diff --git a/include/crm/pengine/pe_types.h b/include/crm/pengine/pe_types.h index e85b4618f0..ee103772c1 100644 --- a/include/crm/pengine/pe_types.h +++ b/include/crm/pengine/pe_types.h @@ -1,487 +1,486 @@ /* * 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__CRM_PENGINE_PE_TYPES__H # define PCMK__CRM_PENGINE_PE_TYPES__H # include // bool # include // time_t # include // xmlNode # include // gboolean, guint, GList, GHashTable # include # include # include #ifdef __cplusplus extern "C" { #endif /*! * \file * \brief Data types for cluster status * \ingroup pengine */ typedef struct pe_node_s pe_node_t; typedef struct pe_action_s pe_action_t; typedef struct pe_resource_s pe_resource_t; typedef struct pe_working_set_s pe_working_set_t; typedef struct resource_object_functions_s { gboolean (*unpack) (pe_resource_t*, pe_working_set_t*); pe_resource_t *(*find_rsc) (pe_resource_t *parent, const char *search, const pe_node_t *node, int flags); /* parameter result must be free'd */ char *(*parameter) (pe_resource_t*, pe_node_t*, gboolean, const char*, pe_working_set_t*); //! \deprecated will be removed in a future release void (*print) (pe_resource_t*, const char*, long, void*); gboolean (*active) (pe_resource_t*, gboolean); enum rsc_role_e (*state) (const pe_resource_t*, gboolean); pe_node_t *(*location) (const pe_resource_t*, GList**, int); void (*free) (pe_resource_t*); void (*count) (pe_resource_t*); gboolean (*is_filtered) (const pe_resource_t*, GList *, gboolean); /*! * \brief Find a node (and optionally count all) where resource is active * * \param[in] rsc Resource to check * \param[out] count_all If not NULL, set this to count of active nodes * \param[out] count_clean If not NULL, set this to count of clean nodes * * \return A node where the resource is active, preferring the source node * if the resource is involved in a partial migration or a clean, * online node if the resource's "requires" is "quorum" or * "nothing", or NULL if the resource is inactive. */ pe_node_t *(*active_node)(const pe_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean); /*! * \brief Get maximum resource instances per node * * \param[in] rsc Resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int (*max_per_node)(const pe_resource_t *rsc); } resource_object_functions_t; typedef struct resource_alloc_functions_s resource_alloc_functions_t; -# define pe_flag_show_scores 0x02000000ULL # define pe_flag_show_utilization 0x04000000ULL /*! * When scheduling, only unpack the CIB (including constraints), calculate * as much cluster status as possible, and apply node health. */ # define pe_flag_check_config 0x08000000ULL struct pe_working_set_s { xmlNode *input; crm_time_t *now; /* options extracted from the input */ char *dc_uuid; pe_node_t *dc_node; const char *stonith_action; const char *placement_strategy; unsigned long long flags; int stonith_timeout; enum pe_quorum_policy no_quorum_policy; GHashTable *config_hash; GHashTable *tickets; // Actions for which there can be only one (e.g. fence nodeX) GHashTable *singletons; GList *nodes; GList *resources; GList *placement_constraints; GList *ordering_constraints; GList *colocation_constraints; GList *ticket_constraints; GList *actions; xmlNode *failed; xmlNode *op_defaults; xmlNode *rsc_defaults; /* stats */ int num_synapse; int max_valid_nodes; //! Deprecated (will be removed in a future release) int order_id; int action_id; /* final output */ xmlNode *graph; GHashTable *template_rsc_sets; const char *localhost; GHashTable *tags; int blocked_resources; int disabled_resources; GList *param_check; // History entries that need to be checked GList *stop_needed; // Containers that need stop actions time_t recheck_by; // Hint to controller to re-run scheduler by this time int ninstances; // Total number of resource instances guint shutdown_lock;// How long (seconds) to lock resources to shutdown node int priority_fencing_delay; // Priority fencing delay void *priv; guint node_pending_timeout; // Node pending timeout }; struct pe_node_shared_s { const char *id; const char *uname; enum node_type type; /* @TODO convert these flags into a bitfield */ gboolean online; gboolean standby; gboolean standby_onfail; gboolean pending; gboolean unclean; gboolean unseen; gboolean shutdown; gboolean expected_up; gboolean is_dc; gboolean maintenance; gboolean rsc_discovery_enabled; gboolean remote_requires_reset; gboolean remote_was_fenced; gboolean remote_maintenance; /* what the remote-rsc is thinking */ gboolean unpacked; int num_resources; pe_resource_t *remote_rsc; GList *running_rsc; /* pe_resource_t* */ GList *allocated_rsc; /* pe_resource_t* */ GHashTable *attrs; /* char* => char* */ GHashTable *utilization; GHashTable *digest_cache; //!< cache of calculated resource digests int priority; // calculated based on the priority of resources running on the node pe_working_set_t *data_set; //!< Cluster that this node is part of }; struct pe_node_s { int weight; gboolean fixed; //!< \deprecated Will be removed in a future release int count; struct pe_node_shared_s *details; int rsc_discover_mode; }; # define pe_rsc_orphan 0x00000001ULL # define pe_rsc_managed 0x00000002ULL # define pe_rsc_block 0x00000004ULL # define pe_rsc_orphan_container_filler 0x00000008ULL # define pe_rsc_notify 0x00000010ULL # define pe_rsc_unique 0x00000020ULL # define pe_rsc_fence_device 0x00000040ULL # define pe_rsc_promotable 0x00000080ULL # define pe_rsc_provisional 0x00000100ULL # define pe_rsc_allocating 0x00000200ULL # define pe_rsc_merging 0x00000400ULL # define pe_rsc_restarting 0x00000800ULL # define pe_rsc_stop 0x00001000ULL # define pe_rsc_reload 0x00002000ULL # define pe_rsc_allow_remote_remotes 0x00004000ULL # define pe_rsc_critical 0x00008000ULL # define pe_rsc_failed 0x00010000ULL # define pe_rsc_detect_loop 0x00020000ULL # define pe_rsc_runnable 0x00040000ULL # define pe_rsc_start_pending 0x00080000ULL //!< \deprecated Do not use # define pe_rsc_starting 0x00100000ULL //!< \deprecated Do not use # define pe_rsc_stopping 0x00200000ULL # define pe_rsc_stop_unexpected 0x00400000ULL # define pe_rsc_allow_migrate 0x00800000ULL # define pe_rsc_failure_ignored 0x01000000ULL # define pe_rsc_replica_container 0x02000000ULL # define pe_rsc_maintenance 0x04000000ULL # define pe_rsc_is_container 0x08000000ULL # define pe_rsc_needs_quorum 0x10000000ULL # define pe_rsc_needs_fencing 0x20000000ULL # define pe_rsc_needs_unfencing 0x40000000ULL /* *INDENT-OFF* */ enum pe_action_flags { pe_action_pseudo = 0x00001, pe_action_runnable = 0x00002, pe_action_optional = 0x00004, pe_action_print_always = 0x00008, pe_action_have_node_attrs = 0x00010, pe_action_implied_by_stonith = 0x00040, pe_action_migrate_runnable = 0x00080, pe_action_dumped = 0x00100, pe_action_processed = 0x00200, #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) pe_action_clear = 0x00400, //! \deprecated Unused #endif pe_action_dangle = 0x00800, /* This action requires one or more of its dependencies to be runnable. * We use this to clear the runnable flag before checking dependencies. */ pe_action_requires_any = 0x01000, pe_action_reschedule = 0x02000, pe_action_tracking = 0x04000, pe_action_dedup = 0x08000, //! Internal state tracking when creating graph pe_action_dc = 0x10000, //! Action may run on DC instead of target }; /* *INDENT-ON* */ struct pe_resource_s { char *id; char *clone_name; xmlNode *xml; xmlNode *orig_xml; xmlNode *ops_xml; pe_working_set_t *cluster; pe_resource_t *parent; enum pe_obj_types variant; void *variant_opaque; resource_object_functions_t *fns; resource_alloc_functions_t *cmds; enum rsc_recovery_type recovery_type; enum pe_restart restart_type; //!< \deprecated will be removed in future release int priority; int stickiness; int sort_index; int failure_timeout; int migration_threshold; guint remote_reconnect_ms; char *pending_task; unsigned long long flags; // @TODO merge these into flags gboolean is_remote_node; gboolean exclusive_discover; /* Pay special attention to whether you want to use rsc_cons_lhs and * rsc_cons directly, which include only colocations explicitly involving * this resource, or call libpacemaker's pcmk__with_this_colocations() and * pcmk__this_with_colocations() functions, which may return relevant * colocations involving the resource's ancestors as well. */ //!@{ //! This field should be treated as internal to Pacemaker GList *rsc_cons_lhs; // List of pcmk__colocation_t* GList *rsc_cons; // List of pcmk__colocation_t* GList *rsc_location; // List of pe__location_t* GList *actions; // List of pe_action_t* GList *rsc_tickets; // List of rsc_ticket* //!@} pe_node_t *allocated_to; pe_node_t *partial_migration_target; pe_node_t *partial_migration_source; GList *running_on; /* pe_node_t* */ GHashTable *known_on; /* pe_node_t* */ GHashTable *allowed_nodes; /* pe_node_t* */ enum rsc_role_e role; enum rsc_role_e next_role; GHashTable *meta; GHashTable *parameters; //! \deprecated Use pe_rsc_params() instead GHashTable *utilization; GList *children; /* pe_resource_t* */ GList *dangling_migrations; /* pe_node_t* */ pe_resource_t *container; GList *fillers; // @COMPAT These should be made const at next API compatibility break pe_node_t *pending_node; // Node on which pending_task is happening pe_node_t *lock_node; // Resource is shutdown-locked to this node time_t lock_time; // When shutdown lock started /* Resource parameters may have node-attribute-based rules, which means the * values can vary by node. This table is a cache of parameter name/value * tables for each node (as needed). Use pe_rsc_params() to get the table * for a given node. */ GHashTable *parameter_cache; // Key = node name, value = parameters table }; struct pe_action_s { int id; int priority; pe_resource_t *rsc; pe_node_t *node; xmlNode *op_entry; char *task; char *uuid; char *cancel_task; char *reason; enum pe_action_flags flags; enum rsc_start_requirement needs; enum action_fail_response on_fail; enum rsc_role_e fail_role; GHashTable *meta; GHashTable *extra; /* * These two varables are associated with the constraint logic * that involves first having one or more actions runnable before * then allowing this action to execute. * * These varables are used with features such as 'clone-min' which * requires at minimum X number of cloned instances to be running * before an order dependency can run. Another option that uses * this is 'require-all=false' in ordering constrants. This option * says "only require one instance of a resource to start before * allowing dependencies to start" -- basically, require-all=false is * the same as clone-min=1. */ /* current number of known runnable actions in the before list. */ int runnable_before; /* the number of "before" runnable actions required for this action * to be considered runnable */ int required_runnable_before; GList *actions_before; /* pe_action_wrapper_t* */ GList *actions_after; /* pe_action_wrapper_t* */ /* Some of the above fields could be moved to the details, * except for API backward compatibility. */ void *action_details; // varies by type of action }; typedef struct pe_ticket_s { char *id; gboolean granted; time_t last_granted; gboolean standby; GHashTable *state; } pe_ticket_t; typedef struct pe_tag_s { char *id; GList *refs; } pe_tag_t; //! Internal tracking for transition graph creation enum pe_link_state { pe_link_not_dumped, //! Internal tracking for transition graph creation pe_link_dumped, //! Internal tracking for transition graph creation pe_link_dup, //! \deprecated No longer used by Pacemaker }; enum pe_discover_e { pe_discover_always = 0, pe_discover_never, pe_discover_exclusive, }; /* *INDENT-OFF* */ enum pe_ordering { pe_order_none = 0x0, /* deleted */ pe_order_optional = 0x1, /* pure ordering, nothing implied */ pe_order_apply_first_non_migratable = 0x2, /* Only apply this constraint's ordering if first is not migratable. */ pe_order_implies_first = 0x10, /* If 'then' is required, ensure 'first' is too */ pe_order_implies_then = 0x20, /* If 'first' is required, ensure 'then' is too */ pe_order_promoted_implies_first = 0x40, /* If 'then' is required and then's rsc is promoted, ensure 'first' becomes required too */ /* first requires then to be both runnable and migrate runnable. */ pe_order_implies_first_migratable = 0x80, pe_order_runnable_left = 0x100, /* 'then' requires 'first' to be runnable */ pe_order_pseudo_left = 0x200, /* 'then' can only be pseudo if 'first' is runnable */ pe_order_implies_then_on_node = 0x400, /* If 'first' is required on 'nodeX', * ensure instances of 'then' on 'nodeX' are too. * Only really useful if 'then' is a clone and 'first' is not */ pe_order_probe = 0x800, /* If 'first->rsc' is * - running but about to stop, ignore the constraint * - otherwise, behave as runnable_left */ pe_order_restart = 0x1000, /* 'then' is runnable if 'first' is optional or runnable */ pe_order_stonith_stop = 0x2000, // #endif #ifdef __cplusplus } #endif #endif // PCMK__CRM_PENGINE_PE_TYPES__H diff --git a/include/crm/pengine/pe_types_compat.h b/include/crm/pengine/pe_types_compat.h index d12b253947..566a5bf5b6 100644 --- a/include/crm/pengine/pe_types_compat.h +++ b/include/crm/pengine/pe_types_compat.h @@ -1,137 +1,140 @@ /* * 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__CRM_PENGINE_PE_TYPES_COMPAT__H # define PCMK__CRM_PENGINE_PE_TYPES_COMPAT__H #include #ifdef __cplusplus extern "C" { #endif /** * \file * \brief Deprecated Pacemaker scheduler API * \ingroup pengine * \deprecated Do not include this header directly. The scheduler APIs in this * header, and the header itself, will be removed in a future * release. */ //! \deprecated Use pcmk_sched_quorate instead #define pe_flag_have_quorum pcmk_sched_quorate //! \deprecated Use pcmk_sched_symmetric_cluster instead #define pe_flag_symmetric_cluster pcmk_sched_symmetric_cluster //! \deprecated Use pcmk_sched_in_maintenance instead #define pe_flag_maintenance_mode pcmk_sched_in_maintenance //! \deprecated Use pcmk_sched_fencing_enabled instead #define pe_flag_stonith_enabled pcmk_sched_fencing_enabled //! \deprecated Use pcmk_sched_have_fencing instead #define pe_flag_have_stonith_resource pcmk_sched_have_fencing //! \deprecated Use pcmk_sched_enable_unfencing instead #define pe_flag_enable_unfencing pcmk_sched_enable_unfencing //! \deprecated Use pcmk_sched_concurrent_fencing instead #define pe_flag_concurrent_fencing pcmk_sched_concurrent_fencing //! \deprecated Use pcmk_sched_stop_removed_resources instead #define pe_flag_stop_rsc_orphans pcmk_sched_stop_removed_resources //! \deprecated Use pcmk_sched_cancel_removed_actions instead #define pe_flag_stop_action_orphans pcmk_sched_cancel_removed_actions //! \deprecated Use pcmk_sched_stop_all instead #define pe_flag_stop_everything pcmk_sched_stop_all //! \deprecated Use pcmk_sched_start_failure_fatal instead #define pe_flag_start_failure_fatal pcmk_sched_start_failure_fatal //! \deprecated Do not use #define pe_flag_remove_after_stop pcmk_sched_remove_after_stop //! \deprecated Use pcmk_sched_startup_fencing instead #define pe_flag_startup_fencing pcmk_sched_startup_fencing //! \deprecated Use pcmk_sched_shutdown_lock instead #define pe_flag_shutdown_lock pcmk_sched_shutdown_lock //! \deprecated Use pcmk_sched_probe_resources instead #define pe_flag_startup_probes pcmk_sched_probe_resources //! \deprecated Use pcmk_sched_have_status instead #define pe_flag_have_status pcmk_sched_have_status //! \deprecated Use pcmk_sched_have_remote_nodes instead #define pe_flag_have_remote_nodes pcmk_sched_have_remote_nodes //! \deprecated Use pcmk_sched_location_only instead #define pe_flag_quick_location pcmk_sched_location_only //! \deprecated Use pcmk_sched_sanitized instead #define pe_flag_sanitized pcmk_sched_sanitized //! \deprecated Do not use #define pe_flag_stdout (1ULL << 22) //! \deprecated Use pcmk_sched_no_counts instead #define pe_flag_no_counts pcmk_sched_no_counts //! \deprecated Use pcmk_sched_no_compat instead #define pe_flag_no_compat pcmk_sched_no_compat +//! \deprecated Use pcmk_sched_output_scores instead +#define pe_flag_show_scores pcmk_sched_output_scores + //!@{ //! \deprecated Do not use (unused by Pacemaker) enum pe_graph_flags { pe_graph_none = 0x00000, pe_graph_updated_first = 0x00001, pe_graph_updated_then = 0x00002, pe_graph_disable = 0x00004, }; //!@} //!@{ //! \deprecated Do not use enum pe_check_parameters { pe_check_last_failure, pe_check_active, }; //!@} //!< \deprecated Use pe_action_t instead typedef struct pe_action_s action_t; //!< \deprecated Use pe_action_wrapper_t instead typedef struct pe_action_wrapper_s action_wrapper_t; //!< \deprecated Use pe_node_t instead typedef struct pe_node_s node_t; //!< \deprecated Use enum pe_quorum_policy instead typedef enum pe_quorum_policy no_quorum_policy_t; //!< \deprecated use pe_resource_t instead typedef struct pe_resource_s resource_t; //!< \deprecated Use pe_tag_t instead typedef struct pe_tag_s tag_t; //!< \deprecated Use pe_ticket_t instead typedef struct pe_ticket_s ticket_t; #ifdef __cplusplus } #endif #endif // PCMK__CRM_PENGINE_PE_TYPES_COMPAT__H diff --git a/lib/pacemaker/pcmk_sched_bundle.c b/lib/pacemaker/pcmk_sched_bundle.c index bd0558a6d3..37666c2da8 100644 --- a/lib/pacemaker/pcmk_sched_bundle.c +++ b/lib/pacemaker/pcmk_sched_bundle.c @@ -1,1052 +1,1053 @@ /* * 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" struct assign_data { const pe_node_t *prefer; bool stop_if_fail; }; /*! * \internal * \brief Assign a single bundle replica's resources (other than container) * * \param[in,out] replica Replica to assign * \param[in] user_data Preferred node, if any * * \return true (to indicate that any further replicas should be processed) */ static bool assign_replica(pe__bundle_replica_t *replica, void *user_data) { pe_node_t *container_host = NULL; struct assign_data *assign_data = user_data; const pe_node_t *prefer = assign_data->prefer; bool stop_if_fail = assign_data->stop_if_fail; const pe_resource_t *bundle = pe__const_top_resource(replica->container, true); if (replica->ip != NULL) { pe_rsc_trace(bundle, "Assigning bundle %s IP %s", bundle->id, replica->ip->id); replica->ip->cmds->assign(replica->ip, prefer, stop_if_fail); } container_host = replica->container->allocated_to; if (replica->remote != NULL) { if (pe__is_guest_or_remote_node(container_host)) { /* REMOTE_CONTAINER_HACK: "Nested" connection resources must be on * the same host because Pacemaker Remote only supports a single * active connection. */ pcmk__new_colocation("#replica-remote-with-host-remote", NULL, INFINITY, replica->remote, container_host->details->remote_rsc, NULL, NULL, pcmk__coloc_influence); } pe_rsc_trace(bundle, "Assigning bundle %s connection %s", bundle->id, replica->remote->id); replica->remote->cmds->assign(replica->remote, prefer, stop_if_fail); } if (replica->child != NULL) { 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 (!pe__same_node(node, replica->node)) { 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(bundle, "Assigning bundle %s replica child %s", bundle->id, replica->child->id); replica->child->cmds->assign(replica->child, replica->node, stop_if_fail); pe__clear_resource_flags(replica->child->parent, pe_rsc_allocating); } return true; } /*! * \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 * \param[in] stop_if_fail If \c true and a primitive descendant of \p rsc * can't be assigned to a node, set the * descendant's next role to stopped and update * existing actions * * \return Node that \p rsc is assigned to, if assigned entirely to one node * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ pe_node_t * pcmk__bundle_assign(pe_resource_t *rsc, const pe_node_t *prefer, bool stop_if_fail) { GList *containers = NULL; pe_resource_t *bundled_resource = NULL; struct assign_data assign_data = { prefer, stop_if_fail }; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe_rsc_trace(rsc, "Assigning bundle %s", rsc->id); pe__set_resource_flags(rsc, pe_rsc_allocating); - pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), + pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, + pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); // Assign all containers first, so we know what nodes the bundle will be on containers = g_list_sort(pe__bundle_containers(rsc), pcmk__cmp_instance); pcmk__assign_instances(rsc, containers, pe__bundle_max(rsc), rsc->fns->max_per_node(rsc)); g_list_free(containers); // Then assign remaining replica resources pe__foreach_bundle_replica(rsc, assign_replica, (void *) &assign_data); // Finally, assign the bundled resources to each bundle node bundled_resource = pe__bundled_resource(rsc); if (bundled_resource != NULL) { pe_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, bundled_resource->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { if (pe__node_is_bundle_instance(rsc, node)) { node->weight = 0; } else { node->weight = -INFINITY; } } bundled_resource->cmds->assign(bundled_resource, prefer, stop_if_fail); } pe__clear_resource_flags(rsc, pe_rsc_allocating|pe_rsc_provisional); return NULL; } /*! * \internal * \brief Create actions for a bundle replica's resources (other than child) * * \param[in,out] replica Replica to create actions for * \param[in] user_data Unused * * \return true (to indicate that any further replicas should be processed) */ static bool create_replica_actions(pe__bundle_replica_t *replica, void *user_data) { if (replica->ip != NULL) { replica->ip->cmds->create_actions(replica->ip); } if (replica->container != NULL) { replica->container->cmds->create_actions(replica->container); } if (replica->remote != NULL) { replica->remote->cmds->create_actions(replica->remote); } return true; } /*! * \internal * \brief Create all actions needed for a given bundle resource * * \param[in,out] rsc Bundle resource to create actions for */ void pcmk__bundle_create_actions(pe_resource_t *rsc) { pe_action_t *action = NULL; GList *containers = NULL; pe_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, create_replica_actions, NULL); containers = pe__bundle_containers(rsc); pcmk__create_instance_actions(rsc, containers); g_list_free(containers); bundled_resource = pe__bundled_resource(rsc); if (bundled_resource != NULL) { bundled_resource->cmds->create_actions(bundled_resource); if (pcmk_is_set(bundled_resource->flags, pe_rsc_promotable)) { pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_PROMOTE, true, true); action = pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_PROMOTED, true, true); action->priority = INFINITY; pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_DEMOTE, true, true); action = pe__new_rsc_pseudo_action(rsc, PCMK_ACTION_DEMOTED, true, true); action->priority = INFINITY; } } } /*! * \internal * \brief Create internal constraints for a bundle replica's resources * * \param[in,out] replica Replica to create internal constraints for * \param[in,out] user_data Replica's parent bundle * * \return true (to indicate that any further replicas should be processed) */ static bool replica_internal_constraints(pe__bundle_replica_t *replica, void *user_data) { pe_resource_t *bundle = user_data; replica->container->cmds->internal_constraints(replica->container); // Start bundle -> start replica container pcmk__order_starts(bundle, replica->container, pe_order_runnable_left|pe_order_implies_first_printed); // Stop bundle -> stop replica child and container if (replica->child != NULL) { pcmk__order_stops(bundle, replica->child, pe_order_implies_first_printed); } pcmk__order_stops(bundle, replica->container, pe_order_implies_first_printed); // Start replica container -> bundle is started pcmk__order_resource_actions(replica->container, PCMK_ACTION_START, bundle, PCMK_ACTION_RUNNING, pe_order_implies_then_printed); // Stop replica container -> bundle is stopped pcmk__order_resource_actions(replica->container, PCMK_ACTION_STOP, bundle, PCMK_ACTION_STOPPED, pe_order_implies_then_printed); if (replica->ip != NULL) { replica->ip->cmds->internal_constraints(replica->ip); // Replica IP address -> replica container (symmetric) 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-container", NULL, INFINITY, replica->ip, replica->container, NULL, NULL, pcmk__coloc_influence); } if (replica->remote != NULL) { /* 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 != NULL) { CRM_ASSERT(replica->remote != NULL); // "Start remote then child" is implicit in scheduler's remote logic } return true; } /*! * \internal * \brief Create implicit constraints needed for a bundle resource * * \param[in,out] rsc Bundle resource to create implicit constraints for */ void pcmk__bundle_internal_constraints(pe_resource_t *rsc) { pe_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, replica_internal_constraints, rsc); bundled_resource = pe__bundled_resource(rsc); if (bundled_resource == NULL) { return; } // Start bundle -> start bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_START, bundled_resource, PCMK_ACTION_START, pe_order_implies_first_printed); // Bundled clone is started -> bundle is started pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_RUNNING, rsc, PCMK_ACTION_RUNNING, pe_order_implies_then_printed); // Stop bundle -> stop bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, bundled_resource, PCMK_ACTION_STOP, pe_order_implies_first_printed); // Bundled clone is stopped -> bundle is stopped pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_STOPPED, rsc, PCMK_ACTION_STOPPED, pe_order_implies_then_printed); bundled_resource->cmds->internal_constraints(bundled_resource); if (!pcmk_is_set(bundled_resource->flags, pe_rsc_promotable)) { return; } pcmk__promotable_restart_ordering(rsc); // Demote bundle -> demote bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_DEMOTE, bundled_resource, PCMK_ACTION_DEMOTE, pe_order_implies_first_printed); // Bundled clone is demoted -> bundle is demoted pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_DEMOTED, rsc, PCMK_ACTION_DEMOTED, pe_order_implies_then_printed); // Promote bundle -> promote bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_PROMOTE, bundled_resource, PCMK_ACTION_PROMOTE, pe_order_implies_first_printed); // Bundled clone is promoted -> bundle is promoted pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_PROMOTED, rsc, PCMK_ACTION_PROMOTED, pe_order_implies_then_printed); } struct match_data { const pe_node_t *node; // Node to compare against replica pe_resource_t *container; // Replica container corresponding to node }; /*! * \internal * \brief Check whether a replica container is assigned to a given node * * \param[in] replica Replica to check * \param[in,out] user_data struct match_data with node to compare against * * \return true if the replica does not match (to indicate further replicas * should be processed), otherwise false */ static bool match_replica_container(const pe__bundle_replica_t *replica, void *user_data) { struct match_data *match_data = user_data; if (pcmk__instance_matches(replica->container, match_data->node, pcmk_role_unknown, false)) { match_data->container = replica->container; return false; // Match found, don't bother searching further replicas } return true; // No match, keep searching } /*! * \internal * \brief Get the host to which a bundle node is assigned * * \param[in] node Possible bundle node to check * * \return Node to which the container for \p node is assigned if \p node is a * bundle node, otherwise \p node itself */ static const pe_node_t * get_bundle_node_host(const pe_node_t *node) { if (pe__is_bundle_node(node)) { const pe_resource_t *container = node->details->remote_rsc->container; return container->fns->location(container, NULL, 0); } return node; } /*! * \internal * \brief Find a bundle container compatible with a dependent resource * * \param[in] dependent Dependent resource in colocation with bundle * \param[in] bundle Bundle that \p dependent is colocated with * * \return A container from \p bundle assigned to the same node as \p dependent * if assigned, otherwise assigned to any of dependent's allowed nodes, * otherwise NULL. */ static pe_resource_t * compatible_container(const pe_resource_t *dependent, const pe_resource_t *bundle) { GList *scratch = NULL; struct match_data match_data = { NULL, NULL }; // If dependent is assigned, only check there match_data.node = dependent->fns->location(dependent, NULL, 0); match_data.node = get_bundle_node_host(match_data.node); if (match_data.node != NULL) { pe__foreach_const_bundle_replica(bundle, match_replica_container, &match_data); return match_data.container; } // Otherwise, check for any of the dependent's allowed nodes scratch = g_hash_table_get_values(dependent->allowed_nodes); scratch = pcmk__sort_nodes(scratch, NULL); for (const GList *iter = scratch; iter != NULL; iter = iter->next) { match_data.node = iter->data; match_data.node = get_bundle_node_host(match_data.node); if (match_data.node == NULL) { continue; } pe__foreach_const_bundle_replica(bundle, match_replica_container, &match_data); if (match_data.container != NULL) { break; } } g_list_free(scratch); return match_data.container; } struct coloc_data { const pcmk__colocation_t *colocation; pe_resource_t *dependent; GList *container_hosts; }; /*! * \internal * \brief Apply a colocation score to replica node scores or resource priority * * \param[in] replica Replica of primary bundle resource in colocation * \param[in,out] user_data struct coloc_data for colocation being applied * * \return true (to indicate that any further replicas should be processed) */ static bool replica_apply_coloc_score(const pe__bundle_replica_t *replica, void *user_data) { struct coloc_data *coloc_data = user_data; pe_node_t *chosen = NULL; if (coloc_data->colocation->score < INFINITY) { replica->container->cmds->apply_coloc_score(coloc_data->dependent, replica->container, coloc_data->colocation, false); return true; } chosen = replica->container->fns->location(replica->container, NULL, 0); if ((chosen == NULL) || is_set_recursive(replica->container, pe_rsc_block, true)) { return true; } if ((coloc_data->colocation->primary_role >= pcmk_role_promoted) && ((replica->child == NULL) || (replica->child->next_role < pcmk_role_promoted))) { return true; } pe_rsc_trace(pe__const_top_resource(replica->container, true), "Allowing mandatory colocation %s using %s @%d", coloc_data->colocation->id, pe__node_name(chosen), chosen->weight); coloc_data->container_hosts = g_list_prepend(coloc_data->container_hosts, chosen); return true; } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__bundle_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { struct coloc_data coloc_data = { colocation, dependent, NULL }; /* This should never be called for the bundle itself as a dependent. * Instead, we add its colocation constraints to its containers and bundled * primitive and call the apply_coloc_score() method for them as dependents. */ CRM_ASSERT((primary != NULL) && (primary->variant == pcmk_rsc_variant_bundle) && (dependent != NULL) && (dependent->variant == pcmk_rsc_variant_primitive) && (colocation != NULL) && !for_dependent); if (pcmk_is_set(primary->flags, pe_rsc_provisional)) { pe_rsc_trace(primary, "Skipping applying colocation %s " "because %s is still provisional", colocation->id, primary->id); return; } pe_rsc_trace(primary, "Applying colocation %s (%s with %s at %s)", colocation->id, dependent->id, primary->id, pcmk_readable_score(colocation->score)); /* If the constraint dependent is a clone or bundle, "dependent" here is one * of its instances. Look for a compatible instance of this bundle. */ if (colocation->dependent->variant > pcmk_rsc_variant_group) { const pe_resource_t *primary_container = compatible_container(dependent, primary); if (primary_container != NULL) { // Success, we found one pe_rsc_debug(primary, "Pairing %s with %s", dependent->id, primary_container->id); dependent->cmds->apply_coloc_score(dependent, primary_container, colocation, true); } else if (colocation->score >= INFINITY) { // Failure, and it's fatal crm_notice("%s cannot run because there is no compatible " "instance of %s to colocate with", dependent->id, primary->id); pcmk__assign_resource(dependent, NULL, true, true); } else { // Failure, but we can ignore it pe_rsc_debug(primary, "%s cannot be colocated with any instance of %s", dependent->id, primary->id); } return; } pe__foreach_const_bundle_replica(primary, replica_apply_coloc_score, &coloc_data); if (colocation->score >= INFINITY) { pcmk__colocation_intersect_nodes(dependent, primary, colocation, coloc_data.container_hosts, false); } g_list_free(coloc_data.container_hosts); } // Bundle implementation of resource_alloc_functions_t:with_this_colocations() void pcmk__with_bundle_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { const pe_resource_t *bundled_rsc = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle) && (orig_rsc != NULL) && (list != NULL)); // The bundle itself and its containers always get its colocations if ((orig_rsc == rsc) || pcmk_is_set(orig_rsc->flags, pe_rsc_replica_container)) { pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); return; } /* The bundled resource gets the colocations if it's promotable and we've * begun choosing roles */ bundled_rsc = pe__bundled_resource(rsc); if ((bundled_rsc == NULL) || !pcmk_is_set(bundled_rsc->flags, pe_rsc_promotable) || (pe__const_top_resource(orig_rsc, false) != bundled_rsc)) { return; } if (orig_rsc == bundled_rsc) { if (pe__clone_flag_is_set(orig_rsc, pe__clone_promotion_constrained)) { /* orig_rsc is the clone and we're setting roles (or have already * done so) */ pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); } } else if (!pcmk_is_set(orig_rsc->flags, pe_rsc_provisional)) { /* orig_rsc is an instance and is already assigned. If something * requests colocations for orig_rsc now, it's for setting roles. */ pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); } } // Bundle implementation of resource_alloc_functions_t:this_with_colocations() void pcmk__bundle_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { const pe_resource_t *bundled_rsc = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle) && (orig_rsc != NULL) && (list != NULL)); // The bundle itself and its containers always get its colocations if ((orig_rsc == rsc) || pcmk_is_set(orig_rsc->flags, pe_rsc_replica_container)) { pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); return; } /* The bundled resource gets the colocations if it's promotable and we've * begun choosing roles */ bundled_rsc = pe__bundled_resource(rsc); if ((bundled_rsc == NULL) || !pcmk_is_set(bundled_rsc->flags, pe_rsc_promotable) || (pe__const_top_resource(orig_rsc, false) != bundled_rsc)) { return; } if (orig_rsc == bundled_rsc) { if (pe__clone_flag_is_set(orig_rsc, pe__clone_promotion_constrained)) { /* orig_rsc is the clone and we're setting roles (or have already * done so) */ pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); } } else if (!pcmk_is_set(orig_rsc->flags, pe_rsc_provisional)) { /* orig_rsc is an instance and is already assigned. If something * requests colocations for orig_rsc now, it's for setting roles. */ pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); } } /*! * \internal * \brief Return action flags for a given bundle resource action * * \param[in,out] action Bundle resource 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 */ uint32_t pcmk__bundle_action_flags(pe_action_t *action, const pe_node_t *node) { GList *containers = NULL; uint32_t flags = 0; pe_resource_t *bundled_resource = NULL; CRM_ASSERT((action != NULL) && (action->rsc != NULL) && (action->rsc->variant == pcmk_rsc_variant_bundle)); bundled_resource = pe__bundled_resource(action->rsc); if (bundled_resource != NULL) { // Clone actions are done on the bundled clone resource, not container switch (get_complex_task(bundled_resource, action->task)) { case pcmk_action_unspecified: case pcmk_action_notify: case pcmk_action_notified: case pcmk_action_promote: case pcmk_action_promoted: case pcmk_action_demote: case pcmk_action_demoted: return pcmk__collective_action_flags(action, bundled_resource->children, node); default: break; } } containers = pe__bundle_containers(action->rsc); flags = pcmk__collective_action_flags(action, containers, node); g_list_free(containers); return flags; } /*! * \internal * \brief Apply a location constraint to a bundle replica * * \param[in,out] replica Replica to apply constraint to * \param[in,out] user_data Location constraint to apply * * \return true (to indicate that any further replicas should be processed) */ static bool apply_location_to_replica(pe__bundle_replica_t *replica, void *user_data) { pe__location_t *location = user_data; if (replica->container != NULL) { replica->container->cmds->apply_location(replica->container, location); } if (replica->ip != NULL) { replica->ip->cmds->apply_location(replica->ip, location); } return true; } /*! * \internal * \brief Apply a location constraint to a bundle resource's allowed node scores * * \param[in,out] rsc Bundle resource to apply constraint to * \param[in,out] location Location constraint to apply */ void pcmk__bundle_apply_location(pe_resource_t *rsc, pe__location_t *location) { pe_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle) && (location != NULL)); pcmk__apply_location(rsc, location); pe__foreach_bundle_replica(rsc, apply_location_to_replica, location); bundled_resource = pe__bundled_resource(rsc); if ((bundled_resource != NULL) && ((location->role_filter == pcmk_role_unpromoted) || (location->role_filter == pcmk_role_promoted))) { bundled_resource->cmds->apply_location(bundled_resource, location); bundled_resource->rsc_location = g_list_prepend( bundled_resource->rsc_location, location); } } #define XPATH_REMOTE "//nvpair[@name='" XML_RSC_ATTR_REMOTE_RA_ADDR "']" /*! * \internal * \brief Add a bundle replica's actions to transition graph * * \param[in,out] replica Replica to add to graph * \param[in] user_data Bundle that replica belongs to (for logging only) * * \return true (to indicate that any further replicas should be processed) */ static bool add_replica_actions_to_graph(pe__bundle_replica_t *replica, void *user_data) { if ((replica->remote != NULL) && (replica->container != NULL) && 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(XPATH_REMOTE, 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, replica->remote->cluster, nvpair, "value"); if (calculated_addr != NULL) { /* 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, replica->remote->cluster); g_hash_table_replace(params, strdup(XML_RSC_ATTR_REMOTE_RA_ADDR), strdup(calculated_addr)); } else { pe_resource_t *bundle = user_data; /* 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). */ pe_rsc_info(bundle, "Unable to determine address for bundle %s " "remote connection", bundle->id); } } if (replica->ip != NULL) { replica->ip->cmds->add_actions_to_graph(replica->ip); } if (replica->container != NULL) { replica->container->cmds->add_actions_to_graph(replica->container); } if (replica->remote != NULL) { replica->remote->cmds->add_actions_to_graph(replica->remote); } return true; } /*! * \internal * \brief Add a bundle resource's actions to the transition graph * * \param[in,out] rsc Bundle resource whose actions should be added */ void pcmk__bundle_add_actions_to_graph(pe_resource_t *rsc) { pe_resource_t *bundled_resource = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); bundled_resource = pe__bundled_resource(rsc); if (bundled_resource != NULL) { bundled_resource->cmds->add_actions_to_graph(bundled_resource); } pe__foreach_bundle_replica(rsc, add_replica_actions_to_graph, rsc); } struct probe_data { pe_resource_t *bundle; // Bundle being probed pe_node_t *node; // Node to create probes on bool any_created; // Whether any probes have been created }; /*! * \internal * \brief Order a bundle replica's start after another replica's probe * * \param[in,out] replica Replica to order start for * \param[in,out] user_data Replica with probe to order after * * \return true (to indicate that any further replicas should be processed) */ static bool order_replica_start_after(pe__bundle_replica_t *replica, void *user_data) { pe__bundle_replica_t *probed_replica = user_data; if ((replica == probed_replica) || (replica->container == NULL)) { return true; } pcmk__new_ordering(probed_replica->container, pcmk__op_key(probed_replica->container->id, PCMK_ACTION_MONITOR, 0), NULL, replica->container, pcmk__op_key(replica->container->id, PCMK_ACTION_START, 0), NULL, pe_order_optional|pe_order_same_node, replica->container->cluster); return true; } /*! * \internal * \brief Create probes for a bundle replica's resources * * \param[in,out] replica Replica to create probes for * \param[in,out] user_data struct probe_data * * \return true (to indicate that any further replicas should be processed) */ static bool create_replica_probes(pe__bundle_replica_t *replica, void *user_data) { struct probe_data *probe_data = user_data; if ((replica->ip != NULL) && replica->ip->cmds->create_probe(replica->ip, probe_data->node)) { probe_data->any_created = true; } if ((replica->child != NULL) && pe__same_node(probe_data->node, replica->node) && replica->child->cmds->create_probe(replica->child, probe_data->node)) { probe_data->any_created = true; } if ((replica->container != NULL) && replica->container->cmds->create_probe(replica->container, probe_data->node)) { probe_data->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 the maximum replicas 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 */ if (probe_data->bundle->fns->max_per_node(probe_data->bundle) == 1) { pe__foreach_bundle_replica(probe_data->bundle, order_replica_start_after, replica); } } if ((replica->container != NULL) && (replica->remote != NULL) && replica->remote->cmds->create_probe(replica->remote, probe_data->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, PCMK_ACTION_MONITOR, 0); pe_action_t *probe = find_first_action(replica->remote->actions, probe_uuid, NULL, probe_data->node); free(probe_uuid); if (probe != NULL) { probe_data->any_created = true; pe_rsc_trace(probe_data->bundle, "Ordering %s probe on %s", replica->remote->id, pe__node_name(probe_data->node)); pcmk__new_ordering(replica->container, pcmk__op_key(replica->container->id, PCMK_ACTION_START, 0), NULL, replica->remote, NULL, probe, pe_order_probe, probe_data->bundle->cluster); } } return true; } /*! * \internal * * \brief Schedule any probes needed for a bundle resource on a node * * \param[in,out] rsc Bundle resource to create probes 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) { struct probe_data probe_data = { rsc, node, false }; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, create_replica_probes, &probe_data); return probe_data.any_created; } /*! * \internal * \brief Output actions for one bundle replica * * \param[in,out] replica Replica to output actions for * \param[in] user_data Unused * * \return true (to indicate that any further replicas should be processed) */ static bool output_replica_actions(pe__bundle_replica_t *replica, void *user_data) { 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); } return true; } /*! * \internal * \brief Output a summary of scheduled actions for a bundle resource * * \param[in,out] rsc Bundle resource to output actions for */ void pcmk__output_bundle_actions(pe_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); pe__foreach_bundle_replica(rsc, output_replica_actions, NULL); } // 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_resource_t *container = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { 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. */ container = pe__first_container(rsc); if (container != NULL) { container->cmds->add_utilization(container, orig_rsc, all_rscs, utilization); } } // Bundle implementation of resource_alloc_functions_t:shutdown_lock() void pcmk__bundle_shutdown_lock(pe_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_bundle)); // Bundles currently don't support shutdown locks } diff --git a/lib/pacemaker/pcmk_sched_clone.c b/lib/pacemaker/pcmk_sched_clone.c index be8c008db0..65ed263b1e 100644 --- a/lib/pacemaker/pcmk_sched_clone.c +++ b/lib/pacemaker/pcmk_sched_clone.c @@ -1,708 +1,709 @@ /* * 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" /*! * \internal * \brief Assign a clone resource's instances to nodes * * \param[in,out] rsc Clone resource to assign * \param[in] prefer Node to prefer, if all else is equal * \param[in] stop_if_fail If \c true and a primitive descendant of \p rsc * can't be assigned to a node, set the * descendant's next role to stopped and update * existing actions * * \return NULL (clones are not assigned to a single node) * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ pe_node_t * pcmk__clone_assign(pe_resource_t *rsc, const pe_node_t *prefer, bool stop_if_fail) { GList *colocations = NULL; CRM_ASSERT(pe_rsc_is_clone(rsc)); if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return NULL; // Assignment has already been done } // 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); // If this clone is promotable, consider nodes' promotion scores if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__add_promotion_scores(rsc); } // If this clone is colocated with any other resources, assign those first colocations = pcmk__this_with_colocations(rsc); for (GList *iter = colocations; iter != NULL; iter = iter->next) { pcmk__colocation_t *constraint = (pcmk__colocation_t *) iter->data; pe_rsc_trace(rsc, "%s: Assigning colocation %s primary %s first", rsc->id, constraint->id, constraint->primary->id); constraint->primary->cmds->assign(constraint->primary, prefer, stop_if_fail); } g_list_free(colocations); // If any resources are colocated with this one, consider their preferences colocations = pcmk__with_this_colocations(rsc); g_list_foreach(colocations, pcmk__add_dependent_scores, rsc); g_list_free(colocations); - pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), + pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, + pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance); pcmk__assign_instances(rsc, rsc->children, pe__clone_max(rsc), pe__clone_node_max(rsc)); 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, "Assigned clone %s", rsc->id); return NULL; } /*! * \internal * \brief Create all actions needed for a given clone resource * * \param[in,out] rsc Clone resource to create actions for */ void pcmk__clone_create_actions(pe_resource_t *rsc) { CRM_ASSERT(pe_rsc_is_clone(rsc)); pe_rsc_trace(rsc, "Creating actions for clone %s", rsc->id); pcmk__create_instance_actions(rsc, rsc->children); if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__create_promotable_actions(rsc); } } /*! * \internal * \brief Create implicit constraints needed for a clone resource * * \param[in,out] rsc Clone resource to create implicit constraints for */ void pcmk__clone_internal_constraints(pe_resource_t *rsc) { bool ordered = false; CRM_ASSERT(pe_rsc_is_clone(rsc)); pe_rsc_trace(rsc, "Creating internal constraints for clone %s", rsc->id); // Restart ordering: Stop -> stopped -> start -> started pcmk__order_resource_actions(rsc, PCMK_ACTION_STOPPED, rsc, PCMK_ACTION_START, pe_order_optional); pcmk__order_resource_actions(rsc, PCMK_ACTION_START, rsc, PCMK_ACTION_RUNNING, pe_order_runnable_left); pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, rsc, PCMK_ACTION_STOPPED, pe_order_runnable_left); // Demoted -> stop and started -> promote if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__order_resource_actions(rsc, PCMK_ACTION_DEMOTED, rsc, PCMK_ACTION_STOP, pe_order_optional); pcmk__order_resource_actions(rsc, PCMK_ACTION_RUNNING, rsc, PCMK_ACTION_PROMOTE, pe_order_runnable_left); } ordered = pe__clone_is_ordered(rsc); if (ordered) { /* Ordered clone instances must start and stop by instance number. The * instances might have been previously shuffled for assignment or * promotion purposes, so re-sort them. */ rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance_number); } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; instance->cmds->internal_constraints(instance); // Start clone -> start instance -> clone started pcmk__order_starts(rsc, instance, pe_order_runnable_left |pe_order_implies_first_printed); pcmk__order_resource_actions(instance, PCMK_ACTION_START, rsc, PCMK_ACTION_RUNNING, pe_order_implies_then_printed); // Stop clone -> stop instance -> clone stopped pcmk__order_stops(rsc, instance, pe_order_implies_first_printed); pcmk__order_resource_actions(instance, PCMK_ACTION_STOP, rsc, PCMK_ACTION_STOPPED, pe_order_implies_then_printed); /* Instances of ordered clones must be started and stopped by instance * number. Since only some instances may be starting or stopping, order * each instance relative to every later instance. */ if (ordered) { for (GList *later = iter->next; later != NULL; later = later->next) { pcmk__order_starts(instance, (pe_resource_t *) later->data, pe_order_optional); pcmk__order_stops((pe_resource_t *) later->data, instance, pe_order_optional); } } } if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__order_promotable_instances(rsc); } } /*! * \internal * \brief Check whether colocated resources can be interleaved * * \param[in] colocation Colocation constraint with clone as primary * * \return true if colocated resources can be interleaved, otherwise false */ static bool can_interleave(const pcmk__colocation_t *colocation) { const pe_resource_t *dependent = colocation->dependent; // Only colocations between clone or bundle resources use interleaving if (dependent->variant <= pcmk_rsc_variant_group) { return false; } // Only the dependent needs to be marked for interleaving if (!crm_is_true(g_hash_table_lookup(dependent->meta, XML_RSC_ATTR_INTERLEAVE))) { return false; } /* @TODO Do we actually care about multiple primary instances sharing a * dependent instance? */ if (dependent->fns->max_per_node(dependent) != colocation->primary->fns->max_per_node(colocation->primary)) { pcmk__config_err("Cannot interleave %s and %s because they do not " "support the same number of instances per node", dependent->id, colocation->primary->id); return false; } return true; } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__clone_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { const GList *iter = 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() method for the instances as dependents. */ CRM_ASSERT(!for_dependent); CRM_ASSERT((colocation != NULL) && pe_rsc_is_clone(primary) && (dependent != NULL) && (dependent->variant == pcmk_rsc_variant_primitive)); if (pcmk_is_set(primary->flags, pe_rsc_provisional)) { pe_rsc_trace(primary, "Delaying processing colocation %s " "because cloned primary %s is still provisional", colocation->id, primary->id); return; } pe_rsc_trace(primary, "Processing colocation %s (%s with clone %s @%s)", colocation->id, dependent->id, primary->id, pcmk_readable_score(colocation->score)); // Apply role-specific colocations if (pcmk_is_set(primary->flags, pe_rsc_promotable) && (colocation->primary_role != pcmk_role_unknown)) { if (pcmk_is_set(dependent->flags, pe_rsc_provisional)) { // We're assigning the dependent to a node pcmk__update_dependent_with_promotable(primary, dependent, colocation); return; } if (colocation->dependent_role == pcmk_role_promoted) { // We're choosing a role for the dependent pcmk__update_promotable_dependent_priority(primary, dependent, colocation); return; } } // Apply interleaved colocations if (can_interleave(colocation)) { const pe_resource_t *primary_instance = NULL; primary_instance = pcmk__find_compatible_instance(dependent, primary, pcmk_role_unknown, false); if (primary_instance != NULL) { pe_rsc_debug(primary, "Interleaving %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("%s cannot run because it cannot interleave with " "any instance of %s", dependent->id, primary->id); pcmk__assign_resource(dependent, NULL, true, true); } else { pe_rsc_debug(primary, "%s will not colocate with %s " "because no instance can interleave with it", dependent->id, primary->id); } return; } // Apply mandatory colocations if (colocation->score >= INFINITY) { GList *primary_nodes = NULL; // Dependent can run only where primary will have unblocked instances for (iter = primary->children; iter != NULL; iter = iter->next) { const pe_resource_t *instance = iter->data; pe_node_t *chosen = instance->fns->location(instance, NULL, 0); if ((chosen != NULL) && !is_set_recursive(instance, pe_rsc_block, TRUE)) { pe_rsc_trace(primary, "Allowing %s: %s %d", colocation->id, pe__node_name(chosen), chosen->weight); primary_nodes = g_list_prepend(primary_nodes, chosen); } } pcmk__colocation_intersect_nodes(dependent, primary, colocation, primary_nodes, false); g_list_free(primary_nodes); return; } // Apply optional colocations for (iter = primary->children; iter != NULL; iter = iter->next) { const pe_resource_t *instance = iter->data; instance->cmds->apply_coloc_score(dependent, instance, colocation, false); } } // Clone implementation of resource_alloc_functions_t:with_this_colocations() void pcmk__with_clone_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_CHECK((rsc != NULL) && (orig_rsc != NULL) && (list != NULL), return); pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); if (rsc->parent != NULL) { rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list); } } // Clone implementation of resource_alloc_functions_t:this_with_colocations() void pcmk__clone_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_CHECK((rsc != NULL) && (orig_rsc != NULL) && (list != NULL), return); pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); if (rsc->parent != NULL) { rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list); } } /*! * \internal * \brief Return action flags for a given clone 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 */ uint32_t pcmk__clone_action_flags(pe_action_t *action, const pe_node_t *node) { CRM_ASSERT((action != NULL) && pe_rsc_is_clone(action->rsc)); return pcmk__collective_action_flags(action, action->rsc->children, node); } /*! * \internal * \brief Apply a location constraint to a clone resource's allowed node scores * * \param[in,out] rsc Clone resource to apply constraint to * \param[in,out] location Location constraint to apply */ void pcmk__clone_apply_location(pe_resource_t *rsc, pe__location_t *location) { CRM_CHECK((location != NULL) && pe_rsc_is_clone(rsc), return); pcmk__apply_location(rsc, location); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; instance->cmds->apply_location(instance, location); } } // GFunc wrapper for calling the action_flags() resource method static void call_action_flags(gpointer data, gpointer user_data) { pe_resource_t *rsc = user_data; rsc->cmds->action_flags((pe_action_t *) data, NULL); } /*! * \internal * \brief Add a clone resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void pcmk__clone_add_actions_to_graph(pe_resource_t *rsc) { CRM_ASSERT(pe_rsc_is_clone(rsc)); g_list_foreach(rsc->actions, call_action_flags, rsc); pe__create_clone_notifications(rsc); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *child_rsc = (pe_resource_t *) iter->data; child_rsc->cmds->add_actions_to_graph(child_rsc); } pcmk__add_rsc_actions_to_graph(rsc); pe__free_clone_notification_data(rsc); } /*! * \internal * \brief Check whether a resource or any children have been probed on a node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p node is in the known_on table of \p rsc or any of its * children, otherwise false */ static bool rsc_probed_on(const pe_resource_t *rsc, const pe_node_t *node) { if (rsc->children != NULL) { 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_probed_on(child, node)) { return true; } } return false; } if (rsc->known_on != NULL) { 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 (pe__same_node(node, known_node)) { return true; } } } return false; } /*! * \internal * \brief Find clone instance that has been probed on given node * * \param[in] clone Clone resource to check * \param[in] node Node to check * * \return Instance of \p clone that has been probed on \p node if any, * otherwise NULL */ static pe_resource_t * find_probed_instance_on(const pe_resource_t *clone, const pe_node_t *node) { for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; if (rsc_probed_on(instance, node)) { return instance; } } return NULL; } /*! * \internal * \brief Probe an anonymous clone on a node * * \param[in,out] clone Anonymous clone to probe * \param[in,out] node Node to probe \p clone on */ static bool probe_anonymous_clone(pe_resource_t *clone, pe_node_t *node) { // Check whether we already probed an instance on this node pe_resource_t *child = find_probed_instance_on(clone, node); // Otherwise, check if we plan to start an instance on this node for (GList *iter = clone->children; (iter != NULL) && (child == NULL); iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; const pe_node_t *instance_node = NULL; instance_node = instance->fns->location(instance, NULL, 0); if (pe__same_node(instance_node, node)) { child = instance; } } // Otherwise, use the first clone instance if (child == NULL) { child = clone->children->data; } // Anonymous clones only need to probe a single instance 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 pcmk__clone_create_probe(pe_resource_t *rsc, pe_node_t *node) { CRM_ASSERT((node != NULL) && pe_rsc_is_clone(rsc)); if (rsc->exclusive_discover) { /* The clone is configured to be probed only where a location constraint * exists with resource-discovery set to exclusive. * * This check is not strictly necessary here since the instance's * create_probe() method would also check, but doing it here is more * efficient (especially for unique clones with a large number of * instances), and affects the CRM_meta_notify_available_uname variable * passed with notify actions. */ pe_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if ((allowed == NULL) || (allowed->rsc_discover_mode != pe_discover_exclusive)) { /* This node is not marked for resource discovery. Remove it from * allowed_nodes so that notifications contain only nodes that the * clone can possibly run on. */ pe_rsc_trace(rsc, "Skipping probe for %s on %s because resource has " "exclusive discovery but is not allowed on node", rsc->id, pe__node_name(node)); g_hash_table_remove(rsc->allowed_nodes, node->details->id); return false; } } rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance_number); if (pcmk_is_set(rsc->flags, pe_rsc_unique)) { return pcmk__probe_resource_list(rsc->children, node); } else { return probe_anonymous_clone(rsc, node); } } /*! * \internal * \brief Add meta-attributes relevant to transition graph actions to XML * * Add clone-specific meta-attributes needed for transition graph actions. * * \param[in] rsc Clone resource whose meta-attributes should be added * \param[in,out] xml Transition graph action attributes XML to add to */ void pcmk__clone_add_graph_meta(const pe_resource_t *rsc, xmlNode *xml) { char *name = NULL; CRM_ASSERT(pe_rsc_is_clone(rsc) && (xml != NULL)); 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, pe__clone_max(rsc)); free(name); name = crm_meta_name(XML_RSC_ATTR_INCARNATION_NODEMAX); crm_xml_add_int(xml, name, pe__clone_node_max(rsc)); 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; CRM_ASSERT(pe_rsc_is_clone(rsc) && (orig_rsc != NULL) && (utilization != 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) { CRM_ASSERT(pe_rsc_is_clone(rsc)); return; // Clones currently don't support shutdown locks } diff --git a/lib/pacemaker/pcmk_sched_group.c b/lib/pacemaker/pcmk_sched_group.c index e363af0a8a..ac18a8281f 100644 --- a/lib/pacemaker/pcmk_sched_group.c +++ b/lib/pacemaker/pcmk_sched_group.c @@ -1,947 +1,948 @@ /* * 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" /*! * \internal * \brief Assign a group resource to a node * * \param[in,out] rsc Group resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * \param[in] stop_if_fail If \c true and a child of \p rsc can't be * assigned to a node, set the child's next role to * stopped and update existing actions * * \return Node that \p rsc is assigned to, if assigned entirely to one node * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ pe_node_t * pcmk__group_assign(pe_resource_t *rsc, const pe_node_t *prefer, bool stop_if_fail) { pe_node_t *first_assigned_node = NULL; pe_resource_t *first_member = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)); if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return rsc->allocated_to; // Assignment already done } if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { pe_rsc_debug(rsc, "Assignment dependency loop detected involving %s", rsc->id); return NULL; } if (rsc->children == NULL) { // No members to assign pe__clear_resource_flags(rsc, pe_rsc_provisional); return NULL; } pe__set_resource_flags(rsc, pe_rsc_allocating); first_member = (pe_resource_t *) rsc->children->data; rsc->role = first_member->role; - pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), + pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, + pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; pe_node_t *node = NULL; pe_rsc_trace(rsc, "Assigning group %s member %s", rsc->id, member->id); node = member->cmds->assign(member, prefer, stop_if_fail); if (first_assigned_node == NULL) { first_assigned_node = node; } } pe__set_next_role(rsc, first_member->next_role, "first group member"); pe__clear_resource_flags(rsc, pe_rsc_allocating|pe_rsc_provisional); if (!pe__group_flag_is_set(rsc, pe__group_colocated)) { return NULL; } return first_assigned_node; } /*! * \internal * \brief Create a pseudo-operation for a group as an ordering point * * \param[in,out] group Group resource to create action for * \param[in] action Action name * * \return Newly created pseudo-operation */ static pe_action_t * create_group_pseudo_op(pe_resource_t *group, const char *action) { pe_action_t *op = custom_action(group, pcmk__op_key(group->id, action, 0), action, NULL, TRUE, TRUE, group->cluster); pe__set_action_flags(op, pe_action_pseudo|pe_action_runnable); return op; } /*! * \internal * \brief Create all actions needed for a given group resource * * \param[in,out] rsc Group resource to create actions for */ void pcmk__group_create_actions(pe_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)); pe_rsc_trace(rsc, "Creating actions for group %s", rsc->id); // Create actions for individual group members for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; member->cmds->create_actions(member); } // Create pseudo-actions for group itself to serve as ordering points create_group_pseudo_op(rsc, PCMK_ACTION_START); create_group_pseudo_op(rsc, PCMK_ACTION_RUNNING); create_group_pseudo_op(rsc, PCMK_ACTION_STOP); create_group_pseudo_op(rsc, PCMK_ACTION_STOPPED); if (crm_is_true(g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_PROMOTABLE))) { create_group_pseudo_op(rsc, PCMK_ACTION_DEMOTE); create_group_pseudo_op(rsc, PCMK_ACTION_DEMOTED); create_group_pseudo_op(rsc, PCMK_ACTION_PROMOTE); create_group_pseudo_op(rsc, PCMK_ACTION_PROMOTED); } } // User data for member_internal_constraints() struct member_data { // These could be derived from member but this avoids some function calls bool ordered; bool colocated; bool promotable; pe_resource_t *last_active; pe_resource_t *previous_member; }; /*! * \internal * \brief Create implicit constraints needed for a group member * * \param[in,out] data Group member to create implicit constraints for * \param[in,out] user_data Member data (struct member_data *) */ static void member_internal_constraints(gpointer data, gpointer user_data) { pe_resource_t *member = (pe_resource_t *) data; struct member_data *member_data = (struct member_data *) user_data; // For ordering demote vs demote or stop vs stop uint32_t down_flags = pe_order_implies_first_printed; // For ordering demote vs demoted or stop vs stopped uint32_t post_down_flags = pe_order_implies_then_printed; // Create the individual member's implicit constraints member->cmds->internal_constraints(member); if (member_data->previous_member == NULL) { // This is first member if (member_data->ordered) { pe__set_order_flags(down_flags, pe_order_optional); post_down_flags = pe_order_implies_then; } } else if (member_data->colocated) { uint32_t flags = pcmk__coloc_none; if (pcmk_is_set(member->flags, pe_rsc_critical)) { flags |= pcmk__coloc_influence; } // Colocate this member with the previous one pcmk__new_colocation("#group-members", NULL, INFINITY, member, member_data->previous_member, NULL, NULL, flags); } if (member_data->promotable) { // Demote group -> demote member -> group is demoted pcmk__order_resource_actions(member->parent, PCMK_ACTION_DEMOTE, member, PCMK_ACTION_DEMOTE, down_flags); pcmk__order_resource_actions(member, PCMK_ACTION_DEMOTE, member->parent, PCMK_ACTION_DEMOTED, post_down_flags); // Promote group -> promote member -> group is promoted pcmk__order_resource_actions(member, PCMK_ACTION_PROMOTE, member->parent, PCMK_ACTION_PROMOTED, pe_order_runnable_left |pe_order_implies_then |pe_order_implies_then_printed); pcmk__order_resource_actions(member->parent, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pe_order_implies_first_printed); } // Stop group -> stop member -> group is stopped pcmk__order_stops(member->parent, member, down_flags); pcmk__order_resource_actions(member, PCMK_ACTION_STOP, member->parent, PCMK_ACTION_STOPPED, post_down_flags); // Start group -> start member -> group is started pcmk__order_starts(member->parent, member, pe_order_implies_first_printed); pcmk__order_resource_actions(member, PCMK_ACTION_START, member->parent, PCMK_ACTION_RUNNING, pe_order_runnable_left |pe_order_implies_then |pe_order_implies_then_printed); if (!member_data->ordered) { pcmk__order_starts(member->parent, member, pe_order_implies_then |pe_order_runnable_left |pe_order_implies_first_printed); if (member_data->promotable) { pcmk__order_resource_actions(member->parent, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pe_order_implies_then |pe_order_runnable_left |pe_order_implies_first_printed); } } else if (member_data->previous_member == NULL) { pcmk__order_starts(member->parent, member, pe_order_none); if (member_data->promotable) { pcmk__order_resource_actions(member->parent, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pe_order_none); } } else { // Order this member relative to the previous one pcmk__order_starts(member_data->previous_member, member, pe_order_implies_then|pe_order_runnable_left); pcmk__order_stops(member, member_data->previous_member, pe_order_optional|pe_order_restart); /* In unusual circumstances (such as adding a new member to the middle * of a group with unmanaged later members), this member may be active * while the previous (new) member is inactive. In this situation, the * usual restart orderings will be irrelevant, so we need to order this * member's stop before the previous member's start. */ if ((member->running_on != NULL) && (member_data->previous_member->running_on == NULL)) { pcmk__order_resource_actions(member, PCMK_ACTION_STOP, member_data->previous_member, PCMK_ACTION_START, pe_order_implies_first |pe_order_runnable_left); } if (member_data->promotable) { pcmk__order_resource_actions(member_data->previous_member, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pe_order_implies_then |pe_order_runnable_left); pcmk__order_resource_actions(member, PCMK_ACTION_DEMOTE, member_data->previous_member, PCMK_ACTION_DEMOTE, pe_order_optional); } } // Make sure partially active groups shut down in sequence if (member->running_on != NULL) { if (member_data->ordered && (member_data->previous_member != NULL) && (member_data->previous_member->running_on == NULL) && (member_data->last_active != NULL) && (member_data->last_active->running_on != NULL)) { pcmk__order_stops(member, member_data->last_active, pe_order_optional); } member_data->last_active = member; } member_data->previous_member = member; } /*! * \internal * \brief Create implicit constraints needed for a group resource * * \param[in,out] rsc Group resource to create implicit constraints for */ void pcmk__group_internal_constraints(pe_resource_t *rsc) { struct member_data member_data = { false, }; const pe_resource_t *top = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)); /* Order group pseudo-actions relative to each other for restarting: * stop group -> group is stopped -> start group -> group is started */ pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, rsc, PCMK_ACTION_STOPPED, pe_order_runnable_left); pcmk__order_resource_actions(rsc, PCMK_ACTION_STOPPED, rsc, PCMK_ACTION_START, pe_order_optional); pcmk__order_resource_actions(rsc, PCMK_ACTION_START, rsc, PCMK_ACTION_RUNNING, pe_order_runnable_left); top = pe__const_top_resource(rsc, false); member_data.ordered = pe__group_flag_is_set(rsc, pe__group_ordered); member_data.colocated = pe__group_flag_is_set(rsc, pe__group_colocated); member_data.promotable = pcmk_is_set(top->flags, pe_rsc_promotable); g_list_foreach(rsc->children, member_internal_constraints, &member_data); } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint for a group with some other resource, apply the * score to the dependent's allowed node scores (if we are still placing * resources) or priority (if we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent group resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply */ static void colocate_group_with(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation) { pe_resource_t *member = NULL; if (dependent->children == NULL) { return; } pe_rsc_trace(primary, "Processing %s (group %s with %s) for dependent", colocation->id, dependent->id, primary->id); if (pe__group_flag_is_set(dependent, pe__group_colocated)) { // Colocate first member (internal colocations will handle the rest) member = (pe_resource_t *) dependent->children->data; member->cmds->apply_coloc_score(member, primary, colocation, true); return; } if (colocation->score >= INFINITY) { pcmk__config_err("%s: Cannot perform mandatory colocation between " "non-colocated group and %s", dependent->id, primary->id); return; } // Colocate each member individually for (GList *iter = dependent->children; iter != NULL; iter = iter->next) { member = (pe_resource_t *) iter->data; member->cmds->apply_coloc_score(member, primary, colocation, true); } } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint for some other resource with a group, apply the * score to the dependent's allowed node scores (if we are still placing * resources) or priority (if we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary group resource in colocation * \param[in] colocation Colocation constraint to apply */ static void colocate_with_group(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation) { const pe_resource_t *member = NULL; pe_rsc_trace(primary, "Processing colocation %s (%s with group %s) for primary", colocation->id, dependent->id, primary->id); if (pcmk_is_set(primary->flags, pe_rsc_provisional)) { return; } if (pe__group_flag_is_set(primary, pe__group_colocated)) { if (colocation->score >= INFINITY) { /* For mandatory colocations, the entire group must be assignable * (and in the specified role if any), so apply the colocation based * on the last member. */ member = pe__last_group_member(primary); } else if (primary->children != NULL) { /* For optional colocations, whether the group is partially or fully * up doesn't matter, so apply the colocation based on the first * member. */ member = (pe_resource_t *) primary->children->data; } if (member == NULL) { return; // Nothing to colocate with } member->cmds->apply_coloc_score(dependent, member, colocation, false); return; } if (colocation->score >= INFINITY) { pcmk__config_err("%s: Cannot perform mandatory colocation with" " non-colocated group %s", dependent->id, primary->id); return; } // Colocate dependent with each member individually for (const GList *iter = primary->children; iter != NULL; iter = iter->next) { member = iter->data; member->cmds->apply_coloc_score(dependent, member, colocation, false); } } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__group_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if (for_dependent) { colocate_group_with(dependent, primary, colocation); } else { // Method should only be called for primitive dependents CRM_ASSERT(dependent->variant == pcmk_rsc_variant_primitive); colocate_with_group(dependent, primary, colocation); } } /*! * \internal * \brief Return action flags for a given group resource action * * \param[in,out] action Group 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 */ uint32_t pcmk__group_action_flags(pe_action_t *action, const pe_node_t *node) { // Default flags for a group action uint32_t flags = pe_action_optional|pe_action_runnable|pe_action_pseudo; CRM_ASSERT(action != NULL); // Update flags considering each member's own flags for same action for (GList *iter = action->rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; // Check whether member has the same action enum action_tasks task = get_complex_task(member, action->task); const char *task_s = task2text(task); pe_action_t *member_action = find_first_action(member->actions, NULL, task_s, node); if (member_action != NULL) { uint32_t member_flags = member->cmds->action_flags(member_action, node); // Group action is mandatory if any member action is if (pcmk_is_set(flags, pe_action_optional) && !pcmk_is_set(member_flags, pe_action_optional)) { pe_rsc_trace(action->rsc, "%s is mandatory because %s is", action->uuid, member_action->uuid); pe__clear_raw_action_flags(flags, "group action", pe_action_optional); pe__clear_action_flags(action, pe_action_optional); } // Group action is unrunnable if any member action is if (!pcmk__str_eq(task_s, action->task, pcmk__str_none) && pcmk_is_set(flags, pe_action_runnable) && !pcmk_is_set(member_flags, pe_action_runnable)) { pe_rsc_trace(action->rsc, "%s is unrunnable because %s is", action->uuid, member_action->uuid); pe__clear_raw_action_flags(flags, "group action", pe_action_runnable); pe__clear_action_flags(action, pe_action_runnable); } /* Group (pseudo-)actions other than stop or demote are unrunnable * unless every member will do it. */ } else if ((task != pcmk_action_stop) && (task != pcmk_action_demote)) { pe_rsc_trace(action->rsc, "%s is not runnable because %s will not %s", action->uuid, member->id, task_s); pe__clear_raw_action_flags(flags, "group action", pe_action_runnable); } } return flags; } /*! * \internal * \brief Update two actions according to an ordering between them * * Given information about an ordering of two actions, update the actions' flags * (and runnable_before members if appropriate) as appropriate for the ordering. * Effects may cascade to other orderings involving the actions as well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this node * (only used when interleaving instances) * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates (may * include 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__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) { uint32_t changed = pcmk__updated_none; // Group method can be called only on behalf of "then" action CRM_ASSERT((first != NULL) && (then != NULL) && (then->rsc != NULL) && (data_set != NULL)); // Update the actions for the group itself changed |= pcmk__update_ordered_actions(first, then, node, flags, filter, type, data_set); // Update the actions for each group member for (GList *iter = then->rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; pe_action_t *member_action = find_first_action(member->actions, NULL, then->task, node); if (member_action != NULL) { changed |= member->cmds->update_ordered_actions(first, member_action, node, flags, filter, type, data_set); } } return changed; } /*! * \internal * \brief Apply a location constraint to a group's allowed node scores * * \param[in,out] rsc Group resource to apply constraint to * \param[in,out] location Location constraint to apply */ void pcmk__group_apply_location(pe_resource_t *rsc, pe__location_t *location) { GList *node_list_orig = NULL; GList *node_list_copy = NULL; bool reset_scores = true; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group) && (location != NULL)); node_list_orig = location->node_list_rh; node_list_copy = pcmk__copy_node_list(node_list_orig, true); reset_scores = pe__group_flag_is_set(rsc, pe__group_colocated); // Apply the constraint for the group itself (updates node scores) pcmk__apply_location(rsc, location); // Apply the constraint for each member for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; member->cmds->apply_location(member, location); if (reset_scores) { /* The first member of colocated groups needs to use the original * node scores, but subsequent members should work on a copy, since * the first member's scores already incorporate theirs. */ reset_scores = false; location->node_list_rh = node_list_copy; } } location->node_list_rh = node_list_orig; g_list_free_full(node_list_copy, free); } // Group implementation of resource_alloc_functions_t:colocated_resources() GList * pcmk__group_colocated_resources(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *colocated_rscs) { const pe_resource_t *member = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)); if (orig_rsc == NULL) { orig_rsc = rsc; } if (pe__group_flag_is_set(rsc, pe__group_colocated) || pe_rsc_is_clone(rsc->parent)) { /* This group has colocated members and/or is cloned -- either way, * add every child's colocated resources to the list. The first and last * members will include the group's own colocations. */ colocated_rscs = g_list_prepend(colocated_rscs, (gpointer) rsc); for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = (const pe_resource_t *) iter->data; colocated_rscs = member->cmds->colocated_resources(member, orig_rsc, colocated_rscs); } } else if (rsc->children != NULL) { /* This group's members are not colocated, and the group is not cloned, * so just add the group's own colocations to the list. */ colocated_rscs = pcmk__colocated_resources(rsc, orig_rsc, colocated_rscs); } return colocated_rscs; } // Group implementation of resource_alloc_functions_t:with_this_colocations() void pcmk__with_group_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group) && (orig_rsc != NULL) && (list != NULL)); // Ignore empty groups if (rsc->children == NULL) { return; } /* "With this" colocations are needed only for the group itself and for its * last member. (Previous members will chain via the group internal * colocations.) */ if ((orig_rsc != rsc) && (orig_rsc != pe__last_group_member(rsc))) { return; } pe_rsc_trace(rsc, "Adding 'with %s' colocations to list for %s", rsc->id, orig_rsc->id); // Add the group's own colocations pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); // If cloned, add any relevant colocations with the clone if (rsc->parent != NULL) { rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list); } if (!pe__group_flag_is_set(rsc, pe__group_colocated)) { // @COMPAT Non-colocated groups are deprecated return; } // Add explicit colocations with the group's (other) children for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { const pe_resource_t *member = iter->data; if (member != orig_rsc) { member->cmds->with_this_colocations(member, orig_rsc, list); } } } // Group implementation of resource_alloc_functions_t:this_with_colocations() void pcmk__group_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { const pe_resource_t *member = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group) && (orig_rsc != NULL) && (list != NULL)); // Ignore empty groups if (rsc->children == NULL) { return; } /* "This with" colocations are normally needed only for the group itself and * for its first member. */ if ((rsc == orig_rsc) || (orig_rsc == (const pe_resource_t *) rsc->children->data)) { pe_rsc_trace(rsc, "Adding '%s with' colocations to list for %s", rsc->id, orig_rsc->id); // Add the group's own colocations pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); // If cloned, add any relevant colocations involving the clone if (rsc->parent != NULL) { rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list); } if (!pe__group_flag_is_set(rsc, pe__group_colocated)) { // @COMPAT Non-colocated groups are deprecated return; } // Add explicit colocations involving the group's (other) children for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = iter->data; if (member != orig_rsc) { member->cmds->this_with_colocations(member, orig_rsc, list); } } return; } /* Later group members honor the group's colocations indirectly, due to the * internal group colocations that chain everything from the first member. * However, if an earlier group member is unmanaged, this chaining will not * happen, so the group's mandatory colocations must be explicitly added. */ for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = iter->data; if (orig_rsc == member) { break; // We've seen all earlier members, and none are unmanaged } if (!pcmk_is_set(member->flags, pe_rsc_managed)) { crm_trace("Adding mandatory '%s with' colocations to list for " "member %s because earlier member %s is unmanaged", rsc->id, orig_rsc->id, member->id); for (const GList *cons_iter = rsc->rsc_cons; cons_iter != NULL; cons_iter = cons_iter->next) { const pcmk__colocation_t *colocation = NULL; colocation = (const pcmk__colocation_t *) cons_iter->data; if (colocation->score == INFINITY) { pcmk__add_this_with(list, colocation, orig_rsc); } } // @TODO Add mandatory (or all?) clone constraints if cloned break; } } } /*! * \internal * \brief Update nodes with scores of colocated resources' nodes * * Given a table of nodes and a resource, update the nodes' scores with the * scores of the best nodes matching the attribute used for each of the * resource's relevant colocations. * * \param[in,out] source_rsc Group resource whose node scores to add * \param[in] target_rsc Resource on whose behalf to update \p *nodes * \param[in] log_id Resource ID for logs (if \c NULL, use * \p source_rsc ID) * \param[in,out] nodes Nodes to update (set initial contents to \c NULL * to copy allowed nodes from \p source_rsc) * \param[in] colocation Original colocation constraint (used to get * configured primary resource's stickiness, and * to get colocation node attribute; if \c NULL, * source_rsc's own matching node scores will * not be added, and \p *nodes must be \c NULL as * well) * \param[in] factor Incorporate scores multiplied by this factor * \param[in] flags Bitmask of enum pcmk__coloc_select values * * \note \c NULL \p target_rsc, \c NULL \p *nodes, \c NULL \p colocation, and * the \c pcmk__coloc_select_this_with flag are used together (and only by * \c cmp_resources()). * \note The caller remains responsible for freeing \p *nodes. * \note This is the group implementation of * \c resource_alloc_functions_t:add_colocated_node_scores(). */ void pcmk__group_add_colocated_node_scores(pe_resource_t *source_rsc, const pe_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags) { pe_resource_t *member = NULL; CRM_ASSERT((source_rsc != NULL) && (source_rsc->variant == pcmk_rsc_variant_group) && (nodes != NULL) && ((colocation != NULL) || ((target_rsc == NULL) && (*nodes == NULL)))); if (log_id == NULL) { log_id = source_rsc->id; } // Avoid infinite recursion if (pcmk_is_set(source_rsc->flags, pe_rsc_merging)) { pe_rsc_info(source_rsc, "%s: Breaking dependency loop at %s", log_id, source_rsc->id); return; } pe__set_resource_flags(source_rsc, pe_rsc_merging); // Ignore empty groups (only possible with schema validation disabled) if (source_rsc->children == NULL) { return; } /* Refer the operation to the first or last member as appropriate. * * cmp_resources() is the only caller that passes a NULL nodes table, * and is also the only caller using pcmk__coloc_select_this_with. * For "this with" colocations, the last member will recursively incorporate * all the other members' "this with" colocations via the internal group * colocations (and via the first member, the group's own colocations). * * For "with this" colocations, the first member works similarly. */ if (*nodes == NULL) { member = pe__last_group_member(source_rsc); } else { member = source_rsc->children->data; } pe_rsc_trace(source_rsc, "%s: Merging scores from group %s using member %s " "(at %.6f)", log_id, source_rsc->id, member->id, factor); member->cmds->add_colocated_node_scores(member, target_rsc, log_id, nodes, colocation, factor, flags); pe__clear_resource_flags(source_rsc, pe_rsc_merging); } // Group implementation of resource_alloc_functions_t:add_utilization() void pcmk__group_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { pe_resource_t *member = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group) && (orig_rsc != NULL) && (utilization != NULL)); if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return; } pe_rsc_trace(orig_rsc, "%s: Adding group %s as colocated utilization", orig_rsc->id, rsc->id); if (pe__group_flag_is_set(rsc, pe__group_colocated) || pe_rsc_is_clone(rsc->parent)) { // Every group member will be on same node, so sum all members for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = (pe_resource_t *) iter->data; if (pcmk_is_set(member->flags, pe_rsc_provisional) && (g_list_find(all_rscs, member) == NULL)) { member->cmds->add_utilization(member, orig_rsc, all_rscs, utilization); } } } else if (rsc->children != NULL) { // Just add first member's utilization member = (pe_resource_t *) rsc->children->data; if ((member != NULL) && pcmk_is_set(member->flags, pe_rsc_provisional) && (g_list_find(all_rscs, member) == NULL)) { member->cmds->add_utilization(member, orig_rsc, all_rscs, utilization); } } } // Group implementation of resource_alloc_functions_t:shutdown_lock() void pcmk__group_shutdown_lock(pe_resource_t *rsc) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_group)); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; member->cmds->shutdown_lock(member); } } diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c index 96be540b0a..104b8d00ab 100644 --- a/lib/pacemaker/pcmk_sched_primitive.c +++ b/lib/pacemaker/pcmk_sched_primitive.c @@ -1,1650 +1,1651 @@ /* * 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 // uint8_t, uint32_t #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); #define RSC_ROLE_MAX (pcmk_role_promoted + 1) static enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = { /* This array lists the immediate next role when transitioning from one role * to a target role. For example, when going from Stopped to Promoted, the * next role is Unpromoted, because the resource must be started before it * can be promoted. The current state then becomes Started, which is fed * into this array again, giving a next role of Promoted. * * Current role Immediate next role Final target role * ------------ ------------------- ----------------- */ /* Unknown */ { pcmk_role_unknown, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_stopped, /* Started */ pcmk_role_stopped, /* Unpromoted */ pcmk_role_stopped, /* Promoted */ }, /* Stopped */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_started, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_unpromoted, /* Promoted */ }, /* Started */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_started, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, /* Unpromoted */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_stopped, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, /* Promoted */ { pcmk_role_stopped, /* Unknown */ pcmk_role_unpromoted, /* Stopped */ pcmk_role_unpromoted, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, }; /*! * \internal * \brief Function to schedule actions needed for a role change * * \param[in,out] rsc Resource whose role is changing * \param[in,out] node Node where resource will be in its next role * \param[in] optional Whether scheduled actions should be optional */ typedef void (*rsc_transition_fn)(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 score * * \param[in] rsc Resource to check * * \return List of allowed nodes sorted by node score */ 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 \c NULL, prefer this node when all else * equal * \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a * node, set next role to stopped and update * existing actions * * \return true if \p rsc could be assigned to a node, otherwise false * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ static bool assign_best_node(pe_resource_t *rsc, const pe_node_t *prefer, bool stop_if_fail) { GList *nodes = NULL; pe_node_t *chosen = NULL; pe_node_t *best = NULL; 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 score 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 score is at least as good as * the best allowed node's. * * An alternative would be to favor the preferred node even if the best * node is better, when the best node's score is less than INFINITY. */ } else if (chosen->weight < best->weight) { 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 * 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 score, 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) { uint8_t log_level = LOG_INFO; if (chosen->weight >= INFINITY) { log_level = LOG_WARNING; } do_crm_log(log_level, "Chose %s for %s from %d nodes with score %s", 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)); } pcmk__assign_resource(rsc, chosen, false, stop_if_fail); g_list_free(nodes); return rsc->allocated_to != NULL; } /*! * \internal * \brief Apply a "this with" colocation to a node's allowed node scores * * \param[in,out] colocation Colocation to apply * \param[in,out] rsc Resource being assigned */ static void apply_this_with(pcmk__colocation_t *colocation, pe_resource_t *rsc) { GHashTable *archive = NULL; pe_resource_t *other = colocation->primary; // In certain cases, we will need to revert the node scores if ((colocation->dependent_role >= pcmk_role_promoted) || ((colocation->score < 0) && (colocation->score > -INFINITY))) { archive = pcmk__copy_node_table(rsc->allowed_nodes); } if (pcmk_is_set(other->flags, pe_rsc_provisional)) { 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, true); } // Apply the colocation score to this resource's allowed node scores rsc->cmds->apply_coloc_score(rsc, other, colocation, true); if ((archive != NULL) && !pcmk__any_node_available(rsc->allowed_nodes)) { 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 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 != pcmk_role_stopped)) { crm_trace("Pacemaker Remote node %s will be