diff --git a/include/crm/common/resources.h b/include/crm/common/resources.h index 8927365706..d5522ec5f6 100644 --- a/include/crm/common/resources.h +++ b/include/crm/common/resources.h @@ -1,163 +1,161 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_RESOURCES__H #define PCMK__CRM_COMMON_RESOURCES__H #include // bool #include // time_t #include // xmlNode #include // gboolean, guint, GList, GHashTable #include // enum rsc_role_e #include // pcmk_resource_t, etc. #ifdef __cplusplus extern "C" { #endif /*! * \file * \brief Scheduler API for resources * \ingroup core */ //!@{ //! \deprecated Do not use // What resource needs before it can be recovered from a failed node enum rsc_start_requirement { pcmk_requires_nothing = 0, // Resource can be recovered immediately pcmk_requires_quorum = 1, // Resource can be recovered if quorate pcmk_requires_fencing = 2, // Resource can be recovered after fencing #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) rsc_req_nothing = pcmk_requires_nothing, rsc_req_quorum = pcmk_requires_quorum, rsc_req_stonith = pcmk_requires_fencing, #endif }; //!@} //! Search options for resources (exact resource ID always matches) enum pe_find { //! Also match clone instance ID from resource history pcmk_rsc_match_history = (1 << 0), //! Also match anonymous clone instances by base name pcmk_rsc_match_anon_basename = (1 << 1), //! Match only clones and their instances, by either clone or instance ID pcmk_rsc_match_clone_only = (1 << 2), //! If matching by node, compare current node instead of assigned node pcmk_rsc_match_current_node = (1 << 3), //! \deprecated Do not use pe_find_inactive = (1 << 4), //! Match clone instances (even unique) by base name as well as exact ID pcmk_rsc_match_basename = (1 << 5), #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) //! \deprecated Use pcmk_rsc_match_history instead pe_find_renamed = pcmk_rsc_match_history, //! \deprecated Use pcmk_rsc_match_anon_basename instead pe_find_anon = pcmk_rsc_match_anon_basename, //! \deprecated Use pcmk_rsc_match_clone_only instead pe_find_clone = pcmk_rsc_match_clone_only, //! \deprecated Use pcmk_rsc_match_current_node instead pe_find_current = pcmk_rsc_match_current_node, //! \deprecated Use pcmk_rsc_match_basename instead pe_find_any = pcmk_rsc_match_basename, #endif }; //! \internal Do not use typedef struct pcmk__resource_private pcmk__resource_private_t; // Implementation of pcmk_resource_t // @COMPAT Make this internal when we can break API backward compatibility //!@{ //! \deprecated Do not use (public access will be removed in a future release) struct pe_resource_s { /* @COMPAT Once all members are moved to pcmk__resource_private_t, * We can make that the pcmk_resource_t implementation and drop this * struct altogether, leaving pcmk_resource_t as an opaque public type. */ pcmk__resource_private_t *private; // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_resource_id() instead char *id; // Resource ID in configuration // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_resource_is_managed() instead unsigned long long flags; // Group of enum pcmk__rsc_flags - pcmk_node_t *allocated_to; // Node resource is assigned to - // The destination node, if migrate_to completed but migrate_from has not pcmk_node_t *partial_migration_target; // The source node, if migrate_to completed but migrate_from has not pcmk_node_t *partial_migration_source; // Nodes where resource may be active GList *running_on; // Nodes where resource has been probed (key is node ID, not name) GHashTable *known_on; // Nodes where resource may run (key is node ID, not name) GHashTable *allowed_nodes; enum rsc_role_e role; // Resource's current role enum rsc_role_e next_role; // Resource's scheduled next role GHashTable *meta; // Resource's meta-attributes GHashTable *parameters; // \deprecated Use pe_rsc_params() instead GHashTable *utilization; // Resource's utilization attributes GList *children; // Resource's child resources, if any // Source nodes where stop is needed after migrate_from and migrate_to GList *dangling_migrations; pcmk_resource_t *container; // Resource containing this one, if any GList *fillers; // Resources contained by this one, if any // @COMPAT These should be made const at next API compatibility break pcmk_node_t *pending_node; // Node on which pending_action is happening pcmk_node_t *lock_node; // Resource 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 has node names as keys and parameter * name/value tables as values. Use pe_rsc_params() to get the table for a * given node rather than use this directly. */ GHashTable *parameter_cache; }; //!@} const char *pcmk_resource_id(const pcmk_resource_t *rsc); bool pcmk_resource_is_managed(const pcmk_resource_t *rsc); #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_RESOURCES__H diff --git a/include/crm/common/resources_internal.h b/include/crm/common/resources_internal.h index 2c03ab5e03..5cdd0b5390 100644 --- a/include/crm/common/resources_internal.h +++ b/include/crm/common/resources_internal.h @@ -1,388 +1,396 @@ /* * Copyright 2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_RESOURCES_INTERNAL__H #define PCMK__CRM_COMMON_RESOURCES_INTERNAL__H #include // gboolean, GList #include // xmlNode #include // pcmk_resource_t #include // enum rsc_role_e #include // pcmk_node_t, etc. #ifdef __cplusplus extern "C" { #endif /*! * \internal * \brief Set resource flags * * \param[in,out] resource Resource to set flags for * \param[in] flags_to_set Group of enum pcmk_rsc_flags to set */ #define pcmk__set_rsc_flags(resource, flags_to_set) do { \ (resource)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Resource", (resource)->id, (resource)->flags, \ (flags_to_set), #flags_to_set); \ } while (0) /*! * \internal * \brief Clear resource flags * * \param[in,out] resource Resource to clear flags for * \param[in] flags_to_clear Group of enum pcmk_rsc_flags to clear */ #define pcmk__clear_rsc_flags(resource, flags_to_clear) do { \ (resource)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Resource", (resource)->id, (resource)->flags, \ (flags_to_clear), #flags_to_clear); \ } while (0) //! Resource variants supported by Pacemaker enum pcmk__rsc_variant { // Order matters: some code compares greater or lesser than pcmk__rsc_variant_unknown = -1, //!