__PCMKI_PCMKI_SCHED_UTILS__H #include // bool #include // GList, GHashTable, gboolean, guint #include // lrmd_event_data_t #include // cib_t #include #include #include #include #include #include /* Constraint helper functions */ GList *pcmk__copy_node_list(const GList *list, bool reset); pe_resource_t *find_compatible_child(const pe_resource_t *local_child, const pe_resource_t *rsc, enum rsc_role_e filter, gboolean current); -pe_resource_t *find_compatible_child_by_node(const pe_resource_t *local_child, - const pe_node_t *local_node, - const pe_resource_t *rsc, - enum rsc_role_e filter, - gboolean current); gboolean is_child_compatible(const pe_resource_t *child_rsc, const pe_node_t *local_node, enum rsc_role_e filter, gboolean current); enum pe_action_flags summary_action_flags(pe_action_t *action, GList *children, const pe_node_t *node); enum action_tasks clone_child_action(pe_action_t * action); int copies_per_node(pe_resource_t * rsc); xmlNode *pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *event, const char *caller_version, int target_rc, const char *node, const char *origin); #endif diff --git a/lib/pacemaker/libpacemaker_private.h b/lib/pacemaker/libpacemaker_private.h index 6f38f58cd2..f003b6b5f7 100644 --- a/lib/pacemaker/libpacemaker_private.h +++ b/lib/pacemaker/libpacemaker_private.h @@ -1,837 +1,849 @@ /* * Copyright 2021-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__LIBPACEMAKER_PRIVATE__H # define PCMK__LIBPACEMAKER_PRIVATE__H /* This header is for the sole use of libpacemaker, so that functions can be * declared with G_GNUC_INTERNAL for efficiency. */ #include // pe_action_t, pe_node_t, pe_working_set_t // Flags to modify the behavior of pcmk__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 allocation methods struct resource_alloc_functions_s { /*! * \internal * \brief Assign a resource to a node * * \param[in,out] rsc Resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * * \return Node that \p rsc is assigned to, if assigned entirely to one node */ pe_node_t *(*assign)(pe_resource_t *rsc, const pe_node_t *prefer); /*! * \internal * \brief Create all actions needed for a given resource * * \param[in,out] rsc Resource to create actions for */ void (*create_actions)(pe_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)(pe_resource_t *rsc, pe_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)(pe_resource_t *rsc); /*! * \internal * \brief Apply a colocation's score to node weights or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node weights (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void (*apply_coloc_score) (pe_resource_t *dependent, const pe_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 indirectly 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 pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *colocated_rscs); /*! * \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)(pe_resource_t *rsc, pe__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. */ enum pe_action_flags (*action_flags)(pe_action_t *action, const pe_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. In some cases, the ordering could be disabled 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 pe_action_optional to affect only * mandatory actions, and pe_action_runnable to * affect only runnable actions) * \param[in] type Group of enum pe_ordering flags to apply * \param[in,out] data_set Cluster working set * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t (*update_ordered_actions)(pe_action_t *first, pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pe_working_set_t *data_set); void (*output_actions)(pe_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)(pe_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 pe_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 allocated to a node. * * \param[in] rsc Resource with utilization to add * \param[in] orig_rsc Resource being allocated (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 pe_resource_t *rsc, const pe_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)(pe_resource_t *rsc); }; // Actions (pcmk_sched_actions.c) G_GNUC_INTERNAL void pcmk__update_action_for_orderings(pe_action_t *action, pe_working_set_t *data_set); G_GNUC_INTERNAL uint32_t pcmk__update_ordered_actions(pe_action_t *first, pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__log_action(const char *pre_text, const pe_action_t *action, bool details); G_GNUC_INTERNAL pe_action_t *pcmk__new_cancel_action(pe_resource_t *rsc, const char *name, guint interval_ms, const pe_node_t *node); G_GNUC_INTERNAL pe_action_t *pcmk__new_shutdown_action(pe_node_t *node); G_GNUC_INTERNAL bool pcmk__action_locks_rsc_to_node(const pe_action_t *action); G_GNUC_INTERNAL void pcmk__deduplicate_action_inputs(pe_action_t *action); G_GNUC_INTERNAL void pcmk__output_actions(pe_working_set_t *data_set); G_GNUC_INTERNAL bool pcmk__check_action_config(pe_resource_t *rsc, pe_node_t *node, const xmlNode *xml_op); G_GNUC_INTERNAL void pcmk__handle_rsc_config_changes(pe_working_set_t *data_set); // Recurring actions (pcmk_sched_recurring.c) G_GNUC_INTERNAL void pcmk__create_recurring_actions(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__schedule_cancel(pe_resource_t *rsc, const char *call_id, const char *task, guint interval_ms, const pe_node_t *node, const char *reason); G_GNUC_INTERNAL void pcmk__reschedule_recurring(pe_resource_t *rsc, const char *task, guint interval_ms, pe_node_t *node); G_GNUC_INTERNAL bool pcmk__action_is_recurring(const pe_action_t *action); // Producing transition graphs (pcmk_graph_producer.c) G_GNUC_INTERNAL bool pcmk__graph_has_loop(const pe_action_t *init_action, const pe_action_t *action, pe_action_wrapper_t *input); G_GNUC_INTERNAL void pcmk__add_rsc_actions_to_graph(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_graph(pe_working_set_t *data_set); // Fencing (pcmk_sched_fencing.c) G_GNUC_INTERNAL void pcmk__order_vs_fence(pe_action_t *stonith_op, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__order_vs_unfence(const pe_resource_t *rsc, pe_node_t *node, pe_action_t *action, enum pe_ordering order); G_GNUC_INTERNAL void pcmk__fence_guest(pe_node_t *node); G_GNUC_INTERNAL bool pcmk__node_unfenced(const pe_node_t *node); G_GNUC_INTERNAL void pcmk__order_restart_vs_unfence(gpointer data, gpointer user_data); // Injected scheduler inputs (pcmk_sched_injections.c) void pcmk__inject_scheduler_input(pe_working_set_t *data_set, cib_t *cib, const pcmk_injections_t *injections); // Constraints of any type (pcmk_sched_constraints.c) G_GNUC_INTERNAL pe_resource_t *pcmk__find_constraint_resource(GList *rsc_list, const char *id); G_GNUC_INTERNAL xmlNode *pcmk__expand_tags_in_sets(xmlNode *xml_obj, const pe_working_set_t *data_set); G_GNUC_INTERNAL bool pcmk__valid_resource_or_tag(const pe_working_set_t *data_set, const char *id, pe_resource_t **rsc, pe_tag_t **tag); G_GNUC_INTERNAL bool pcmk__tag_to_set(xmlNode *xml_obj, xmlNode **rsc_set, const char *attr, bool convert_rsc, const pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__create_internal_constraints(pe_working_set_t *data_set); // Location constraints G_GNUC_INTERNAL void pcmk__unpack_location(xmlNode *xml_obj, pe_working_set_t *data_set); G_GNUC_INTERNAL pe__location_t *pcmk__new_location(const char *id, pe_resource_t *rsc, int node_weight, const char *discover_mode, pe_node_t *foo_node, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__apply_locations(pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__apply_location(pe_resource_t *rsc, pe__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 enum pcmk__coloc_affects pcmk__colocation_affects(const pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool preview); G_GNUC_INTERNAL void pcmk__apply_coloc_to_weights(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__apply_coloc_to_priority(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__add_colocated_node_scores(pe_resource_t *rsc, const char *log_id, GHashTable **nodes, const char *attr, float factor, uint32_t flags); G_GNUC_INTERNAL void pcmk__unpack_colocation(xmlNode *xml_obj, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__add_this_with(pe_resource_t *rsc, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__add_with_this(pe_resource_t *rsc, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__new_colocation(const char *id, const char *node_attr, int score, pe_resource_t *dependent, pe_resource_t *primary, const char *dependent_role, const char *primary_role, bool influence, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__block_colocation_dependents(pe_action_t *action, pe_working_set_t *data_set); /*! * \internal * \brief Check whether colocation's dependent preferences should be considered * * \param[in] colocation Colocation constraint * \param[in] rsc Primary instance (normally this will be * colocation->primary, which NULL will be treated as, * but for clones or bundles with multiple instances * this can be a particular instance) * * \return true if colocation influence should be effective, otherwise false */ static inline bool pcmk__colocation_has_influence(const pcmk__colocation_t *colocation, const pe_resource_t *rsc) { if (rsc == NULL) { rsc = colocation->primary; } /* A bundle replica colocates its remote connection with its container, * using a finite score so that the container can run on Pacemaker Remote * nodes. * * Moving a connection is lightweight and does not interrupt the service, * while moving a container is heavyweight and does interrupt the service, * so don't move a clean, active container based solely on the preferences * of its connection. * * This also avoids problematic scenarios where two containers want to * perpetually swap places. */ if (pcmk_is_set(colocation->dependent->flags, pe_rsc_allow_remote_remotes) && !pcmk_is_set(rsc->flags, pe_rsc_failed) && pcmk__list_of_1(rsc->running_on)) { return false; } /* The dependent in a colocation influences the primary's location * if the influence option is true or the primary is not yet active. */ return colocation->influence || (rsc->running_on == NULL); } // Ordering constraints (pcmk_sched_ordering.c) G_GNUC_INTERNAL void pcmk__new_ordering(pe_resource_t *first_rsc, char *first_task, pe_action_t *first_action, pe_resource_t *then_rsc, char *then_task, pe_action_t *then_action, uint32_t flags, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__unpack_ordering(xmlNode *xml_obj, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__disable_invalid_orderings(pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__order_stops_before_shutdown(pe_node_t *node, pe_action_t *shutdown_op); G_GNUC_INTERNAL void pcmk__apply_orderings(pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__order_after_each(pe_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 Bitmask of enum pe_ordering flags */ #define pcmk__order_resource_actions(first_rsc, first_task, \ then_rsc, then_task, flags) \ pcmk__new_ordering((first_rsc), \ pcmk__op_key((first_rsc)->id, (first_task), 0), \ NULL, \ (then_rsc), \ pcmk__op_key((then_rsc)->id, (then_task), 0), \ NULL, (flags), (first_rsc)->cluster) #define pcmk__order_starts(rsc1, rsc2, flags) \ pcmk__order_resource_actions((rsc1), CRMD_ACTION_START, \ (rsc2), CRMD_ACTION_START, (flags)) #define pcmk__order_stops(rsc1, rsc2, flags) \ pcmk__order_resource_actions((rsc1), CRMD_ACTION_STOP, \ (rsc2), CRMD_ACTION_STOP, (flags)) // Ticket constraints (pcmk_sched_tickets.c) G_GNUC_INTERNAL void pcmk__unpack_rsc_ticket(xmlNode *xml_obj, pe_working_set_t *data_set); // Promotable clone resources (pcmk_sched_promotable.c) G_GNUC_INTERNAL void pcmk__add_promotion_scores(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__require_promotion_tickets(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__set_instance_roles(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_promotable_actions(pe_resource_t *clone); G_GNUC_INTERNAL void pcmk__promotable_restart_ordering(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__order_promotable_instances(pe_resource_t *clone); G_GNUC_INTERNAL void pcmk__update_dependent_with_promotable(const pe_resource_t *primary, pe_resource_t *dependent, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__update_promotable_dependent_priority(const pe_resource_t *primary, pe_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 pe_node_t *node); G_GNUC_INTERNAL void pcmk__order_remote_connection_actions(pe_working_set_t *data_set); G_GNUC_INTERNAL bool pcmk__rsc_corresponds_to_guest(const pe_resource_t *rsc, const pe_node_t *node); G_GNUC_INTERNAL pe_node_t *pcmk__connection_host_for_action(const pe_action_t *action); G_GNUC_INTERNAL void pcmk__substitute_remote_addr(pe_resource_t *rsc, GHashTable *params); G_GNUC_INTERNAL void pcmk__add_bundle_meta_to_xml(xmlNode *args_xml, const pe_action_t *action); // Primitives (pcmk_sched_primitive.c) G_GNUC_INTERNAL pe_node_t *pcmk__primitive_assign(pe_resource_t *rsc, const pe_node_t *prefer); G_GNUC_INTERNAL void pcmk__primitive_create_actions(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__primitive_internal_constraints(pe_resource_t *rsc); G_GNUC_INTERNAL enum pe_action_flags pcmk__primitive_action_flags(pe_action_t *action, const pe_node_t *node); G_GNUC_INTERNAL void pcmk__primitive_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__schedule_cleanup(pe_resource_t *rsc, const pe_node_t *node, bool optional); G_GNUC_INTERNAL void pcmk__primitive_add_graph_meta(const pe_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__primitive_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__primitive_shutdown_lock(pe_resource_t *rsc); // Groups (pcmk_sched_group.c) G_GNUC_INTERNAL pe_node_t *pcmk__group_assign(pe_resource_t *rsc, const pe_node_t *prefer); G_GNUC_INTERNAL void pcmk__group_create_actions(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__group_internal_constraints(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__group_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__group_apply_location(pe_resource_t *rsc, pe__location_t *location); G_GNUC_INTERNAL enum pe_action_flags pcmk__group_action_flags(pe_action_t *action, const pe_node_t *node); G_GNUC_INTERNAL uint32_t pcmk__group_update_ordered_actions(pe_action_t *first, pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pe_working_set_t *data_set); G_GNUC_INTERNAL GList *pcmk__group_colocated_resources(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *colocated_rscs); G_GNUC_INTERNAL void pcmk__group_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__group_shutdown_lock(pe_resource_t *rsc); // Clones (pcmk_sched_clone.c) G_GNUC_INTERNAL void pcmk__clone_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); // Bundles (pcmk_sched_bundle.c) +G_GNUC_INTERNAL +GList *pcmk__bundle_containers(const pe_resource_t *bundle); + +G_GNUC_INTERNAL +const pe_resource_t *pcmk__get_rsc_in_container(const pe_resource_t *instance, + const pe_node_t *node); + G_GNUC_INTERNAL void pcmk__bundle_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__output_bundle_actions(pe_resource_t *rsc); // Clone instances or bundle replica containers (pcmk_sched_instances.c) G_GNUC_INTERNAL -void distribute_children(pe_resource_t *rsc, GList *children, int max, - int per_host_max, pe_working_set_t *data_set); +void pcmk__assign_instances(pe_resource_t *collective, GList *instances, + int max_total, int max_per_node); + +G_GNUC_INTERNAL +void pcmk__create_instance_actions(pe_resource_t *rsc, GList *instances, + notify_data_t **start_notify, + notify_data_t **stop_notify); // Injections (pcmk_injections.c) G_GNUC_INTERNAL xmlNode *pcmk__inject_node(cib_t *cib_conn, const char *node, const char *uuid); G_GNUC_INTERNAL xmlNode *pcmk__inject_node_state_change(cib_t *cib_conn, const char *node, bool up); G_GNUC_INTERNAL xmlNode *pcmk__inject_resource_history(pcmk__output_t *out, xmlNode *cib_node, const char *resource, const char *lrm_name, const char *rclass, const char *rtype, const char *rprovider); G_GNUC_INTERNAL void pcmk__inject_failcount(pcmk__output_t *out, xmlNode *cib_node, const char *resource, const char *task, guint interval_ms, int rc); G_GNUC_INTERNAL xmlNode *pcmk__inject_action_result(xmlNode *cib_resource, lrmd_event_data_t *op, int target_rc); // Nodes (pcmk_sched_nodes.c) G_GNUC_INTERNAL bool pcmk__node_available(const pe_node_t *node, bool consider_score, bool consider_guest); G_GNUC_INTERNAL bool pcmk__any_node_available(GHashTable *nodes); G_GNUC_INTERNAL GHashTable *pcmk__copy_node_table(GHashTable *nodes); G_GNUC_INTERNAL GList *pcmk__sort_nodes(GList *nodes, pe_node_t *active_node); G_GNUC_INTERNAL void pcmk__apply_node_health(pe_working_set_t *data_set); G_GNUC_INTERNAL pe_node_t *pcmk__top_allowed_node(const pe_resource_t *rsc, const pe_node_t *node); // Functions applying to more than one variant (pcmk_sched_resource.c) G_GNUC_INTERNAL void pcmk__set_allocation_methods(pe_working_set_t *data_set); G_GNUC_INTERNAL bool pcmk__rsc_agent_changed(pe_resource_t *rsc, pe_node_t *node, const xmlNode *rsc_entry, bool active_on_node); G_GNUC_INTERNAL GList *pcmk__rscs_matching_id(const char *id, const pe_working_set_t *data_set); G_GNUC_INTERNAL GList *pcmk__colocated_resources(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *colocated_rscs); G_GNUC_INTERNAL void pcmk__noop_add_graph_meta(const pe_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__output_resource_actions(pe_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__finalize_assignment(pe_resource_t *rsc, pe_node_t *chosen, bool force); G_GNUC_INTERNAL