online", remote_node->details->id); remote_node->details->online = TRUE; if (remote_node->details->unseen) { // Avoid unnecessary fence, since we will attempt connection remote_node->details->unclean = FALSE; } } else { crm_trace("Pacemaker Remote node %s will be shut down " "(%sassigned connection's next role is %s)", remote_node->details->id, ((connection->allocated_to == NULL)? "un" : ""), 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 * \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a * node, set next role to stopped and update * existing actions * * \return Node that \p rsc is assigned to, if assigned entirely to one node * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ pe_node_t * pcmk__primitive_assign(pe_resource_t *rsc, const pe_node_t *prefer, bool stop_if_fail) { GList *this_with_colocations = NULL; GList *with_this_colocations = NULL; GList *iter = NULL; pcmk__colocation_t *colocation = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); // Never assign a child without parent being assigned first if ((rsc->parent != NULL) && !pcmk_is_set(rsc->parent->flags, pe_rsc_allocating)) { pe_rsc_debug(rsc, "%s: Assigning parent %s first", rsc->id, rsc->parent->id); rsc->parent->cmds->assign(rsc->parent, prefer, stop_if_fail); } if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { // Assignment has already been done const char *node_name = "no node"; if (rsc->allocated_to != NULL) { node_name = pe__node_name(rsc->allocated_to); } pe_rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name); return rsc->allocated_to; } // Ensure we detect assignment loops if (pcmk_is_set(rsc->flags, 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_scores(true, rsc, "Pre-assignment", rsc->allowed_nodes, rsc->cluster); this_with_colocations = pcmk__this_with_colocations(rsc); with_this_colocations = pcmk__with_this_colocations(rsc); // Apply mandatory colocations first, to satisfy as many as possible for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score <= -CRM_SCORE_INFINITY) || (colocation->score >= CRM_SCORE_INFINITY)) { apply_this_with(colocation, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score <= -CRM_SCORE_INFINITY) || (colocation->score >= CRM_SCORE_INFINITY)) { pcmk__add_dependent_scores(colocation, rsc); } } pe__show_node_scores(true, rsc, "Mandatory-colocations", rsc->allowed_nodes, rsc->cluster); // Then apply optional colocations for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score > -CRM_SCORE_INFINITY) && (colocation->score < CRM_SCORE_INFINITY)) { apply_this_with(colocation, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score > -CRM_SCORE_INFINITY) && (colocation->score < CRM_SCORE_INFINITY)) { pcmk__add_dependent_scores(colocation, rsc); } } g_list_free(this_with_colocations); g_list_free(with_this_colocations); if (rsc->next_role == pcmk_role_stopped) { 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, pcmk_sched_quorate) && (rsc->cluster->no_quorum_policy == pcmk_no_quorum_freeze)) { crm_notice("Resource %s cannot be elevated from %s to %s due to " "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_scores(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), + pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags, + pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); // Unmanage resource if fencing is enabled but no device is configured if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled) && !pcmk_is_set(rsc->cluster->flags, pcmk_sched_have_fencing)) { 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 == pcmk_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__assign_resource(rsc, assign_to, true, stop_if_fail); } else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_stop_all)) { // Must stop at some point, but be consistent with stop_if_fail if (stop_if_fail) { pe_rsc_debug(rsc, "Forcing %s to stop: stop-all-resources", rsc->id); } pcmk__assign_resource(rsc, NULL, true, stop_if_fail); } else if (!assign_best_node(rsc, prefer, stop_if_fail)) { // Assignment failed if (!pcmk_is_set(rsc->flags, pe_rsc_orphan)) { pe_rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id); } else if ((rsc->running_on != NULL) && stop_if_fail) { pe_rsc_info(rsc, "Stopping orphan resource %s", rsc->id); } } 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 != pcmk_role_stopped) { next_role = rsc_state_matrix[role][pcmk_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 == pcmk_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 != pcmk_role_unknown) { return "explicit"; } if (rsc->allocated_to == NULL) { pe__set_next_role(rsc, pcmk_role_stopped, "assignment"); } else { pe__set_next_role(rsc, pcmk_role_started, "assignment"); } return "implicit"; } /*! * \internal * \brief Create an action to represent an already pending start * * \param[in,out] rsc Resource to create start action for */ static void create_pending_start(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) && (rsc->variant == pcmk_rsc_variant_primitive)); 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 = rsc->fns->active_node(rsc, &num_all_active, &num_clean_active); g_list_foreach(rsc->dangling_migrations, pcmk__abort_dangling_migration, rsc); if ((current != NULL) && (rsc->allocated_to != NULL) && !pe__same_node(current, rsc->allocated_to) && (rsc->next_role >= pcmk_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 pcmk_multiply_active_restart: need_stop = true; break; case pcmk_multiply_active_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 == pcmk_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 > pcmk_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) && (rsc->variant == pcmk_rsc_variant_primitive)); 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, pcmk_sched_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, PCMK_ACTION_STOP, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_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 > pcmk_role_unpromoted)) { pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_DEMOTE, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, pe_order_promoted_implies_first, rsc->cluster); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_PROMOTE, 0), NULL, 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, PCMK_ACTION_LRM_DELETE, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 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, PCMK_ACTION_MONITOR, rsc, PCMK_ACTION_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, PCMK_ACTION_START, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, pe_order_implies_then|pe_order_runnable_left, rsc->cluster); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, rsc->container, pcmk__op_key(rsc->container->id, PCMK_ACTION_STOP, 0), NULL, 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, pcmk__coloc_influence); } } 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 scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__primitive_apply_coloc_score(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { enum pcmk__coloc_affects filter_results; CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if (for_dependent) { // Always process on behalf of primary resource primary->cmds->apply_coloc_score(dependent, primary, colocation, false); return; } filter_results = pcmk__colocation_affects(dependent, primary, colocation, false); 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_scores(dependent, primary, colocation); break; default: // pcmk__coloc_affects_nothing return; } } /* Primitive implementation of * resource_alloc_functions_t:with_this_colocations() */ void pcmk__with_primitive_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (list != NULL)); if (rsc == orig_rsc) { /* For the resource itself, add all of its own colocations and relevant * colocations from its parent (if any). */ pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc); if (rsc->parent != NULL) { rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list); } } else { // For an ancestor, add only explicitly configured constraints for (GList *iter = rsc->rsc_cons_lhs; iter != NULL; iter = iter->next) { pcmk__colocation_t *colocation = iter->data; if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) { pcmk__add_with_this(list, colocation, orig_rsc); } } } } /* Primitive implementation of * resource_alloc_functions_t:this_with_colocations() */ void pcmk__primitive_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (list != NULL)); if (rsc == orig_rsc) { /* For the resource itself, add all of its own colocations and relevant * colocations from its parent (if any). */ pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc); if (rsc->parent != NULL) { rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list); } } else { // For an ancestor, add only explicitly configured constraints for (GList *iter = rsc->rsc_cons; iter != NULL; iter = iter->next) { pcmk__colocation_t *colocation = iter->data; if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) { pcmk__add_this_with(list, colocation, orig_rsc); } } } } /*! * \internal * \brief Return action flags for a given primitive resource action * * \param[in,out] action Action to get flags for * \param[in] node If not NULL, limit effects to this node (ignored) * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__primitive_action_flags(pe_action_t *action, const pe_node_t *node) { CRM_ASSERT(action != NULL); return (uint32_t) action->flags; } /*! * \internal * \brief Check whether a node is a multiply active resource's expected node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return 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 > pcmk_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, pcmk_sched_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, PCMK_ACTION_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, PCMK_ACTION_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, PCMK_ACTION_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, PCMK_ACTION_STOP, rsc, PCMK_ACTION_DELETE, flag); pcmk__order_resource_actions(rsc, PCMK_ACTION_DELETE, rsc, PCMK_ACTION_START, flag); } /*! * \internal * \brief Add primitive meta-attributes relevant to graph actions to XML * * \param[in] rsc Primitive resource whose meta-attributes should be added * \param[in,out] xml Transition graph action attributes XML to add to */ void pcmk__primitive_add_graph_meta(const pe_resource_t *rsc, xmlNode *xml) { char *name = NULL; char *value = NULL; const pe_resource_t *parent = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (xml != NULL)); /* Clone instance numbers get set internally as meta-attributes, and are * needed in the transition graph (for example, to tell unique clone * instances apart). */ value = g_hash_table_lookup(rsc->meta, 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) { CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive) && (orig_rsc != NULL) && (utilization != NULL)); 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 = NULL; CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)); 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)) { 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); } diff --git a/lib/pacemaker/pcmk_sched_promotable.c b/lib/pacemaker/pcmk_sched_promotable.c index 2da1c2686c..dddad856fe 100644 --- a/lib/pacemaker/pcmk_sched_promotable.c +++ b/lib/pacemaker/pcmk_sched_promotable.c @@ -1,1299 +1,1299 @@ /* * 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" /*! * \internal * \brief Add implicit promotion ordering for a promotable instance * * \param[in,out] clone Clone resource * \param[in,out] child Instance of \p clone being ordered * \param[in,out] last Previous instance ordered (NULL if \p child is first) */ static void order_instance_promotion(pe_resource_t *clone, pe_resource_t *child, pe_resource_t *last) { // "Promote clone" -> promote instance -> "clone promoted" pcmk__order_resource_actions(clone, PCMK_ACTION_PROMOTE, child, PCMK_ACTION_PROMOTE, pe_order_optional); pcmk__order_resource_actions(child, PCMK_ACTION_PROMOTE, clone, PCMK_ACTION_PROMOTED, pe_order_optional); // If clone is ordered, order this instance relative to last if ((last != NULL) && pe__clone_is_ordered(clone)) { pcmk__order_resource_actions(last, PCMK_ACTION_PROMOTE, child, PCMK_ACTION_PROMOTE, pe_order_optional); } } /*! * \internal * \brief Add implicit demotion ordering for a promotable instance * * \param[in,out] clone Clone resource * \param[in,out] child Instance of \p clone being ordered * \param[in] last Previous instance ordered (NULL if \p child is first) */ static void order_instance_demotion(pe_resource_t *clone, pe_resource_t *child, pe_resource_t *last) { // "Demote clone" -> demote instance -> "clone demoted" pcmk__order_resource_actions(clone, PCMK_ACTION_DEMOTE, child, PCMK_ACTION_DEMOTE, pe_order_implies_first_printed); pcmk__order_resource_actions(child, PCMK_ACTION_DEMOTE, clone, PCMK_ACTION_DEMOTED, pe_order_implies_then_printed); // If clone is ordered, order this instance relative to last if ((last != NULL) && pe__clone_is_ordered(clone)) { pcmk__order_resource_actions(child, PCMK_ACTION_DEMOTE, last, PCMK_ACTION_DEMOTE, pe_order_optional); } } /*! * \internal * \brief Check whether an instance will be promoted or demoted * * \param[in] rsc Instance to check * \param[out] demoting If \p rsc will be demoted, this will be set to true * \param[out] promoting If \p rsc will be promoted, this will be set to true */ static void check_for_role_change(const pe_resource_t *rsc, bool *demoting, bool *promoting) { const GList *iter = NULL; // If this is a cloned group, check group members recursively if (rsc->children != NULL) { for (iter = rsc->children; iter != NULL; iter = iter->next) { check_for_role_change((const pe_resource_t *) iter->data, demoting, promoting); } return; } for (iter = rsc->actions; iter != NULL; iter = iter->next) { const pe_action_t *action = (const pe_action_t *) iter->data; if (*promoting && *demoting) { return; } else if (pcmk_is_set(action->flags, pe_action_optional)) { continue; } else if (pcmk__str_eq(PCMK_ACTION_DEMOTE, action->task, pcmk__str_none)) { *demoting = true; } else if (pcmk__str_eq(PCMK_ACTION_PROMOTE, action->task, pcmk__str_none)) { *promoting = true; } } } /*! * \internal * \brief Add promoted-role location constraint scores to an instance's priority * * Adjust a promotable clone instance's promotion priority by the scores of any * location constraints in a list that are both limited to the promoted role and * for the node where the instance will be placed. * * \param[in,out] child Promotable clone instance * \param[in] location_constraints List of location constraints to apply * \param[in] chosen Node where \p child will be placed */ static void apply_promoted_locations(pe_resource_t *child, const GList *location_constraints, const pe_node_t *chosen) { for (const GList *iter = location_constraints; iter; iter = iter->next) { const pe__location_t *location = iter->data; const pe_node_t *constraint_node = NULL; if (location->role_filter == pcmk_role_promoted) { constraint_node = pe_find_node_id(location->node_list_rh, chosen->details->id); } if (constraint_node != NULL) { int new_priority = pcmk__add_scores(child->priority, constraint_node->weight); pe_rsc_trace(child, "Applying location %s to %s promotion priority on %s: " "%s + %s = %s", location->id, child->id, pe__node_name(constraint_node), pcmk_readable_score(child->priority), pcmk_readable_score(constraint_node->weight), pcmk_readable_score(new_priority)); child->priority = new_priority; } } } /*! * \internal * \brief Get the node that an instance will be promoted on * * \param[in] rsc Promotable clone instance to check * * \return Node that \p rsc will be promoted on, or NULL if none */ static pe_node_t * node_to_be_promoted_on(const pe_resource_t *rsc) { pe_node_t *node = NULL; pe_node_t *local_node = NULL; const pe_resource_t *parent = NULL; // If this is a cloned group, bail if any group member can't be promoted for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *child = (pe_resource_t *) iter->data; if (node_to_be_promoted_on(child) == NULL) { pe_rsc_trace(rsc, "%s can't be promoted because member %s can't", rsc->id, child->id); return NULL; } } node = rsc->fns->location(rsc, NULL, FALSE); if (node == NULL) { pe_rsc_trace(rsc, "%s can't be promoted because it won't be active", rsc->id); return NULL; } else if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { if (rsc->fns->state(rsc, TRUE) == pcmk_role_promoted) { crm_notice("Unmanaged instance %s will be left promoted on %s", rsc->id, pe__node_name(node)); } else { pe_rsc_trace(rsc, "%s can't be promoted because it is unmanaged", rsc->id); return NULL; } } else if (rsc->priority < 0) { pe_rsc_trace(rsc, "%s can't be promoted because its promotion priority %d " "is negative", rsc->id, rsc->priority); return NULL; } else if (!pcmk__node_available(node, false, true)) { pe_rsc_trace(rsc, "%s can't be promoted because %s can't run resources", rsc->id, pe__node_name(node)); return NULL; } parent = pe__const_top_resource(rsc, false); local_node = g_hash_table_lookup(parent->allowed_nodes, node->details->id); if (local_node == NULL) { /* It should not be possible for the scheduler to have assigned the * instance to a node where its parent is not allowed, but it's good to * have a fail-safe. */ if (pcmk_is_set(rsc->flags, pe_rsc_managed)) { crm_warn("%s can't be promoted because %s is not allowed on %s " "(scheduler bug?)", rsc->id, parent->id, pe__node_name(node)); } // else the instance is unmanaged and already promoted return NULL; } else if ((local_node->count >= pe__clone_promoted_node_max(parent)) && pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe_rsc_trace(rsc, "%s can't be promoted because %s has " "maximum promoted instances already", rsc->id, pe__node_name(node)); return NULL; } return local_node; } /*! * \internal * \brief Compare two promotable clone instances by promotion priority * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a has higher promotion priority, * a positive number if \p b has higher promotion priority, * or 0 if promotion priorities are equal */ static gint cmp_promotable_instance(gconstpointer a, gconstpointer b) { const pe_resource_t *rsc1 = (const pe_resource_t *) a; const pe_resource_t *rsc2 = (const pe_resource_t *) b; enum rsc_role_e role1 = pcmk_role_unknown; enum rsc_role_e role2 = pcmk_role_unknown; CRM_ASSERT((rsc1 != NULL) && (rsc2 != NULL)); // Check sort index set by pcmk__set_instance_roles() if (rsc1->sort_index > rsc2->sort_index) { pe_rsc_trace(rsc1, "%s has higher promotion priority than %s " "(sort index %d > %d)", rsc1->id, rsc2->id, rsc1->sort_index, rsc2->sort_index); return -1; } else if (rsc1->sort_index < rsc2->sort_index) { pe_rsc_trace(rsc1, "%s has lower promotion priority than %s " "(sort index %d < %d)", rsc1->id, rsc2->id, rsc1->sort_index, rsc2->sort_index); return 1; } // If those are the same, prefer instance whose current role is higher role1 = rsc1->fns->state(rsc1, TRUE); role2 = rsc2->fns->state(rsc2, TRUE); if (role1 > role2) { pe_rsc_trace(rsc1, "%s has higher promotion priority than %s " "(higher current role)", rsc1->id, rsc2->id); return -1; } else if (role1 < role2) { pe_rsc_trace(rsc1, "%s has lower promotion priority than %s " "(lower current role)", rsc1->id, rsc2->id); return 1; } // Finally, do normal clone instance sorting return pcmk__cmp_instance(a, b); } /*! * \internal * \brief Add a promotable clone instance's sort index to its node's score * * Add a promotable clone instance's sort index (which sums its promotion * preferences and scores of relevant location constraints for the promoted * role) to the node score of the instance's assigned node. * * \param[in] data Promotable clone instance * \param[in,out] user_data Clone parent of \p data */ static void add_sort_index_to_node_score(gpointer data, gpointer user_data) { const pe_resource_t *child = (const pe_resource_t *) data; pe_resource_t *clone = (pe_resource_t *) user_data; pe_node_t *node = NULL; const pe_node_t *chosen = NULL; if (child->sort_index < 0) { pe_rsc_trace(clone, "Not adding sort index of %s: negative", child->id); return; } chosen = child->fns->location(child, NULL, FALSE); if (chosen == NULL) { pe_rsc_trace(clone, "Not adding sort index of %s: inactive", child->id); return; } node = g_hash_table_lookup(clone->allowed_nodes, chosen->details->id); CRM_ASSERT(node != NULL); node->weight = pcmk__add_scores(child->sort_index, node->weight); pe_rsc_trace(clone, "Added cumulative priority of %s (%s) to score on %s (now %s)", child->id, pcmk_readable_score(child->sort_index), pe__node_name(node), pcmk_readable_score(node->weight)); } /*! * \internal * \brief Apply colocation to dependent's node scores if for promoted role * * \param[in,out] data Colocation constraint to apply * \param[in,out] user_data Promotable clone that is constraint's dependent */ static void apply_coloc_to_dependent(gpointer data, gpointer user_data) { pcmk__colocation_t *colocation = data; pe_resource_t *clone = user_data; pe_resource_t *primary = colocation->primary; uint32_t flags = pcmk__coloc_select_default; float factor = colocation->score / (float) INFINITY; if (colocation->dependent_role != pcmk_role_promoted) { return; } if (colocation->score < INFINITY) { flags = pcmk__coloc_select_active; } pe_rsc_trace(clone, "Applying colocation %s (promoted %s with %s) @%s", colocation->id, colocation->dependent->id, colocation->primary->id, pcmk_readable_score(colocation->score)); primary->cmds->add_colocated_node_scores(primary, clone, clone->id, &clone->allowed_nodes, colocation, factor, flags); } /*! * \internal * \brief Apply colocation to primary's node scores if for promoted role * * \param[in,out] data Colocation constraint to apply * \param[in,out] user_data Promotable clone that is constraint's primary */ static void apply_coloc_to_primary(gpointer data, gpointer user_data) { pcmk__colocation_t *colocation = data; pe_resource_t *clone = user_data; pe_resource_t *dependent = colocation->dependent; const float factor = colocation->score / (float) INFINITY; const uint32_t flags = pcmk__coloc_select_active |pcmk__coloc_select_nonnegative; if ((colocation->primary_role != pcmk_role_promoted) || !pcmk__colocation_has_influence(colocation, NULL)) { return; } pe_rsc_trace(clone, "Applying colocation %s (%s with promoted %s) @%s", colocation->id, colocation->dependent->id, colocation->primary->id, pcmk_readable_score(colocation->score)); dependent->cmds->add_colocated_node_scores(dependent, clone, clone->id, &clone->allowed_nodes, colocation, factor, flags); } /*! * \internal * \brief Set clone instance's sort index to its node's