diff --git a/include/crm/pengine/pe_types.h b/include/crm/pengine/pe_types.h index 77171e4808..e3ecaa823b 100644 --- a/include/crm/pengine/pe_types.h +++ b/include/crm/pengine/pe_types.h @@ -1,546 +1,552 @@ /* - * Copyright 2004-2021 the Pacemaker project contributors + * Copyright 2004-2022 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__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 #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; enum pe_obj_types { pe_unknown = -1, pe_native = 0, pe_group = 1, pe_clone = 2, pe_container = 3, }; 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) (pe_resource_t*, GList *, gboolean); } resource_object_functions_t; typedef struct resource_alloc_functions_s resource_alloc_functions_t; enum pe_quorum_policy { no_quorum_freeze, no_quorum_stop, no_quorum_ignore, no_quorum_suicide, no_quorum_demote }; enum node_type { node_ping, node_member, node_remote }; //! \deprecated will be removed in a future release enum pe_restart { pe_restart_restart, //! \deprecated will be removed in a future release pe_restart_ignore //! \deprecated will be removed in a future release }; //! Determine behavior of pe_find_resource_with_flags() enum pe_find { pe_find_renamed = 0x001, //!< match resource ID or LRM history ID pe_find_anon = 0x002, //!< match base name of anonymous clone instances pe_find_clone = 0x004, //!< match only clone instances pe_find_current = 0x008, //!< match resource active on specified node pe_find_inactive = 0x010, //!< match resource not running anywhere pe_find_any = 0x020, //!< match base name of any clone instance }; // @TODO Make these an enum # define pe_flag_have_quorum 0x00000001ULL # define pe_flag_symmetric_cluster 0x00000002ULL # define pe_flag_maintenance_mode 0x00000008ULL # define pe_flag_stonith_enabled 0x00000010ULL # define pe_flag_have_stonith_resource 0x00000020ULL # define pe_flag_enable_unfencing 0x00000040ULL # define pe_flag_concurrent_fencing 0x00000080ULL # define pe_flag_stop_rsc_orphans 0x00000100ULL # define pe_flag_stop_action_orphans 0x00000200ULL # define pe_flag_stop_everything 0x00000400ULL # define pe_flag_start_failure_fatal 0x00001000ULL //! \deprecated # define pe_flag_remove_after_stop 0x00002000ULL # define pe_flag_startup_fencing 0x00004000ULL # define pe_flag_shutdown_lock 0x00008000ULL # define pe_flag_startup_probes 0x00010000ULL # define pe_flag_have_status 0x00020000ULL # define pe_flag_have_remote_nodes 0x00040000ULL # define pe_flag_quick_location 0x00100000ULL # define pe_flag_sanitized 0x00200000ULL //! \deprecated # define pe_flag_stdout 0x00400000ULL //! Don't count total, disabled and blocked resource instances # define pe_flag_no_counts 0x00800000ULL /*! Skip deprecated code that is kept solely for backward API compatibility. * (Internal code should always set this.) */ # define pe_flag_no_compat 0x01000000ULL # 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; }; enum pe_check_parameters { /* Clear fail count if parameters changed for un-expired start or monitor * last_failure. */ pe_check_last_failure, /* Clear fail count if parameters changed for start, monitor, promote, or * migrate_from actions for active resources. */ pe_check_active, }; 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 }; struct pe_node_s { int weight; gboolean fixed; 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_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_runnable 0x00040000ULL # define pe_rsc_start_pending 0x00080000ULL # define pe_rsc_starting 0x00100000ULL # define pe_rsc_stopping 0x00200000ULL # define pe_rsc_allow_migrate 0x00800000ULL # define pe_rsc_failure_ignored 0x01000000ULL # 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 enum pe_graph_flags { pe_graph_none = 0x00000, pe_graph_updated_first = 0x00001, pe_graph_updated_then = 0x00002, pe_graph_disable = 0x00004, }; /* *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; //!