bool pcmk__assign_resource(pe_resource_t *rsc, pe_node_t *node, bool force); G_GNUC_INTERNAL void pcmk__unassign_resource(pe_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__threshold_reached(pe_resource_t *rsc, const pe_node_t *node, pe_resource_t **failed); G_GNUC_INTERNAL void pcmk__sort_resources(pe_working_set_t *data_set); 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(pe_resource_t *rsc, pe_node_t *node); G_GNUC_INTERNAL void pcmk__order_probes(pe_working_set_t *data_set); G_GNUC_INTERNAL bool pcmk__probe_resource_list(GList *rscs, pe_node_t *node); G_GNUC_INTERNAL void pcmk__schedule_probes(pe_working_set_t *data_set); // Functions related to live migration (pcmk_sched_migration.c) void pcmk__create_migration_actions(pe_resource_t *rsc, const pe_node_t *current); void pcmk__abort_dangling_migration(void *data, void *user_data); bool pcmk__rsc_can_migrate(const pe_resource_t *rsc, const pe_node_t *current); void pcmk__order_migration_equivalents(pe__ordering_t *order); // Functions related to node utilization (pcmk_sched_utilization.c) G_GNUC_INTERNAL int pcmk__compare_node_capacities(const pe_node_t *node1, const pe_node_t *node2); G_GNUC_INTERNAL void pcmk__consume_node_capacity(GHashTable *current_utilization, const pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__release_node_capacity(GHashTable *current_utilization, const pe_resource_t *rsc); G_GNUC_INTERNAL const pe_node_t *pcmk__ban_insufficient_capacity(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_utilization_constraints(pe_resource_t *rsc, const GList *allowed_nodes); G_GNUC_INTERNAL void pcmk__show_node_capacities(const char *desc, pe_working_set_t *data_set); #endif // PCMK__LIBPACEMAKER_PRIVATE__H diff --git a/lib/pacemaker/pcmk_sched_bundle.c b/lib/pacemaker/pcmk_sched_bundle.c index cb04ad4f1c..49ca4be0f3 100644 --- a/lib/pacemaker/pcmk_sched_bundle.c +++ b/lib/pacemaker/pcmk_sched_bundle.c @@ -1,1149 +1,864 @@ /* * Copyright 2004-2023 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 "libpacemaker_private.h" #define PE__VARIANT_BUNDLE 1 #include static bool is_bundle_node(pe__bundle_variant_data_t *data, pe_node_t *node) { for (GList *gIter = data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (node->details == replica->node->details) { return TRUE; } } return FALSE; } -static GList * -get_container_list(const pe_resource_t *rsc) +/*! + * \internal + * \brief Get a list of a bundle's containers + * + * \param[in] bundle Bundle resource + * + * \return Newly created list of \p bundle's containers + * \note It is the caller's responsibility to free the result with + * g_list_free(). + */ +GList * +pcmk__bundle_containers(const pe_resource_t *bundle) { GList *containers = NULL; + const pe__bundle_variant_data_t *data = NULL; - if (rsc->variant == pe_container) { - pe__bundle_variant_data_t *data = NULL; - - get_bundle_variant_data(data, rsc); - for (GList *gIter = data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; + get_bundle_variant_data(data, bundle); + for (GList *iter = data->replicas; iter != NULL; iter = iter->next) { + pe__bundle_replica_t *replica = iter->data; - containers = g_list_append(containers, replica->container); - } + containers = g_list_append(containers, replica->container); } return containers; } -static inline GList * -get_containers_or_children(const pe_resource_t *rsc) -{ - return (rsc->variant == pe_container)? - get_container_list(rsc) : rsc->children; -} - /*! * \internal * \brief Assign a bundle 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 * * \return Node that \p rsc is assigned to, if assigned entirely to one node */ pe_node_t * pcmk__bundle_allocate(pe_resource_t *rsc, const pe_node_t *prefer) { GList *containers = NULL; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return NULL); get_bundle_variant_data(bundle_data, rsc); pe__set_resource_flags(rsc, pe_rsc_allocating); - containers = get_container_list(rsc); + containers = pcmk__bundle_containers(rsc); pe__show_node_weights(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); containers = g_list_sort(containers, pcmk__cmp_instance); - distribute_children(rsc, containers, bundle_data->nreplicas, - bundle_data->nreplicas_per_host, rsc->cluster); + pcmk__assign_instances(rsc, containers, bundle_data->nreplicas, + bundle_data->nreplicas_per_host); g_list_free(containers); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; pe_node_t *container_host = NULL; CRM_ASSERT(replica); if (replica->ip) { pe_rsc_trace(rsc, "Allocating bundle %s IP %s", rsc->id, replica->ip->id); replica->ip->cmds->assign(replica->ip, prefer); } container_host = replica->container->allocated_to; if (replica->remote && pe__is_guest_or_remote_node(container_host)) { /* We need 'nested' connection resources to be on the same * host because pacemaker-remoted only supports a single * active connection */ pcmk__new_colocation("child-remote-with-docker-remote", NULL, INFINITY, replica->remote, container_host->details->remote_rsc, NULL, NULL, true, rsc->cluster); } if (replica->remote) { pe_rsc_trace(rsc, "Allocating bundle %s connection %s", rsc->id, replica->remote->id); replica->remote->cmds->assign(replica->remote, prefer); } // Explicitly allocate replicas' children before bundle child if (replica->child) { pe_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, replica->child->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { if (node->details != replica->node->details) { node->weight = -INFINITY; } else if (!pcmk__threshold_reached(replica->child, node, NULL)) { node->weight = INFINITY; } } pe__set_resource_flags(replica->child->parent, pe_rsc_allocating); pe_rsc_trace(rsc, "Allocating bundle %s replica child %s", rsc->id, replica->child->id); replica->child->cmds->assign(replica->child, replica->node); pe__clear_resource_flags(replica->child->parent, pe_rsc_allocating); } } if (bundle_data->child) { pe_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, bundle_data->child->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { if (is_bundle_node(bundle_data, node)) { node->weight = 0; } else { node->weight = -INFINITY; } } pe_rsc_trace(rsc, "Allocating bundle %s child %s", rsc->id, bundle_data->child->id); bundle_data->child->cmds->assign(bundle_data->child, prefer); } pe__clear_resource_flags(rsc, pe_rsc_allocating|pe_rsc_provisional); return NULL; } void pcmk__bundle_create_actions(pe_resource_t *rsc) { pe_action_t *action = NULL; GList *containers = NULL; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); - containers = get_container_list(rsc); + containers = pcmk__bundle_containers(rsc); get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->ip) { replica->ip->cmds->create_actions(replica->ip); } if (replica->container) { replica->container->cmds->create_actions(replica->container); } if (replica->remote) { replica->remote->cmds->create_actions(replica->remote); } } - clone_create_pseudo_actions(rsc, containers, NULL, NULL); + pcmk__create_instance_actions(rsc, containers, NULL, NULL); if (bundle_data->child) { bundle_data->child->cmds->create_actions(bundle_data->child); if (pcmk_is_set(bundle_data->child->flags, pe_rsc_promotable)) { /* promote */ pe__new_rsc_pseudo_action(rsc, RSC_PROMOTE, true, true); action = pe__new_rsc_pseudo_action(rsc, RSC_PROMOTED, true, true); action->priority = INFINITY; /* demote */ pe__new_rsc_pseudo_action(rsc, RSC_DEMOTE, true, true); action = pe__new_rsc_pseudo_action(rsc, RSC_DEMOTED, true, true); action->priority = INFINITY; } } g_list_free(containers); } void pcmk__bundle_internal_constraints(pe_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); if (bundle_data->child) { pcmk__order_resource_actions(rsc, RSC_START, bundle_data->child, RSC_START, pe_order_implies_first_printed); pcmk__order_resource_actions(rsc, RSC_STOP, bundle_data->child, RSC_STOP, pe_order_implies_first_printed); if (bundle_data->child->children) { pcmk__order_resource_actions(bundle_data->child, RSC_STARTED, rsc, RSC_STARTED, pe_order_implies_then_printed); pcmk__order_resource_actions(bundle_data->child, RSC_STOPPED, rsc, RSC_STOPPED, pe_order_implies_then_printed); } else { pcmk__order_resource_actions(bundle_data->child, RSC_START, rsc, RSC_STARTED, pe_order_implies_then_printed); pcmk__order_resource_actions(bundle_data->child, RSC_STOP, rsc, RSC_STOPPED, pe_order_implies_then_printed); } } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); CRM_ASSERT(replica->container); replica->container->cmds->internal_constraints(replica->container); pcmk__order_starts(rsc, replica->container, pe_order_runnable_left|pe_order_implies_first_printed); if (replica->child) { pcmk__order_stops(rsc, replica->child, pe_order_implies_first_printed); } pcmk__order_stops(rsc, replica->container, pe_order_implies_first_printed); pcmk__order_resource_actions(replica->container, RSC_START, rsc, RSC_STARTED, pe_order_implies_then_printed); pcmk__order_resource_actions(replica->container, RSC_STOP, rsc, RSC_STOPPED, pe_order_implies_then_printed); if (replica->ip) { replica->ip->cmds->internal_constraints(replica->ip); // Start IP then container pcmk__order_starts(replica->ip, replica->container, pe_order_runnable_left|pe_order_preserve); pcmk__order_stops(replica->container, replica->ip, pe_order_implies_first|pe_order_preserve); pcmk__new_colocation("ip-with-docker", NULL, INFINITY, replica->ip, replica->container, NULL, NULL, true, rsc->cluster); } if (replica->remote) { /* This handles ordering and colocating remote relative to container * (via "resource-with-container"). Since IP is also ordered and * colocated relative to the container, we don't need to do anything * explicit here with IP. */ replica->remote->cmds->internal_constraints(replica->remote); } if (replica->child) { CRM_ASSERT(replica->remote); // "Start remote then child" is implicit in scheduler's remote logic } } if (bundle_data->child) { bundle_data->child->cmds->internal_constraints(bundle_data->child); if (pcmk_is_set(bundle_data->child->flags, pe_rsc_promotable)) { pcmk__promotable_restart_ordering(rsc); /* child demoted before global demoted */ pcmk__order_resource_actions(bundle_data->child, RSC_DEMOTED, rsc, RSC_DEMOTED, pe_order_implies_then_printed); /* global demote before child demote */ pcmk__order_resource_actions(rsc, RSC_DEMOTE, bundle_data->child, RSC_DEMOTE, pe_order_implies_first_printed); /* child promoted before global promoted */ pcmk__order_resource_actions(bundle_data->child, RSC_PROMOTED, rsc, RSC_PROMOTED, pe_order_implies_then_printed); /* global promote before child promote */ pcmk__order_resource_actions(rsc, RSC_PROMOTE, bundle_data->child, RSC_PROMOTE, pe_order_implies_first_printed); } } } static pe_resource_t * compatible_replica_for_node(const pe_resource_t *rsc_lh, const pe_node_t *candidate, const pe_resource_t *rsc, enum rsc_role_e filter, gboolean current) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(candidate != NULL, return NULL); get_bundle_variant_data(bundle_data, rsc); crm_trace("Looking for compatible child from %s for %s on %s", rsc_lh->id, rsc->id, pe__node_name(candidate)); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (is_child_compatible(replica->container, candidate, filter, current)) { crm_trace("Pairing %s with %s on %s", rsc_lh->id, replica->container->id, pe__node_name(candidate)); return replica->container; } } crm_trace("Can't pair %s with %s", rsc_lh->id, rsc->id); return NULL; } static pe_resource_t * compatible_replica(const pe_resource_t *rsc_lh, const pe_resource_t *rsc, enum rsc_role_e filter, gboolean current, pe_working_set_t *data_set) { GList *scratch = NULL; pe_resource_t *pair = NULL; pe_node_t *active_node_lh = NULL; active_node_lh = rsc_lh->fns->location(rsc_lh, NULL, current); if (active_node_lh) { return compatible_replica_for_node(rsc_lh, active_node_lh, rsc, filter, current); } scratch = g_hash_table_get_values(rsc_lh->allowed_nodes); scratch = pcmk__sort_nodes(scratch, NULL); for (GList *gIter = scratch; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; pair = compatible_replica_for_node(rsc_lh, node, rsc, filter, current); if (pair) { goto done; } } pe_rsc_debug(rsc, "Can't pair %s with %s", rsc_lh->id, (rsc? rsc->id : "none")); done: g_list_free(scratch); return pair; } int copies_per_node(pe_resource_t * rsc) { /* Strictly speaking, there should be a 'copies_per_node' addition * to the resource function table and each case would be a * function. However that would be serious overkill to return an * int. In fact, it seems to me that both function tables * could/should be replaced by resources.{c,h} full of * rsc_{some_operation} functions containing a switch as below * which calls out to functions named {variant}_{some_operation} * as needed. */ switch(rsc->variant) { case pe_unknown: return 0; case pe_native: case pe_group: return 1; case pe_clone: { const char *max_clones_node = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INCARNATION_NODEMAX); if (max_clones_node == NULL) { return 1; } else { int max_i; pcmk__scan_min_int(max_clones_node, &max_i, 0); return max_i; } } case pe_container: { pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, rsc); return data->nreplicas_per_host; } } return 0; } /*! * \internal * \brief Apply a colocation's score to node weights or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node weights (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__bundle_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { GList *allocated_primaries = NULL; pe__bundle_variant_data_t *bundle_data = NULL; /* This should never be called for the bundle itself as a dependent. * Instead, we add its colocation constraints to its replicas and call the * apply_coloc_score() for the replicas as dependents. */ CRM_ASSERT(!for_dependent); CRM_CHECK((colocation != NULL) && (dependent != NULL) && (primary != NULL), return); CRM_ASSERT(dependent->variant == pe_native); if (pcmk_is_set(primary->flags, pe_rsc_provisional)) { pe_rsc_trace(primary, "%s is still provisional", primary->id); return; } else if (colocation->dependent->variant > pe_group) { pe_resource_t *primary_replica = compatible_replica(dependent, primary, RSC_ROLE_UNKNOWN, FALSE, dependent->cluster); if (primary_replica) { pe_rsc_debug(primary, "Pairing %s with %s", dependent->id, primary_replica->id); dependent->cmds->apply_coloc_score(dependent, primary_replica, colocation, true); } else if (colocation->score >= INFINITY) { crm_notice("Cannot pair %s with instance of %s", dependent->id, primary->id); pcmk__assign_resource(dependent, NULL, true); } else { pe_rsc_debug(primary, "Cannot pair %s with instance of %s", dependent->id, primary->id); } return; } get_bundle_variant_data(bundle_data, primary); pe_rsc_trace(primary, "Processing constraint %s: %s -> %s %d", colocation->id, dependent->id, primary->id, colocation->score); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (colocation->score < INFINITY) { replica->container->cmds->apply_coloc_score(dependent, replica->container, colocation, false); } else { pe_node_t *chosen = replica->container->fns->location(replica->container, NULL, FALSE); if ((chosen == NULL) || is_set_recursive(replica->container, pe_rsc_block, TRUE)) { continue; } if ((colocation->primary_role >= RSC_ROLE_PROMOTED) && (replica->child == NULL)) { continue; } if ((colocation->primary_role >= RSC_ROLE_PROMOTED) && (replica->child->next_role < RSC_ROLE_PROMOTED)) { continue; } pe_rsc_trace(primary, "Allowing %s: %s %d", colocation->id, pe__node_name(chosen), chosen->weight); allocated_primaries = g_list_prepend(allocated_primaries, chosen); } } if (colocation->score >= INFINITY) { node_list_exclude(dependent->allowed_nodes, allocated_primaries, FALSE); } g_list_free(allocated_primaries); } enum pe_action_flags pcmk__bundle_action_flags(pe_action_t *action, const pe_node_t *node) { GList *containers = NULL; enum pe_action_flags flags = 0; pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, action->rsc); if(data->child) { enum action_tasks task = get_complex_task(data->child, action->task, TRUE); switch(task) { case no_action: case action_notify: case action_notified: case action_promote: case action_promoted: case action_demote: case action_demoted: return summary_action_flags(action, data->child->children, node); default: break; } } - containers = get_container_list(action->rsc); + containers = pcmk__bundle_containers(action->rsc); flags = summary_action_flags(action, containers, node); g_list_free(containers); return flags; } -pe_resource_t * -find_compatible_child_by_node(const pe_resource_t *local_child, - const pe_node_t *local_node, - const pe_resource_t *rsc, enum rsc_role_e filter, - gboolean current) -{ - GList *gIter = NULL; - GList *children = NULL; - - if (local_node == NULL) { - crm_err("Can't colocate unrunnable child %s with %s", local_child->id, rsc->id); - return NULL; - } - - crm_trace("Looking for compatible child from %s for %s on %s", - local_child->id, rsc->id, pe__node_name(local_node)); - - children = get_containers_or_children(rsc); - for (gIter = children; gIter != NULL; gIter = gIter->next) { - pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; - - if(is_child_compatible(child_rsc, local_node, filter, current)) { - crm_trace("Pairing %s with %s on %s", - local_child->id, child_rsc->id, pe__node_name(local_node)); - return child_rsc; - } - } - - crm_trace("Can't pair %s with %s", local_child->id, rsc->id); - if(children != rsc->children) { - g_list_free(children); - } - return NULL; -} - -static pe__bundle_replica_t * -replica_for_container(const