< Unknown resource variant pcmk__rsc_variant_primitive = 0, //!< Primitive resource pcmk__rsc_variant_group = 1, //!< Group resource pcmk__rsc_variant_clone = 2, //!< Clone resource pcmk__rsc_variant_bundle = 3, //!< Bundle resource }; //! How to recover a resource that is incorrectly active on multiple nodes enum pcmk__multiply_active { pcmk__multiply_active_restart, //!< Stop on all, start on desired pcmk__multiply_active_stop, //!< Stop on all and leave stopped pcmk__multiply_active_block, //!< Do nothing to resource pcmk__multiply_active_unexpected, //!< Stop unexpected instances }; //! \deprecated enum pcmk__restart { pcmk__restart_restart, pcmk__restart_ignore, }; //! Resource scheduling flags enum pcmk__rsc_flags { // No resource flags set (compare with equality rather than bit set) pcmk__no_rsc_flags = 0ULL, // Whether resource has been removed from the configuration pcmk__rsc_removed = (1ULL << 0), /* NOTE: sbd (at least as of 1.5.2) uses pe_rsc_managed which equates to * this value, so the value should not be changed */ // Whether resource is managed pcmk__rsc_managed = (1ULL << 1), // Whether resource is blocked from further action pcmk__rsc_blocked = (1ULL << 2), // Whether resource has been removed but has a container pcmk__rsc_removed_filler = (1ULL << 3), // Whether resource has clone notifications enabled pcmk__rsc_notify = (1ULL << 4), // Whether resource is not an anonymous clone instance pcmk__rsc_unique = (1ULL << 5), // Whether resource's class is "stonith" pcmk__rsc_fence_device = (1ULL << 6), // Whether resource can be promoted and demoted pcmk__rsc_promotable = (1ULL << 7), // Whether resource has not yet been assigned to a node pcmk__rsc_unassigned = (1ULL << 8), // Whether resource is in the process of being assigned to a node pcmk__rsc_assigning = (1ULL << 9), // Whether resource is in the process of modifying allowed node scores pcmk__rsc_updating_nodes = (1ULL << 10), // Whether resource is in the process of scheduling actions to restart pcmk__rsc_restarting = (1ULL << 11), // Whether resource must be stopped (instead of demoted) if it is failed pcmk__rsc_stop_if_failed = (1ULL << 12), // Whether a reload action has been scheduled for resource pcmk__rsc_reload = (1ULL << 13), // Whether resource is a remote connection allowed to run on a remote node pcmk__rsc_remote_nesting_allowed = (1ULL << 14), // Whether resource has \c PCMK_META_CRITICAL meta-attribute enabled pcmk__rsc_critical = (1ULL << 15), // Whether resource is considered failed pcmk__rsc_failed = (1ULL << 16), // Flag for non-scheduler code to use to detect recursion loops pcmk__rsc_detect_loop = (1ULL << 17), // Whether resource is a Pacemaker Remote connection pcmk__rsc_is_remote_connection = (1ULL << 18), // Whether resource has pending start action in history pcmk__rsc_start_pending = (1ULL << 19), // Whether resource is probed only on nodes marked exclusive pcmk__rsc_exclusive_probes = (1ULL << 20), /* * Whether resource is multiply active with recovery set to * \c PCMK_VALUE_STOP_UNEXPECTED */ pcmk__rsc_stop_unexpected = (1ULL << 22), // Whether resource is allowed to live-migrate pcmk__rsc_migratable = (1ULL << 23), // Whether resource has an ignorable failure pcmk__rsc_ignore_failure = (1ULL << 24), // Whether resource is an implicit container resource for a bundle replica pcmk__rsc_replica_container = (1ULL << 25), // Whether resource, its node, or entire cluster is in maintenance mode pcmk__rsc_maintenance = (1ULL << 26), // Whether resource can be started or promoted only on quorate nodes pcmk__rsc_needs_quorum = (1ULL << 28), // Whether resource requires fencing before recovery if on unclean node pcmk__rsc_needs_fencing = (1ULL << 29), // Whether resource can be started or promoted only on unfenced nodes pcmk__rsc_needs_unfencing = (1ULL << 30), }; //! Resource assignment methods (implementation defined by libpacemaker) typedef struct pcmk__assignment_methods pcmk__assignment_methods_t; //! Resource object methods typedef struct { /*! * \internal * \brief Parse variant-specific resource XML from CIB into struct members * * \param[in,out] rsc Partially unpacked resource * \param[in,out] scheduler Scheduler data * * \return TRUE if resource was unpacked successfully, otherwise FALSE */ gboolean (*unpack)(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler); /*! * \internal * \brief Search for a resource ID in a resource and its children * * \param[in] rsc Search this resource and its children * \param[in] id Search for this resource ID * \param[in] on_node If not NULL, limit search to resources on this node * \param[in] flags Group of enum pe_find flags * * \return Resource that matches search criteria if any, otherwise NULL */ pcmk_resource_t *(*find_rsc)(pcmk_resource_t *rsc, const char *search, const pcmk_node_t *node, int