@{ //! 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; 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 #if ENABLE_VERSIONED_ATTRS xmlNode *versioned_parameters; #endif }; #if ENABLE_VERSIONED_ATTRS // Used as action->action_details if action->rsc is not NULL typedef struct pe_rsc_action_details_s { xmlNode *versioned_parameters; xmlNode *versioned_meta; } pe_rsc_action_details_t; #endif 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/lib/pacemaker/pcmk_sched_allocate.c b/lib/pacemaker/pcmk_sched_allocate.c index b6ec19d6e6..4e73f6a841 100644 --- a/lib/pacemaker/pcmk_sched_allocate.c +++ b/lib/pacemaker/pcmk_sched_allocate.c @@ -1,803 +1,806 @@ /* * Copyright 2004-2022 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include "libpacemaker_private.h" CRM_TRACE_INIT_DATA(pacemaker); /*! * \internal * \brief Do deferred action checks after allocation * * When unpacking the resource history, the scheduler checks for resource * configurations that have changed since an action was run. However, at that * time, bundles using the REMOTE_CONTAINER_HACK don't have their final * parameter information, so instead they add a deferred check to a list. This * function processes one entry in that list. * * \param[in] rsc Resource that action history is for * \param[in] node Node that action history is for * \param[in] rsc_op Action history entry * \param[in] check Type of deferred check to do * \param[in] data_set Working set for cluster */ static void check_params(pe_resource_t *rsc, pe_node_t *node, xmlNode *rsc_op, enum pe_check_parameters check, pe_working_set_t *data_set) { const char *reason = NULL; op_digest_cache_t *digest_data = NULL; switch (check) { case pe_check_active: if (pcmk__check_action_config(rsc, node, rsc_op) && pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL, data_set)) { reason = "action definition changed"; } break; case pe_check_last_failure: digest_data = rsc_action_digest_cmp(rsc, rsc_op, node, data_set); switch (digest_data->rc) { case RSC_DIGEST_UNKNOWN: crm_trace("Resource %s history entry %s on %s has " "no digest to compare", rsc->id, ID(rsc_op), node->details->id); break; case RSC_DIGEST_MATCH: break; default: reason = "resource parameters have changed"; break; } break; } if (reason != NULL) { pe__clear_failcount(rsc, node, reason, data_set); } } /*! * \internal * \brief Check whether a resource has failcount clearing scheduled on a node * * \param[in] node Node to check * \param[in] rsc Resource to check * * \return true if \p rsc has failcount clearing scheduled on \p node, * otherwise false */ static bool failcount_clear_action_exists(pe_node_t *node, pe_resource_t *rsc) { GList *list = pe__resource_actions(rsc, node, CRM_OP_CLEAR_FAILCOUNT, TRUE); if (list != NULL) { g_list_free(list); return true; } return false; } /*! * \internal * \brief Ban a resource from a node if it reached its failure threshold there * * \param[in] rsc Resource to check failure threshold for * \param[in] node Node to check \p rsc on */ static void check_failure_threshold(pe_resource_t *rsc, pe_node_t *node) { // If this is a collective resource, apply recursively to children instead if (rsc->children != NULL) { g_list_foreach(rsc->children, (GFunc) check_failure_threshold, node); return; } else if (failcount_clear_action_exists(node, rsc)) { /* Don't force the resource away from this node due to a failcount * that's going to be cleared. * * @TODO Failcount clearing can be scheduled in * pcmk__handle_rsc_config_changes() via process_rsc_history(), or in * stage5() via check_params(). This runs well before then, so it cannot * detect those, meaning we might check the migration threshold when we * shouldn't. Worst case, we stop or move the resource, then move it * back in the next transition. */ return; } else { pe_resource_t *failed = NULL; if (pcmk__threshold_reached(rsc, node, &failed)) { resource_location(failed, node, -INFINITY, "__fail_limit__", rsc->cluster); } } } /*! * \internal * \brief Apply stickiness to a resource if appropriate * * \param[in] rsc Resource to check for stickiness * \param[in] data_set Cluster working set */ static void apply_stickiness(pe_resource_t *rsc, pe_working_set_t *data_set) { pe_node_t *node = NULL; // If this is a collective resource, apply recursively to children instead if (rsc->children != NULL) { g_list_foreach(rsc->children, (GFunc) apply_stickiness, data_set); return; } /* A resource is sticky if it is managed, has stickiness configured, and is * active on a single node. */ if (!pcmk_is_set(rsc->flags, pe_rsc_managed) || (rsc->stickiness < 1) || !pcmk__list_of_1(rsc->running_on)) { return; } node = rsc->running_on->data; /* In a symmetric cluster, stickiness can always be used. In an * asymmetric cluster, we have to check whether the resource is still * allowed on the node, so we don't keep the resource somewhere it is no * longer explicitly enabled. */ if (!pcmk_is_set(rsc->cluster->flags, pe_flag_symmetric_cluster) && (pe_hash_table_lookup(rsc->allowed_nodes, node->details->id) == NULL)) { pe_rsc_debug(rsc, "Ignoring %s stickiness because the cluster is " "asymmetric and node %s is not explicitly allowed", rsc->id, node->details->uname); return; } pe_rsc_debug(rsc, "Resource %s has %d stickiness on node %s", rsc->id, rsc->stickiness, node->details->uname); resource_location(rsc, node, rsc->stickiness, "stickiness", rsc->cluster); } gboolean stage0(pe_working_set_t * data_set) { if (data_set->input == NULL) { return FALSE; } if (!pcmk_is_set(data_set->flags, pe_flag_have_status)) { crm_trace("Calculating status"); cluster_status(data_set); } pcmk__set_allocation_methods(data_set); pcmk__apply_node_health(data_set); pcmk__unpack_constraints(data_set); return TRUE; } static void rsc_discover_filter(pe_resource_t *rsc, pe_node_t *node) { pe_resource_t *top = uber_parent(rsc); pe_node_t *match; if (rsc->exclusive_discover == FALSE && top->exclusive_discover == FALSE) { return; } g_list_foreach(rsc->children, (GFunc) rsc_discover_filter, node); match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (match && match->rsc_discover_mode != pe_discover_exclusive) { match->weight = -INFINITY; } } static time_t shutdown_time(pe_node_t *node, pe_working_set_t *data_set) { const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN); time_t result = 0; if (shutdown) { long long result_ll; if (pcmk__scan_ll(shutdown, &result_ll, 0LL) == pcmk_rc_ok) { result = (time_t) result_ll; } } return result? result : get_effective_time(data_set); } static void apply_shutdown_lock(pe_resource_t *rsc, pe_working_set_t *data_set) { const char *class; // Only primitives and (uncloned) groups may be locked if (rsc->variant == pe_group) { g_list_foreach(rsc->children, (GFunc) apply_shutdown_lock, data_set); } else if (rsc->variant != pe_native) { return; } // Fence devices and remote connections can't be locked class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches) || pe__resource_is_remote_conn(rsc, data_set)) { 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, data_set); 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, node->details->uname); } else { rsc->lock_node = node; rsc->lock_time = shutdown_time(node, data_set); } } } if (rsc->lock_node == NULL) { // No lock needed return; } if (data_set->shutdown_lock > 0) { time_t lock_expiration = rsc->lock_time + data_set->shutdown_lock; pe_rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)", rsc->id, rsc->lock_node->details->uname, (long long) lock_expiration); pe__update_recheck_time(++lock_expiration, data_set); } else { pe_rsc_info(rsc, "Locking %s to %s due to shutdown", rsc->id, rsc->lock_node->details->uname); } // If resource is locked to one node, ban it from all other nodes for (GList *item = data_set->nodes; item != NULL; item = item->next) { pe_node_t *node = item->data; if (strcmp(node->details->uname, rsc->lock_node->details->uname)) { resource_location(rsc, node, -CRM_SCORE_INFINITY, XML_CONFIG_ATTR_SHUTDOWN_LOCK, data_set); } } } /* * \internal * \brief Stage 2 of cluster status: apply node-specific criteria * * Count known nodes, and apply location constraints, stickiness, and exclusive * resource discovery. */ gboolean stage2(pe_working_set_t * data_set) { GList *gIter = NULL; if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { g_list_foreach(data_set->resources, (GFunc) apply_shutdown_lock, data_set); } if (!pcmk_is_set(data_set->flags, pe_flag_no_compat)) { // @COMPAT API backward compatibility for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; if (node && (node->weight >= 0) && node->details->online && (node->details->type != node_ping)) { data_set->max_valid_nodes++; } } crm_trace("Online node count: %d", data_set->max_valid_nodes); } pcmk__apply_locations(data_set); g_list_foreach(data_set->resources, (GFunc) apply_stickiness, data_set); gIter = data_set->nodes; for (; gIter != NULL; gIter = gIter->next) { GList *gIter2 = NULL; pe_node_t *node = (pe_node_t *) gIter->data; gIter2 = data_set->resources; for (; gIter2 != NULL; gIter2 = gIter2->next) { pe_resource_t *rsc = (pe_resource_t *) gIter2->data; check_failure_threshold(rsc, node); rsc_discover_filter(rsc, node); } } return TRUE; } static void allocate_resources(pe_working_set_t * data_set) { GList *gIter = NULL; if (pcmk_is_set(data_set->flags, pe_flag_have_remote_nodes)) { /* Allocate remote connection resources first (which will also allocate * any colocation dependencies). If the connection is migrating, always * prefer the partial migration target. */ for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; if (rsc->is_remote_node == FALSE) { continue; } pe_rsc_trace(rsc, "Allocating remote connection resource '%s'", rsc->id); rsc->cmds->allocate(rsc, rsc->partial_migration_target, data_set); } } /* now do the rest of the resources */ for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; if (rsc->is_remote_node == TRUE) { continue; } pe_rsc_trace(rsc, "Allocating %s resource '%s'", crm_element_name(rsc->xml), rsc->id); rsc->cmds->allocate(rsc, NULL, data_set); } } // Clear fail counts for orphaned rsc on all online nodes static void cleanup_orphans(pe_resource_t * rsc, pe_working_set_t * data_set) { GList *gIter = NULL; for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; if (node->details->online && pe_get_failcount(node, rsc, NULL, pe_fc_effective, NULL, data_set)) { pe_action_t *clear_op = NULL; clear_op = pe__clear_failcount(rsc, node, "it is orphaned", data_set); /* We can't use order_action_then_stop() here because its * pe_order_preserve breaks things */ pcmk__new_ordering(clear_op->rsc, NULL, clear_op, rsc, stop_key(rsc), NULL, pe_order_optional, data_set); } } } gboolean stage5(pe_working_set_t * data_set) { pcmk__output_t *out = data_set->priv; GList *gIter = NULL; if (!pcmk__str_eq(data_set->placement_strategy, "default", pcmk__str_casei)) { pcmk__sort_resources(data_set); } gIter = data_set->nodes; for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; if (pcmk_is_set(data_set->flags, pe_flag_show_utilization)) { out->message(out, "node-capacity", node, "Original"); } } crm_trace("Allocating services"); /* Take (next) highest resource, assign it and create its actions */ allocate_resources(data_set); gIter = data_set->nodes; for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; if (pcmk_is_set(data_set->flags, pe_flag_show_utilization)) { out->message(out, "node-capacity", node, "Remaining"); } } // Process deferred action checks pe__foreach_param_check(data_set, check_params); pe__free_param_checks(data_set); if (pcmk_is_set(data_set->flags, pe_flag_startup_probes)) { crm_trace("Calculating needed probes"); pcmk__schedule_probes(data_set); } crm_trace("Handle orphans"); if (pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)) { for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; /* There's no need to recurse into rsc->children because those * should just be unallocated clone instances. */ if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) { cleanup_orphans(rsc, data_set); } } } crm_trace("Creating actions"); for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; rsc->cmds->create_actions(rsc, data_set); } crm_trace("Creating done"); return TRUE; } static gboolean is_managed(const pe_resource_t * rsc) { GList *gIter = rsc->children; if (pcmk_is_set(rsc->flags, pe_rsc_managed)) { return TRUE; } for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; if (is_managed(child_rsc)) { return TRUE; } } return FALSE; } static gboolean any_managed_resources(pe_working_set_t * data_set) { GList *gIter = data_set->resources; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; if (is_managed(rsc)) { return TRUE; } } return FALSE; } /* * Create dependencies for stonith and shutdown operations */ gboolean stage6(pe_working_set_t * data_set) { pe_action_t *dc_down = NULL; pe_action_t *stonith_op = NULL; gboolean integrity_lost = FALSE; gboolean need_stonith = TRUE; GList *gIter; GList *stonith_ops = NULL; GList *shutdown_ops = NULL; /* Remote ordering constraints need to happen prior to calculating fencing * because it is one more place we can mark nodes as needing fencing. */ pcmk__order_remote_connection_actions(data_set); crm_trace("Processing fencing and shutdown cases"); if (any_managed_resources(data_set) == FALSE) { crm_notice("Delaying fencing operations until there are resources to manage"); need_stonith = FALSE; } /* Check each node for stonith/shutdown */ for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; /* Guest nodes are "fenced" by recovering their container resource, * so handle them separately. */ if (pe__is_guest_node(node)) { if (node->details->remote_requires_reset && need_stonith && pe_can_fence(data_set, node)) { pcmk__fence_guest(node, data_set); } continue; } stonith_op = NULL; if (node->details->unclean && need_stonith && pe_can_fence(data_set, node)) { stonith_op = pe_fence_op(node, NULL, FALSE, "node is unclean", FALSE, data_set); pe_warn("Scheduling Node %s for STONITH", node->details->uname); pcmk__order_vs_fence(stonith_op, data_set); if (node->details->is_dc) { // Remember if the DC is being fenced dc_down = stonith_op; } else { if (!pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing) && (stonith_ops != NULL)) { /* Concurrent fencing is disabled, so order each non-DC * fencing in a chain. If there is any DC fencing or * shutdown, it will be ordered after the last action in the * chain later. */ order_actions((pe_action_t *) stonith_ops->data, stonith_op, pe_order_optional); } // Remember all non-DC fencing actions in a separate list stonith_ops = g_list_prepend(stonith_ops, stonith_op); } } else if (node->details->online && node->details->shutdown && /* TODO define what a shutdown op means for a remote node. * For now we do not send shutdown operations for remote nodes, but * if we can come up with a good use for this in the future, we will. */ pe__is_guest_or_remote_node(node) == FALSE) { pe_action_t *down_op = pcmk__new_shutdown_action(node, data_set); if (node->details->is_dc) { // Remember if the DC is being shut down dc_down = down_op; } else { // Remember non-DC shutdowns for later ordering shutdown_ops = g_list_prepend(shutdown_ops, down_op); } } if (node->details->unclean && stonith_op == NULL) { integrity_lost = TRUE; pe_warn("Node %s is unclean!", node->details->uname); } } if (integrity_lost) { if (!pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { pe_warn("YOUR RESOURCES ARE NOW LIKELY COMPROMISED"); pe_err("ENABLE STONITH TO KEEP YOUR RESOURCES SAFE"); } else if (!pcmk_is_set(data_set->flags, pe_flag_have_quorum)) { crm_notice("Cannot fence