pe_resource_t *rsc, const pe_resource_t *container, - const pe_node_t *node) -{ - if (rsc->variant == pe_container) { - const pe__bundle_variant_data_t *data = NULL; - - get_bundle_variant_data(data, rsc); - for (GList *gIter = data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; - - if (replica->child - && (container == replica->container) - && pe__same_node(node, replica->node)) { - return replica; - } - } - } - return NULL; -} - -static uint32_t -multi_update_interleave_actions(pe_action_t *first, pe_action_t *then, - const pe_node_t *node, uint32_t filter, - uint32_t type, pe_working_set_t *data_set) -{ - GList *gIter = NULL; - GList *children = NULL; - gboolean current = FALSE; - uint32_t changed = pcmk__updated_none; - - /* Fix this - lazy */ - if (pcmk__ends_with(first->uuid, "_stopped_0") - || pcmk__ends_with(first->uuid, "_demoted_0")) { - current = TRUE; - } - - children = get_containers_or_children(then->rsc); - for (gIter = children; gIter != NULL; gIter = gIter->next) { - pe_resource_t *then_child = gIter->data; - pe_resource_t *first_child = find_compatible_child(then_child, - first->rsc, - RSC_ROLE_UNKNOWN, - current); - if (first_child == NULL && current) { - crm_trace("Ignore"); - - } else if (first_child == NULL) { - crm_debug("No match found for %s (%d / %s / %s)", then_child->id, current, first->uuid, then->uuid); - - /* Me no like this hack - but what else can we do? - * - * If there is no-one active or about to be active - * on the same node as then_child, then they must - * not be allowed to start - */ - if (pcmk_any_flags_set(type, pe_order_runnable_left|pe_order_implies_then) /* Mandatory */ ) { - pe_rsc_info(then->rsc, "Inhibiting %s from being active", then_child->id); - if (pcmk__assign_resource(then_child, NULL, true)) { - pcmk__set_updated_flags(changed, first, pcmk__updated_then); - } - } - - } else { - pe_action_t *first_action = NULL; - pe_action_t *then_action = NULL; - - enum action_tasks task = clone_child_action(first); - const char *first_task = task2text(task); - - pe__bundle_replica_t *first_replica = NULL; - pe__bundle_replica_t *then_replica = NULL; - - first_replica = replica_for_container(first->rsc, first_child, - node); - if (strstr(first->task, "stop") && first_replica && first_replica->child) { - /* Except for 'stopped' we should be looking at the - * in-container resource, actions for the child will - * happen later and are therefor more likely to align - * with the user's intent. - */ - first_action = find_first_action(first_replica->child->actions, - NULL, task2text(task), node); - } else { - first_action = find_first_action(first_child->actions, NULL, task2text(task), node); - } - - then_replica = replica_for_container(then->rsc, then_child, node); - if (strstr(then->task, "mote") - && then_replica && then_replica->child) { - /* Promote/demote actions will never be found for the - * container resource, look in the child instead - * - * Alternatively treat: - * 'XXXX then promote YYYY' as 'XXXX then start container for YYYY', and - * 'demote XXXX then stop YYYY' as 'stop container for XXXX then stop YYYY' - */ - then_action = find_first_action(then_replica->child->actions, - NULL, then->task, node); - } else { - then_action = find_first_action(then_child->actions, NULL, then->task, node); - } - - if (first_action == NULL) { - if (!pcmk_is_set(first_child->flags, pe_rsc_orphan) - && !pcmk__str_any_of(first_task, RSC_STOP, RSC_DEMOTE, NULL)) { - crm_err("Internal error: No action found for %s in %s (first)", - first_task, first_child->id); - - } else { - crm_trace("No action found for %s in %s%s (first)", - first_task, first_child->id, - pcmk_is_set(first_child->flags, pe_rsc_orphan)? " (ORPHAN)" : ""); - } - continue; - } - - /* We're only interested if 'then' is neither stopping nor being demoted */ - if (then_action == NULL) { - if (!pcmk_is_set(then_child->flags, pe_rsc_orphan) - && !pcmk__str_any_of(then->task, RSC_STOP, RSC_DEMOTE, NULL)) { - crm_err("Internal error: No action found for %s in %s (then)", - then->task, then_child->id); - - } else { - crm_trace("No action found for %s in %s%s (then)", - then->task, then_child->id, - pcmk_is_set(then_child->flags, pe_rsc_orphan)? " (ORPHAN)" : ""); - } - continue; - } - - if (order_actions(first_action, then_action, type)) { - crm_debug("Created constraint for %s (%d) -> %s (%d) %.6x", - first_action->uuid, - pcmk_is_set(first_action->flags, pe_action_optional), - then_action->uuid, - pcmk_is_set(then_action->flags, pe_action_optional), - type); - pcmk__set_updated_flags(changed, first, - pcmk__updated_first|pcmk__updated_then); - } - if(first_action && then_action) { - changed |= then_child->cmds->update_ordered_actions(first_action, - then_action, - node, - first_child->cmds->action_flags(first_action, node), - filter, - type, - data_set); - } else { - crm_err("Nothing found either for %s (%p) or %s (%p) %s", - first_child->id, first_action, - then_child->id, then_action, task2text(task)); - } - } - } - - if(children != then->rsc->children) { - g_list_free(children); - } - return changed; -} - -static bool -can_interleave_actions(pe_action_t *first, pe_action_t *then) -{ - bool interleave = FALSE; - pe_resource_t *rsc = NULL; - const char *interleave_s = NULL; - - if(first->rsc == NULL || then->rsc == NULL) { - crm_trace("Not interleaving %s with %s (both must be resources)", first->uuid, then->uuid); - return FALSE; - } else if(first->rsc == then->rsc) { - crm_trace("Not interleaving %s with %s (must belong to different resources)", first->uuid, then->uuid); - return FALSE; - } else if(first->rsc->variant < pe_clone || then->rsc->variant < pe_clone) { - crm_trace("Not interleaving %s with %s (both sides must be clones or bundles)", first->uuid, then->uuid); - return FALSE; - } - - if (pcmk__ends_with(then->uuid, "_stop_0") - || pcmk__ends_with(then->uuid, "_demote_0")) { - rsc = first->rsc; - } else { - rsc = then->rsc; - } - - interleave_s = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERLEAVE); - interleave = crm_is_true(interleave_s); - crm_trace("Interleave %s -> %s: %s (based on %s)", - first->uuid, then->uuid, interleave ? "yes" : "no", rsc->id); - - return interleave; -} - /*! * \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. In some cases, the ordering could be disabled as well. + * \brief Get containerized resource corresponding to a given bundle container * - * \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 pe_action_optional to affect only mandatory - * actions, and pe_action_runnable to affect only - * runnable actions) - * \param[in] type Group of enum pe_ordering flags to apply - * \param[in,out] data_set Cluster working set + * \param[in] instance Collective instance that might be a bundle container + * \param[in] node Node that \p instance might be running on * - * \return Group of enum pcmk__updated flags indicating what was updated + * \return Bundled resource instance inside \p instance if it is a bundle + * container instance running on \p node, otherwise NULL */ -uint32_t -pcmk__multi_update_actions(pe_action_t *first, pe_action_t *then, - const pe_node_t *node, uint32_t flags, - uint32_t filter, uint32_t type, - pe_working_set_t *data_set) +const pe_resource_t * +pcmk__get_rsc_in_container(const pe_resource_t *instance, const pe_node_t *node) { - uint32_t changed = pcmk__updated_none; - - crm_trace("%s -> %s", first->uuid, then->uuid); - - if(can_interleave_actions(first, then)) { - changed = multi_update_interleave_actions(first, then, node, filter, - type, data_set); - - } else if(then->rsc) { - GList *gIter = NULL; - GList *children = NULL; - - // Handle the 'primitive' ordering case - changed |= pcmk__update_ordered_actions(first, then, node, flags, - filter, type, data_set); - - // Now any children (or containers in the case of a bundle) - children = get_containers_or_children(then->rsc); - for (gIter = children; gIter != NULL; gIter = gIter->next) { - pe_resource_t *then_child = (pe_resource_t *) gIter->data; - uint32_t then_child_changed = pcmk__updated_none; - pe_action_t *then_child_action = find_first_action(then_child->actions, NULL, then->task, node); - - if (then_child_action) { - uint32_t then_child_flags = then_child->cmds->action_flags(then_child_action, - node); - - if (pcmk_is_set(then_child_flags, pe_action_runnable)) { - then_child_changed |= then_child->cmds->update_ordered_actions(first, - then_child_action, - node, - flags, - filter, - type, - data_set); - } - changed |= then_child_changed; - if (pcmk_is_set(then_child_changed, pcmk__updated_then)) { - for (GList *lpc = then_child_action->actions_after; lpc != NULL; lpc = lpc->next) { - pe_action_wrapper_t *next = (pe_action_wrapper_t *) lpc->data; - - pcmk__update_action_for_orderings(next->action, - data_set); - } - } - } - } + const pe__bundle_variant_data_t *data = NULL; + const pe_resource_t *top = pe__const_top_resource(instance, true); - if(children != then->rsc->children) { - g_list_free(children); + if ((node == NULL) || (top == NULL) || (top->variant != pe_container)) { + return NULL; + } + get_bundle_variant_data(data, top); + + for (const GList *iter = data->replicas; iter != NULL; iter = iter->next) { + const pe__bundle_replica_t *replica = iter->data; + + if ((instance == replica->container) + && pe__same_node(node, replica->node)) { + return replica->child; } } - return changed; + return NULL; } void pcmk__bundle_rsc_location(pe_resource_t *rsc, pe__location_t *constraint) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); pcmk__apply_location(rsc, constraint); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (replica->container) { replica->container->cmds->apply_location(replica->container, constraint); } if (replica->ip) { replica->ip->cmds->apply_location(replica->ip, constraint); } } if (bundle_data->child && ((constraint->role_filter == RSC_ROLE_UNPROMOTED) || (constraint->role_filter == RSC_ROLE_PROMOTED))) { bundle_data->child->cmds->apply_location(bundle_data->child, constraint); bundle_data->child->rsc_location = g_list_prepend(bundle_data->child->rsc_location, constraint); } } /*! * \internal * \brief Add a resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void pcmk__bundle_expand(pe_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); if (bundle_data->child) { bundle_data->child->cmds->add_actions_to_graph(bundle_data->child); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->remote && replica->container && pe__bundle_needs_remote_name(replica->remote)) { /* REMOTE_CONTAINER_HACK: Allow remote nodes to run containers that * run pacemaker-remoted inside, without needing a separate IP for * the container. This is done by configuring the inner remote's * connection host as the magic string "#uname", then * replacing it with the underlying host when needed. */ xmlNode *nvpair = get_xpath_object("//nvpair[@name='" XML_RSC_ATTR_REMOTE_RA_ADDR "']", replica->remote->xml, LOG_ERR); const char *calculated_addr = NULL; // Replace the value in replica->remote->xml (if appropriate) calculated_addr = pe__add_bundle_remote_name(replica->remote, rsc->cluster, nvpair, "value"); if (calculated_addr) { /* Since this is for the bundle as a resource, and not any * particular action, replace the value in the default * parameters (not evaluated for node). create_graph_action() * will grab it from there to replace it in node-evaluated * parameters. */ GHashTable *params = pe_rsc_params(replica->remote, NULL, rsc->cluster); g_hash_table_replace(params, strdup(XML_RSC_ATTR_REMOTE_RA_ADDR), strdup(calculated_addr)); } else { /* The only way to get here is if the remote connection is * neither currently running nor scheduled to run. That means we * won't be doing any operations that require addr (only start * requires it; we additionally use it to compare digests when * unpacking status, promote, and migrate_from history, but * that's already happened by this point). */ crm_info("Unable to determine address for bundle %s remote connection", rsc->id); } } if (replica->ip) { replica->ip->cmds->add_actions_to_graph(replica->ip); } if (replica->container) { replica->container->cmds->add_actions_to_graph(replica->container); } if (replica->remote) { replica->remote->cmds->add_actions_to_graph(replica->remote); } } } /*! * \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 pcmk__bundle_create_probe(pe_resource_t *rsc, pe_node_t *node) { bool any_created = false; pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return false); get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if ((replica->ip != NULL) && replica->ip->cmds->create_probe(replica->ip, node)) { any_created = true; } if ((replica->child != NULL) && (node->details == replica->node->details) && replica->child->cmds->create_probe(replica->child, node)) { any_created = true; } if ((replica->container != NULL) && replica->container->cmds->create_probe(replica->container, node)) { any_created = true; /* If we're limited to one replica per host (due to * the lack of an IP range probably), then we don't * want any of our peer containers starting until * we've established that no other copies are already * running. * * Partly this is to ensure that nreplicas_per_host is * observed, but also to ensure that the containers * don't fail to start because the necessary port * mappings (which won't include an IP for uniqueness) * are already taken */ for (GList *tIter = bundle_data->replicas; tIter && (bundle_data->nreplicas_per_host == 1); tIter = tIter->next) { pe__bundle_replica_t *other = tIter->data; if ((other != replica) && (other != NULL) && (other->container != NULL)) { pcmk__new_ordering(replica->container, pcmk__op_key(replica->container->id, RSC_STATUS, 0), NULL, other->container, pcmk__op_key(other->container->id, RSC_START, 0), NULL, pe_order_optional|pe_order_same_node, rsc->cluster); } } } if ((replica->container != NULL) && (replica->remote != NULL) && replica->remote->cmds->create_probe(replica->remote, node)) { /* Do not probe the remote resource until we know where the * container is running. This is required for REMOTE_CONTAINER_HACK * to correctly probe remote resources. */ char *probe_uuid = pcmk__op_key(replica->remote->id, RSC_STATUS, 0); pe_action_t *probe = find_first_action(replica->remote->actions, probe_uuid, NULL, node); free(probe_uuid); if (probe != NULL) { any_created = true; crm_trace("Ordering %s probe on %s", replica->remote->id, pe__node_name(node)); pcmk__new_ordering(replica->container, pcmk__op_key(replica->container->id, RSC_START, 0), NULL, replica->remote, NULL, probe, pe_order_probe, rsc->cluster); } } } return any_created; } void pcmk__output_bundle_actions(pe_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (replica->ip != NULL) { replica->ip->cmds->output_actions(replica->ip); } if (replica->container != NULL) { replica->container->cmds->output_actions(replica->container); } if (replica->remote != NULL) { replica->remote->cmds->output_actions(replica->remote); } if (replica->child != NULL) { replica->child->cmds->output_actions(replica->child); } } } // Bundle implementation of resource_alloc_functions_t:add_utilization() void pcmk__bundle_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { pe__bundle_variant_data_t *bundle_data = NULL; pe__bundle_replica_t *replica = NULL; if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return; } get_bundle_variant_data(bundle_data, rsc); if (bundle_data->replicas == NULL) { return; } /* All bundle replicas are identical, so using the utilization of the first * is sufficient for any. Only the implicit container resource can have * utilization values. */ replica = (pe__bundle_replica_t *) bundle_data->replicas->data; if (replica->container != NULL) { replica->container->cmds->add_utilization(replica->container, orig_rsc, all_rscs, utilization); } } // Bundle implementation of resource_alloc_functions_t:shutdown_lock() void pcmk__bundle_shutdown_lock(pe_resource_t *rsc) { return; // Bundles currently don't support shutdown locks } diff --git a/lib/pacemaker/pcmk_sched_clone.c b/lib/pacemaker/pcmk_sched_clone.c index fb2705e0f3..d369dd838f 100644 --- a/lib/pacemaker/pcmk_sched_clone.c +++ b/lib/pacemaker/pcmk_sched_clone.c @@ -1,645 +1,645 @@ /* * Copyright 2004-2023 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 "libpacemaker_private.h" #define VARIANT_CLONE 1 #include /*! * \internal * \brief Assign a clone 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 * * \return Node that \p rsc is assigned to, if assigned entirely to one node */ pe_node_t * pcmk__clone_allocate(pe_resource_t *rsc, const pe_node_t *prefer) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return NULL; } else if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { pe_rsc_debug(rsc, "Dependency loop detected involving %s", rsc->id); return NULL; } if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__add_promotion_scores(rsc); } pe__set_resource_flags(rsc, pe_rsc_allocating); /* This information is used by pcmk__cmp_instance() when deciding the order * in which to assign clone instances to nodes. */ for (GList *gIter = rsc->rsc_cons; gIter != NULL; gIter = gIter->next) { pcmk__colocation_t *constraint = (pcmk__colocation_t *) gIter->data; pe_rsc_trace(rsc, "%s: Allocating %s first", rsc->id, constraint->primary->id); constraint->primary->cmds->assign(constraint->primary, prefer); } for (GList *gIter = rsc->rsc_cons_lhs; gIter != NULL; gIter = gIter->next) { pcmk__colocation_t *constraint = (pcmk__colocation_t *) gIter->data; if (pcmk__colocation_has_influence(constraint, NULL)) { pe_resource_t *dependent = constraint->dependent; const char *attr = constraint->node_attribute; const float factor = constraint->score / (float) INFINITY; const uint32_t flags = pcmk__coloc_select_active |pcmk__coloc_select_nonnegative; pcmk__add_colocated_node_scores(dependent, rsc->id, &rsc->allowed_nodes, attr, factor, flags); } } pe__show_node_weights(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance); - distribute_children(rsc, rsc->children, clone_data->clone_max, - clone_data->clone_node_max, rsc->cluster); + pcmk__assign_instances(rsc, rsc->children, clone_data->clone_max, + clone_data->clone_node_max); if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__set_instance_roles(rsc); } pe__clear_resource_flags(rsc, pe_rsc_provisional|pe_rsc_allocating); pe_rsc_trace(rsc, "Done allocating %s", rsc->id); return NULL; } static pe_action_t * find_rsc_action(pe_resource_t *rsc, const char *task) { pe_action_t *match = NULL; GList *actions = pe__resource_actions(rsc, NULL, task, FALSE); for (GList *item = actions; item != NULL; item = item->next) { pe_action_t *op = (pe_action_t *) item->data; if (!pcmk_is_set(op->flags, pe_action_optional)) { if (match != NULL) { // More than one match, don't return any match = NULL; break; } match = op; } } g_list_free(actions); return match; } static void child_ordering_constraints(pe_resource_t * rsc, pe_working_set_t * data_set) { pe_action_t *stop = NULL; pe_action_t *start = NULL; pe_action_t *last_stop = NULL; pe_action_t *last_start = NULL; GList *gIter = NULL; if (!pe__clone_is_ordered(rsc)) { return; } /* we have to maintain a consistent sorted child list when building order constraints */ rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance_number); for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child = (pe_resource_t *) gIter->data; stop = find_rsc_action(child, RSC_STOP); if (stop) { if (last_stop) { /* child/child relative stop */ order_actions(stop, last_stop, pe_order_optional); } last_stop = stop; } start = find_rsc_action(child, RSC_START); if (start) { if (last_start) { /* child/child relative start */ order_actions(last_start, start, pe_order_optional); } last_start = start; } } } void clone_create_actions(pe_resource_t *rsc) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); pe_rsc_debug(rsc, "Creating actions for clone %s", rsc->id); - clone_create_pseudo_actions(rsc, rsc->children, &clone_data->start_notify, - &clone_data->stop_notify); + pcmk__create_instance_actions(rsc, rsc->children, &clone_data->start_notify, + &clone_data->stop_notify); child_ordering_constraints(rsc, rsc->cluster); if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__create_promotable_actions(rsc); } } void clone_internal_constraints(pe_resource_t *rsc) { pe_resource_t *last_rsc = NULL; GList *gIter; bool ordered = pe__clone_is_ordered(rsc); pe_rsc_trace(rsc, "Internal constraints for %s", rsc->id); pcmk__order_resource_actions(rsc, RSC_STOPPED, rsc, RSC_START, pe_order_optional); pcmk__order_resource_actions(rsc, RSC_START, rsc, RSC_STARTED, pe_order_runnable_left); pcmk__order_resource_actions(rsc, RSC_STOP, rsc, RSC_STOPPED, pe_order_runnable_left); if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__order_resource_actions(rsc, RSC_DEMOTED, rsc, RSC_STOP, pe_order_optional); pcmk__order_resource_actions(rsc, RSC_STARTED, rsc, RSC_PROMOTE, pe_order_runnable_left); } if (ordered) { /* we have to maintain a consistent sorted child list when building order constraints */ rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance_number); } for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; child_rsc->cmds->internal_constraints(child_rsc); pcmk__order_starts(rsc, child_rsc, pe_order_runnable_left|pe_order_implies_first_printed); pcmk__order_resource_actions(child_rsc, RSC_START, rsc, RSC_STARTED, pe_order_implies_then_printed); if (ordered && (last_rsc != NULL)) { pcmk__order_starts(last_rsc, child_rsc, pe_order_optional); } pcmk__order_stops(rsc, child_rsc, pe_order_implies_first_printed); pcmk__order_resource_actions(child_rsc, RSC_STOP, rsc, RSC_STOPPED, pe_order_implies_then_printed); if (ordered && (last_rsc != NULL)) { pcmk__order_stops(child_rsc, last_rsc, pe_order_optional); } last_rsc = child_rsc; } if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__order_promotable_instances(rsc); } } /*! * \internal * \brief Apply a colocation's score to node weights or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node weights (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__clone_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { GList *gIter = NULL; gboolean do_interleave = FALSE; const char *interleave_s = NULL; /* This should never be called for the clone itself as a dependent. Instead, * we add its colocation constraints to its instances and call the * apply_coloc_score() for the instances as dependents. */ CRM_ASSERT(!for_dependent); CRM_CHECK((colocation != NULL) && (dependent != NULL) && (primary != NULL), return); CRM_CHECK(dependent->variant == pe_native, return); pe_rsc_trace(primary, "Processing constraint %s: %s -> %s %d", colocation->id, dependent->id, primary->id, colocation->score); if (pcmk_is_set(primary->flags, pe_rsc_promotable)) { if (pcmk_is_set(primary->flags, pe_rsc_provisional)) { // We haven't placed the primary yet, so we can't apply colocation pe_rsc_trace(primary, "%s is still provisional", primary->id); return; } else if (colocation->primary_role == RSC_ROLE_UNKNOWN) { // This isn't a role-specfic colocation, so handle normally pe_rsc_trace(primary, "Handling %s as a clone colocation", colocation->id); } else if (pcmk_is_set(dependent->flags, pe_rsc_provisional)) { // We're placing the dependent pcmk__update_dependent_with_promotable(primary, dependent, colocation); return; } else if (colocation->dependent_role == RSC_ROLE_PROMOTED) { // We're choosing roles for the dependent pcmk__update_promotable_dependent_priority(primary, dependent, colocation); return; } } // Only the dependent needs to be marked for interleave interleave_s = g_hash_table_lookup(colocation->dependent->meta, XML_RSC_ATTR_INTERLEAVE); if (crm_is_true(interleave_s) && (colocation->dependent->variant > pe_group)) { /* @TODO Do we actually care about multiple primary copies sharing a * dependent copy anymore? */ if (copies_per_node(colocation->dependent) != copies_per_node(colocation->primary)) { pcmk__config_err("Cannot interleave %s and %s because they do not " "support the same number of instances per node", colocation->dependent->id, colocation->primary->id); } else { do_interleave = TRUE; } } if (pcmk_is_set(primary->flags, pe_rsc_provisional)) { pe_rsc_trace(primary, "%s is still provisional", primary->id); return; } else if (do_interleave) { pe_resource_t *primary_instance = NULL; primary_instance = find_compatible_child(dependent, primary, RSC_ROLE_UNKNOWN, FALSE); if (primary_instance != NULL) { pe_rsc_debug(primary, "Pairing %s with %s", dependent->id, primary_instance->id); dependent->cmds->apply_coloc_score(dependent, primary_instance, colocation, true); } else if (colocation->score >= INFINITY) { crm_notice("Cannot pair %s with instance of %s", dependent->id, primary->id); pcmk__assign_resource(dependent, NULL, true); } else { pe_rsc_debug(primary, "Cannot pair %s with instance of %s", dependent->id, primary->id); } return; } else if (colocation->score >= INFINITY) { GList *affected_nodes = NULL; gIter = primary->children; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; pe_node_t *chosen = child_rsc->fns->location(child_rsc, NULL, FALSE); if (chosen != NULL && is_set_recursive(child_rsc, pe_rsc_block, TRUE) == FALSE) { pe_rsc_trace(primary, "Allowing %s: %s %d", colocation->id, pe__node_name(chosen), chosen->weight); affected_nodes = g_list_prepend(affected_nodes, chosen); } } node_list_exclude(dependent->allowed_nodes, affected_nodes, FALSE); g_list_free(affected_nodes); return; } gIter = primary->children; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; child_rsc->cmds->apply_coloc_score(dependent, child_rsc, colocation, false); } } enum pe_action_flags clone_action_flags(pe_action_t *action, const pe_node_t *node) { return summary_action_flags(action, action->rsc->children, node); } void clone_rsc_location(pe_resource_t *rsc, pe__location_t *constraint) { GList *gIter = rsc->children; pe_rsc_trace(rsc, "Processing location constraint %s for %s", constraint->id, rsc->id); pcmk__apply_location(rsc, constraint); for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; child_rsc->cmds->apply_location(child_rsc, constraint); } } /*! * \internal * \brief Add a resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void clone_expand(pe_resource_t *rsc) { GList *gIter = NULL; clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); g_list_foreach(rsc->actions, (GFunc) rsc->cmds->action_flags, NULL); pe__create_notifications(rsc, clone_data->start_notify); pe__create_notifications(rsc, clone_data->stop_notify); pe__create_notifications(rsc, clone_data->promote_notify); pe__create_notifications(rsc, clone_data->demote_notify); /* Now that the notifcations have been created we can expand the children */ gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; child_rsc->cmds->add_actions_to_graph(child_rsc); } pcmk__add_rsc_actions_to_graph(rsc); /* The notifications are in the graph now, we can destroy the notify_data */ pe__free_notification_data(clone_data->demote_notify); clone_data->demote_notify = NULL; pe__free_notification_data(clone_data->stop_notify); clone_data->stop_notify = NULL; pe__free_notification_data(clone_data->start_notify); clone_data->start_notify = NULL; pe__free_notification_data(clone_data->promote_notify); clone_data->promote_notify = NULL; } // Check whether a resource or any of its children is known on node static bool rsc_known_on(const pe_resource_t *rsc, const pe_node_t *node) { if (rsc->children) { for (GList *child_iter = rsc->children; child_iter != NULL; child_iter = child_iter->next) { pe_resource_t *child = (pe_resource_t *) child_iter->data; if (rsc_known_on(child, node)) { return TRUE; } } } else if (rsc->known_on) { GHashTableIter iter; pe_node_t *known_node = NULL; g_hash_table_iter_init(&iter, rsc->known_on); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &known_node)) { if (node->details == known_node->details) { return TRUE; } } } return FALSE; } // Look for an instance of clone that is known on node static pe_resource_t * find_instance_on(const pe_resource_t *clone, const pe_node_t *node) { for (GList *gIter = clone->children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child = (pe_resource_t *) gIter->data; if (rsc_known_on(child, node)) { return child; } } return NULL; } // For anonymous clones, only a single instance needs to be probed static bool probe_anonymous_clone(pe_resource_t *rsc, pe_node_t *node, pe_working_set_t *data_set) { // First, check if we probed an instance on this node last time pe_resource_t *child = find_instance_on(rsc, node); // Otherwise, check if we plan to start an instance on this node if (child == NULL) { for (GList *child_iter = rsc->children; child_iter && !child; child_iter = child_iter->next) { pe_node_t *local_node = NULL; pe_resource_t *child_rsc = (pe_resource_t *) child_iter->data; if (child_rsc) { /* make clang analyzer happy */ local_node = child_rsc->fns->location(child_rsc, NULL, FALSE); if (local_node && (local_node->details == node->details)) { child = child_rsc; } } } } // Otherwise, use the first clone instance if (child == NULL) { child = rsc->children->data; } CRM_ASSERT(child); return child->cmds->create_probe(child, node); } /*! * \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 clone_create_probe(pe_resource_t *rsc, pe_node_t *node) { CRM_ASSERT(rsc); rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance_number); if (rsc->children == NULL) { pe_warn("Clone %s has no children", rsc->id); return false; } if (rsc->exclusive_discover) { pe_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (allowed && allowed->rsc_discover_mode != pe_discover_exclusive) { /* exclusive discover is enabled and this node is not marked * as a node this resource should be discovered on * * remove the node from allowed_nodes so that the * notification contains only nodes that we might ever run * on */ g_hash_table_remove(rsc->allowed_nodes, node->details->id); /* Bit of a shortcut - might as well take it */ return false; } } if (pcmk_is_set(rsc->flags, pe_rsc_unique)) { return pcmk__probe_resource_list(rsc->children, node); } else { return probe_anonymous_clone(rsc, node, rsc->cluster); } } void clone_append_meta(const pe_resource_t *rsc, xmlNode *xml) { char *name = NULL; clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); name = crm_meta_name(XML_RSC_ATTR_UNIQUE); crm_xml_add(xml, name, pe__rsc_bool_str(rsc, pe_rsc_unique)); free(name); name = crm_meta_name(XML_RSC_ATTR_NOTIFY); crm_xml_add(xml, name, pe__rsc_bool_str(rsc, pe_rsc_notify)); free(name); name = crm_meta_name(XML_RSC_ATTR_INCARNATION_MAX); crm_xml_add_int(xml, name, clone_data->clone_max); free(name); name = crm_meta_name(XML_RSC_ATTR_INCARNATION_NODEMAX); crm_xml_add_int(xml, name, clone_data->clone_node_max); free(name); if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { int promoted_max = pe__clone_promoted_max(rsc); int promoted_node_max = pe__clone_promoted_node_max(rsc); name = crm_meta_name(XML_RSC_ATTR_PROMOTED_MAX); crm_xml_add_int(xml, name, promoted_max); free(name); name = crm_meta_name(XML_RSC_ATTR_PROMOTED_NODEMAX); crm_xml_add_int(xml, name, promoted_node_max); free(name); /* @COMPAT Maintain backward compatibility with resource agents that * expect the old names (deprecated since 2.0.0). */ name = crm_meta_name(PCMK_XA_PROMOTED_MAX_LEGACY); crm_xml_add_int(xml, name, promoted_max); free(name); name = crm_meta_name(PCMK_XA_PROMOTED_NODE_MAX_LEGACY); crm_xml_add_int(xml, name, promoted_node_max); free(name); } } // Clone implementation of resource_alloc_functions_t:add_utilization() void pcmk__clone_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { bool existing = false; pe_resource_t *child = NULL; if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return; } // Look for any child already existing in the list for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { child = (pe_resource_t *) iter->data; if (g_list_find(all_rscs, child)) { existing = true; // Keep checking remaining children } else { // If this is a clone of a group, look for group's members for (GList *member_iter = child->children; member_iter != NULL; member_iter = member_iter->next) { pe_resource_t *member = (pe_resource_t *) member_iter->data; if (g_list_find(all_rscs, member) != NULL) { // Add *child's* utilization, not group member's child->cmds->add_utilization(child, orig_rsc, all_rscs, utilization); existing = true; break; } } } } if (!existing && (rsc->children != NULL)) { // If nothing was found, still add first child's utilization child = (pe_resource_t *) rsc->children->data; child->cmds->add_utilization(child, orig_rsc, all_rscs, utilization); } } // Clone implementation of resource_alloc_functions_t:shutdown_lock() void pcmk__clone_shutdown_lock(pe_resource_t *rsc) { return; // Clones currently don't support shutdown locks } diff --git a/lib/pacemaker/pcmk_sched_instances.c b/lib/pacemaker/pcmk_sched_instances.c index 420edeef0c..b344d9e2a0 100644 --- a/lib/pacemaker/pcmk_sched_instances.c +++ b/lib/pacemaker/pcmk_sched_instances.c @@ -1,654 +1,1020 @@ /* * Copyright 2004-2023 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. */ /* This file is intended for code usable with both clone instances and bundle * replica containers. */ #include #include #include #include "libpacemaker_private.h" static void append_parent_colocation(pe_resource_t * rsc, pe_resource_t * child, gboolean all); /*! * \internal * \brief Check whether a node is allowed to run an instance * * \param[in] instance Clone instance or bundle container to check * \param[in] node Node to check * \param[in] max_per_node Maximum number of instances allowed to run on a node * * \return true if \p node is allowed to run \p instance, otherwise false */ static bool can_run_instance(const pe_resource_t *instance, const pe_node_t *node, int max_per_node) { pe_node_t *allowed_node = NULL; if (pcmk_is_set(instance->flags, pe_rsc_orphan)) { pe_rsc_trace(instance, "%s cannot run on %s: orphaned", instance->id, pe__node_name(node)); return false; } if (!pcmk__node_available(node, false, false)) { pe_rsc_trace(instance, "%s cannot run on %s: node cannot run resources", instance->id, pe__node_name(node)); return false; } allowed_node = pcmk__top_allowed_node(instance, node); if (allowed_node == NULL) { crm_warn("%s cannot run on %s: node not allowed", instance->id, pe__node_name(node)); return false; } if (allowed_node->weight < 0) { pe_rsc_trace(instance, "%s cannot run on %s: parent score is %s there", instance->id, pe__node_name(node), pcmk_readable_score(allowed_node->weight)); return false; } if (allowed_node->count >= max_per_node) { pe_rsc_trace(instance, "%s cannot run on %s: node already has %d instance%s", instance->id, pe__node_name(node), max_per_node, pcmk__plural_s(max_per_node)); return false; } pe_rsc_trace(instance, "%s can run on %s (%d already running)", instance->id, pe__node_name(node), allowed_node->count); return true; } /*! * \internal * \brief Ban a clone instance or bundle replica from unavailable allowed nodes * * \param[in,out] instance Clone instance or bundle replica to ban * \param[in] max_per_node