flags); /*! * \internal * \brief Get value of a resource instance attribute * * \param[in,out] rsc Resource to check * \param[in] node Node to use to evaluate rules * \param[in] create Ignored * \param[in] name Name of instance attribute to check * \param[in,out] scheduler Scheduler data * * \return Value of requested attribute if available, otherwise NULL * \note The caller is responsible for freeing the result using free(). */ char *(*parameter)(pcmk_resource_t *rsc, pcmk_node_t *node, gboolean create, const char *name, pcmk_scheduler_t *scheduler); /*! * \internal * \brief Check whether a resource is active * * \param[in] rsc Resource to check * \param[in] all If \p rsc is collective, all instances must be active * * \return TRUE if \p rsc is active, otherwise FALSE */ gboolean (*active)(pcmk_resource_t *rsc, gboolean all); /*! * \internal * \brief Get resource's current or assigned role * * \param[in] rsc Resource to check * \param[in] current If TRUE, check current role, otherwise assigned role * * \return Current or assigned role of \p rsc */ enum rsc_role_e (*state)(const pcmk_resource_t *rsc, gboolean current); /*! * \internal * \brief List nodes where a resource (or any of its children) is * * \param[in] rsc Resource to check * \param[out] list List to add result to * \param[in] current If 0, list nodes where \p rsc is assigned; * if 1, where active; if 2, where active or pending * * \return If list contains only one node, that node, otherwise NULL */ pcmk_node_t *(*location)(const pcmk_resource_t *rsc, GList **list, int current); /*! * \internal * \brief Free all memory used by a resource * * \param[in,out] rsc Resource to free */ void (*free)(pcmk_resource_t *rsc); /*! * \internal * \brief Increment cluster's instance counts for a resource * * Given a resource, increment its cluster's ninstances, disabled_resources, * and blocked_resources counts for the resource and its descendants. * * \param[in,out] rsc Resource to count */ void (*count)(pcmk_resource_t *rsc); /*! * \internal * \brief Check whether a given resource is in a list of resources * * \param[in] rsc Resource ID to check for * \param[in] only_rsc List of resource IDs to check * \param[in] check_parent If TRUE, check top ancestor as well * * \return TRUE if \p rsc, its top parent if requested, or '*' is in * \p only_rsc, otherwise FALSE */ gboolean (*is_filtered)(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent); /*! * \internal * \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 \c PCMK_META_REQUIRES is * \c PCMK_VALUE_QUORUM or \c PCMK_VALUE_NOTHING, otherwise \c NULL. */ pcmk_node_t *(*active_node)(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean); /*! * \internal * \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 pcmk_resource_t *rsc); } pcmk__rsc_methods_t; // Implementation of pcmk__resource_private_t struct pcmk__resource_private { enum pcmk__rsc_variant variant; // Resource variant void *variant_opaque; // Variant-specific data char *history_id; // Resource instance ID in history int priority; // Priority relative other resources int promotion_priority; // Promotion priority on assigned node int stickiness; // Extra preference for current node guint failure_expiration_ms; // Failures expire after this much time int ban_after_failures; // Ban from node after this many failures guint remote_reconnect_ms; // Retry interval for remote connections char *pending_action; // Pending action in history, if any GList *actions; // Actions scheduled for resource pcmk_resource_t *parent; // Resource's parent resource, if any pcmk_scheduler_t *scheduler; // Scheduler data containing resource enum pcmk__restart restart_type; // Deprecated // Resource configuration (possibly expanded from template) xmlNode *xml; // Original resource configuration, if using template xmlNode *orig_xml; // Configuration of resource operations (possibly expanded from template) xmlNode *ops_xml; // What to do if the resource is incorrectly active on multiple nodes enum pcmk__multiply_active multiply_active_policy; + /* The assigned node (if not NULL) is the one where the resource *should* + * be active by the end of the current scheduler transition. Only primitive + * resources have an assigned node. + * + * @TODO This should probably be part of the primitive variant data. + */ + pcmk_node_t *assigned_node; + /* Pay special attention to whether you want to use with_this_colocations * and this_with_colocations 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. */ // Colocations of other resources with this one GList *with_this_colocations; // Colocations of this resource with others GList *this_with_colocations; GList *location_constraints; // Location constraints for resource GList *ticket_constraints; // Ticket constraints for resource const pcmk__rsc_methods_t *fns; // Resource object methods const pcmk__assignment_methods_t *cmds; // Resource assignment methods }; const char *pcmk__multiply_active_text(const pcmk_resource_t *rsc); /*! * \internal * \brief Get node where resource is currently active (if any) * * \param[in] rsc Resource to check * * \return Node that \p rsc is active on, if any, otherwise NULL */ static inline pcmk_node_t * pcmk__current_node(const pcmk_resource_t *rsc) { if (rsc == NULL) { return NULL; } return rsc->private->fns->active_node(rsc, NULL, NULL); } #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_RESOURCES_INTERNAL__H diff --git a/lib/common/attrs.c b/lib/common/attrs.c index 35715df91e..9466e6c7aa 100644 --- a/lib/common/attrs.c +++ b/lib/common/attrs.c @@ -1,183 +1,183 @@ /* * Copyright 2011-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef _GNU_SOURCE # define _GNU_SOURCE #endif #include #include #include #include #include #define OCF_RESKEY_PREFIX "OCF_RESKEY_" #define LRM_TARGET_ENV OCF_RESKEY_PREFIX CRM_META "_" PCMK__META_ON_NODE /*! * \internal * \brief Get the node name that should be used to set node attributes * * If given NULL, "auto", or "localhost" as an argument, check the environment * to detect the node name that should be used to set node attributes. (The * caller might not know the correct name, for example if the target is part of * a bundle with \c PCMK_META_CONTAINER_ATTRIBUTE_TARGET set to * \c PCMK_VALUE_HOST.) * * \param[in] name NULL, "auto" or "localhost" to check environment variables, * or anything else to return NULL * * \return Node name that should be used for node attributes based on the * environment if known, otherwise NULL */ const char * pcmk__node_attr_target(const char *name) { if (name == NULL || pcmk__strcase_any_of(name, "auto", "localhost", NULL)) { char buf[128] = OCF_RESKEY_PREFIX; size_t offset = sizeof(OCF_RESKEY_PREFIX) - 1; char *target_var = crm_meta_name(PCMK_META_CONTAINER_ATTRIBUTE_TARGET); char *phys_var = crm_meta_name(PCMK__META_PHYSICAL_HOST); const char *target = NULL; const char *host_physical = NULL; snprintf(buf + offset, sizeof(buf) - offset, "%s", target_var); target = getenv(buf); snprintf(buf + offset, sizeof(buf) - offset, "%s", phys_var); host_physical = getenv(buf); // It is important to use the name by which the scheduler knows us if (host_physical && pcmk__str_eq(target, PCMK_VALUE_HOST, pcmk__str_casei)) { name = host_physical; } else { const char *host_pcmk = getenv(LRM_TARGET_ENV); if (host_pcmk) { name = host_pcmk; } } free(target_var); free(phys_var); // TODO? Call pcmk__cluster_local_node_name() if name == NULL // (currently would require linkage against libcrmcluster) return name; } else { return NULL; } } /*! * \brief Return the name of the node attribute used as a promotion score * * \param[in] rsc_id Resource ID that promotion score is for (or NULL to * check the OCF_RESOURCE_INSTANCE environment variable) * * \return Newly allocated string with the node attribute name (or NULL on * error, including no ID or environment variable specified) * \note It is the caller's responsibility to free() the result. */ char * pcmk_promotion_score_name(const char *rsc_id) { if (pcmk__str_empty(rsc_id)) { rsc_id = getenv("OCF_RESOURCE_INSTANCE"); if (pcmk__str_empty(rsc_id)) { return NULL; } } return crm_strdup_printf("master-%s", rsc_id); } /*! * \internal * \brief Get the value of a node attribute * * \param[in] node Node to get attribute for * \param[in] name Name of node attribute to get * \param[in] target If this is \c PCMK_VALUE_HOST and \p node is a guest * (bundle) node, get the value from the guest's host, * otherwise get the value from \p node itself * \param[in] node_type If getting the value from \p node's host, this * indicates whether to check the current or assigned host * * \return Value of \p name attribute for \p node */ const char * pcmk__node_attr(const pcmk_node_t *node, const char *name, const char *target, enum pcmk__rsc_node node_type) { const char *value = NULL; // Attribute value to return const char *node_type_s = NULL; // Readable equivalent of node_type const pcmk_node_t *host = NULL; const pcmk_resource_t *container = NULL; if ((node == NULL) || (name == NULL)) { return NULL; } /* Check the node's own attributes unless this is a guest (bundle) node with * the container host as the attribute target. */ if (!pcmk__is_guest_or_bundle_node(node) || !pcmk__str_eq(target, PCMK_VALUE_HOST, pcmk__str_casei)) { value = g_hash_table_lookup(node->details->attrs, name); crm_trace("%s='%s' on %s", name, pcmk__s(value, ""), pcmk__node_name(node)); return value; } /* This resource needs attributes set for the container's host instead of * for the container itself (useful when the container uses the host's * storage). */ container = node->details->remote_rsc->container; switch (node_type) { case pcmk__rsc_node_assigned: - host = container->allocated_to; + host = container->private->assigned_node; if (host == NULL) { crm_trace("Skipping %s lookup for %s because " "its container %s is unassigned", name, pcmk__node_name(node), container->id); return NULL; } node_type_s = "assigned"; break; case pcmk__rsc_node_current: if (container->running_on != NULL) { host = container->running_on->data; } if (host == NULL) { crm_trace("Skipping %s lookup for %s because " "its container %s is inactive", name, pcmk__node_name(node), container->id); return NULL; } node_type_s = "current"; break; default: // Add support for other enum pcmk__rsc_node values if needed CRM_ASSERT(false); break; } value = g_hash_table_lookup(host->details->attrs, name); crm_trace("%s='%s' for %s on %s container host %s", name, pcmk__s(value, ""), pcmk__node_name(node), node_type_s, pcmk__node_name(host)); return value; } diff --git a/lib/pacemaker/pcmk_graph_producer.c b/lib/pacemaker/pcmk_graph_producer.c index f77fd25caf..75289d4fdb 100644 --- a/lib/pacemaker/pcmk_graph_producer.c +++ b/lib/pacemaker/pcmk_graph_producer.c @@ -1,1098 +1,1098 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include "libpacemaker_private.h" // Convenience macros for logging action properties #define action_type_str(flags) \ (pcmk_is_set((flags), pcmk_action_pseudo)? "pseudo-action" : "action") #define action_optional_str(flags) \ (pcmk_is_set((flags), pcmk_action_optional)? "optional" : "required") #define