unclean nodes until quorum is" " attained (or no-quorum-policy is set to ignore)"); } } if (dc_down != NULL) { /* Order any non-DC shutdowns before any DC shutdown, to avoid repeated * DC elections. However, we don't want to order non-DC shutdowns before * a DC *fencing*, because even though we don't want a node that's * shutting down to become DC, the DC fencing could be ordered before a * clone stop that's also ordered before the shutdowns, thus leading to * a graph loop. */ if (pcmk__str_eq(dc_down->task, CRM_OP_SHUTDOWN, pcmk__str_casei)) { for (gIter = shutdown_ops; gIter != NULL; gIter = gIter->next) { pe_action_t *node_stop = (pe_action_t *) gIter->data; crm_debug("Ordering shutdown on %s before %s on DC %s", node_stop->node->details->uname, dc_down->task, dc_down->node->details->uname); order_actions(node_stop, dc_down, pe_order_optional); } } // Order any non-DC fencing before any DC fencing or shutdown if (pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)) { /* With concurrent fencing, order each non-DC fencing action * separately before any DC fencing or shutdown. */ for (gIter = stonith_ops; gIter != NULL; gIter = gIter->next) { order_actions((pe_action_t *) gIter->data, dc_down, pe_order_optional); } } else if (stonith_ops) { /* Without concurrent fencing, the non-DC fencing actions are * already ordered relative to each other, so we just need to order * the DC fencing after the last action in the chain (which is the * first item in the list). */ order_actions((pe_action_t *) stonith_ops->data, dc_down, pe_order_optional); } } g_list_free(stonith_ops); g_list_free(shutdown_ops); return TRUE; } static void log_resource_details(pe_working_set_t *data_set) { pcmk__output_t *out = data_set->priv; GList *all = NULL; /* We need a list of nodes that we are allowed to output information for. * This is necessary because out->message for all the resource-related * messages expects such a list, due to the `crm_mon --node=` feature. Here, * we just make it a list of all the nodes. */ all = g_list_prepend(all, (gpointer) "*"); for (GList *item = data_set->resources; item != NULL; item = item->next) { pe_resource_t *rsc = (pe_resource_t *) item->data; // Log all resources except inactive orphans if (!pcmk_is_set(rsc->flags, pe_rsc_orphan) || (rsc->role != RSC_ROLE_STOPPED)) { out->message(out, crm_map_element_name(rsc->xml), 0, rsc, all, all); } } g_list_free(all); } static void log_all_actions(pe_working_set_t *data_set) { /* This only ever outputs to the log, so ignore whatever output object was * previously set and just log instead. */ pcmk__output_t *prev_out = data_set->priv; pcmk__output_t *out = pcmk__new_logger(); if (out == NULL) { return; } pcmk__output_set_log_level(out, LOG_NOTICE); data_set->priv = out; out->begin_list(out, NULL, NULL, "Actions"); pcmk__output_actions(data_set); out->end_list(out); out->finish(out, CRM_EX_OK, true, NULL); pcmk__output_free(out); data_set->priv = prev_out; } /*! * \internal * \brief Run the scheduler for a given CIB * * \param[in,out] data_set Cluster working set * \param[in] xml_input CIB XML to use as scheduler input */ xmlNode * pcmk__schedule_actions(pe_working_set_t *data_set, xmlNode *xml_input) { GList *gIter = NULL; CRM_ASSERT(xml_input || pcmk_is_set(data_set->flags, pe_flag_have_status)); if (!pcmk_is_set(data_set->flags, pe_flag_have_status)) { set_working_set_defaults(data_set); data_set->input = xml_input; } else { crm_trace("Already have status - reusing"); } crm_trace("Calculate cluster status"); stage0(data_set); + if (pcmk_is_set(data_set->flags, pe_flag_check_config)) { + return data_set->graph; + } if (!pcmk_is_set(data_set->flags, pe_flag_quick_location) && pcmk__is_daemon) { log_resource_details(data_set); } crm_trace("Applying location constraints"); stage2(data_set); if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) { return