Maximum instances allowed to run on a node */ static void ban_unavailable_allowed_nodes(pe_resource_t *instance, int max_per_node) { if (instance->allowed_nodes != NULL) { GHashTableIter iter; const pe_node_t *allowed_node = NULL; g_hash_table_iter_init(&iter, instance->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &allowed_node)) { if (!can_run_instance(instance, allowed_node, max_per_node)) { // Ban instance (and all its children) from node common_update_score(instance, allowed_node->details->id, -INFINITY); } } } } /*! * \internal * \brief Choose a node for an instance * * \param[in,out] instance Clone instance or bundle replica container * \param[in] prefer If not NULL, attempt early assignment to this * node, if still the best choice; otherwise, * perform final assignment * \param[in] all_coloc If true (indicating that there are more * available nodes than instances), add all parent * colocations to instance, otherwise add only * negative (and for "this with" colocations, * infinite) colocations to avoid needless * shuffling of instances among nodes * \param[in] max_per_node Assign at most this many instances to one node * * \return true if \p instance could be assigned to a node, otherwise false */ static bool assign_instance(pe_resource_t *instance, const pe_node_t *prefer, bool all_coloc, int max_per_node) { pe_node_t *chosen = NULL; pe_node_t *allowed = NULL; CRM_ASSERT(instance != NULL); pe_rsc_trace(instance, "Assigning %s (preferring %s, using %s parent colocations)", instance->id, ((prefer == NULL)? "no node" : prefer->details->uname), (all_coloc? "all" : "essential")); if (!pcmk_is_set(instance->flags, pe_rsc_provisional)) { // Instance is already assigned return instance->fns->location(instance, NULL, FALSE) != NULL; } if (pcmk_is_set(instance->flags, pe_rsc_allocating)) { pe_rsc_debug(instance, "Assignment loop detected involving %s colocations", instance->id); return false; } if (prefer != NULL) { // Possible early assignment to preferred node // Get preferred node with instance's scores allowed = g_hash_table_lookup(instance->allowed_nodes, prefer->details->id); if ((allowed == NULL) || (allowed->weight < 0)) { pe_rsc_trace(instance, "Not assigning %s to preferred node %s: unavailable", instance->id, pe__node_name(prefer)); return false; } } ban_unavailable_allowed_nodes(instance, max_per_node); if (prefer == NULL) { // Final assignment chosen = instance->cmds->assign(instance, NULL); } else { // Possible early assignment to preferred node GHashTable *backup = pcmk__copy_node_table(instance->allowed_nodes); chosen = instance->cmds->assign(instance, prefer); // Revert nodes if preferred node won't be assigned if ((chosen != NULL) && (chosen->details != prefer->details)) { crm_info("Not assigning %s to preferred node %s: %s is better", instance->id, pe__node_name(prefer), pe__node_name(chosen)); g_hash_table_destroy(instance->allowed_nodes); instance->allowed_nodes = backup; pcmk__unassign_resource(instance); chosen = NULL; } else if (backup != NULL) { g_hash_table_destroy(backup); } } // The parent tracks how many instances have been assigned to each node if (chosen != NULL) { allowed = pcmk__top_allowed_node(instance, chosen); if (allowed == NULL) { /* The instance is allowed on the node, but its parent isn't. This * shouldn't be possible if the resource is managed, and we won't be * able to limit the number of instances assigned to the node. */ CRM_LOG_ASSERT(!pcmk_is_set(instance->flags, pe_rsc_managed)); } else { allowed->count++; } } return chosen != NULL; } static void append_parent_colocation(pe_resource_t * rsc, pe_resource_t * child, gboolean all) { GList *gIter = NULL; gIter = rsc->rsc_cons; for (; gIter != NULL; gIter = gIter->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) gIter->data; if (all || cons->score < 0 || cons->score == INFINITY) { pcmk__add_this_with(child, cons); } } gIter = rsc->rsc_cons_lhs; for (; gIter != NULL; gIter = gIter->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) gIter->data; if (!pcmk__colocation_has_influence(cons, child)) { continue; } if (all || cons->score < 0) { pcmk__add_with_this(child, cons); } } } /*! * \internal * \brief Reset the node counts of a resource's allowed nodes to zero * * \param[in,out] rsc Resource to reset * * \return Number of nodes that are available to run resources */ static unsigned int reset_allowed_node_counts(pe_resource_t *rsc) { unsigned int available_nodes = 0; pe_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { node->count = 0; if (pcmk__node_available(node, false, false)) { available_nodes++; } } return available_nodes; } /*! * \internal * \brief Check whether an instance has a preferred node * * \param[in] rsc Clone or bundle being assigned (for logs only) * \param[in] instance Clone instance or bundle replica container * \param[in] optimal_per_node Optimal number of instances per node * * \return Instance's current node if still available, otherwise NULL */ static const pe_node_t * preferred_node(const pe_resource_t *rsc, const pe_resource_t *instance, int optimal_per_node) { const pe_node_t *node = NULL; const pe_node_t *parent_node = NULL; // Check whether instance is active, healthy, and not yet assigned if ((instance->running_on == NULL) || !pcmk_is_set(instance->flags, pe_rsc_provisional) || pcmk_is_set(instance->flags, pe_rsc_failed)) { return NULL; } // Check whether instance's current node can run resources node = pe__current_node(instance); if (!pcmk__node_available(node, true, false)) { pe_rsc_trace(rsc, "Not assigning %s to %s early (unavailable)", instance->id, pe__node_name(node)); return NULL; } // Check whether node already has optimal number of instances assigned parent_node = pcmk__top_allowed_node(instance, node); if ((parent_node != NULL) && (parent_node->count >= optimal_per_node)) { pe_rsc_trace(rsc, "Not assigning %s to %s early " "(optimal instances already assigned)", instance->id, pe__node_name(node)); return NULL; } return node; } +/*! + * \internal + * \brief Assign collective instances to nodes + * + * \param[in,out] collective Clone or bundle resource being assigned + * \param[in,out] instances List of clone instances or bundle containers + * \param[in] max_total Maximum instances to assign in total + * \param[in] max_per_node Maximum instances to assign to any one node + */ void -distribute_children(pe_resource_t *rsc, GList *children, int max, - int per_host_max, pe_working_set_t *data_set) +pcmk__assign_instances(pe_resource_t *collective, GList *instances, + int max_total, int max_per_node) { // Reuse node count to track number of assigned instances - unsigned int available_nodes = reset_allowed_node_counts(rsc); + unsigned int available_nodes = reset_allowed_node_counts(collective); /* Include finite positive preferences of the collective's * colocation dependents only if not every node will get an instance. */ - bool all_coloc = (max < available_nodes); + bool all_coloc = (max_total < available_nodes); - int loop_max = 0; - int allocated = 0; + int optimal_per_node = 0; + int assigned = 0; + GList *iter = NULL; + pe_resource_t *instance = NULL; + const pe_node_t *current = NULL; - if(available_nodes) { - loop_max = max / available_nodes; + if (available_nodes > 0) { + optimal_per_node = max_total / available_nodes; } - if (loop_max < 1) { - loop_max = 1; + if (optimal_per_node < 1) { + optimal_per_node = 1; } - pe_rsc_debug(rsc, - "Allocating up to %d %s instances to a possible %u nodes " + pe_rsc_debug(collective, + "Assigning up to %d %s instance%s to up to %u node%s " "(at most %d per host, %d optimal)", - max, rsc->id, available_nodes, per_host_max, loop_max); - - /* Pre-allocate as many instances as we can to their current location */ - for (GList *gIter = children; gIter != NULL && allocated < max; gIter = gIter->next) { - pe_resource_t *child = (pe_resource_t *) gIter->data; - const pe_node_t *child_node = NULL; - - append_parent_colocation(child->parent, child, all_coloc); - - child_node = preferred_node(rsc, child, loop_max); - if ((child_node != NULL) - && assign_instance(child, child_node, all_coloc, per_host_max)) { - pe_rsc_trace(rsc, "Pre-allocated %s to %s", child->id, - pe__node_name(child_node)); - allocated++; + max_total, collective->id, pcmk__plural_s(max_total), + available_nodes, pcmk__plural_s(available_nodes), + max_per_node, optimal_per_node); + + // Assign as many instances as possible to their current location + for (iter = instances; (iter != NULL) && (assigned < max_total); + iter = iter->next) { + instance = (pe_resource_t *) iter->data; + + append_parent_colocation(instance->parent, instance, all_coloc); + + current = preferred_node(collective, instance, optimal_per_node); + if ((current != NULL) + && assign_instance(instance, current, all_coloc, max_per_node)) { + pe_rsc_trace(collective, "Assigned %s to current node %s", + instance->id, pe__node_name(current)); + assigned++; } } - pe_rsc_trace(rsc, "Done pre-allocating (%d of %d)", allocated, max); + pe_rsc_trace(collective, "Assigned %d of %d instance%s to current node", + assigned, max_total, pcmk__plural_s(max_total)); - for (GList *gIter = children; gIter != NULL; gIter = gIter->next) { - pe_resource_t *child = (pe_resource_t *) gIter->data; + for (iter = instances; iter != NULL; iter = iter->next) { + instance = (pe_resource_t *) iter->data; + + if (!pcmk_is_set(instance->flags, pe_rsc_provisional)) { + continue; // Already assigned + } - if (child->running_on != NULL) { - pe_node_t *child_node = pe__current_node(child); - pe_node_t *local_node = pcmk__top_allowed_node(child, child_node); + if (instance->running_on != NULL) { + current = pe__current_node(instance); + if (pcmk__top_allowed_node(instance, current) == NULL) { + const char *unmanaged = ""; - if (local_node == NULL) { - crm_err("%s is running on %s which isn't allowed", - child->id, pe__node_name(child_node)); + if (!pcmk_is_set(instance->flags, pe_rsc_managed)) { + unmanaged = "Unmanaged resource "; + } + crm_notice("%s%s is running on %s which is no longer allowed", + unmanaged, instance->id, pe__node_name(current)); } } - if (!pcmk_is_set(child->flags, pe_rsc_provisional)) { - } else if (allocated >= max) { - pe_rsc_debug(rsc, "Child %s not allocated - limit reached %d %d", child->id, allocated, max); - resource_location(child, NULL, -INFINITY, "clone:limit_reached", data_set); - } else { - if (assign_instance(child, NULL, all_coloc, per_host_max)) { - allocated++; - } + if (assigned >= max_total) { + pe_rsc_debug(collective, + "Not assigning %s because maximum %d instances " + "already assigned", + instance->id, max_total); + resource_location(instance, NULL, -INFINITY, + "collective_limit_reached", collective->cluster); + + } else if (assign_instance(instance, NULL, all_coloc, max_per_node)) { + assigned++; } } - pe_rsc_debug(rsc, "Allocated %d %s instances of a possible %d", - allocated, rsc->id, max); + pe_rsc_debug(collective, "Assigned %d of %d possible instance%s of %s", + assigned, max_total, pcmk__plural_s(max_total), + collective->id); } +enum instance_state { + instance_starting = (1 << 0), + instance_stopping = (1 << 1), + + /* This indicates that some instance is restarting. It's not the same as + * instance_starting|instance_stopping, which would indicate that some + * instance is starting, and some instance (not necessarily the same one) is + * stopping. + */ + instance_restarting = (1 << 2), + + instance_active = (1 << 3), + + instance_all = instance_starting|instance_stopping + |instance_restarting|instance_active, +}; + +/*! + * \internal + * \brief Check whether an instance is active, starting, and/or stopping + * + * \param[in] instance Clone instance or bundle replica container + * \param[in,out] state Whether any instance is starting, stopping, etc. + */ static void -clone_update_pseudo_status(pe_resource_t * rsc, gboolean * stopping, gboolean * starting, - gboolean * active) +check_instance_state(const pe_resource_t *instance, uint32_t *state) { - GList *gIter = NULL; + const GList *iter = NULL; + uint32_t instance_state = 0; // State of just this instance - if (rsc->children) { - - gIter = rsc->children; - for (; gIter != NULL; gIter = gIter->next) { - pe_resource_t *child = (pe_resource_t *) gIter->data; + // No need to check further if all conditions have already been detected + if (pcmk_all_flags_set(*state, instance_all)) { + return; + } - clone_update_pseudo_status(child, stopping, starting, active); + // If instance is a collective (a cloned group), check its children instead + if (instance->variant > pe_native) { + for (iter = instance->children; + (iter != NULL) && !pcmk_all_flags_set(*state, instance_all); + iter = iter->next) { + check_instance_state((const pe_resource_t *) iter->data, state); } - return; } - CRM_ASSERT(active != NULL); - CRM_ASSERT(starting != NULL); - CRM_ASSERT(stopping != NULL); + // If we get here, instance is a primitive - if (rsc->running_on) { - *active = TRUE; + if (instance->running_on != NULL) { + instance_state |= instance_active; } - gIter = rsc->actions; - for (; gIter != NULL; gIter = gIter->next) { - pe_action_t *action = (pe_action_t *) gIter->data; - - if (*starting && *stopping) { - return; - - } else if (pcmk_is_set(action->flags, pe_action_optional)) { - pe_rsc_trace(rsc, "Skipping optional: %s", action->uuid); - continue; - - } else if (!pcmk_any_flags_set(action->flags, - pe_action_pseudo|pe_action_runnable)) { - pe_rsc_trace(rsc, "Skipping unrunnable: %s", action->uuid); - continue; - - } else if (pcmk__str_eq(RSC_STOP, action->task, pcmk__str_casei)) { - pe_rsc_trace(rsc, "Stopping due to: %s", action->uuid); - *stopping = TRUE; - - } else if (pcmk__str_eq(RSC_START, action->task, pcmk__str_casei)) { - if (!pcmk_is_set(action->flags, pe_action_runnable)) { - pe_rsc_trace(rsc, "Skipping pseudo-op: %s run=%d, pseudo=%d", - action->uuid, - pcmk_is_set(action->flags, pe_action_runnable), - pcmk_is_set(action->flags, pe_action_pseudo)); + // Check each of the instance's actions for runnable start or stop + for (iter = instance->actions; + (iter != NULL) && !pcmk_all_flags_set(instance_state, + instance_starting + |instance_stopping); + iter = iter->next) { + + const pe_action_t *action = (const pe_action_t *) iter->data; + const bool optional = pcmk_is_set(action->flags, pe_action_optional); + + if (pcmk__str_eq(RSC_START, action->task, pcmk__str_none)) { + if (!optional && pcmk_is_set(action->flags, pe_action_runnable)) { + pe_rsc_trace(instance, "Instance is starting due to %s", + action->uuid); + instance_state |= instance_starting; } else { - pe_rsc_trace(rsc, "Starting due to: %s", action->uuid); - pe_rsc_trace(rsc, "%s run=%d, pseudo=%d", - action->uuid, - pcmk_is_set(action->flags, pe_action_runnable), - pcmk_is_set(action->flags, pe_action_pseudo)); - *starting = TRUE; + pe_rsc_trace(instance, "%s doesn't affect %s state (%s)", + action->uuid, instance->id, + (optional? "optional" : "unrunnable")); + } + + } else if (pcmk__str_eq(RSC_STOP, action->task, pcmk__str_none)) { + /* Only stop actions can be pseudo-actions for primitives. That + * indicates that the node they are on is being fenced, so the stop + * is implied rather than actually executed. + */ + if (!optional + && pcmk_any_flags_set(action->flags, + pe_action_pseudo|pe_action_runnable)) { + pe_rsc_trace(instance, "Instance is stopping due to %s", + action->uuid); + instance_state |= instance_stopping; + } else { + pe_rsc_trace(instance, "%s doesn't affect %s state (%s)", + action->uuid, instance->id, + (optional? "optional" : "unrunnable")); } } } + + if (pcmk_all_flags_set(instance_state, + instance_starting|instance_stopping)) { + instance_state |= instance_restarting; + } + *state |= instance_state; } +/*! + * \internal + * \brief Create actions for collective resource instances + * + * \param[in,out] collective Clone or bundle resource to create actions for + * \param[in,out] instances List of clone instances or bundle containers + * \param[in,out] start_notify If not NULL, create start notification actions + * \param[in,out] stop_notify If not NULL, create stop notification actions + */ void -clone_create_pseudo_actions(pe_resource_t *rsc, GList *children, - notify_data_t **start_notify, - notify_data_t **stop_notify) +pcmk__create_instance_actions(pe_resource_t *collective, GList *instances, + notify_data_t **start_notify, + notify_data_t **stop_notify) { - gboolean child_active = FALSE; - gboolean child_starting = FALSE; - gboolean child_stopping = FALSE; - gboolean allow_dependent_migrations = TRUE; + uint32_t state = 0; pe_action_t *stop = NULL; pe_action_t *stopped = NULL; pe_action_t *start = NULL; pe_action_t *started = NULL; - pe_rsc_trace(rsc, "Creating actions for %s", rsc->id); + pe_rsc_trace(collective, "Creating collective instance actions for %s", + collective->id); - for (GList *gIter = children; gIter != NULL; gIter = gIter->next) { - pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; - gboolean starting = FALSE; - gboolean stopping = FALSE; + // Create actions for each instance appropriate to its variant + for (GList *iter = instances; iter != NULL; iter = iter->next) { + pe_resource_t *instance = (pe_resource_t *) iter->data; - child_rsc->cmds->create_actions(child_rsc); - clone_update_pseudo_status(child_rsc, &stopping, &starting, &child_active); - if (stopping && starting) { - allow_dependent_migrations = FALSE; - } - - child_stopping |= stopping; - child_starting |= starting; + instance->cmds->create_actions(instance); + check_instance_state(instance, &state); } - /* start */ - start = pe__new_rsc_pseudo_action(rsc, RSC_START, !child_starting, true); - started = pe__new_rsc_pseudo_action(rsc, RSC_STARTED, !child_starting, + // Create pseudo-actions for rsc start and started + start = pe__new_rsc_pseudo_action(collective, RSC_START, + !pcmk_is_set(state, instance_starting), + true); + started = pe__new_rsc_pseudo_action(collective, RSC_STARTED, + !pcmk_is_set(state, instance_starting), false); started->priority = INFINITY; - - if (child_active || child_starting) { + if (pcmk_any_flags_set(state, instance_active|instance_starting)) { pe__set_action_flags(started, pe_action_runnable); } - if (start_notify != NULL && *start_notify == NULL) { - *start_notify = pe__clone_notif_pseudo_ops(rsc, RSC_START, start, + if ((start_notify != NULL) && (*start_notify == NULL)) { + *start_notify = pe__clone_notif_pseudo_ops(collective, RSC_START, start, started); } - /* stop */ - stop = pe__new_rsc_pseudo_action(rsc, RSC_STOP, !child_stopping, true); - stopped = pe__new_rsc_pseudo_action(rsc, RSC_STOPPED, !child_stopping, + // Create pseudo-actions for rsc stop and stopped + stop = pe__new_rsc_pseudo_action(collective, RSC_STOP, + !pcmk_is_set(state, instance_stopping), + true); + stopped = pe__new_rsc_pseudo_action(collective, RSC_STOPPED, + !pcmk_is_set(state, instance_stopping), true); stopped->priority = INFINITY; - if (allow_dependent_migrations) { + if (!pcmk_is_set(state, instance_restarting)) { pe__set_action_flags(stop, pe_action_migrate_runnable); } - if (stop_notify != NULL && *stop_notify == NULL) { - *stop_notify = pe__clone_notif_pseudo_ops(rsc, RSC_STOP, stop, stopped); - - if (start_notify && *start_notify && *stop_notify) { - order_actions((*stop_notify)->post_done, (*start_notify)->pre, pe_order_optional); + if ((stop_notify != NULL) && (*stop_notify == NULL)) { + *stop_notify = pe__clone_notif_pseudo_ops(collective, RSC_STOP, stop, + stopped); + if ((start_notify != NULL) && (*start_notify != NULL) + && (*stop_notify != NULL)) { + order_actions((*stop_notify)->post_done, (*start_notify)->pre, + pe_order_optional); } } } +static inline GList * +get_containers_or_children(const pe_resource_t *rsc) +{ + return (rsc->variant == pe_container)? + pcmk__bundle_containers(rsc) : rsc->children; +} + gboolean is_child_compatible(const pe_resource_t *child_rsc, const pe_node_t *local_node, enum rsc_role_e filter, gboolean current) { pe_node_t *node = NULL; enum rsc_role_e next_role = child_rsc->fns->state(child_rsc, current); CRM_CHECK(child_rsc && local_node, return FALSE); if (is_set_recursive(child_rsc, pe_rsc_block, TRUE) == FALSE) { /* We only want instances that haven't failed */ node = child_rsc->fns->location(child_rsc, NULL, current); } if (filter != RSC_ROLE_UNKNOWN && next_role != filter) { crm_trace("Filtered %s", child_rsc->id); return FALSE; } if (node && (node->details == local_node->details)) { return TRUE; } else if (node) { crm_trace("%s - %s vs %s", child_rsc->id, pe__node_name(node), pe__node_name(local_node)); } else { crm_trace("%s - not allocated %d", child_rsc->id, current); } return FALSE; } +static pe_resource_t * +find_compatible_child_by_node(const pe_resource_t *local_child, + const pe_node_t *local_node, + const pe_resource_t *rsc, enum rsc_role_e filter, + gboolean current) +{ + GList *gIter = NULL; + GList *children = NULL; + + if (local_node == NULL) { + crm_err("Can't colocate unrunnable child %s with %s", local_child->id, rsc->id); + return NULL; + } + + crm_trace("Looking for compatible child from %s for %s on %s", + local_child->id, rsc->id, pe__node_name(local_node)); + + children = get_containers_or_children(rsc); + for (gIter = children; gIter != NULL; gIter = gIter->next) { + pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; + + if(is_child_compatible(child_rsc, local_node, filter, current)) { + crm_trace("Pairing %s with %s on %s", + local_child->id, child_rsc->id, pe__node_name(local_node)); + return child_rsc; + } + } + + crm_trace("Can't pair %s with %s", local_child->id, rsc->id); + if(children != rsc->children) { + g_list_free(children); + } + return NULL; +} + pe_resource_t * find_compatible_child(const pe_resource_t *local_child, const pe_resource_t *rsc, enum rsc_role_e filter, gboolean current) { pe_resource_t *pair = NULL; GList *gIter = NULL; GList *scratch = NULL; pe_node_t *local_node = NULL; local_node = local_child->fns->location(local_child, NULL, current); if (local_node) { return find_compatible_child_by_node(local_child, local_node, rsc, filter, current); } scratch = g_hash_table_get_values(local_child->allowed_nodes); scratch = pcmk__sort_nodes(scratch, NULL); gIter = scratch; for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; pair = find_compatible_child_by_node(local_child, node, rsc, filter, current); if (pair) { goto done; } } pe_rsc_debug(rsc, "Can't pair %s with %s", local_child->id, rsc->id); done: g_list_free(scratch); return pair; } +static uint32_t +multi_update_interleave_actions(pe_action_t *first, pe_action_t *then, + const pe_node_t *node, uint32_t filter, + uint32_t type, pe_working_set_t *data_set) +{ + GList *gIter = NULL; + GList *children = NULL; + gboolean current = FALSE; + uint32_t changed = pcmk__updated_none; + + /* Fix this - lazy */ + if (pcmk__ends_with(first->uuid, "_stopped_0") + || pcmk__ends_with(first->uuid, "_demoted_0")) { + current = TRUE; + } + + children = get_containers_or_children(then->rsc); + for (gIter = children; gIter != NULL; gIter = gIter->next) { + pe_resource_t *then_child = gIter->data; + pe_resource_t *first_child = find_compatible_child(then_child, + first->rsc, + RSC_ROLE_UNKNOWN, + current); + if (first_child == NULL && current) { + crm_trace("Ignore"); + + } else if (first_child == NULL) { + crm_debug("No match found for %s (%d / %s / %s)", then_child->id, current, first->uuid, then->uuid); + + /* Me no like this hack - but what else can we do? + * + * If there is no-one active or about to be active + * on the same node as then_child, then they must + * not be allowed to start + */ + if (pcmk_any_flags_set(type, pe_order_runnable_left|pe_order_implies_then) /* Mandatory */ ) { + pe_rsc_info(then->rsc, "Inhibiting %s from being active", then_child->id); + if (pcmk__assign_resource(then_child, NULL, true)) { + pcmk__set_updated_flags(changed, first, pcmk__updated_then); + } + } + + } else { + pe_action_t *first_action = NULL; + pe_action_t *then_action = NULL; + + enum action_tasks task = clone_child_action(first); + const char *first_task = task2text(task); + + const pe_resource_t *first_rsc = NULL; + const pe_resource_t *then_rsc = NULL; + + first_rsc = pcmk__get_rsc_in_container(first_child, node); + if ((first_rsc != NULL) + && pcmk__str_any_of(first->task, CRMD_ACTION_STOP, + CRMD_ACTION_STOPPED, NULL)) { + /* Use the containerized resource since its actions will happen + * later and are more likely to align with the user's intent. + */ + first_action = find_first_action(first_rsc->actions, NULL, + first_task, node); + } else { + first_action = find_first_action(first_child->actions, NULL, + first_task, node); + } + + then_rsc = pcmk__get_rsc_in_container(then_child, node); + if ((then_rsc != NULL) + && pcmk__str_any_of(then->task, CRMD_ACTION_PROMOTE, + CRMD_ACTION_PROMOTED, CRMD_ACTION_DEMOTE, + CRMD_ACTION_DEMOTED, NULL)) { + /* Role actions apply only to the containerized resource, not + * the container itself. + */ + then_action = find_first_action(then_rsc->actions, NULL, + then->task, node); + } else { + then_action = find_first_action(then_child->actions, NULL, + then->task, node); + } + + if (first_action == NULL) { + if (!pcmk_is_set(first_child->flags, pe_rsc_orphan) + && !pcmk__str_any_of(first_task, RSC_STOP, RSC_DEMOTE, NULL)) { + crm_err("Internal error: No action found for %s in %s (first)", + first_task, first_child->id); + + } else { + crm_trace("No action found for %s in %s%s (first)", + first_task, first_child->id, + pcmk_is_set(first_child->flags, pe_rsc_orphan)? " (ORPHAN)" : ""); + } + continue; + } + + /* We're only interested if 'then' is neither stopping nor being demoted */ + if (then_action == NULL) { + if (!pcmk_is_set(then_child->flags, pe_rsc_orphan) + && !pcmk__str_any_of(then->task, RSC_STOP, RSC_DEMOTE, NULL)) { + crm_err("Internal error: No action found for %s in %s (then)", + then->task, then_child->id); + + } else { + crm_trace("No action found for %s in %s%s (then)", + then->task, then_child->id, + pcmk_is_set(then_child->flags, pe_rsc_orphan)? " (ORPHAN)" : ""); + } + continue; + } + + if (order_actions(first_action, then_action, type)) { + crm_debug("Created constraint for %s (%d) -> %s (%d) %.6x", + first_action->uuid, + pcmk_is_set(first_action->flags, pe_action_optional), + then_action->uuid, + pcmk_is_set(then_action->flags, pe_action_optional), + type); + pcmk__set_updated_flags(changed, first, + pcmk__updated_first|pcmk__updated_then); + } + if(first_action && then_action) { + changed |= then_child->cmds->update_ordered_actions(first_action, + then_action, + node, + first_child->cmds->action_flags(first_action, node), + filter, + type, + data_set); + } else { + crm_err("Nothing found either for %s (%p) or %s (%p) %s", + first_child->id, first_action, + then_child->id, then_action, task2text(task)); + } + } + } + + if(children != then->rsc->children) { + g_list_free(children); + } + return changed; +} + +static bool +can_interleave_actions(pe_action_t *first, pe_action_t *then) +{ + bool interleave = FALSE; + pe_resource_t *rsc = NULL; + const char *interleave_s = NULL; + + if(first->rsc == NULL || then->rsc == NULL) { + crm_trace("Not interleaving %s with %s (both must be resources)", first->uuid, then->uuid); + return FALSE; + } else if(first->rsc == then->rsc) { + crm_trace("Not interleaving %s with %s (must belong to different resources)", first->uuid, then->uuid); + return FALSE; + } else if(first->rsc->variant < pe_clone || then->rsc->variant < pe_clone) { + crm_trace("Not interleaving %s with %s (both sides must be clones or bundles)", first->uuid, then->uuid); + return FALSE; + } + + if (pcmk__ends_with(then->uuid, "_stop_0") + || pcmk__ends_with(then->uuid, "_demote_0")) { + rsc = first->rsc; + } else { + rsc = then->rsc; + } + + interleave_s = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERLEAVE); + interleave = crm_is_true(interleave_s); + crm_trace("Interleave %s -> %s: %s (based on %s)", + first->uuid, then->uuid, interleave ? "yes" : "no", rsc->id); + + return interleave; +} + +/*! + * \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. In some cases, the ordering could be disabled 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 pe_action_optional to affect only mandatory + * actions, and pe_action_runnable to affect only + * runnable actions) + * \param[in] type Group of enum pe_ordering flags to apply + * \param[in,out] data_set Cluster working set + * + * \return Group of enum pcmk__updated flags indicating what was updated + */ +uint32_t +pcmk__multi_update_actions(pe_action_t *first, pe_action_t *then, + const pe_node_t *node, uint32_t flags, + uint32_t filter, uint32_t type, + pe_working_set_t *data_set) +{ + uint32_t changed = pcmk__updated_none; + + crm_trace("%s -> %s", first->uuid, then->uuid); + + if(can_interleave_actions(first, then)) { + changed = multi_update_interleave_actions(first, then, node, filter, + type, data_set); + + } else if(then->rsc) { + GList *gIter = NULL; + GList *children = NULL; + + // Handle the 'primitive' ordering case + changed |= pcmk__update_ordered_actions(first, then, node, flags, + filter, type, data_set); + + // Now any children (or containers in the case of a bundle) + children = get_containers_or_children(then->rsc); + for (gIter = children; gIter != NULL; gIter = gIter->next) { + pe_resource_t *then_child = (pe_resource_t *) gIter->data; + uint32_t then_child_changed = pcmk__updated_none; + pe_action_t *then_child_action = find_first_action(then_child->actions, NULL, then->task, node); + + if (then_child_action) { + uint32_t then_child_flags = then_child->cmds->action_flags(then_child_action, + node); + + if (pcmk_is_set(then_child_flags, pe_action_runnable)) { + then_child_changed |= then_child->cmds->update_ordered_actions(first, + then_child_action, + node, + flags, + filter, + type, + data_set); + } + changed |= then_child_changed; + if (pcmk_is_set(then_child_changed, pcmk__updated_then)) { + for (GList *lpc = then_child_action->actions_after; lpc != NULL; lpc = lpc->next) { + pe_action_wrapper_t *next = (pe_action_wrapper_t *) lpc->data; + + pcmk__update_action_for_orderings(next->action, + data_set); + } + } + } + } + + if(children != then->rsc->children) { + g_list_free(children); + } + } + return changed; +} + enum action_tasks clone_child_action(pe_action_t * action) { enum action_tasks result = no_action; pe_resource_t *child = (pe_resource_t *) action->rsc->children->data; if (pcmk__strcase_any_of(action->task, "notify", "notified", NULL)) { /* Find the action we're notifying about instead */ int stop = 0; char *key = action->uuid; int lpc = strlen(key); for (; lpc > 0; lpc--) { if (key[lpc] == '_' && stop == 0) { stop = lpc; } else if (key[lpc] == '_') { char *task_mutable = NULL; lpc++; task_mutable = strdup(key + lpc); task_mutable[stop - lpc] = 0; crm_trace("Extracted action '%s' from '%s'", task_mutable, key); result = get_complex_task(child, task_mutable, TRUE); free(task_mutable); break; } } } else { result = get_complex_task(child, action->task, TRUE); } return result; } #define pe__clear_action_summary_flags(flags, action, flag) do { \ flags = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Action summary", action->rsc->id, \ flags, flag, #flag); \ } while (0) enum pe_action_flags summary_action_flags(pe_action_t *action, GList *children, const pe_node_t *node) { GList *gIter = NULL; gboolean any_runnable = FALSE; gboolean check_runnable = TRUE; enum action_tasks task = clone_child_action(action); enum pe_action_flags flags = (pe_action_optional | pe_action_runnable | pe_action_pseudo); const char *task_s = task2text(task); for (gIter = children; gIter != NULL; gIter = gIter->next) { pe_action_t *child_action = NULL; pe_resource_t *child = (pe_resource_t *) gIter->data; child_action = find_first_action(child->actions, NULL, task_s, child->children ? NULL : node); pe_rsc_trace(action->rsc, "Checking for %s in %s on %s (%s)", task_s, child->id, pe__node_name(node), child_action?child_action->uuid:"NA"); if (child_action) { enum pe_action_flags child_flags = child->cmds->action_flags(child_action, node); if (pcmk_is_set(flags, pe_action_optional) && !pcmk_is_set(child_flags, pe_action_optional)) { pe_rsc_trace(child, "%s is mandatory because of %s", action->uuid, child_action->uuid); pe__clear_action_summary_flags(flags, action, pe_action_optional); pe__clear_action_flags(action, pe_action_optional); } if (pcmk_is_set(child_flags, pe_action_runnable)) { any_runnable = TRUE; } } } if (check_runnable && any_runnable == FALSE) { pe_rsc_trace(action->rsc, "%s is not runnable because no children are", action->uuid); pe__clear_action_summary_flags(flags, action, pe_action_runnable); if (node == NULL) { pe__clear_action_flags(action, pe_action_runnable); } } return flags; } diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c index f8ba29337a..8927ec6c0a 100644 --- a/lib/pacemaker/pcmk_sched_primitive.c +++ b/lib/pacemaker/pcmk_sched_primitive.c @@ -1,1518 +1,1518 @@ /* * Copyright 2004-2023 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 "libpacemaker_private.h" static void stop_resource(pe_resource_t *rsc, pe_node_t *node, bool optional); static void start_resource(pe_resource_t *rsc, pe_node_t *node, bool optional); static void demote_resource(pe_resource_t *rsc, pe_node_t *node, bool optional); static void promote_resource(pe_resource_t *rsc, pe_node_t *node, bool optional); static void assert_role_error(pe_resource_t *rsc, pe_node_t *node, bool optional); 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 */ { RSC_ROLE_UNKNOWN, /* Unknown */ RSC_ROLE_STOPPED, /* Stopped */ RSC_ROLE_STOPPED, /* Started */ RSC_ROLE_STOPPED, /* Unpromoted */ RSC_ROLE_STOPPED, /* Promoted */ }, /* Stopped */ { RSC_ROLE_STOPPED, /* Unknown */ RSC_ROLE_STOPPED, /* Stopped */ RSC_ROLE_STARTED, /* Started */ RSC_ROLE_UNPROMOTED, /* Unpromoted */ RSC_ROLE_UNPROMOTED, /* Promoted */ }, /* Started */ { RSC_ROLE_STOPPED, /* Unknown */ RSC_ROLE_STOPPED, /* Stopped */ RSC_ROLE_STARTED, /* Started */ RSC_ROLE_UNPROMOTED, /* Unpromoted */ RSC_ROLE_PROMOTED, /* Promoted */ }, /* Unpromoted */ { RSC_ROLE_STOPPED, /* Unknown */ RSC_ROLE_STOPPED, /* Stopped */ RSC_ROLE_STOPPED, /* Started */ RSC_ROLE_UNPROMOTED, /* Unpromoted */ RSC_ROLE_PROMOTED, /* Promoted */ }, /* Promoted */ { RSC_ROLE_STOPPED, /* Unknown */ RSC_ROLE_UNPROMOTED, /* Stopped */ RSC_ROLE_UNPROMOTED, /* Started */ RSC_ROLE_UNPROMOTED, /* Unpromoted */ RSC_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)(pe_resource_t *rsc, pe_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 weight * * \param[in] rsc Resource to check * * \return List of allowed nodes sorted by node weight */ static GList * sorted_allowed_nodes(const pe_resource_t *rsc) { if (rsc->allowed_nodes != NULL) { GList *nodes = g_hash_table_get_values(rsc->allowed_nodes); if (nodes != NULL) { return pcmk__sort_nodes(nodes, pe__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 NULL, prefer this node when all else equal * * \return true if \p rsc could be assigned to a node, otherwise false */ static bool assign_best_node(pe_resource_t *rsc, const pe_node_t *prefer) { GList *nodes = NULL; pe_node_t *chosen = NULL; pe_node_t *best = NULL; bool result = false; const pe_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc); if (prefer == NULL) { prefer = most_free_node; } if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { // We've already finished assignment of resources to nodes return rsc->allocated_to != NULL; } // Sort allowed nodes by weight nodes = sorted_allowed_nodes(rsc); if (nodes != NULL) { best = (pe_node_t *) nodes->data; // First node has best score } if ((prefer != NULL) && (nodes != NULL)) { // Get the allowed node version of prefer chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id); if (chosen == NULL) { pe_rsc_trace(rsc, "Preferred node %s for %s was unknown", pe__node_name(prefer), rsc->id); /* Favor the preferred node as long as its weight 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 weight is less than INFINITY. */ } else if (chosen->weight < best->weight) { pe_rsc_trace(rsc, "Preferred node %s for %s was unsuitable", pe__node_name(chosen), rsc->id); chosen = NULL; } else if (!pcmk__node_available(chosen, true, false)) { pe_rsc_trace(rsc, "Preferred node %s for %s was unavailable", pe__node_name(chosen), rsc->id); chosen = NULL; } else { pe_rsc_trace(rsc, "Chose preferred node %s for %s (ignoring %d candidates)", pe__node_name(chosen), rsc->id, g_list_length(nodes)); } } if ((chosen == NULL) && (best != NULL)) { /* Either there is no preferred node, or the preferred node is not * suitable, but another node is allowed to run the resource. */ chosen = best; if (!pe_rsc_is_unique_clone(rsc->parent) && (chosen->weight > 0) // Zero not acceptable && pcmk__node_available(chosen, false, false)) { /* If the resource is already running on a node, prefer that node if * it is just as good as the chosen node. * * We don't do this for unique clone instances, because - * distribute_children() has already assigned instances to their + * 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. */ pe_node_t *running = pe__current_node(rsc); if (running == NULL) { // Nothing to do } else if (!pcmk__node_available(running, true, false)) { pe_rsc_trace(rsc, "Current node for %s (%s) can't run resources", rsc->id, pe__node_name(running)); } else { int nodes_with_best_score = 1; for (GList *iter = nodes->next; iter; iter = iter->next) { pe_node_t *allowed = (pe_node_t *) iter->data; if (allowed->weight != chosen->weight) { // The nodes are sorted by weight, so no more are equal break; } if (pe__same_node(allowed, running)) { // Scores are equal, so prefer the current node chosen = allowed; } nodes_with_best_score++; } if (nodes_with_best_score > 1) { do_crm_log(((chosen->weight >= INFINITY)? LOG_WARNING : LOG_INFO), "Chose %s for %s from %d nodes with score %s", pe__node_name(chosen), rsc->id, nodes_with_best_score, pcmk_readable_score(chosen->weight)); } } } pe_rsc_trace(rsc, "Chose %s for %s from %d candidates", pe__node_name(chosen), rsc->id, g_list_length(nodes)); } result = pcmk__finalize_assignment(rsc, chosen, false); g_list_free(nodes); return result; } /*! * \internal * \brief Apply a "this with" colocation to a node's allowed node scores * * \param[in,out] data Colocation to apply * \param[in,out] user_data Resource being assigned */ static void apply_this_with(gpointer data, gpointer user_data) { pcmk__colocation_t *colocation = (pcmk__colocation_t *) data; pe_resource_t *rsc = (pe_resource_t *) user_data; GHashTable *archive = NULL; pe_resource_t *other = colocation->primary; // In certain cases, we will need to revert the node scores if ((colocation->dependent_role >= RSC_ROLE_PROMOTED) || ((colocation->score < 0) && (colocation->score > -INFINITY))) { archive = pcmk__copy_node_table(rsc->allowed_nodes); } pe_rsc_trace(rsc, "%s: Assigning colocation %s primary %s first" "(score=%d role=%s)", rsc->id, colocation->id, other->id, colocation->score, role2text(colocation->dependent_role)); other->cmds->assign(other, NULL); // Apply the colocation score to this resource's allowed node scores rsc->cmds->apply_coloc_score(rsc, other, colocation, true); if ((archive != NULL) && !pcmk__any_node_available(rsc->allowed_nodes)) { pe_rsc_info(rsc, "%s: Reverting scores from colocation with %s " "because no nodes allowed", rsc->id, other->id); g_hash_table_destroy(rsc->allowed_nodes); rsc->allowed_nodes = archive; archive = NULL; } if (archive != NULL) { g_hash_table_destroy(archive); } } /*! * \internal * \brief Apply a "with this" colocation to a node's allowed node scores * * \param[in,out] data Colocation to apply * \param[in,out] user_data Resource being assigned */ static void apply_with_this(void *data, void *user_data) { pcmk__colocation_t *colocation = (pcmk__colocation_t *) data; pe_resource_t *rsc = (pe_resource_t *) user_data; pe_resource_t *other = colocation->dependent; const float factor = colocation->score / (float) INFINITY; if (!pcmk__colocation_has_influence(colocation, NULL)) { return; } pe_rsc_trace(rsc, "%s: Incorporating attenuated %s assignment scores due " "to colocation %s", rsc->id, other->id, colocation->id); pcmk__add_colocated_node_scores(other, rsc->id, &rsc->allowed_nodes, colocation->node_attribute, factor, pcmk__coloc_select_active); } /*! * \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 pe_resource_t *connection) { pe_node_t *remote_node = pe_find_node(connection->cluster->nodes, connection->id); CRM_CHECK(remote_node != NULL, return); if ((connection->allocated_to != NULL) && (connection->next_role != RSC_ROLE_STOPPED)) { crm_trace("Pacemaker Remote node %s will be online", remote_node->details->id); remote_node->details->online = TRUE; if (remote_node->details->unseen) { // Avoid unnecessary fence, since we will attempt connection remote_node->details->unclean = FALSE; } } else { crm_trace("Pacemaker Remote node %s will be shut down " "(%sassigned connection's next role is %s)", remote_node->details->id, ((connection->allocated_to == NULL)? "un" : ""), role2text(connection->next_role)); remote_node->details->shutdown = TRUE; } } /*! * \internal * \brief Assign a primitive resource to a node * * \param[in,out] rsc Resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * * \return Node that \p rsc is assigned to, if assigned entirely to one node */ pe_node_t * pcmk__primitive_assign(pe_resource_t *rsc, const pe_node_t *prefer) { CRM_ASSERT(rsc != NULL); // Never assign a child without parent being assigned first if ((rsc->parent != NULL) && !pcmk_is_set(rsc->parent->flags, pe_rsc_allocating)) { pe_rsc_debug(rsc, "%s: Assigning parent %s first", rsc->id, rsc->parent->id); rsc->parent->cmds->assign(rsc->parent, prefer); } if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return rsc->allocated_to; // Assignment has already been done } // Ensure we detect assignment loops if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { pe_rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id); return NULL; } pe__set_resource_flags(rsc, pe_rsc_allocating); pe__show_node_weights(true, rsc, "Pre-assignment", rsc->allowed_nodes, rsc->cluster); g_list_foreach(rsc->rsc_cons, apply_this_with, rsc); pe__show_node_weights(true, rsc, "Post-this-with", rsc->allowed_nodes, rsc->cluster); g_list_foreach(rsc->rsc_cons_lhs, apply_with_this, rsc); if (rsc->next_role == RSC_ROLE_STOPPED) { pe_rsc_trace(rsc, "Banning %s from all nodes because it will be stopped", rsc->id); resource_location(rsc, NULL, -INFINITY, XML_RSC_ATTR_TARGET_ROLE, rsc->cluster); } else if ((rsc->next_role > rsc->role) && !pcmk_is_set(rsc->cluster->flags, pe_flag_have_quorum) && (rsc->cluster->no_quorum_policy == no_quorum_freeze)) { crm_notice("Resource %s cannot be elevated from %s to %s due to " "no-quorum-policy=freeze", rsc->id, role2text(rsc->role), role2text(rsc->next_role)); pe__set_next_role(rsc, rsc->role, "no-quorum-policy=freeze"); } pe__show_node_weights(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); // Unmanage resource if fencing is enabled but no device is configured if (pcmk_is_set(rsc->cluster->flags, pe_flag_stonith_enabled) && !pcmk_is_set(rsc->cluster->flags, pe_flag_have_stonith_resource)) { pe__clear_resource_flags(rsc, pe_rsc_managed); } if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { // Unmanaged resources stay on their current node const char *reason = NULL; pe_node_t *assign_to = NULL; pe__set_next_role(rsc, rsc->role, "unmanaged"); assign_to = pe__current_node(rsc); if (assign_to == NULL) { reason = "inactive"; } else if (rsc->role == RSC_ROLE_PROMOTED) { reason = "promoted"; } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { reason = "failed"; } else { reason = "active"; } pe_rsc_info(rsc, "Unmanaged resource %s assigned to %s: %s", rsc->id, (assign_to? assign_to->details->uname : "no node"), reason); pcmk__finalize_assignment(rsc, assign_to, true); } else if (pcmk_is_set(rsc->cluster->flags, pe_flag_stop_everything)) { pe_rsc_debug(rsc, "Forcing %s to stop: stop-all-resources", rsc->id); pcmk__finalize_assignment(rsc, NULL, true); } else if (pcmk_is_set(rsc->flags, pe_rsc_provisional) && assign_best_node(rsc, prefer)) { // Assignment successful } else if (rsc->allocated_to == NULL) { if (!pcmk_is_set(rsc->flags, pe_rsc_orphan)) { pe_rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id); } else if (rsc->running_on != NULL) { pe_rsc_info(rsc, "Stopping orphan resource %s", rsc->id); } } else { pe_rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, pe__node_name(rsc->allocated_to)); } pe__clear_resource_flags(rsc, pe_rsc_allocating); if (rsc->is_remote_node) { remote_connection_assigned(rsc); } return rsc->allocated_to; } /*! * \internal * \brief Schedule actions to bring resource down and back to current role * * \param[in,out] rsc Resource to restart * \param[in,out] current Node that resource should be brought down on * \param[in] need_stop Whether the resource must be stopped * \param[in] need_promote Whether the resource must be promoted * * \return Role that resource would have after scheduled actions are taken */ static void schedule_restart_actions(pe_resource_t *rsc, pe_node_t *current, bool need_stop, bool need_promote) { enum rsc_role_e role = rsc->role; enum rsc_role_e next_role; rsc_transition_fn fn = NULL; pe__set_resource_flags(rsc, pe_rsc_restarting); // Bring resource down to a stop on its current node while (role != RSC_ROLE_STOPPED) { next_role = rsc_state_matrix[role][RSC_ROLE_STOPPED]; pe_rsc_trace(rsc, "Creating %s action to take %s down from %s to %s", (need_stop? "required" : "optional"), rsc->id, role2text(role), role2text(next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, current, !need_stop); role = next_role; } // Bring resource up to its next role on its next node while ((rsc->role <= rsc->next_role) && (role != rsc->role) && !pcmk_is_set(rsc->flags, pe_rsc_block)) { bool required = need_stop; next_role = rsc_state_matrix[role][rsc->role]; if ((next_role == RSC_ROLE_PROMOTED) && need_promote) { required = true; } pe_rsc_trace(rsc, "Creating %s action to take %s up from %s to %s", (required? "required" : "optional"), rsc->id, role2text(role), role2text(next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, rsc->allocated_to, !required); role = next_role; } pe__clear_resource_flags(rsc, pe_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(pe_resource_t *rsc) { if (rsc->next_role != RSC_ROLE_UNKNOWN) { return "explicit"; } if (rsc->allocated_to == NULL) { pe__set_next_role(rsc, RSC_ROLE_STOPPED, "assignment"); } else { pe__set_next_role(rsc, RSC_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(pe_resource_t *rsc) { pe_action_t *start = NULL; pe_rsc_trace(rsc, "Creating action for %s to represent already pending start", rsc->id); start = start_action(rsc, rsc->allocated_to, TRUE); pe__set_action_flags(start, pe_action_print_always); } /*! * \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(pe_resource_t *rsc) { enum rsc_role_e role = rsc->role; while (role != rsc->next_role) { enum rsc_role_e next_role = rsc_state_matrix[role][rsc->next_role]; rsc_transition_fn fn = NULL; pe_rsc_trace(rsc, "Creating action to take %s from %s to %s (ending at %s)", rsc->id, role2text(role), role2text(next_role), role2text(rsc->next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, rsc->allocated_to, false); role = next_role; } } /*! * \internal * \brief Create all actions needed for a given primitive resource * * \param[in,out] rsc Primitive resource to create actions for */ void pcmk__primitive_create_actions(pe_resource_t *rsc) { bool need_stop = false; bool need_promote = false; bool is_moving = false; bool allow_migrate = false; bool multiply_active = false; pe_node_t *current = NULL; unsigned int num_all_active = 0; unsigned int num_clean_active = 0; const char *next_role_source = NULL; CRM_ASSERT(rsc != NULL); next_role_source = set_default_next_role(rsc); pe_rsc_trace(rsc, "Creating all actions for %s transition from %s to %s " "(%s) on %s", rsc->id, role2text(rsc->role), role2text(rsc->next_role), next_role_source, pe__node_name(rsc->allocated_to)); current = pe__find_active_on(rsc, &num_all_active, &num_clean_active); g_list_foreach(rsc->dangling_migrations, pcmk__abort_dangling_migration, rsc); if ((current != NULL) && (rsc->allocated_to != NULL) && (current->details != rsc->allocated_to->details) && (rsc->next_role >= RSC_ROLE_STARTED)) { pe_rsc_trace(rsc, "Moving %s from %s to %s", rsc->id, pe__node_name(current), pe__node_name(rsc->allocated_to)); is_moving = true; allow_migrate = pcmk__rsc_can_migrate(rsc, current); // This is needed even if migrating (though I'm not sure why ...) need_stop = true; } // Check whether resource is partially migrated and/or multiply active if ((rsc->partial_migration_source != NULL) && (rsc->partial_migration_target != NULL) && allow_migrate && (num_all_active == 2) && pe__same_node(current, rsc->partial_migration_source) && pe__same_node(rsc->allocated_to, rsc->partial_migration_target)) { /* A partial migration is in progress, and the migration target remains * the same as when the migration began. */ pe_rsc_trace(rsc, "Partial migration of %s from %s to %s will continue", rsc->id, pe__node_name(rsc->partial_migration_source), pe__node_name(rsc->partial_migration_target)); } else if ((rsc->partial_migration_source != NULL) || (rsc->partial_migration_target != NULL)) { // A partial migration is in progress but can't be continued if (num_all_active > 2) { // The resource is migrating *and* multiply active! crm_notice("Forcing recovery of %s because it is migrating " "from %s to %s and possibly active elsewhere", rsc->id, pe__node_name(rsc->partial_migration_source), pe__node_name(rsc->partial_migration_target)); } else { // The migration source or target isn't available crm_notice("Forcing recovery of %s because it can no longer " "migrate from %s to %s", rsc->id, pe__node_name(rsc->partial_migration_source), pe__node_name(rsc->partial_migration_target)); } need_stop = true; rsc->partial_migration_source = rsc->partial_migration_target = NULL; allow_migrate = false; } else if (pcmk_is_set(rsc->flags, pe_rsc_needs_fencing)) { multiply_active = (num_all_active > 1); } else { /* If a resource has "requires" set to nothing or quorum, don't consider * it active on unclean nodes (similar to how all resources behave when * stonith-enabled is false). We can start such resources elsewhere * before fencing completes, and if we considered the resource active on * the failed node, we would attempt recovery for being active on * multiple nodes. */ multiply_active = (num_clean_active > 1); } if (multiply_active) { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); // Resource was (possibly) incorrectly multiply active pe_proc_err("%s resource %s might be active on %u nodes (%s)", pcmk__s(class, "Untyped"), rsc->id, num_all_active, recovery2text(rsc->recovery_type)); crm_notice("See https://wiki.clusterlabs.org/wiki/FAQ" "#Resource_is_Too_Active for more information"); switch (rsc->recovery_type) { case recovery_stop_start: need_stop = true; break; case recovery_stop_unexpected: need_stop = true; // stop_resource() will skip expected node pe__set_resource_flags(rsc, pe_rsc_stop_unexpected); break; default: break; } } else { pe__clear_resource_flags(rsc, pe_rsc_stop_unexpected); } if (pcmk_is_set(rsc->flags, pe_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, pe_rsc_failed)) { if (pcmk_is_set(rsc->flags, pe_rsc_stop)) { need_stop = true; pe_rsc_trace(rsc, "Recovering %s", rsc->id); } else { pe_rsc_trace(rsc, "Recovering %s by demotion", rsc->id); if (rsc->next_role == RSC_ROLE_PROMOTED) { need_promote = true; } } } else if (pcmk_is_set(rsc->flags, pe_rsc_block)) { pe_rsc_trace(rsc, "Blocking further actions on %s", rsc->id); need_stop = true; } else if ((rsc->role > RSC_ROLE_STARTED) && (current != NULL) && (rsc->allocated_to != NULL)) { pe_action_t *start = NULL; pe_rsc_trace(rsc, "Creating start action for promoted resource %s", rsc->id); start = start_action(rsc, rsc->allocated_to, TRUE); if (!pcmk_is_set(start->flags, pe_action_optional)) { // Recovery of a promoted resource pe_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 pe_resource_t *rsc) { GHashTableIter iter; pe_node_t *node = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (node->details->remote_rsc != NULL) { node->weight = -INFINITY; } } } /*! * \internal * \brief Return allowed nodes as (possibly sorted) list * * Convert a resource's hash table of allowed nodes to a list. If printing to * stdout, sort the list, to keep action ID numbers consistent for regression * test output (while avoiding the performance