score * * \param[in,out] data Promotable clone instance * \param[in] user_data Parent clone of \p data */ static void set_sort_index_to_node_score(gpointer data, gpointer user_data) { pe_resource_t *child = (pe_resource_t *) data; const pe_resource_t *clone = (const pe_resource_t *) user_data; pe_node_t *chosen = child->fns->location(child, NULL, FALSE); if (!pcmk_is_set(child->flags, pe_rsc_managed) && (child->next_role == pcmk_role_promoted)) { child->sort_index = INFINITY; pe_rsc_trace(clone, "Final sort index for %s is INFINITY (unmanaged promoted)", child->id); } else if ((chosen == NULL) || (child->sort_index < 0)) { pe_rsc_trace(clone, "Final sort index for %s is %d (ignoring node score)", child->id, child->sort_index); } else { const pe_node_t *node = g_hash_table_lookup(clone->allowed_nodes, chosen->details->id); CRM_ASSERT(node != NULL); child->sort_index = node->weight; pe_rsc_trace(clone, "Adding scores for %s: final sort index for %s is %d", clone->id, child->id, child->sort_index); } } /*! * \internal * \brief Sort a promotable clone's instances by descending promotion priority * * \param[in,out] clone Promotable clone to sort */ static void sort_promotable_instances(pe_resource_t *clone) { GList *colocations = NULL; if (pe__set_clone_flag(clone, pe__clone_promotion_constrained) == pcmk_rc_already) { return; } pe__set_resource_flags(clone, pe_rsc_merging); for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pe_resource_t *child = (pe_resource_t *) iter->data; pe_rsc_trace(clone, "Adding scores for %s: initial sort index for %s is %d", clone->id, child->id, child->sort_index); } pe__show_node_scores(true, clone, "Before", clone->allowed_nodes, clone->cluster); g_list_foreach(clone->children, add_sort_index_to_node_score, clone); colocations = pcmk__this_with_colocations(clone); g_list_foreach(colocations, apply_coloc_to_dependent, clone); g_list_free(colocations); colocations = pcmk__with_this_colocations(clone); g_list_foreach(colocations, apply_coloc_to_primary, clone); g_list_free(colocations); // Ban resource from all nodes if it needs a ticket but doesn't have it pcmk__require_promotion_tickets(clone); pe__show_node_scores(true, clone, "After", clone->allowed_nodes, clone->cluster); // Reset sort indexes to final node scores g_list_foreach(clone->children, set_sort_index_to_node_score, clone); // Finally, sort instances in descending order of promotion priority clone->children = g_list_sort(clone->children, cmp_promotable_instance); pe__clear_resource_flags(clone, pe_rsc_merging); } /*! * \internal * \brief Find the active instance (if any) of an anonymous clone on a node * * \param[in] clone Anonymous clone to check * \param[in] id Instance ID (without instance number) to check * \param[in] node Node to check * * \return */ static pe_resource_t * find_active_anon_instance(const pe_resource_t *clone, const char *id, const pe_node_t *node) { for (GList *iter = clone->children; iter; iter = iter->next) { pe_resource_t *child = iter->data; pe_resource_t *active = NULL; // Use ->find_rsc() in case this is a cloned group active = clone->fns->find_rsc(child, id, node, pcmk_rsc_match_clone_only |pcmk_rsc_match_current_node); if (active != NULL) { return active; } } return NULL; } /* * \brief Check whether an anonymous clone instance is known on a node * * \param[in] clone Anonymous clone to check * \param[in] id Instance ID (without instance number) to check * \param[in] node Node to check * * \return true if \p id instance of \p clone is known on \p node, * otherwise false */ static bool anonymous_known_on(const pe_resource_t *clone, const char *id, const pe_node_t *node) { for (GList *iter = clone->children; iter; iter = iter->next) { pe_resource_t *child = iter->data; /* Use ->find_rsc() because this might be a cloned group, and knowing * that other members of the group are known here implies nothing. */ child = clone->fns->find_rsc(child, id, NULL, pcmk_rsc_match_clone_only); CRM_LOG_ASSERT(child != NULL); if (child != NULL) { if (g_hash_table_lookup(child->known_on, node->details->id)) { return true; } } } return false; } /*! * \internal * \brief Check whether a node is allowed to run a resource * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p node is allowed to run \p rsc, otherwise false */ static bool is_allowed(const pe_resource_t *rsc, const pe_node_t *node) { pe_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); return (allowed != NULL) && (allowed->weight >= 0); } /*! * \brief Check whether a clone instance's promotion score should be considered * * \param[in] rsc Promotable clone instance to check * \param[in] node Node where score would be applied * * \return true if \p rsc's promotion score should be considered on \p node, * otherwise false */ static bool promotion_score_applies(const pe_resource_t *rsc, const pe_node_t *node) { char *id = clone_strip(rsc->id); const pe_resource_t *parent = pe__const_top_resource(rsc, false); pe_resource_t *active = NULL; const char *reason = "allowed"; // Some checks apply only to anonymous clone instances if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) { // If instance is active on the node, its score definitely applies active = find_active_anon_instance(parent, id, node); if (active == rsc) { reason = "active"; goto check_allowed; } /* If *no* instance is active on this node, this instance's score will * count if it has been probed on this node. */ if ((active == NULL) && anonymous_known_on(parent, id, node)) { reason = "probed"; goto check_allowed; } } /* If this clone's status is unknown on *all* nodes (e.g. cluster startup), * take all instances' scores into account, to make sure we use any * permanent promotion scores. */ if ((rsc->running_on == NULL) && (g_hash_table_size(rsc->known_on) == 0)) { reason = "none probed"; goto check_allowed; } /* Otherwise, we've probed and/or started the resource *somewhere*, so * consider promotion scores on nodes where we know the status. */ if ((g_hash_table_lookup(rsc->known_on, node->details->id) != NULL) || (pe_find_node_id(rsc->running_on, node->details->id) != NULL)) { reason = "known"; } else { pe_rsc_trace(rsc, "Ignoring %s promotion score (for %s) on %s: not probed", rsc->id, id, pe__node_name(node)); free(id); return false; } check_allowed: if (is_allowed(rsc, node)) { pe_rsc_trace(rsc, "Counting %s promotion score (for %s) on %s: %s", rsc->id, id, pe__node_name(node), reason); free(id); return true; } pe_rsc_trace(rsc, "Ignoring %s promotion score (for %s) on %s: not allowed", rsc->id, id, pe__node_name(node)); free(id); return false; } /*! * \internal * \brief Get the value of a promotion score node attribute * * \param[in] rsc Promotable clone instance to get promotion score for * \param[in] node Node to get promotion score for * \param[in] name Resource name to use in promotion score attribute name * * \return Value of promotion score node attribute for \p rsc on \p node */ static const char * promotion_attr_value(const pe_resource_t *rsc, const pe_node_t *node, const char *name) { char *attr_name = NULL; const char *attr_value = NULL; enum pe__rsc_node node_type = pe__rsc_node_assigned; if (pcmk_is_set(rsc->flags, pe_rsc_provisional)) { // Not assigned yet node_type = pe__rsc_node_current; } attr_name = pcmk_promotion_score_name(name); attr_value = pe__node_attribute_calculated(node, attr_name, rsc, node_type, false); free(attr_name); return attr_value; } /*! * \internal * \brief Get the promotion score for a clone instance on a node * * \param[in] rsc Promotable clone instance to get score for * \param[in] node Node to get score for * \param[out] is_default If non-NULL, will be set true if no score available * * \return Promotion score for \p rsc on \p node (or 0 if none) */ static int promotion_score(const pe_resource_t *rsc, const pe_node_t *node, bool *is_default) { char *name = NULL; const char *attr_value = NULL; if (is_default != NULL) { *is_default = true; } CRM_CHECK((rsc != NULL) && (node != NULL), return 0); /* If this is an instance of a cloned group, the promotion score is the sum * of all members' promotion scores. */ if (rsc->children != NULL) { int score = 0; for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { const pe_resource_t *child = (const pe_resource_t *) iter->data; bool child_default = false; int child_score = promotion_score(child, node, &child_default); if (!child_default && (is_default != NULL)) { *is_default = false; } score += child_score; } return score; } if (!promotion_score_applies(rsc, node)) { return 0; } /* For the promotion score attribute name, use the name the resource is * known as in resource history, since that's what crm_attribute --promotion * would have used. */ name = (rsc->clone_name == NULL)? rsc->id : rsc->clone_name; attr_value = promotion_attr_value(rsc, node, name); if (attr_value != NULL) { pe_rsc_trace(rsc, "Promotion score for %s on %s = %s", name, pe__node_name(node), pcmk__s(attr_value, "(unset)")); } else if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) { /* If we don't have any resource history yet, we won't have clone_name. * In that case, for anonymous clones, try the resource name without * any instance number. */ name = clone_strip(rsc->id); if (strcmp(rsc->id, name) != 0) { attr_value = promotion_attr_value(rsc, node, name); pe_rsc_trace(rsc, "Promotion score for %s on %s (for %s) = %s", name, pe__node_name(node), rsc->id, pcmk__s(attr_value, "(unset)")); } free(name); } if (attr_value == NULL) { return 0; } if (is_default != NULL) { *is_default = false; } return char2score(attr_value); } /*! * \internal * \brief Include promotion scores in instances' node scores and priorities * * \param[in,out] rsc Promotable clone resource to update */ void pcmk__add_promotion_scores(pe_resource_t *rsc) { if (pe__set_clone_flag(rsc, pe__clone_promotion_added) == pcmk_rc_already) { return; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *child_rsc = (pe_resource_t *) iter->data; GHashTableIter iter; pe_node_t *node = NULL; int score, new_score; g_hash_table_iter_init(&iter, child_rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (!pcmk__node_available(node, false, false)) { /* This node will never be promoted, so don't apply the * promotion score, as that may lead to clone shuffling. */ continue; } score = promotion_score(child_rsc, node, NULL); if (score > 0) { new_score = pcmk__add_scores(node->weight, score); if (new_score != node->weight) { // Could remain INFINITY node->weight = new_score; pe_rsc_trace(rsc, "Added %s promotion priority (%s) to score " "on %s (now %s)", child_rsc->id, pcmk_readable_score(score), pe__node_name(node), pcmk_readable_score(new_score)); } } if (score > child_rsc->priority) { pe_rsc_trace(rsc, "Updating %s priority to promotion score (%d->%d)", child_rsc->id, child_rsc->priority, score); child_rsc->priority = score; } } } } /*! * \internal * \brief If a resource's current role is started, change it to unpromoted * * \param[in,out] data Resource to update * \param[in] user_data Ignored */ static void set_current_role_unpromoted(void *data, void *user_data) { pe_resource_t *rsc = (pe_resource_t *) data; if (rsc->role == pcmk_role_started) { // Promotable clones should use unpromoted role instead of started rsc->role = pcmk_role_unpromoted; } g_list_foreach(rsc->children, set_current_role_unpromoted, NULL); } /*! * \internal * \brief Set a resource's next role to unpromoted (or stopped if unassigned) * * \param[in,out] data Resource to update * \param[in] user_data Ignored */ static void set_next_role_unpromoted(void *data, void *user_data) { pe_resource_t *rsc = (pe_resource_t *) data; GList *assigned = NULL; rsc->fns->location(rsc, &assigned, FALSE); if (assigned == NULL) { pe__set_next_role(rsc, pcmk_role_stopped, "stopped instance"); } else { pe__set_next_role(rsc, pcmk_role_unpromoted, "unpromoted instance"); g_list_free(assigned); } g_list_foreach(rsc->children, set_next_role_unpromoted, NULL); } /*! * \internal * \brief Set a resource's next role to promoted if not already set * * \param[in,out] data Resource to update * \param[in] user_data Ignored */ static void set_next_role_promoted(void *data, gpointer user_data) { pe_resource_t *rsc = (pe_resource_t *) data; if (rsc->next_role == pcmk_role_unknown) { pe__set_next_role(rsc, pcmk_role_promoted, "promoted instance"); } g_list_foreach(rsc->children, set_next_role_promoted, NULL); } /*! * \internal * \brief Show instance's promotion score on node where it will be active * * \param[in,out] instance Promotable clone instance to show */ static void show_promotion_score(pe_resource_t *instance) { pe_node_t *chosen = instance->fns->location(instance, NULL, FALSE); - if (pcmk_is_set(instance->cluster->flags, pe_flag_show_scores) + if (pcmk_is_set(instance->cluster->flags, pcmk_sched_output_scores) && !pcmk__is_daemon && (instance->cluster->priv != NULL)) { pcmk__output_t *out = instance->cluster->priv; out->message(out, "promotion-score", instance, chosen, pcmk_readable_score(instance->sort_index)); } else { pe_rsc_debug(pe__const_top_resource(instance, false), "%s promotion score on %s: sort=%s priority=%s", instance->id, ((chosen == NULL)? "none" : pe__node_name(chosen)), pcmk_readable_score(instance->sort_index), pcmk_readable_score(instance->priority)); } } /*! * \internal * \brief Set a clone instance's promotion priority * * \param[in,out] data Promotable clone instance to update * \param[in] user_data Instance's parent clone */ static void set_instance_priority(gpointer data, gpointer user_data) { pe_resource_t *instance = (pe_resource_t *) data; const pe_resource_t *clone = (const pe_resource_t *) user_data; const pe_node_t *chosen = NULL; enum rsc_role_e next_role = pcmk_role_unknown; GList *list = NULL; pe_rsc_trace(clone, "Assigning priority for %s: %s", instance->id, role2text(instance->next_role)); if (instance->fns->state(instance, TRUE) == pcmk_role_started) { set_current_role_unpromoted(instance, NULL); } // Only an instance that will be active can be promoted chosen = instance->fns->location(instance, &list, FALSE); if (pcmk__list_of_multiple(list)) { pcmk__config_err("Cannot promote non-colocated child %s", instance->id); } g_list_free(list); if (chosen == NULL) { return; } next_role = instance->fns->state(instance, FALSE); switch (next_role) { case pcmk_role_started: case pcmk_role_unknown: // Set instance priority to its promotion score (or -1 if none) { bool is_default = false; instance->priority = promotion_score(instance, chosen, &is_default); if (is_default) { /* * Default to -1 if no value is set. This allows * instances eligible for promotion to be specified * based solely on rsc_location constraints, but * prevents any instance from being promoted if neither * a constraint nor a promotion score is present */ instance->priority = -1; } } break; case pcmk_role_unpromoted: case pcmk_role_stopped: // Instance can't be promoted instance->priority = -INFINITY; break; case pcmk_role_promoted: // Nothing needed (re-creating actions after scheduling fencing) break; default: CRM_CHECK(FALSE, crm_err("Unknown resource role %d for %s", next_role, instance->id)); } // Add relevant location constraint scores for promoted role apply_promoted_locations(instance, instance->rsc_location, chosen); apply_promoted_locations(instance, clone->rsc_location, chosen); // Consider instance's role-based colocations with other resources list = pcmk__this_with_colocations(instance); for (GList *iter = list; iter != NULL; iter = iter->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) iter->data; instance->cmds->apply_coloc_score(instance, cons->primary, cons, true); } g_list_free(list); instance->sort_index = instance->priority; if (next_role == pcmk_role_promoted) { instance->sort_index = INFINITY; } pe_rsc_trace(clone, "Assigning %s priority = %d", instance->id, instance->priority); } /*! * \internal * \brief Set a promotable clone instance's role * * \param[in,out] data Promotable clone instance to update * \param[in,out] user_data Pointer to count of instances chosen for promotion */ static void set_instance_role(gpointer data, gpointer user_data) { pe_resource_t *instance = (pe_resource_t *) data; int *count = (int *) user_data; const pe_resource_t *clone = pe__const_top_resource(instance, false); pe_node_t *chosen = NULL; show_promotion_score(instance); if (instance->sort_index < 0) { pe_rsc_trace(clone, "Not supposed to promote instance %s", instance->id); } else if ((*count < pe__clone_promoted_max(instance)) || !pcmk_is_set(clone->flags, pe_rsc_managed)) { chosen = node_to_be_promoted_on(instance); } if (chosen == NULL) { set_next_role_unpromoted(instance, NULL); return; } if ((instance->role < pcmk_role_promoted) && !pcmk_is_set(instance->cluster->flags, pcmk_sched_quorate) && (instance->cluster->no_quorum_policy == pcmk_no_quorum_freeze)) { crm_notice("Clone instance %s cannot be promoted without quorum", instance->id); set_next_role_unpromoted(instance, NULL); return; } chosen->count++; pe_rsc_info(clone, "Choosing %s (%s) on %s for promotion", instance->id, role2text(instance->role), pe__node_name(chosen)); set_next_role_promoted(instance, NULL); (*count)++; } /*! * \internal * \brief Set roles for all instances of a promotable clone * * \param[in,out] rsc Promotable clone resource to update */ void pcmk__set_instance_roles(pe_resource_t *rsc) { int promoted = 0; GHashTableIter iter; pe_node_t *node = NULL; // Repurpose count to track the number of promoted instances assigned g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { node->count = 0; } // Set instances' promotion priorities and sort by highest priority first g_list_foreach(rsc->children, set_instance_priority, rsc); sort_promotable_instances(rsc); // Choose the first N eligible instances to be promoted g_list_foreach(rsc->children, set_instance_role, &promoted); pe_rsc_info(rsc, "%s: Promoted %d instances of a possible %d", rsc->id, promoted, pe__clone_promoted_max(rsc)); } /*! * * \internal * \brief Create actions for promotable clone instances * * \param[in,out] clone Promotable clone to create actions for * \param[out] any_promoting Will be set true if any instance is promoting * \param[out] any_demoting Will be set true if any instance is demoting */ static void create_promotable_instance_actions(pe_resource_t *clone, bool *any_promoting, bool *any_demoting) { for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; instance->cmds->create_actions(instance); check_for_role_change(instance, any_demoting, any_promoting); } } /*! * \internal * \brief Reset each promotable instance's resource priority * * Reset the priority of each instance of a promotable clone to the clone's * priority (after promotion actions are scheduled, when instance priorities * were repurposed as promotion scores). * * \param[in,out] clone Promotable clone to reset */ static void reset_instance_priorities(pe_resource_t *clone) { for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; instance->priority = clone->priority; } } /*! * \internal * \brief Create actions specific to promotable clones * * \param[in,out] clone Promotable clone to create actions for */ void pcmk__create_promotable_actions(pe_resource_t *clone) { bool any_promoting = false; bool any_demoting = false; // Create actions for each clone instance individually create_promotable_instance_actions(clone, &any_promoting, &any_demoting); // Create pseudo-actions for clone as a whole pe__create_promotable_pseudo_ops(clone, any_promoting, any_demoting); // Undo our temporary repurposing of resource priority for instances reset_instance_priorities(clone); } /*! * \internal * \brief Create internal orderings for a promotable clone's instances * * \param[in,out] clone Promotable clone instance to order */ void pcmk__order_promotable_instances(pe_resource_t *clone) { pe_resource_t *previous = NULL; // Needed for ordered clones pcmk__promotable_restart_ordering(clone); for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; // Demote before promote pcmk__order_resource_actions(instance, PCMK_ACTION_DEMOTE, instance, PCMK_ACTION_PROMOTE, pe_order_optional); order_instance_promotion(clone, instance, previous); order_instance_demotion(clone, instance, previous); previous = instance; } } /*! * \internal * \brief Update dependent's allowed nodes for colocation with promotable * * \param[in,out] dependent Dependent resource to update * \param[in] primary Primary resource * \param[in] primary_node Node where an instance of the primary will be * \param[in] colocation Colocation constraint to apply */ static void update_dependent_allowed_nodes(pe_resource_t *dependent, const pe_resource_t *primary, const pe_node_t *primary_node, const pcmk__colocation_t *colocation) { GHashTableIter iter; pe_node_t *node = NULL; const char *primary_value = NULL; const char *attr = colocation->node_attribute; if (colocation->score >= INFINITY) { return; // Colocation is mandatory, so allowed node scores don't matter } primary_value = pcmk__colocation_node_attr(primary_node, attr, primary); pe_rsc_trace(colocation->primary, "Applying %s (%s with %s on %s by %s @%d) to %s", colocation->id, colocation->dependent->id, colocation->primary->id, pe__node_name(primary_node), attr, colocation->score, dependent->id); g_hash_table_iter_init(&iter, dependent->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { const char *dependent_value = pcmk__colocation_node_attr(node, attr, dependent); if (pcmk__str_eq(primary_value, dependent_value, pcmk__str_casei)) { node->weight = pcmk__add_scores(node->weight, colocation->score); pe_rsc_trace(colocation->primary, "Added %s score (%s) to %s (now %s)", colocation->id, pcmk_readable_score(colocation->score), pe__node_name(node), pcmk_readable_score(node->weight)); } } } /*! * \brief Update dependent for a colocation with a promotable clone * * \param[in] primary Primary resource in the colocation * \param[in,out] dependent Dependent resource in the colocation * \param[in] colocation Colocation constraint to apply */ void pcmk__update_dependent_with_promotable(const pe_resource_t *primary, pe_resource_t *dependent, const pcmk__colocation_t *colocation) { GList *affected_nodes = NULL; /* Build a list of all nodes where an instance of the primary will be, and * (for optional colocations) update the dependent's allowed node scores for * each one. */ for (GList *iter = primary->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; pe_node_t *node = instance->fns->location(instance, NULL, FALSE); if (node == NULL) { continue; } if (instance->fns->state(instance, FALSE) == colocation->primary_role) { update_dependent_allowed_nodes(dependent, primary, node, colocation); affected_nodes = g_list_prepend(affected_nodes, node); } } /* For mandatory colocations, add the primary's node score to the * dependent's node score for each affected node, and ban the dependent * from all other nodes. * * However, skip this for promoted-with-promoted colocations, otherwise * inactive dependent instances can't start (in the unpromoted role). */ if ((colocation->score >= INFINITY) && ((colocation->dependent_role != pcmk_role_promoted) || (colocation->primary_role != pcmk_role_promoted))) { pe_rsc_trace(colocation->primary, "Applying %s (mandatory %s with %s) to %s", colocation->id, colocation->dependent->id, colocation->primary->id, dependent->id); pcmk__colocation_intersect_nodes(dependent, primary, colocation, affected_nodes, true); } g_list_free(affected_nodes); } /*! * \internal * \brief Update dependent priority for colocation with promotable * * \param[in] primary Primary resource in the colocation * \param[in,out] dependent Dependent resource in the colocation * \param[in] colocation Colocation constraint to apply */ void pcmk__update_promotable_dependent_priority(const pe_resource_t *primary, pe_resource_t *dependent, const pcmk__colocation_t *colocation) { pe_resource_t *primary_instance = NULL; // Look for a primary instance where dependent will be primary_instance = pcmk__find_compatible_instance(dependent, primary, colocation->primary_role, false); if (primary_instance != NULL) { // Add primary instance's priority to dependent's int new_priority = pcmk__add_scores(dependent->priority, colocation->score); pe_rsc_trace(colocation->primary, "Applying %s (%s with %s) to %s priority (%s + %s = %s)", colocation->id, colocation->dependent->id, colocation->primary->id, dependent->id, pcmk_readable_score(dependent->priority), pcmk_readable_score(colocation->score), pcmk_readable_score(new_priority)); dependent->priority = new_priority; } else if (colocation->score >= INFINITY) { // Mandatory colocation, but primary won't be here pe_rsc_trace(colocation->primary, "Applying %s (%s with %s) to %s: can't be promoted", colocation->id, colocation->dependent->id, colocation->primary->id, dependent->id); dependent->priority = -INFINITY; } } diff --git a/lib/pacemaker/pcmk_simulate.c b/lib/pacemaker/pcmk_simulate.c index 235faa299d..fb7a4a55a4 100644 --- a/lib/pacemaker/pcmk_simulate.c +++ b/lib/pacemaker/pcmk_simulate.c @@ -1,1004 +1,1005 @@ /* * 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. */ #include #include #include #include #include #include #include #include #include #include #include #include "libpacemaker_private.h" static pcmk__output_t *out = NULL; static cib_t *fake_cib = NULL; static GList *fake_resource_list = NULL; static const GList *fake_op_fail_list = NULL; static void set_effective_date(pe_working_set_t *data_set, bool print_original, const char *use_date); /*! * \internal * \brief Create an action name for use in a dot graph * * \param[in] action Action to create name for * \param[in] verbose If true, add action ID to name * * \return Newly allocated string with action name * \note It is the caller's responsibility to free the result. */ static char * create_action_name(const pe_action_t *action, bool verbose) { char *action_name = NULL; const char *prefix = ""; const char *action_host = NULL; const char *clone_name = NULL; const char *task = action->task; if (action->node != NULL) { action_host = action->node->details->uname; } else if (!pcmk_is_set(action->flags, pe_action_pseudo)) { action_host = ""; } if (pcmk__str_eq(action->task, PCMK_ACTION_CANCEL, pcmk__str_none)) { prefix = "Cancel "; task = action->cancel_task; } if (action->rsc != NULL) { clone_name = action->rsc->clone_name; } if (clone_name != NULL) { char *key = NULL; guint interval_ms = 0; if (pcmk__guint_from_hash(action->meta, XML_LRM_ATTR_INTERVAL_MS, 0, &interval_ms) != pcmk_rc_ok) { interval_ms = 0; } if (pcmk__strcase_any_of(action->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_NOTIFIED, NULL)) { const char *n_type = g_hash_table_lookup(action->meta, "notify_key_type"); const char *n_task = g_hash_table_lookup(action->meta, "notify_key_operation"); CRM_ASSERT(n_type != NULL); CRM_ASSERT(n_task != NULL); key = pcmk__notify_key(clone_name, n_type, n_task); } else { key = pcmk__op_key(clone_name, task, interval_ms); } if (action_host != NULL) { action_name = crm_strdup_printf("%s%s %s", prefix, key, action_host); } else { action_name = crm_strdup_printf("%s%s", prefix, key); } free(key); } else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { const char *op = g_hash_table_lookup(action->meta, "stonith_action"); action_name = crm_strdup_printf("%s%s '%s' %s", prefix, action->task, op, action_host); } else if (action->rsc && action_host) { action_name = crm_strdup_printf("%s%s %s", prefix, action->uuid, action_host); } else if (action_host) { action_name = crm_strdup_printf("%s%s %s", prefix, action->task, action_host); } else { action_name = crm_strdup_printf("%s", action->uuid); } if (verbose) { char *with_id = crm_strdup_printf("%s (%d)", action_name, action->id); free(action_name); action_name = with_id; } return action_name; } /*! * \internal * \brief Display the status of a cluster * * \param[in,out] data_set Cluster working set * \param[in] show_opts How to modify display (as pcmk_show_opt_e flags) * \param[in] section_opts Sections to display (as pcmk_section_e flags) * \param[in] title What to use as list title * \param[in] print_spacer Whether to display a spacer first */ static void print_cluster_status(pe_working_set_t *data_set, uint32_t show_opts, uint32_t section_opts, const char *title, bool print_spacer) { pcmk__output_t *out = data_set->priv; GList *all = NULL; crm_exit_t stonith_rc = 0; enum pcmk_pacemakerd_state state = pcmk_pacemakerd_state_invalid; section_opts |= pcmk_section_nodes | pcmk_section_resources; show_opts |= pcmk_show_inactive_rscs | pcmk_show_failed_detail; all = g_list_prepend(all, (gpointer) "*"); PCMK__OUTPUT_SPACER_IF(out, print_spacer); out->begin_list(out, NULL, NULL, "%s", title); out->message(out, "cluster-status", data_set, state, stonith_rc, NULL, false, section_opts, show_opts, NULL, all, all); out->end_list(out); g_list_free(all); } /*! * \internal * \brief Display a summary of all actions scheduled in a transition * * \param[in,out] data_set Cluster working set (fully scheduled) * \param[in] print_spacer Whether to display a spacer first */ static void print_transition_summary(pe_working_set_t *data_set, bool print_spacer) { pcmk__output_t *out = data_set->priv; PCMK__OUTPUT_SPACER_IF(out, print_spacer); out->begin_list(out, NULL, NULL, "Transition Summary"); pcmk__output_actions(data_set); out->end_list(out); } /*! * \internal * \brief Reset a cluster working set's input, output, date, and flags * * \param[in,out] data_set Cluster working set * \param[in] input What to set as cluster input * \param[in] out What to set as cluster output object * \param[in] use_date What to set as cluster's current timestamp * \param[in] flags Group of enum pcmk_scheduler_flags to set */ static void reset(pe_working_set_t *data_set, xmlNodePtr input, pcmk__output_t *out, const char *use_date, unsigned int flags) { data_set->input = input; data_set->priv = out; set_effective_date(data_set, true, use_date); if (pcmk_is_set(flags, pcmk_sim_sanitized)) { pe__set_working_set_flags(data_set, pcmk_sched_sanitized); } if (pcmk_is_set(flags, pcmk_sim_show_scores)) { - pe__set_working_set_flags(data_set, pe_flag_show_scores); + pe__set_working_set_flags(data_set, pcmk_sched_output_scores); } if (pcmk_is_set(flags, pcmk_sim_show_utilization)) { pe__set_working_set_flags(data_set, pe_flag_show_utilization); } } /*! * \brief Write out a file in dot(1) format describing the actions that will * be taken by the scheduler in response to an input CIB file. * * \param[in,out] data_set Working set for the cluster * \param[in] dot_file The filename to write * \param[in] all_actions Write all actions, even those that are optional * or are on unmanaged resources * \param[in] verbose Add extra information, such as action IDs, to the * output * * \return Standard Pacemaker return code */ static int write_sim_dotfile(pe_working_set_t *data_set, const char *dot_file, bool all_actions, bool verbose) { GList *iter = NULL; FILE *dot_strm = fopen(dot_file, "w"); if (dot_strm == NULL) { return errno; } fprintf(dot_strm, " digraph \"g\" {\n"); for (iter = data_set->actions; iter != NULL; iter = iter->next) { pe_action_t *action = (pe_action_t *) iter->data; const char *style = "dashed"; const char *font = "black"; const char *color = "black"; char *action_name = create_action_name(action, verbose); if (pcmk_is_set(action->flags, pe_action_pseudo)) { font = "orange"; } if (pcmk_is_set(action->flags, pe_action_dumped)) { style = "bold"; color = "green"; } else if ((action->rsc != NULL) && !pcmk_is_set(action->rsc->flags, pe_rsc_managed)) { color = "red"; font = "purple"; if (!all_actions) { goto do_not_write; } } else if (pcmk_is_set(action->flags, pe_action_optional)) { color = "blue"; if (!all_actions) { goto do_not_write; } } else { color = "red"; CRM_LOG_ASSERT(!pcmk_is_set(action->flags, pe_action_runnable)); } pe__set_action_flags(action, pe_action_dumped); fprintf(dot_strm, "\"%s\" [ style=%s color=\"%s\" fontcolor=\"%s\"]\n", action_name, style, color, font); do_not_write: free(action_name); } for (iter = data_set->actions; iter != NULL; iter = iter->next) { pe_action_t *action = (pe_action_t *) iter->data; for (GList *before_iter = action->actions_before; before_iter != NULL; before_iter = before_iter->next) { pe_action_wrapper_t *before = before_iter->data; char *before_name = NULL; char *after_name = NULL; const char *style = "dashed"; bool optional = true; if (before->state == pe_link_dumped) { optional = false; style = "bold"; } else if (before->type == pe_order_none) { continue; } else if (pcmk_is_set(before->action->flags, pe_action_dumped) && pcmk_is_set(action->flags, pe_action_dumped) && before->type != pe_order_load) { optional = false; } if (all_actions || !optional) { before_name = create_action_name(before->action, verbose); after_name = create_action_name(action, verbose); fprintf(dot_strm, "\"%s\" -> \"%s\" [ style = %s]\n", before_name, after_name, style); free(before_name); free(after_name); } } } fprintf(dot_strm, "}\n"); fflush(dot_strm); fclose(dot_strm); return pcmk_rc_ok; } /*! * \brief Profile the configuration updates and scheduler actions in a single * CIB file, printing the profiling timings. * * \note \p data_set->priv must have been set to a valid \p pcmk__output_t * object before this function is called. * * \param[in] xml_file The CIB file to profile * \param[in] repeat Number of times to run * \param[in,out] data_set Working set for the cluster * \param[in] use_date The date to set the cluster's time to (may be NULL) */ static void profile_file(const char *xml_file, long long repeat, pe_working_set_t *data_set, const char *use_date) { pcmk__output_t *out = data_set->priv; xmlNode *cib_object = NULL; clock_t start = 0; clock_t end; unsigned long long data_set_flags = pcmk_sched_no_compat; CRM_ASSERT(out != NULL); cib_object = filename2xml(xml_file); start = clock(); if (pcmk_find_cib_element(cib_object, XML_CIB_TAG_STATUS) == NULL) { create_xml_node(cib_object, XML_CIB_TAG_STATUS); } if (cli_config_update(&cib_object, NULL, FALSE) == FALSE) { free_xml(cib_object); return; } if (validate_xml(cib_object, NULL, FALSE) != TRUE) { free_xml(cib_object); return; } - if (pcmk_is_set(data_set->flags, pe_flag_show_scores)) { - data_set_flags |= pe_flag_show_scores; + if (pcmk_is_set(data_set->flags, pcmk_sched_output_scores)) { + data_set_flags |= pcmk_sched_output_scores; } if (pcmk_is_set(data_set->flags, pe_flag_show_utilization)) { data_set_flags |= pe_flag_show_utilization; } for (int i = 0; i < repeat; ++i) { xmlNode *input = (repeat == 1)? cib_object : copy_xml(cib_object); data_set->input = input; set_effective_date(data_set, false, use_date); pcmk__schedule_actions(input, data_set_flags, data_set); pe_reset_working_set(data_set); } end = clock(); out->message(out, "profile", xml_file, start, end); } void pcmk__profile_dir(const char *dir, long long repeat, pe_working_set_t *data_set, const char *use_date) { pcmk__output_t *out = data_set->priv; struct dirent **namelist; int file_num = scandir(dir, &namelist, 0, alphasort); CRM_ASSERT(out != NULL); if (file_num > 0) { struct stat prop; char buffer[FILENAME_MAX]; out->begin_list(out, NULL, NULL, "Timings"); while (file_num--) { if ('.' == namelist[file_num]->d_name[0]) { free(namelist[file_num]); continue; } else if (!pcmk__ends_with_ext(namelist[file_num]->d_name, ".xml")) { free(namelist[file_num]); continue; } snprintf(buffer, sizeof(buffer), "%s/%s", dir, namelist[file_num]->d_name); if (stat(buffer, &prop) == 0 && S_ISREG(prop.st_mode)) { profile_file(buffer, repeat, data_set, use_date); } free(namelist[file_num]); } free(namelist); out->end_list(out); } } /*! * \brief Set the date of the cluster, either to the value given by * \p use_date, or to the "execution-date" value in the CIB. * * \note \p data_set->priv must have been set to a valid \p pcmk__output_t * object before this function is called. * * \param[in,out] data_set Working set for the cluster * \param[in] print_original If \p true, the "execution-date" should * also be printed * \param[in] use_date The date to set the cluster's time to * (may be NULL) */ static void set_effective_date(pe_working_set_t *data_set, bool print_original, const char *use_date) { pcmk__output_t *out = data_set->priv; time_t original_date = 0; CRM_ASSERT(out != NULL); crm_element_value_epoch(data_set->input, "execution-date", &original_date); if (use_date) { data_set->now = crm_time_new(use_date); out->info(out, "Setting effective cluster time: %s", use_date); crm_time_log(LOG_NOTICE, "Pretending 'now' is", data_set->now, crm_time_log_date | crm_time_log_timeofday); } else if (original_date != 0) { data_set->now = pcmk__copy_timet(original_date); if (print_original) { char *when = crm_time_as_string(data_set->now, crm_time_log_date|crm_time_log_timeofday); out->info(out, "Using the original execution date of: %s", when); free(when); } } } /*! * \internal * \brief Simulate successfully executing a pseudo-action in a graph * * \param[in,out] graph Graph to update with pseudo-action result * \param[in,out] action Pseudo-action to simulate executing * * \return Standard Pacemaker return code */ static int simulate_pseudo_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); out->message(out, "inject-pseudo-action", node, task); pcmk__update_graph(graph, action); return pcmk_rc_ok; } /*! * \internal * \brief Simulate executing a resource action in a graph * * \param[in,out] graph Graph to update with resource action result * \param[in,out] action Resource action to simulate executing * * \return Standard Pacemaker return code */ static int simulate_resource_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { int rc; lrmd_event_data_t *op = NULL; int target_outcome = PCMK_OCF_OK; const char *rtype = NULL; const char *rclass = NULL; const char *resource = NULL; const char *rprovider = NULL; const char *resource_config_name = NULL; const char *operation = crm_element_value(action->xml, "operation"); const char *target_rc_s = crm_meta_value(action->params, XML_ATTR_TE_TARGET_RC); xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; xmlNode *action_rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE); char *node = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET); char *uuid = NULL; const char *router_node = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); // Certain actions don't need to be displayed or history entries if (pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_none)) { crm_debug("No history injection for %s op on %s", operation, node); goto done; // Confirm action and update graph } if (action_rsc == NULL) { // Shouldn't be possible crm_log_xml_err(action->xml, "Bad"); free(node); return EPROTO; } /* A resource might be known by different names in the configuration and in * the action (for example, a clone instance). Grab the configuration name * (which is preferred when writing history), and if necessary, the instance * name. */ resource_config_name = crm_element_value(action_rsc, XML_ATTR_ID); if (resource_config_name == NULL) { // Shouldn't be possible crm_log_xml_err(action->xml, "No ID"); free(node); return EPROTO; } resource = resource_config_name; if (pe_find_resource(fake_resource_list, resource) == NULL) { const char *longname = crm_element_value(action_rsc, XML_ATTR_ID_LONG); if ((longname != NULL) && (pe_find_resource(fake_resource_list, longname) != NULL)) { resource = longname; } } // Certain actions need to be displayed but don't need history entries if (pcmk__strcase_any_of(operation, PCMK_ACTION_DELETE, PCMK_ACTION_META_DATA, NULL)) { out->message(out, "inject-rsc-action", resource, operation, node, (guint) 0); goto done; // Confirm action and update graph } rclass = crm_element_value(action_rsc, XML_AGENT_ATTR_CLASS); rtype = crm_element_value(action_rsc, XML_ATTR_TYPE); rprovider = crm_element_value(action_rsc, XML_AGENT_ATTR_PROVIDER); pcmk__scan_min_int(target_rc_s, &target_outcome, 0); CRM_ASSERT(fake_cib->cmds->query(fake_cib, NULL, NULL, cib_sync_call|cib_scope_local) == pcmk_ok); // Ensure the action node is in the CIB uuid = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET_UUID); cib_node = pcmk__inject_node(fake_cib, node, ((router_node == NULL)? uuid: node)); free(uuid); CRM_ASSERT(cib_node != NULL); // Add a history entry for the action cib_resource = pcmk__inject_resource_history(out, cib_node, resource, resource_config_name, rclass, rtype, rprovider); if (cib_resource == NULL) { crm_err("Could not simulate action %d history for resource %s", action->id, resource); free(node); free_xml(cib_node); return EINVAL; } // Simulate and display an executor event for the action result op = pcmk__event_from_graph_action(cib_resource, action, PCMK_EXEC_DONE, target_outcome, "User-injected result"); out->message(out, "inject-rsc-action", resource, op->op_type, node, op->interval_ms); // Check whether action is in a list of desired simulated failures for (const GList *iter = fake_op_fail_list; iter != NULL; iter = iter->next) { const char *spec = (const char *) iter->data; char *key = NULL; const char *match_name = NULL; // Allow user to specify anonymous clone with or without instance number key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = resource; } free(key); // If not found, try the resource's name in the configuration if ((match_name == NULL) && (strcmp(resource, resource_config_name) != 0)) { key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource_config_name, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = resource_config_name; } free(key); } if (match_name == NULL) { continue; // This failed action entry doesn't match } // ${match_name}_${task}_${interval_in_ms}@${node}=${rc} rc = sscanf(spec, "%*[^=]=%d", (int *) &op->rc); if (rc != 1) { out->err(out, "Invalid failed operation '%s' " "(result code must be integer)", spec); continue; // Keep checking other list entries } out->info(out, "Pretending action %d failed with rc=%d", action->id, op->rc); pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); graph->abort_priority = INFINITY; pcmk__inject_failcount(out, cib_node, match_name, op->op_type, op->interval_ms, op->rc); break; } pcmk__inject_action_result(cib_resource, op, target_outcome); lrmd_free_event(op); rc = fake_cib->cmds->modify(fake_cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); done: free(node); free_xml(cib_node); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); pcmk__update_graph(graph, action); return pcmk_rc_ok; } /*! * \internal * \brief Simulate successfully executing a cluster action * * \param[in,out] graph Graph to update with action result * \param[in,out] action Cluster action to simulate * * \return Standard Pacemaker return code */ static int simulate_cluster_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); xmlNode *rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); out->message(out, "inject-cluster-action", node, task, rsc); pcmk__update_graph(graph, action); return pcmk_rc_ok; } /*! * \internal * \brief Simulate successfully executing a fencing action * * \param[in,out] graph Graph to update with action result * \param[in,out] action Fencing action to simulate * * \return Standard Pacemaker return code */ static int simulate_fencing_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *op = crm_meta_value(action->params, "stonith_action"); char *target = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET); out->message(out, "inject-fencing-action", target, op); if (!pcmk__str_eq(op, PCMK_ACTION_ON, pcmk__str_casei)) { int rc = pcmk_ok; GString *xpath = g_string_sized_new(512); // Set node state to offline xmlNode *cib_node = pcmk__inject_node_state_change(fake_cib, target, false); CRM_ASSERT(cib_node != NULL); crm_xml_add(cib_node, XML_ATTR_ORIGIN, __func__); rc = fake_cib->cmds->replace(fake_cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); // Simulate controller clearing node's resource history and attributes pcmk__g_strcat(xpath, "//" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='", target, "']/" XML_CIB_TAG_LRM, NULL); fake_cib->cmds->remove(fake_cib, (const char *) xpath->str, NULL, cib_xpath|cib_sync_call|cib_scope_local); g_string_truncate(xpath, 0); pcmk__g_strcat(xpath, "//" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='", target, "']" "/" XML_TAG_TRANSIENT_NODEATTRS, NULL); fake_cib->cmds->remove(fake_cib, (const char *) xpath->str, NULL, cib_xpath|cib_sync_call|cib_scope_local); free_xml(cib_node); g_string_free(xpath, TRUE); } pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); pcmk__update_graph(graph, action); free(target); return pcmk_rc_ok; } enum pcmk__graph_status pcmk__simulate_transition(pe_working_set_t *data_set, cib_t *cib, const GList *op_fail_list) { pcmk__graph_t *transition = NULL; enum pcmk__graph_status graph_rc; pcmk__graph_functions_t simulation_fns = { simulate_pseudo_action, simulate_resource_action, simulate_cluster_action, simulate_fencing_action, }; out = data_set->priv; fake_cib = cib; fake_op_fail_list = op_fail_list; if (!out->is_quiet(out)) { out->begin_list(out, NULL, NULL, "Executing Cluster Transition"); } pcmk__set_graph_functions(&simulation_fns); transition = pcmk__unpack_graph(data_set->graph, crm_system_name); pcmk__log_graph(LOG_DEBUG, transition); fake_resource_list = data_set->resources; do { graph_rc = pcmk__execute_graph(transition); } while (graph_rc == pcmk__graph_active); fake_resource_list = NULL; if (graph_rc != pcmk__graph_complete) { out->err(out, "Transition failed: %s", pcmk__graph_status2text(graph_rc)); pcmk__log_graph(LOG_ERR, transition); out->err(out, "An invalid transition was produced"); } pcmk__free_graph(transition); if (!out->is_quiet(out)) { // If not quiet, we'll need the resulting CIB for later display xmlNode *cib_object = NULL; int rc = fake_cib->cmds->query(fake_cib, NULL, &cib_object, cib_sync_call|cib_scope_local); CRM_ASSERT(rc == pcmk_ok); pe_reset_working_set(data_set); data_set->input = cib_object; out->end_list(out); } return graph_rc; } int pcmk__simulate(pe_working_set_t *data_set, pcmk__output_t *out, const pcmk_injections_t *injections, unsigned int flags, uint32_t section_opts, const char *use_date, const char *input_file, const char *graph_file, const char *dot_file) { int printed = pcmk_rc_no_output; int rc = pcmk_rc_ok; xmlNodePtr input = NULL; cib_t *cib = NULL; rc = cib__signon_query(out, &cib, &input); if (rc != pcmk_rc_ok) { goto simulate_done; } reset(data_set, input, out, use_date, flags); cluster_status(data_set); if ((cib->variant == cib_native) && pcmk_is_set(section_opts, pcmk_section_times)) { if (pcmk__our_nodename == NULL) { // Currently used only in the times section pcmk__query_node_name(out, 0, &pcmk__our_nodename, 0); } data_set->localhost = pcmk__our_nodename; } if (!out->is_quiet(out)) { const bool show_pending = pcmk_is_set(flags, pcmk_sim_show_pending); if (pcmk_is_set(data_set->flags, pcmk_sched_in_maintenance)) { printed = out->message(out, "maint-mode", data_set->flags); } if (data_set->disabled_resources || data_set->blocked_resources) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); printed = out->info(out, "%d of %d resource instances DISABLED and " "%d BLOCKED from further action due to failure", data_set->disabled_resources, data_set->ninstances, data_set->blocked_resources); } /* Most formatted output headers use caps for each word, but this one * only has the first word capitalized for compatibility with pcs. */ print_cluster_status(data_set, (show_pending? pcmk_show_pending : 0), section_opts, "Current cluster status", (printed == pcmk_rc_ok)); printed = pcmk_rc_ok; } // If the user requested any injections, handle them if ((injections->node_down != NULL) || (injections->node_fail != NULL) || (injections->node_up != NULL) || (injections->op_inject != NULL) || (injections->ticket_activate != NULL) || (injections->ticket_grant != NULL) || (injections->ticket_revoke != NULL) || (injections->ticket_standby != NULL) || (injections->watchdog != NULL)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); pcmk__inject_scheduler_input(data_set, cib, injections); printed = pcmk_rc_ok; rc = cib->cmds->query(cib, NULL, &input, cib_sync_call); if (rc != pcmk_rc_ok) { rc = pcmk_legacy2rc(rc); goto simulate_done; } cleanup_calculations(data_set); reset(data_set, input, out, use_date, flags); cluster_status(data_set); } if (input_file != NULL) { rc = write_xml_file(input, input_file, FALSE); if (rc < 0) { rc = pcmk_legacy2rc(rc); goto simulate_done; } } if (pcmk_any_flags_set(flags, pcmk_sim_process | pcmk_sim_simulate)) { pcmk__output_t *logger_out = NULL; unsigned long long data_set_flags = pcmk_sched_no_compat; - if (pcmk_is_set(data_set->flags, pe_flag_show_scores)) { - data_set_flags |= pe_flag_show_scores; + if (pcmk_is_set(data_set->flags, pcmk_sched_output_scores)) { + data_set_flags |= pcmk_sched_output_scores; } if (pcmk_is_set(data_set->flags, pe_flag_show_utilization)) { data_set_flags |= pe_flag_show_utilization; } if (pcmk_all_flags_set(data_set->flags, - pe_flag_show_scores|pe_flag_show_utilization)) { + pcmk_sched_output_scores + |pe_flag_show_utilization)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Assignment Scores and Utilization Information"); printed = pcmk_rc_ok; - } else if (pcmk_is_set(data_set->flags, pe_flag_show_scores)) { + } else if (pcmk_is_set(data_set->flags, pcmk_sched_output_scores)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Assignment Scores"); printed = pcmk_rc_ok; } else if (pcmk_is_set(data_set->flags, pe_flag_show_utilization)) { PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Utilization Information"); printed = pcmk_rc_ok; } else { rc = pcmk__log_output_new(&logger_out); if (rc != pcmk_rc_ok) { goto simulate_done; } pe__register_messages(logger_out); pcmk__register_lib_messages(logger_out); data_set->priv = logger_out; } pcmk__schedule_actions(input, data_set_flags, data_set); if (logger_out == NULL) { out->end_list(out); } else { logger_out->finish(logger_out, CRM_EX_OK, true, NULL); pcmk__output_free(logger_out); data_set->priv = out; } input = NULL; /* Don't try and free it twice */ if (graph_file != NULL) { rc = write_xml_file(data_set->graph, graph_file, FALSE); if (rc < 0) { rc = pcmk_rc_graph_error; goto simulate_done; } } if (dot_file != NULL) { rc = write_sim_dotfile(data_set, dot_file, pcmk_is_set(flags, pcmk_sim_all_actions), pcmk_is_set(flags, pcmk_sim_verbose)); if (rc != pcmk_rc_ok) { rc = pcmk_rc_dot_error; goto simulate_done; } } if (!out->is_quiet(out)) { print_transition_summary(data_set, printed == pcmk_rc_ok); } } rc = pcmk_rc_ok; if (!pcmk_is_set(flags, pcmk_sim_simulate)) { goto simulate_done; } PCMK__OUTPUT_SPACER_IF(out, printed == pcmk_rc_ok); if (pcmk__simulate_transition(data_set, cib, injections->op_fail) != pcmk__graph_complete) { rc = pcmk_rc_invalid_transition; } if (out->is_quiet(out)) { goto simulate_done; } set_effective_date(data_set, true, use_date); if (pcmk_is_set(flags, pcmk_sim_show_scores)) { - pe__set_working_set_flags(data_set, pe_flag_show_scores); + pe__set_working_set_flags(data_set, pcmk_sched_output_scores); } if (pcmk_is_set(flags, pcmk_sim_show_utilization)) { pe__set_working_set_flags(data_set, pe_flag_show_utilization); } cluster_status(data_set); print_cluster_status(data_set, 0, section_opts, "Revised Cluster Status", true); simulate_done: cib__clean_up_connection(&cib); return rc; } int pcmk_simulate(xmlNodePtr *xml, pe_working_set_t *data_set, const pcmk_injections_t *injections, unsigned int flags, unsigned int section_opts, const char *use_date, const char *input_file, const char *graph_file, const char *dot_file) { pcmk__output_t *out = NULL; int rc = pcmk_rc_ok; rc = pcmk__xml_output_new(&out, xml); if (rc != pcmk_rc_ok) { return rc; } pe__register_messages(out); pcmk__register_lib_messages(out); rc = pcmk__simulate(data_set, out, injections, flags, section_opts, use_date, input_file, graph_file, dot_file); pcmk__xml_output_finish(out, xml); return rc; } diff --git a/tools/crm_simulate.c b/tools/crm_simulate.c index 3a3f665e8e..d7369a9b4f 100644 --- a/tools/crm_simulate.c +++ b/tools/crm_simulate.c @@ -1,588 +1,588 @@ /* * Copyright 2009-2023 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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define SUMMARY "crm_simulate - simulate a Pacemaker cluster's response to events" struct { char *dot_file; char *graph_file; gchar *input_file; pcmk_injections_t *injections; unsigned int flags; gchar *output_file; long long repeat; gboolean store; gchar *test_dir; char *use_date; char *xml_file; } options = { .flags = pcmk_sim_show_pending | pcmk_sim_sanitized, .repeat = 1 }; uint32_t section_opts = 0; char *temp_shadow = NULL; crm_exit_t exit_code = CRM_EX_OK; #define INDENT " " static pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; static gboolean all_actions_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.flags |= pcmk_sim_all_actions; return TRUE; } static gboolean attrs_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { section_opts |= pcmk_section_attributes; return TRUE; } static gboolean failcounts_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { section_opts |= pcmk_section_failcounts | pcmk_section_failures; return TRUE; } static gboolean in_place_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.store = TRUE; options.flags |= pcmk_sim_process | pcmk_sim_simulate; return TRUE; } static gboolean live_check_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (options.xml_file) { free(options.xml_file); } options.xml_file = NULL; options.flags &= ~pcmk_sim_sanitized; return TRUE; } static gboolean node_down_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.injections->node_down = g_list_append(options.injections->node_down, g_strdup(optarg)); return TRUE; } static gboolean node_fail_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.injections->node_fail = g_list_append(options.injections->node_fail, g_strdup(optarg)); return TRUE; } static gboolean node_up_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { pcmk__simulate_node_config = true; options.injections->node_up = g_list_append(options.injections->node_up, g_strdup(optarg)); return TRUE; } static gboolean op_fail_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.flags |= pcmk_sim_process | pcmk_sim_simulate; options.injections->op_fail = g_list_append(options.injections->op_fail, g_strdup(optarg)); return TRUE; } static gboolean op_inject_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.injections->op_inject = g_list_append(options.injections->op_inject, g_strdup(optarg)); return TRUE; } static gboolean pending_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.flags |= pcmk_sim_show_pending; return TRUE; } static gboolean process_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.flags |= pcmk_sim_process; return TRUE; } static gboolean quorum_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { pcmk__str_update(&options.injections->quorum, optarg); return TRUE; } static gboolean save_dotfile_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.flags |= pcmk_sim_process; pcmk__str_update(&options.dot_file, optarg); return TRUE; } static gboolean save_graph_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.flags |= pcmk_sim_process; pcmk__str_update(&options.graph_file, optarg); return TRUE; } static gboolean show_scores_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.flags |= pcmk_sim_process | pcmk_sim_show_scores; return TRUE; } static gboolean simulate_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.flags |= pcmk_sim_process | pcmk_sim_simulate; return TRUE; } static gboolean ticket_activate_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.injections->ticket_activate = g_list_append(options.injections->ticket_activate, g_strdup(optarg)); return TRUE; } static gboolean ticket_grant_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.injections->ticket_grant = g_list_append(options.injections->ticket_grant, g_strdup(optarg)); return TRUE; } static gboolean ticket_revoke_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.injections->ticket_revoke = g_list_append(options.injections->ticket_revoke, g_strdup(optarg)); return TRUE; } static gboolean ticket_standby_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.injections->ticket_standby = g_list_append(options.injections->ticket_standby, g_strdup(optarg)); return TRUE; } static gboolean utilization_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { options.flags |= pcmk_sim_process | pcmk_sim_show_utilization; return TRUE; } static gboolean watchdog_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { pcmk__str_update(&options.injections->watchdog, optarg); return TRUE; } static gboolean xml_file_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { pcmk__str_update(&options.xml_file, optarg); options.flags |= pcmk_sim_sanitized; return TRUE; } static gboolean xml_pipe_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { pcmk__str_update(&options.xml_file, "-"); options.flags |= pcmk_sim_sanitized; return TRUE; } static GOptionEntry operation_entries[] = { { "run", 'R', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, process_cb, "Process the supplied input and show what actions the cluster will take in response", NULL }, { "simulate", 'S', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, simulate_cb, "Like --run, but also simulate taking those actions and show the resulting new status", NULL }, { "in-place", 'X', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, in_place_cb, "Like --simulate, but also store the results back to the input file", NULL }, { "show-attrs", 'A', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, attrs_cb, "Show node attributes", NULL }, { "show-failcounts", 'c', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, failcounts_cb, "Show resource fail counts", NULL }, { "show-scores", 's', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, show_scores_cb, "Show allocation scores", NULL }, { "show-utilization", 'U', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, utilization_cb, "Show utilization information", NULL }, { "profile", 'P', 0, G_OPTION_ARG_FILENAME, &options.test_dir, "Process all the XML files in the named directory to create profiling data", "DIR" }, { "repeat", 'N', 0, G_OPTION_ARG_INT, &options.repeat, "With --profile, repeat each test N times and print timings", "N" }, /* Deprecated */ { "pending", 'j', G_OPTION_FLAG_NO_ARG|G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_CALLBACK, pending_cb, "Display pending state if 'record-pending' is enabled", NULL }, { NULL } }; static GOptionEntry synthetic_entries[] = { { "node-up", 'u', 0, G_OPTION_ARG_CALLBACK, node_up_cb, "Simulate bringing a node online", "NODE" }, { "node-down", 'd', 0, G_OPTION_ARG_CALLBACK, node_down_cb, "Simulate taking a node offline", "NODE" }, { "node-fail", 'f', 0, G_OPTION_ARG_CALLBACK, node_fail_cb, "Simulate a node failing", "NODE" }, { "op-inject", 'i', 0, G_OPTION_ARG_CALLBACK, op_inject_cb, "Generate a failure for the cluster to react to in the simulation.\n" INDENT "See `Operation Specification` help for more information.", "OPSPEC" }, { "op-fail", 'F', 0, G_OPTION_ARG_CALLBACK, op_fail_cb, "If the specified task occurs during the simulation, have it fail with return code ${rc}.\n" INDENT "The transition will normally stop at the failed action.\n" INDENT "Save the result with --save-output and re-run with --xml-file.\n" INDENT "See `Operation Specification` help for more information.", "OPSPEC" }, { "set-datetime", 't', 0, G_OPTION_ARG_STRING, &options.use_date, "Set date/time (ISO 8601 format, see https://en.wikipedia.org/wiki/ISO_8601)", "DATETIME" }, { "quorum", 'q', 0, G_OPTION_ARG_CALLBACK, quorum_cb, "Set to '1' (or 'true') to indicate cluster has quorum", "QUORUM" }, { "watchdog", 'w', 0, G_OPTION_ARG_CALLBACK, watchdog_cb, "Set to '1' (or 'true') to indicate cluster has an active watchdog device", "DEVICE" }, { "ticket-grant", 'g', 0, G_OPTION_ARG_CALLBACK, ticket_grant_cb, "Simulate granting a ticket", "TICKET" }, { "ticket-revoke", 'r', 0, G_OPTION_ARG_CALLBACK, ticket_revoke_cb, "Simulate revoking a ticket", "TICKET" }, { "ticket-standby", 'b', 0, G_OPTION_ARG_CALLBACK, ticket_standby_cb, "Simulate making a ticket standby", "TICKET" }, { "ticket-activate", 'e', 0, G_OPTION_ARG_CALLBACK, ticket_activate_cb, "Simulate activating a ticket", "TICKET" }, { NULL } }; static GOptionEntry artifact_entries[] = { { "save-input", 'I', 0, G_OPTION_ARG_FILENAME, &options.input_file, "Save the input configuration to the named file", "FILE" }, { "save-output", 'O', 0, G_OPTION_ARG_FILENAME, &options.output_file, "Save the output configuration to the named file", "FILE" }, { "save-graph", 'G', 0, G_OPTION_ARG_CALLBACK, save_graph_cb, "Save the transition graph (XML format) to the named file", "FILE" }, { "save-dotfile", 'D', 0, G_OPTION_ARG_CALLBACK, save_dotfile_cb, "Save the transition graph (DOT format) to the named file", "FILE" }, { "all-actions", 'a', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, all_actions_cb, "Display all possible actions in DOT graph (even if not part of transition)", NULL }, { NULL } }; static GOptionEntry source_entries[] = { { "live-check", 'L', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, live_check_cb, "Connect to CIB manager and use the current CIB contents as input", NULL }, { "xml-file", 'x', 0, G_OPTION_ARG_CALLBACK, xml_file_cb, "Retrieve XML from the named file", "FILE" }, { "xml-pipe", 'p', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, xml_pipe_cb, "Retrieve XML from stdin", NULL }, { NULL } }; static int setup_input(pcmk__output_t *out, const char *input, const char *output, GError **error) { int rc = pcmk_rc_ok; xmlNode *cib_object = NULL; char *local_output = NULL; if (input == NULL) { /* Use live CIB */ rc = cib__signon_query(out, NULL, &cib_object); if (rc != pcmk_rc_ok) { // cib__signon_query() outputs any relevant error return rc; } } else if (pcmk__str_eq(input, "-", pcmk__str_casei)) { cib_object = filename2xml(NULL); } else { cib_object = filename2xml(input); } if (pcmk_find_cib_element(cib_object, XML_CIB_TAG_STATUS) == NULL) { create_xml_node(cib_object, XML_CIB_TAG_STATUS); } if (cli_config_update(&cib_object, NULL, FALSE) == FALSE) { free_xml(cib_object); return pcmk_rc_transform_failed; } if (validate_xml(cib_object, NULL, FALSE) != TRUE) { free_xml(cib_object); return pcmk_rc_schema_validation; } if (output == NULL) { char *pid = pcmk__getpid_s(); local_output = get_shadow_file(pid); temp_shadow = strdup(local_output); output = local_output; free(pid); } rc = write_xml_file(cib_object, output, FALSE); free_xml(cib_object); cib_object = NULL; if (rc < 0) { rc = pcmk_legacy2rc(rc); g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_CANTCREAT, "Could not create '%s': %s", output, pcmk_rc_str(rc)); return rc; } else { setenv("CIB_file", output, 1); free(local_output); return pcmk_rc_ok; } } static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; GOptionEntry extra_prog_entries[] = { { "quiet", 'Q', 0, G_OPTION_ARG_NONE, &(args->quiet), "Display only essential output", NULL }, { NULL } }; const char *description = "Operation Specification:\n\n" "The OPSPEC in any command line option is of the form\n" "${resource}_${task}_${interval_in_ms}@${node}=${rc}\n" "(memcached_monitor_20000@bart.example.com=7, for example).\n" "${rc} is an OCF return code. For more information on these\n" "return codes, refer to https://clusterlabs.org/pacemaker/doc/2.1/Pacemaker_Administration/html/agents.html#ocf-return-codes\n\n" "Examples:\n\n" "Pretend a recurring monitor action found memcached stopped on node\n" "fred.example.com and, during recovery, that the memcached stop\n" "action failed:\n\n" "\tcrm_simulate -LS --op-inject memcached:0_monitor_20000@bart.example.com=7 " "--op-fail memcached:0_stop_0@fred.example.com=1 --save-output /tmp/memcached-test.xml\n\n" "Now see what the reaction to the stop failed would be:\n\n" "\tcrm_simulate -S --xml-file /tmp/memcached-test.xml\n\n"; context = pcmk__build_arg_context(args, "text (default), xml", group, NULL); pcmk__add_main_args(context, extra_prog_entries); g_option_context_set_description(context, description); pcmk__add_arg_group(context, "operations", "Operations:", "Show operations options", operation_entries); pcmk__add_arg_group(context, "synthetic", "Synthetic Cluster Events:", "Show synthetic cluster event options", synthetic_entries); pcmk__add_arg_group(context, "artifact", "Artifact Options:", "Show artifact options", artifact_entries); pcmk__add_arg_group(context, "source", "Data Source:", "Show data source options", source_entries); return context; } int main(int argc, char **argv) { int rc = pcmk_rc_ok; pe_working_set_t *data_set = NULL; pcmk__output_t *out = NULL; GError *error = NULL; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); gchar **processed_args = pcmk__cmdline_preproc(argv, "bdefgiqrtuwxDFGINOP"); GOptionContext *context = build_arg_context(args, &output_group); options.injections = calloc(1, sizeof(pcmk_injections_t)); if (options.injections == NULL) { rc = ENOMEM; goto done; } /* This must come before g_option_context_parse_strv. */ options.xml_file = strdup("-"); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } pcmk__cli_init_logging("crm_simulate", args->verbosity); rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { fprintf(stderr, "Error creating output format %s: %s\n", args->output_ty, pcmk_rc_str(rc)); exit_code = CRM_EX_ERROR; goto done; } if (pcmk__str_eq(args->output_ty, "text", pcmk__str_null_matches) && !pcmk_is_set(options.flags, pcmk_sim_show_scores) && !pcmk_is_set(options.flags, pcmk_sim_show_utilization)) { pcmk__force_args(context, &error, "%s --text-fancy", g_get_prgname()); } else if (pcmk__str_eq(args->output_ty, "xml", pcmk__str_none)) { pcmk__force_args(context, &error, "%s --xml-simple-list --xml-substitute", g_get_prgname()); } pe__register_messages(out); pcmk__register_lib_messages(out); out->quiet = args->quiet; if (args->version) { out->version(out, false); goto done; } if (args->verbosity > 0) { options.flags |= pcmk_sim_verbose; #ifdef PCMK__COMPAT_2_0 /* Redirect stderr to stdout so we can grep the output */ close(STDERR_FILENO); dup2(STDOUT_FILENO, STDERR_FILENO); #endif } data_set = pe_new_working_set(); if (data_set == NULL) { rc = ENOMEM; g_set_error(&error, PCMK__RC_ERROR, rc, "Could not allocate working set"); goto done; } if (pcmk_is_set(options.flags, pcmk_sim_show_scores)) { - pe__set_working_set_flags(data_set, pe_flag_show_scores); + pe__set_working_set_flags(data_set, pcmk_sched_output_scores); } if (pcmk_is_set(options.flags, pcmk_sim_show_utilization)) { pe__set_working_set_flags(data_set, pe_flag_show_utilization); } pe__set_working_set_flags(data_set, pcmk_sched_no_compat); if (options.test_dir != NULL) { data_set->priv = out; pcmk__profile_dir(options.test_dir, options.repeat, data_set, options.use_date); rc = pcmk_rc_ok; goto done; } rc = setup_input(out, options.xml_file, options.store? options.xml_file : options.output_file, &error); if (rc != pcmk_rc_ok) { goto done; } rc = pcmk__simulate(data_set, out, options.injections, options.flags, section_opts, options.use_date, options.input_file, options.graph_file, options.dot_file); done: pcmk__output_and_clear_error(&error, NULL); /* There sure is a lot to free in options. */ free(options.dot_file); free(options.graph_file); g_free(options.input_file); g_free(options.output_file); g_free(options.test_dir); free(options.use_date); free(options.xml_file); pcmk_free_injections(options.injections); pcmk__free_arg_context(context); g_strfreev(processed_args); if (data_set) { pe_free_working_set(data_set); } fflush(stderr); if (temp_shadow) { unlink(temp_shadow); free(temp_shadow); } if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); } if (out != NULL) { out->finish(out, exit_code, true, NULL); pcmk__output_free(out); } pcmk__unregister_formats(); crm_exit(exit_code); }