action_runnable_str(flags) \ (pcmk_is_set((flags), pcmk_action_runnable)? "runnable" : "unrunnable") #define action_node_str(a) \ (((a)->node == NULL)? "no node" : (a)->node->details->uname) /*! * \internal * \brief Add an XML node tag for a specified ID * * \param[in] id Node UUID to add * \param[in,out] xml Parent XML tag to add to */ static xmlNode* add_node_to_xml_by_id(const char *id, xmlNode *xml) { xmlNode *node_xml; node_xml = pcmk__xe_create(xml, PCMK_XE_NODE); crm_xml_add(node_xml, PCMK_XA_ID, id); return node_xml; } /*! * \internal * \brief Add an XML node tag for a specified node * * \param[in] node Node to add * \param[in,out] xml XML to add node to */ static void add_node_to_xml(const pcmk_node_t *node, void *xml) { add_node_to_xml_by_id(node->details->id, (xmlNode *) xml); } /*! * \internal * \brief Count (optionally add to XML) nodes needing maintenance state update * * \param[in,out] xml Parent XML tag to add to, if any * \param[in] scheduler Scheduler data * * \return Count of nodes added * \note Only Pacemaker Remote nodes are considered currently */ static int add_maintenance_nodes(xmlNode *xml, const pcmk_scheduler_t *scheduler) { xmlNode *maintenance = NULL; int count = 0; if (xml != NULL) { maintenance = pcmk__xe_create(xml, PCMK__XE_MAINTENANCE); } for (const GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { const pcmk_node_t *node = iter->data; if (pcmk__is_pacemaker_remote_node(node) && (node->details->maintenance != node->details->remote_maintenance)) { if (maintenance != NULL) { crm_xml_add(add_node_to_xml_by_id(node->details->id, maintenance), PCMK__XA_NODE_IN_MAINTENANCE, (node->details->maintenance? "1" : "0")); } count++; } } crm_trace("%s %d nodes in need of maintenance mode update in state", ((maintenance == NULL)? "Counted" : "Added"), count); return count; } /*! * \internal * \brief Add pseudo action with nodes needing maintenance state update * * \param[in,out] scheduler Scheduler data */ static void add_maintenance_update(pcmk_scheduler_t *scheduler) { pcmk_action_t *action = NULL; if (add_maintenance_nodes(NULL, scheduler) != 0) { action = get_pseudo_op(PCMK_ACTION_MAINTENANCE_NODES, scheduler); pcmk__set_action_flags(action, pcmk_action_always_in_graph); } } /*! * \internal * \brief Add XML with nodes that an action is expected to bring down * * If a specified action is expected to bring any nodes down, add an XML block * with their UUIDs. When a node is lost, this allows the controller to * determine whether it was expected. * * \param[in,out] xml Parent XML tag to add to * \param[in] action Action to check for downed nodes */ static void add_downed_nodes(xmlNode *xml, const pcmk_action_t *action) { CRM_CHECK((xml != NULL) && (action != NULL) && (action->node != NULL), return); if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { /* Shutdown makes the action's node down */ xmlNode *downed = pcmk__xe_create(xml, PCMK__XE_DOWNED); add_node_to_xml_by_id(action->node->details->id, downed); } else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { /* Fencing makes the action's node and any hosted guest nodes down */ const char *fence = g_hash_table_lookup(action->meta, PCMK__META_STONITH_ACTION); if (pcmk__is_fencing_action(fence)) { xmlNode *downed = pcmk__xe_create(xml, PCMK__XE_DOWNED); add_node_to_xml_by_id(action->node->details->id, downed); pe_foreach_guest_node(action->node->details->data_set, action->node, add_node_to_xml, downed); } } else if ((action->rsc != NULL) && pcmk_is_set(action->rsc->flags, pcmk__rsc_is_remote_connection) && pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_none)) { /* Stopping a remote connection resource makes connected node down, * unless it's part of a migration */ GList *iter; pcmk_action_t *input; bool migrating = false; for (iter = action->actions_before; iter != NULL; iter = iter->next) { input = ((pcmk__related_action_t *) iter->data)->action; if ((input->rsc != NULL) && pcmk__str_eq(action->rsc->id, input->rsc->id, pcmk__str_none) && pcmk__str_eq(input->task, PCMK_ACTION_MIGRATE_FROM, pcmk__str_none)) { migrating = true; break; } } if (!migrating) { xmlNode *downed = pcmk__xe_create(xml, PCMK__XE_DOWNED); add_node_to_xml_by_id(action->rsc->id, downed); } } } /*! * \internal * \brief Create a transition graph operation key for a clone action * * \param[in] action Clone action * \param[in] interval_ms Action interval in milliseconds * * \return Newly allocated string with transition graph operation key */ static char * clone_op_key(const pcmk_action_t *action, guint interval_ms) { if (pcmk__str_eq(action->task, PCMK_ACTION_NOTIFY, pcmk__str_none)) { const char *n_type = g_hash_table_lookup(action->meta, "notify_type"); const char *n_task = g_hash_table_lookup(action->meta, "notify_operation"); return pcmk__notify_key(action->rsc->private->history_id, n_type, n_task); } return pcmk__op_key(action->rsc->private->history_id, pcmk__s(action->cancel_task, action->task), interval_ms); } /*! * \internal * \brief Add node details to transition graph action XML * * \param[in] action Scheduled action * \param[in,out] xml Transition graph action XML for \p action */ static void add_node_details(const pcmk_action_t *action, xmlNode *xml) { pcmk_node_t *router_node = pcmk__connection_host_for_action(action); crm_xml_add(xml, PCMK__META_ON_NODE, action->node->details->uname); crm_xml_add(xml, PCMK__META_ON_NODE_UUID, action->node->details->id); if (router_node != NULL) { crm_xml_add(xml, PCMK__XA_ROUTER_NODE, router_node->details->uname); } } /*! * \internal * \brief