hit on a live cluster). * * \param[in] rsc Resource to check for allowed nodes * * \return List of resource's allowed nodes * \note Callers should take care not to rely on the list being sorted. */ static GList * allowed_nodes_as_list(const pe_resource_t *rsc) { GList *allowed_nodes = NULL; if (rsc->allowed_nodes) { allowed_nodes = g_hash_table_get_values(rsc->allowed_nodes); } if (!pcmk__is_daemon) { allowed_nodes = g_list_sort(allowed_nodes, pe__cmp_node_name); } return allowed_nodes; } /*! * \internal * \brief Create implicit constraints needed for a primitive resource * * \param[in,out] rsc Primitive resource to create implicit constraints for */ void pcmk__primitive_internal_constraints(pe_resource_t *rsc) { GList *allowed_nodes = NULL; bool check_unfencing = false; bool check_utilization = false; CRM_ASSERT(rsc != NULL); if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe_rsc_trace(rsc, "Skipping implicit constraints for unmanaged resource %s", rsc->id); return; } // Whether resource requires unfencing check_unfencing = !pcmk_is_set(rsc->flags, pe_rsc_fence_device) && pcmk_is_set(rsc->cluster->flags, pe_flag_enable_unfencing) && pcmk_is_set(rsc->flags, pe_rsc_needs_unfencing); // Whether a non-default placement strategy is used check_utilization = (g_hash_table_size(rsc->utilization) > 0) && !pcmk__str_eq(rsc->cluster->placement_strategy, "default", pcmk__str_casei); // Order stops before starts (i.e. restart) pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, pe_order_optional|pe_order_implies_then|pe_order_restart, rsc->cluster); // Promotable ordering: demote before stop, start before promote if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pe_rsc_promotable) || (rsc->role > RSC_ROLE_UNPROMOTED)) { pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, RSC_DEMOTE, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, pe_order_promoted_implies_first, rsc->cluster); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_PROMOTE, 0), NULL, pe_order_runnable_left, rsc->cluster); } // Don't clear resource history if probing on same node pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, CRM_OP_LRM_DELETE, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_STATUS, 0), NULL, pe_order_same_node|pe_order_then_cancels_first, rsc->cluster); // Certain checks need allowed nodes if (check_unfencing || check_utilization || (rsc->container != NULL)) { allowed_nodes = allowed_nodes_as_list(rsc); } if (check_unfencing) { g_list_foreach(allowed_nodes, pcmk__order_restart_vs_unfence, rsc); } if (check_utilization) { pcmk__create_utilization_constraints(rsc, allowed_nodes); } if (rsc->container != NULL) { pe_resource_t *remote_rsc = NULL; if (rsc->is_remote_node) { // rsc is the implicit remote connection for a guest or bundle node /* Guest resources are not allowed to run on Pacemaker Remote nodes, * to avoid nesting remotes. However, bundles are allowed. */ if (!pcmk_is_set(rsc->flags, pe_rsc_allow_remote_remotes)) { rsc_avoids_remote_nodes(rsc->container); } /* If someone cleans up a guest or bundle node's container, we will * likely schedule a (re-)probe of the container and recovery of the * connection. Order the connection stop after the container probe, * so that if we detect the container running, we will trigger a new * transition and avoid the unnecessary recovery. */ pcmk__order_resource_actions(rsc->container, RSC_STATUS, rsc, RSC_STOP, pe_order_optional); /* A user can specify that a resource must start on a Pacemaker Remote * node by explicitly configuring it with the container=NODENAME * meta-attribute. This is of questionable merit, since location * constraints can accomplish the same thing. But we support it, so here * we check whether a resource (that is not itself a remote connection) * has container set to a remote node or guest node resource. */ } else if (rsc->container->is_remote_node) { remote_rsc = rsc->container; } else { remote_rsc = pe__resource_contains_guest_node(rsc->cluster, rsc->container); } if (remote_rsc != NULL) { /* Force the resource on the Pacemaker Remote node instead of * colocating the resource with the container resource. */ for (GList *item = allowed_nodes; item; item = item->next) { pe_node_t *node = item->data; if (node->details->remote_rsc != remote_rsc) { node->weight = -INFINITY; } } } else { /* This resource is either a filler for a container that does NOT * represent a Pacemaker Remote node, or a Pacemaker Remote * connection resource for a guest node or bundle. */ int score; crm_trace("Order and colocate %s relative to its container %s", rsc->id, rsc->container->id); pcmk__new_ordering(rsc->container, pcmk__op_key(rsc->container->id, RSC_START, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, pe_order_implies_then|pe_order_runnable_left, rsc->cluster); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, rsc->container, pcmk__op_key(rsc->container->id, RSC_STOP, 0), NULL, pe_order_implies_first, rsc->cluster); if (pcmk_is_set(rsc->flags, pe_rsc_allow_remote_remotes)) { score = 10000; /* Highly preferred but not essential */ } else { score = INFINITY; /* Force them to run on the same host */ } pcmk__new_colocation("resource-with-container", NULL, score, rsc, rsc->container, NULL, NULL, true, rsc->cluster); } } if (rsc->is_remote_node || pcmk_is_set(rsc->flags, pe_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 weights or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node weights (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__primitive_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { enum pcmk__coloc_affects filter_results; CRM_CHECK((colocation != NULL) && (dependent != NULL) && (primary != NULL), return); if (for_dependent) { // Always process on behalf of primary resource primary->cmds->apply_coloc_score(dependent, primary, colocation, false); return; } filter_results = pcmk__colocation_affects(dependent, primary, colocation, false); pe_rsc_trace(dependent, "%s %s with %s (%s, score=%d, filter=%d)", ((colocation->score > 0)? "Colocating" : "Anti-colocating"), dependent->id, primary->id, colocation->id, colocation->score, filter_results); switch (filter_results) { case pcmk__coloc_affects_role: pcmk__apply_coloc_to_priority(dependent, primary, colocation); break; case pcmk__coloc_affects_location: pcmk__apply_coloc_to_weights(dependent, primary, colocation); break; default: // pcmk__coloc_affects_nothing return; } } /*! * \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 */ enum pe_action_flags pcmk__primitive_action_flags(pe_action_t *action, const pe_node_t *node) { CRM_ASSERT(action != NULL); return 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 true if \p rsc is multiply active with multiple-active set to * 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 pe_resource_t *rsc, const pe_node_t *node) { return pcmk_all_flags_set(rsc->flags, pe_rsc_stop_unexpected|pe_rsc_restarting) && (rsc->next_role > RSC_ROLE_STOPPED) && pe__same_node(rsc->allocated_to, node); } /*! * \internal * \brief Schedule actions needed to stop a resource wherever it is active * * \param[in,out] rsc Resource being stopped * \param[in] node Node where resource is being stopped (ignored) * \param[in] optional Whether actions should be optional */ static void stop_resource(pe_resource_t *rsc, pe_node_t *node, bool optional) { for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pe_node_t *current = (pe_node_t *) iter->data; pe_action_t *stop = NULL; if (is_expected_node(rsc, current)) { /* We are scheduling restart actions for a multiply active resource * with multiple-active=stop_unexpected, and this is where it should * not be stopped. */ pe_rsc_trace(rsc, "Skipping stop of multiply active resource %s " "on expected node %s", rsc->id, pe__node_name(current)); continue; } if (rsc->partial_migration_target != NULL) { // Continue migration if node originally was and remains target if (pe__same_node(current, rsc->partial_migration_target) && pe__same_node(current, rsc->allocated_to)) { pe_rsc_trace(rsc, "Skipping stop of %s on %s " "because partial migration there will continue", rsc->id, pe__node_name(current)); continue; } else { pe_rsc_trace(rsc, "Forcing stop of %s on %s " "because migration target changed", rsc->id, pe__node_name(current)); optional = false; } } pe_rsc_trace(rsc, "Scheduling stop of %s on %s", rsc->id, pe__node_name(current)); stop = stop_action(rsc, current, optional); if (rsc->allocated_to == NULL) { pe_action_set_reason(stop, "node availability", true); } else if (pcmk_all_flags_set(rsc->flags, pe_rsc_restarting |pe_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, pe_rsc_managed)) { pe__clear_action_flags(stop, pe_action_runnable); } if (pcmk_is_set(rsc->cluster->flags, pe_flag_remove_after_stop)) { pcmk__schedule_cleanup(rsc, current, optional); } if (pcmk_is_set(rsc->flags, pe_rsc_needs_unfencing)) { pe_action_t *unfence = pe_fence_op(current, "on", true, NULL, false, rsc->cluster); order_actions(stop, unfence, pe_order_implies_first); if (!pcmk__node_unfenced(current)) { pe_proc_err("Stopping %s until %s can be unfenced", rsc->id, pe__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(pe_resource_t *rsc, pe_node_t *node, bool optional) { pe_action_t *start = NULL; CRM_ASSERT(node != NULL); pe_rsc_trace(rsc, "Scheduling %s start of %s on %s (score %d)", (optional? "optional" : "required"), rsc->id, pe__node_name(node), node->weight); start = start_action(rsc, node, TRUE); pcmk__order_vs_unfence(rsc, node, start, pe_order_implies_then); if (pcmk_is_set(start->flags, pe_action_runnable) && !optional) { pe__clear_action_flags(start, pe_action_optional); } if (is_expected_node(rsc, node)) { /* This could be a problem if the start becomes necessary for other * reasons later. */ pe_rsc_trace(rsc, "Start of multiply active resouce %s " "on expected node %s will be a pseudo-action", rsc->id, pe__node_name(node)); pe__set_action_flags(start, pe_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(pe_resource_t *rsc, pe_node_t *node, bool optional) { GList *iter = NULL; GList *action_list = NULL; bool runnable = true; CRM_ASSERT(node != NULL); // Any start must be runnable for promotion to be runnable action_list = pe__resource_actions(rsc, node, RSC_START, true); for (iter = action_list; iter != NULL; iter = iter->next) { pe_action_t *start = (pe_action_t *) iter->data; if (!pcmk_is_set(start->flags, pe_action_runnable)) { runnable = false; } } g_list_free(action_list); if (runnable) { pe_action_t *promote = promote_action(rsc, node, optional); pe_rsc_trace(rsc, "Scheduling %s promotion of %s on %s", (optional? "optional" : "required"), rsc->id, pe__node_name(node)); if (is_expected_node(rsc, node)) { /* This could be a problem if the promote becomes necessary for * other reasons later. */ pe_rsc_trace(rsc, "Promotion of multiply active resouce %s " "on expected node %s will be a pseudo-action", rsc->id, pe__node_name(node)); pe__set_action_flags(promote, pe_action_pseudo); } } else { pe_rsc_trace(rsc, "Not promoting %s on %s: start unrunnable", rsc->id, pe__node_name(node)); action_list = pe__resource_actions(rsc, node, RSC_PROMOTE, true); for (iter = action_list; iter != NULL; iter = iter->next) { pe_action_t *promote = (pe_action_t *) iter->data; pe__clear_action_flags(promote, pe_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(pe_resource_t *rsc, pe_node_t *node, bool optional) { /* Since this will only be called for a primitive (possibly as an instance * of a collective resource), the resource is multiply active if it is * running on more than one node, so we want to demote on all of them as * part of recovery, regardless of which one is the desired node. */ for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pe_node_t *current = (pe_node_t *) iter->data; if (is_expected_node(rsc, current)) { pe_rsc_trace(rsc, "Skipping demote of multiply active resource %s " "on expected node %s", rsc->id, pe__node_name(current)); } else { pe_rsc_trace(rsc, "Scheduling %s demotion of %s on %s", (optional? "optional" : "required"), rsc->id, pe__node_name(current)); demote_action(rsc, current, optional); } } } static void assert_role_error(pe_resource_t *rsc, pe_node_t *node, bool optional) { CRM_ASSERT(false); } /*! * \internal * \brief Schedule cleanup of a resource * * \param[in,out] rsc Resource to clean up * \param[in] node Node to clean up on * \param[in] optional Whether clean-up should be optional */ void pcmk__schedule_cleanup(pe_resource_t *rsc, const pe_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? pe_order_implies_then : pe_order_optional; CRM_CHECK((rsc != NULL) && (node != NULL), return); if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { pe_rsc_trace(rsc, "Skipping clean-up of %s on %s: resource failed", rsc->id, pe__node_name(node)); return; } if (node->details->unclean || !node->details->online) { pe_rsc_trace(rsc, "Skipping clean-up of %s on %s: node unavailable", rsc->id, pe__node_name(node)); return; } crm_notice("Scheduling clean-up of %s on %s", rsc->id, pe__node_name(node)); delete_action(rsc, node, optional); // stop -> clean-up -> start pcmk__order_resource_actions(rsc, RSC_STOP, rsc, RSC_DELETE, flag); pcmk__order_resource_actions(rsc, RSC_DELETE, rsc, RSC_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 pe_resource_t *rsc, xmlNode *xml) { char *name = NULL; char *value = NULL; const pe_resource_t *parent = NULL; CRM_ASSERT((rsc != NULL) && (xml != NULL)); /* Clone instance numbers get set internally as meta-attributes, and are * needed in the transition graph (for example, to tell unique clone * instances apart). */ value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INCARNATION); if (value != NULL) { name = crm_meta_name(XML_RSC_ATTR_INCARNATION); crm_xml_add(xml, name, value); free(name); } // Not sure if this one is really needed ... value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_REMOTE_NODE); if (value != NULL) { name = crm_meta_name(XML_RSC_ATTR_REMOTE_NODE); crm_xml_add(xml, name, value); free(name); } /* The container meta-attribute can be set on the primitive itself or one of * its parents (for example, a group inside a container resource), so check * them all, and keep the highest one found. */ for (parent = rsc; parent != NULL; parent = parent->parent) { if (parent->container != NULL) { crm_xml_add(xml, CRM_META "_" XML_RSC_ATTR_CONTAINER, parent->container->id); } } /* Bundle replica children will get their external-ip set internally as a * meta-attribute. The graph action needs it, but under a different naming * convention than other meta-attributes. */ value = g_hash_table_lookup(rsc->meta, "external-ip"); if (value != NULL) { crm_xml_add(xml, "pcmk_external_ip", value); } } // Primitive implementation of resource_alloc_functions_t:add_utilization() void pcmk__primitive_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return; } pe_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(pe_node_t *node) { const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN); time_t result = 0; if (shutdown != NULL) { long long result_ll; if (pcmk__scan_ll(shutdown, &result_ll, 0LL) == pcmk_rc_ok) { result = (time_t) result_ll; } } return (result == 0)? get_effective_time(node->details->data_set) : result; } /*! * \internal * \brief Ban a resource from a node if it's not locked to the node * * \param[in] data Node to check * \param[in,out] user_data Resource to check */ static void ban_if_not_locked(gpointer data, gpointer user_data) { const pe_node_t *node = (const pe_node_t *) data; pe_resource_t *rsc = (pe_resource_t *) user_data; if (strcmp(node->details->uname, rsc->lock_node->details->uname) != 0) { resource_location(rsc, node, -CRM_SCORE_INFINITY, XML_CONFIG_ATTR_SHUTDOWN_LOCK, rsc->cluster); } } // Primitive implementation of resource_alloc_functions_t:shutdown_lock() void pcmk__primitive_shutdown_lock(pe_resource_t *rsc) { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); // Fence devices and remote connections can't be locked if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches) || pe__resource_is_remote_conn(rsc, rsc->cluster)) { return; } if (rsc->lock_node != NULL) { // The lock was obtained from resource history if (rsc->running_on != NULL) { /* The resource was started elsewhere even though it is now * considered locked. This shouldn't be possible, but as a * failsafe, we don't want to disturb the resource now. */ pe_rsc_info(rsc, "Cancelling shutdown lock because %s is already active", rsc->id); pe__clear_resource_history(rsc, rsc->lock_node, rsc->cluster); rsc->lock_node = NULL; rsc->lock_time = 0; } // Only a resource active on exactly one node can be locked } else if (pcmk__list_of_1(rsc->running_on)) { pe_node_t *node = rsc->running_on->data; if (node->details->shutdown) { if (node->details->unclean) { pe_rsc_debug(rsc, "Not locking %s to unclean %s for shutdown", rsc->id, pe__node_name(node)); } else { rsc->lock_node = node; rsc->lock_time = shutdown_time(node); } } } if (rsc->lock_node == NULL) { // No lock needed return; } if (rsc->cluster->shutdown_lock > 0) { time_t lock_expiration = rsc->lock_time + rsc->cluster->shutdown_lock; pe_rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)", rsc->id, pe__node_name(rsc->lock_node), (long long) lock_expiration); pe__update_recheck_time(++lock_expiration, rsc->cluster); } else { pe_rsc_info(rsc, "Locking %s to %s due to shutdown", rsc->id, pe__node_name(rsc->lock_node)); } // If resource is locked to one node, ban it from all other nodes g_list_foreach(rsc->cluster->nodes, ban_if_not_locked, rsc); }