NULL; } pcmk__create_internal_constraints(data_set); pcmk__handle_rsc_config_changes(data_set); crm_trace("Allocate resources"); stage5(data_set); crm_trace("Processing fencing and shutdown cases"); stage6(data_set); pcmk__apply_orderings(data_set); log_all_actions(data_set); crm_trace("Create transition graph"); pcmk__create_graph(data_set); crm_trace("=#=#=#=#= Summary =#=#=#=#="); crm_trace("\t========= Set %d (Un-runnable) =========", -1); if (get_crm_log_level() == LOG_TRACE) { gIter = data_set->actions; for (; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (!pcmk_any_flags_set(action->flags, pe_action_optional |pe_action_runnable |pe_action_pseudo)) { pcmk__log_action("\t", action, true); } } } return data_set->graph; } diff --git a/tools/crm_verify.c b/tools/crm_verify.c index 881fa203d5..b20370ae0b 100644 --- a/tools/crm_verify.c +++ b/tools/crm_verify.c @@ -1,283 +1,283 @@ /* - * Copyright 2004-2021 the Pacemaker project contributors + * Copyright 2004-2022 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU 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 const char *SUMMARY = "Check a Pacemaker configuration for errors\n\n" "Check the well-formedness of a complete Pacemaker XML configuration,\n" "its conformance to the configured schema, and the presence of common\n" "misconfigurations. Problems reported as errors must be fixed before the\n" "cluster will work properly. It is left to the administrator to decide\n" "whether to fix problems reported as warnings."; struct { char *cib_save; gboolean use_live_cib; char *xml_file; gboolean xml_stdin; char *xml_string; } options; -extern gboolean stage0(pe_working_set_t * data_set); - static GOptionEntry data_entries[] = { { "live-check", 'L', 0, G_OPTION_ARG_NONE, &options.use_live_cib, "Check the configuration used by the running cluster", NULL }, { "xml-file", 'x', 0, G_OPTION_ARG_FILENAME, &options.xml_file, "Check the configuration in the named file", "FILE" }, { "xml-pipe", 'p', 0, G_OPTION_ARG_NONE, &options.xml_stdin, "Check the configuration piped in via stdin", NULL }, { "xml-text", 'X', 0, G_OPTION_ARG_STRING, &options.xml_string, "Check the configuration in the supplied string", "XML" }, { NULL } }; static GOptionEntry addl_entries[] = { { "save-xml", 'S', G_OPTION_FLAG_NONE, G_OPTION_ARG_FILENAME, &options.cib_save, "Save verified XML to named file (most useful with -L)", "FILE" }, { NULL } }; static pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; const char *description = "Examples:\n\n" "Check the consistency of the configuration in the running cluster:\n\n" "\tcrm_verify --live-check\n\n" "Check the consistency of the configuration in a given file and " "produce verbose output:\n\n" "\tcrm_verify --xml-file file.xml --verbose\n\n"; context = pcmk__build_arg_context(args, "text (default), xml", group, NULL); g_option_context_set_description(context, description); pcmk__add_arg_group(context, "data", "Data sources:", "Show data options", data_entries); pcmk__add_arg_group(context, "additional", "Additional options:", "Show additional options", addl_entries); return context; } int main(int argc, char **argv) { xmlNode *cib_object = NULL; xmlNode *status = NULL; pe_working_set_t *data_set = NULL; const char *xml_tag = NULL; int rc = pcmk_rc_ok; crm_exit_t exit_code = CRM_EX_OK; GError *error = NULL; pcmk__output_t *out = NULL; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); gchar **processed_args = pcmk__cmdline_preproc(argv, "xSX"); GOptionContext *context = build_arg_context(args, &output_group); 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_verify", args->verbosity); rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_ERROR; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); goto done; } if (args->version) { out->version(out, false); goto done; } pcmk__register_lib_messages(out); crm_info("=#=#=#=#= Getting XML =#=#=#=#="); if (options.use_live_cib) { crm_info("Reading