Add resource details to transition graph action XML * * \param[in] action Scheduled action * \param[in,out] action_xml Transition graph action XML for \p action */ static void add_resource_details(const pcmk_action_t *action, xmlNode *action_xml) { xmlNode *rsc_xml = NULL; const char *attr_list[] = { PCMK_XA_CLASS, PCMK_XA_PROVIDER, PCMK_XA_TYPE, }; /* If a resource is locked to a node via PCMK_OPT_SHUTDOWN_LOCK, mark its * actions so the controller can preserve the lock when the action * completes. */ if (pcmk__action_locks_rsc_to_node(action)) { crm_xml_add_ll(action_xml, PCMK_OPT_SHUTDOWN_LOCK, (long long) action->rsc->lock_time); } // List affected resource rsc_xml = pcmk__xe_create(action_xml, (const char *) action->rsc->private->xml->name); if (pcmk_is_set(action->rsc->flags, pcmk__rsc_removed) && (action->rsc->private->history_id != NULL)) { /* Use the numbered instance name here, because if there is more * than one instance on a node, we need to make sure the command * goes to the right one. * * This is important even for anonymous clones, because the clone's * unique meta-attribute might have just been toggled from on to * off. */ crm_debug("Using orphan clone name %s instead of history ID %s", action->rsc->id, action->rsc->private->history_id); crm_xml_add(rsc_xml, PCMK_XA_ID, action->rsc->private->history_id); crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->id); } else if (!pcmk_is_set(action->rsc->flags, pcmk__rsc_unique)) { const char *xml_id = pcmk__xe_id(action->rsc->private->xml); crm_debug("Using anonymous clone name %s for %s (aka %s)", xml_id, action->rsc->id, action->rsc->private->history_id); /* ID is what we'd like client to use * LONG_ID is what they might know it as instead * * LONG_ID is only strictly needed /here/ during the * transition period until all nodes in the cluster * are running the new software /and/ have rebooted * once (meaning that they've only ever spoken to a DC * supporting this feature). * * If anyone toggles the unique flag to 'on', the * 'instance free' name will correspond to an orphan * and fall into the clause above instead */ crm_xml_add(rsc_xml, PCMK_XA_ID, xml_id); if ((action->rsc->private->history_id != NULL) && !pcmk__str_eq(xml_id, action->rsc->private->history_id, pcmk__str_none)) { crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->private->history_id); } else { crm_xml_add(rsc_xml, PCMK__XA_LONG_ID, action->rsc->id); } } else { CRM_ASSERT(action->rsc->private->history_id == NULL); crm_xml_add(rsc_xml, PCMK_XA_ID, action->rsc->id); } for (int lpc = 0; lpc < PCMK__NELEM(attr_list); lpc++) { crm_xml_add(rsc_xml, attr_list[lpc], g_hash_table_lookup(action->rsc->meta, attr_list[lpc])); } } /*! * \internal * \brief Add action attributes to transition graph action XML * * \param[in,out] action Scheduled action * \param[in,out] action_xml Transition graph action XML for \p action */ static void add_action_attributes(pcmk_action_t *action, xmlNode *action_xml) { xmlNode *args_xml = NULL; pcmk_resource_t *rsc = action->rsc; /* We create free-standing XML to start, so we can sort the attributes * before adding it to action_xml, which keeps the scheduler regression * test graphs comparable. */ args_xml = pcmk__xe_create(action_xml, PCMK__XE_ATTRIBUTES); crm_xml_add(args_xml, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); g_hash_table_foreach(action->extra, hash2field, args_xml); if ((rsc != NULL) && (action->node != NULL)) { // Get the resource instance attributes, evaluated properly for node GHashTable *params = pe_rsc_params(rsc, action->node, rsc->private->scheduler); pcmk__substitute_remote_addr(rsc, params); g_hash_table_foreach(params, hash2smartfield, args_xml); } else if ((rsc != NULL) && (rsc->private->variant <= pcmk__rsc_variant_primitive)) { GHashTable *params = pe_rsc_params(rsc, NULL, rsc->private->scheduler); g_hash_table_foreach(params, hash2smartfield, args_xml); } g_hash_table_foreach(action->meta, hash2metafield, args_xml); if (rsc != NULL) { pcmk_resource_t *parent = rsc; while (parent != NULL) { parent->private->cmds->add_graph_meta(parent, args_xml); parent = parent->private->parent; } pcmk__add_guest_meta_to_xml(args_xml, action); } else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none) && (action->node != NULL)) { /* Pass the node's attributes as meta-attributes. * * @TODO: Determine whether it is still necessary to do this. It was * added in 33d99707, probably for the libfence-based implementation in * c9a90bd, which is no longer used. */ g_hash_table_foreach(action->node->details->attrs, hash2metafield, args_xml); } pcmk__xe_sort_attrs(args_xml); } /*! * \internal * \brief Create the transition graph XML for a scheduled action * * \param[in,out] parent Parent XML element to add action to * \param[in,out] action Scheduled action * \param[in] skip_details If false, add action details as sub-elements * \param[in] scheduler Scheduler data */ static void create_graph_action(xmlNode *parent, pcmk_action_t *action, bool skip_details, const pcmk_scheduler_t *scheduler) { bool needs_node_info = true; bool needs_maintenance_info = false; xmlNode *action_xml = NULL; if ((action == NULL) || (scheduler == NULL)) { return; } // Create the top-level element based on task if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { /* All fences need node info; guest node fences are pseudo-events */ if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { action_xml = pcmk__xe_create(parent, PCMK__XE_PSEUDO_EVENT); } else { action_xml = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT); } } else