XML from: live cluster"); rc = cib__signon_query(NULL, &cib_object); if (rc != pcmk_rc_ok) { g_set_error(&error, PCMK__RC_ERROR, rc, "CIB query failed: %s", pcmk_rc_str(rc)); goto done; } } else if (options.xml_file != NULL) { cib_object = filename2xml(options.xml_file); if (cib_object == NULL) { rc = ENODATA; g_set_error(&error, PCMK__RC_ERROR, rc, "Couldn't parse input file: %s", options.xml_file); goto done; } } else if (options.xml_string != NULL) { cib_object = string2xml(options.xml_string); if (cib_object == NULL) { rc = ENODATA; g_set_error(&error, PCMK__RC_ERROR, rc, "Couldn't parse input string: %s", options.xml_string); goto done; } } else if (options.xml_stdin) { cib_object = stdin2xml(); if (cib_object == NULL) { rc = ENODATA; g_set_error(&error, PCMK__RC_ERROR, rc, "Couldn't parse input from STDIN."); goto done; } } else { rc = ENODATA; g_set_error(&error, PCMK__RC_ERROR, rc, "No configuration source specified. Use --help for usage information."); goto done; } xml_tag = crm_element_name(cib_object); if (!pcmk__str_eq(xml_tag, XML_TAG_CIB, pcmk__str_casei)) { rc = EBADMSG; g_set_error(&error, PCMK__RC_ERROR, rc, "This tool can only check complete configurations (i.e. those starting with )."); goto done; } if (options.cib_save != NULL) { write_xml_file(cib_object, options.cib_save, FALSE); } status = pcmk_find_cib_element(cib_object, XML_CIB_TAG_STATUS); if (status == NULL) { create_xml_node(cib_object, XML_CIB_TAG_STATUS); } if (validate_xml(cib_object, NULL, FALSE) == FALSE) { pcmk__config_err("CIB did not pass schema validation"); free_xml(cib_object); cib_object = NULL; } else if (cli_config_update(&cib_object, NULL, FALSE) == FALSE) { crm_config_error = TRUE; free_xml(cib_object); cib_object = NULL; out->err(out, "The cluster will NOT be able to use this configuration.\n" "Please manually update the configuration to conform to the %s syntax.", xml_latest_schema()); } data_set = pe_new_working_set(); if (data_set == NULL) { rc = errno; crm_perror(LOG_CRIT, "Unable to allocate working set"); goto done; } pe__set_working_set_flags(data_set, pe_flag_no_counts|pe_flag_no_compat); data_set->priv = out; - if (cib_object == NULL) { - } else if (status != NULL || options.use_live_cib) { - /* live queries will always have a status section and can do a full simulation */ + /* Process the configuration to set crm_config_error/crm_config_warning. + * + * @TODO Some parts of the configuration are unpacked only when needed (for + * example, action configuration), so we aren't necessarily checking those. + */ + if (cib_object != NULL) { + if ((status == NULL) && !options.use_live_cib) { + // No status available, so do minimal checks + pe__set_working_set_flags(data_set, pe_flag_check_config); + } pcmk__schedule_actions(data_set, cib_object); - - } else { - data_set->now = crm_time_new(NULL); - data_set->input = cib_object; - stage0(data_set); } pe_free_working_set(data_set); if (crm_config_error) { rc = pcmk_rc_schema_validation; if (args->verbosity > 0) { g_set_error(&error, PCMK__RC_ERROR, rc, "Errors found during check: config not valid"); } else { g_set_error(&error, PCMK__RC_ERROR, rc, "Errors found during check: config not valid\n-V may provide more details"); } } else if (crm_config_warning) { rc = pcmk_rc_schema_validation; if (args->verbosity > 0) { g_set_error(&error, PCMK__RC_ERROR, rc, "Warnings found during check: config may not be valid"); } else { g_set_error(&error, PCMK__RC_ERROR, rc, "Warnings found during check: config may not be valid\n-V may provide more details"); } } done: g_strfreev(processed_args); pcmk__free_arg_context(context); free(options.cib_save); free(options.xml_file); free(options.xml_string); if (exit_code == CRM_EX_OK) { exit_code = pcmk_rc2exitc(rc); } pcmk__output_and_clear_error(error, NULL); if (out != NULL) { out->finish(out, exit_code, true, NULL); pcmk__output_free(out); } crm_exit(exit_code); }