if (pcmk__str_any_of(action->task, PCMK_ACTION_DO_SHUTDOWN, PCMK_ACTION_CLEAR_FAILCOUNT, NULL)) { action_xml = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT); } else if (pcmk__str_eq(action->task, PCMK_ACTION_LRM_DELETE, pcmk__str_none)) { // CIB-only clean-up for shutdown locks action_xml = pcmk__xe_create(parent, PCMK__XE_CRM_EVENT); crm_xml_add(action_xml, PCMK__XA_MODE, PCMK__VALUE_CIB); } else if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { if (pcmk__str_eq(action->task, PCMK_ACTION_MAINTENANCE_NODES, pcmk__str_none)) { needs_maintenance_info = true; } action_xml = pcmk__xe_create(parent, PCMK__XE_PSEUDO_EVENT); needs_node_info = false; } else { action_xml = pcmk__xe_create(parent, PCMK__XE_RSC_OP); } crm_xml_add_int(action_xml, PCMK_XA_ID, action->id); crm_xml_add(action_xml, PCMK_XA_OPERATION, action->task); if ((action->rsc != NULL) && (action->rsc->private->history_id != NULL)) { char *clone_key = NULL; guint interval_ms; if (pcmk__guint_from_hash(action->meta, PCMK_META_INTERVAL, 0, &interval_ms) != pcmk_rc_ok) { interval_ms = 0; } clone_key = clone_op_key(action, interval_ms); crm_xml_add(action_xml, PCMK__XA_OPERATION_KEY, clone_key); crm_xml_add(action_xml, "internal_" PCMK__XA_OPERATION_KEY, action->uuid); free(clone_key); } else { crm_xml_add(action_xml, PCMK__XA_OPERATION_KEY, action->uuid); } if (needs_node_info && (action->node != NULL)) { add_node_details(action, action_xml); pcmk__insert_dup(action->meta, PCMK__META_ON_NODE, action->node->details->uname); pcmk__insert_dup(action->meta, PCMK__META_ON_NODE_UUID, action->node->details->id); } if (skip_details) { return; } if ((action->rsc != NULL) && !pcmk_is_set(action->flags, pcmk_action_pseudo)) { // This is a real resource action, so add resource details add_resource_details(action, action_xml); } /* List any attributes in effect */ add_action_attributes(action, action_xml); /* List any nodes this action is expected to make down */ if (needs_node_info && (action->node != NULL)) { add_downed_nodes(action_xml, action); } if (needs_maintenance_info) { add_maintenance_nodes(action_xml, scheduler); } } /*! * \internal * \brief Check whether an action should be added to the transition graph * * \param[in] action Action to check * * \return true if action should be added to graph, otherwise false */ static bool should_add_action_to_graph(const pcmk_action_t *action) { if (!pcmk_is_set(action->flags, pcmk_action_runnable)) { crm_trace("Ignoring action %s (%d): unrunnable", action->uuid, action->id); return false; } if (pcmk_is_set(action->flags, pcmk_action_optional) && !pcmk_is_set(action->flags, pcmk_action_always_in_graph)) { crm_trace("Ignoring action %s (%d): optional", action->uuid, action->id); return false; } /* Actions for unmanaged resources should be excluded from the graph, * with the exception of monitors and cancellation of recurring monitors. */ if ((action->rsc != NULL) && !pcmk_is_set(action->rsc->flags, pcmk__rsc_managed) && !pcmk__str_eq(action->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { const char *interval_ms_s; /* A cancellation of a recurring monitor will get here because the task * is cancel rather than monitor, but the interval can still be used to * recognize it. The interval has been normalized to milliseconds by * this point, so a string comparison is sufficient. */ interval_ms_s = g_hash_table_lookup(action->meta, PCMK_META_INTERVAL); if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches)) { crm_trace("Ignoring action %s (%d): for unmanaged resource (%s)", action->uuid, action->id, action->rsc->id); return false; } } /* Always add pseudo-actions, fence actions, and shutdown actions (already * determined to be required and runnable by this point) */ if (pcmk_is_set(action->flags, pcmk_action_pseudo) || pcmk__strcase_any_of(action->task, PCMK_ACTION_STONITH, PCMK_ACTION_DO_SHUTDOWN, NULL)) { return true; } if (action->node == NULL) { pcmk__sched_err("Skipping action %s (%d) " "because it was not assigned to a node (bug?)", action->uuid, action->id); pcmk__log_action("Unassigned", action, false); return false; } if (pcmk_is_set(action->flags, pcmk_action_on_dc)) { crm_trace("Action %s (%d) should be dumped: " "can run on DC instead of %s", action->uuid, action->id, pcmk__node_name(action->node)); } else if (pcmk__is_guest_or_bundle_node(action->node) && !action->node->details->remote_requires_reset) { crm_trace("Action %s (%d) should be dumped: " "assuming will be runnable on guest %s", action->uuid, action->id, pcmk__node_name(action->node)); } else if (!action->node->details->online) { pcmk__sched_err("Skipping action %s (%d) " "because it was scheduled for offline node (bug?)", action->uuid, action->id); pcmk__log_action("Offline node", action, false); return false; } else if (action->node->details->unclean) { pcmk__sched_err("Skipping action %s (%d) " "because it was scheduled for unclean node (bug?)", action->uuid, action->id); pcmk__log_action("Unclean node", action, false); return false; } return true; } /*! * \internal * \brief Check whether an ordering's flags can change an action * * \param[in] ordering Ordering to check * * \return true if ordering has flags that can change an action, false otherwise */ static bool ordering_can_change_actions(const pcmk__related_action_t *ordering) { return pcmk_any_flags_set(ordering->type, ~(pcmk__ar_then_implies_first_graphed |pcmk__ar_first_implies_then_graphed |pcmk__ar_ordered)); } /*! * \internal * \brief Check whether an action input should be in the transition graph * * \param[in] action Action to check * \param[in,out] input Action input to check * * \return true if input should be in graph, false otherwise * \note This function may not only check an input, but disable it under certian * circumstances (load or anti-colocation orderings that are not needed). */ static bool should_add_input_to_graph(const pcmk_action_t *action, pcmk__related_action_t *input) { if (input->state == pe_link_dumped) { return true; } if ((uint32_t) input->type == pcmk__ar_none) { crm_trace("Ignoring %s (%d) input %s (%d): " "ordering disabled", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (!pcmk_is_set(input->action->flags, pcmk_action_runnable) && !ordering_can_change_actions(input)) { crm_trace("Ignoring %s (%d) input %s (%d): " "optional and input unrunnable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (!pcmk_is_set(input->action->flags, pcmk_action_runnable) && pcmk_is_set(input->type, pcmk__ar_min_runnable)) { crm_trace("Ignoring %s (%d) input %s (%d): " "minimum number of instances required but input unrunnable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (pcmk_is_set(input->type, pcmk__ar_unmigratable_then_blocks) && !pcmk_is_set(input->action->flags, pcmk_action_runnable)) { crm_trace("Ignoring %s (%d) input %s (%d): " "input blocked if 'then' unmigratable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (pcmk_is_set(input->type, pcmk__ar_if_first_unmigratable) && pcmk_is_set(input->action->flags, pcmk_action_migratable)) { crm_trace("Ignoring %s (%d) input %s (%d): ordering applies " "only if input is unmigratable, but it is migratable", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if (((uint32_t) input->type == pcmk__ar_ordered) && pcmk_is_set(input->action->flags, pcmk_action_migratable) && pcmk__ends_with(input->action->uuid, "_stop_0")) { crm_trace("Ignoring %s (%d) input %s (%d): " "optional but stop in migration", action->uuid, action->id, input->action->uuid, input->action->id); return false; } else if ((uint32_t) input->type == pcmk__ar_if_on_same_node_or_target) { pcmk_node_t *input_node = input->action->node; if ((action->rsc != NULL) && pcmk__str_eq(action->task, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)) { - pcmk_node_t *assigned = action->rsc->allocated_to; + pcmk_node_t *assigned = action->rsc->private->assigned_node; /* For load_stopped -> migrate_to orderings, we care about where * the resource has been assigned, not where migrate_to will be * executed. */ if (!pcmk__same_node(input_node, assigned)) { crm_trace("Ignoring %s (%d) input %s (%d): " "migration target %s is not same as input node %s", action->uuid, action->id, input->action->uuid, input->action->id, (assigned? assigned->details->uname : ""), (input_node? input_node->details->uname : "")); input->type = (enum pe_ordering) pcmk__ar_none; return false; } } else if (!pcmk__same_node(input_node, action->node)) { crm_trace("Ignoring %s (%d) input %s (%d): " "not on same node (%s vs %s)", action->uuid, action->id, input->action->uuid, input->action->id, (action->node? action->node->details->uname : ""), (input_node? input_node->details->uname : "")); input->type = (enum pe_ordering) pcmk__ar_none; return false; } else if (pcmk_is_set(input->action->flags, pcmk_action_optional)) { crm_trace("Ignoring %s (%d) input %s (%d): " "ordering optional", action->uuid, action->id, input->action->uuid, input->action->id); input->type = (enum pe_ordering) pcmk__ar_none; return false; } } else if ((uint32_t) input->type == pcmk__ar_if_required_on_same_node) { if (input->action->node && action->node && !pcmk__same_node(input->action->node, action->node)) { crm_trace("Ignoring %s (%d) input %s (%d): " "not on same node (%s vs %s)", action->uuid, action->id, input->action->uuid, input->action->id, pcmk__node_name(action->node), pcmk__node_name(input->action->node)); input->type = (enum pe_ordering) pcmk__ar_none; return false; } else if (pcmk_is_set(input->action->flags, pcmk_action_optional)) { crm_trace("Ignoring %s (%d) input %s (%d): optional", action->uuid, action->id, input->action->uuid, input->action->id); input->type = (enum pe_ordering) pcmk__ar_none; return false; } } else if (input->action->rsc && input->action->rsc != action->rsc && pcmk_is_set(input->action->rsc->flags, pcmk__rsc_failed) && !pcmk_is_set(input->action->rsc->flags, pcmk__rsc_managed) && pcmk__ends_with(input->action->uuid, "_stop_0") && pcmk__is_clone(action->rsc)) { crm_warn("Ignoring requirement that %s complete before %s:" " unmanaged failed resources cannot prevent clone shutdown", input->action->uuid, action->uuid); return false; } else if (pcmk_is_set(input->action->flags, pcmk_action_optional) && !pcmk_any_flags_set(input->action->flags, pcmk_action_always_in_graph |pcmk_action_added_to_graph) && !should_add_action_to_graph(input->action)) { crm_trace("Ignoring %s (%d) input %s (%d): " "input optional", action->uuid, action->id, input->action->uuid, input->action->id); return false; } crm_trace("%s (%d) input %s %s (%d) on %s should be dumped: %s %s %#.6x", action->uuid, action->id, action_type_str(input->action->flags), input->action->uuid, input->action->id, action_node_str(input->action), action_runnable_str(input->action->flags), action_optional_str(input->action->flags), input->type); return true; } /*! * \internal * \brief Check whether an ordering creates an ordering loop * * \param[in] init_action "First" action in ordering * \param[in] action Callers should always set this the same as * \p init_action (this function may use a different * value for recursive calls) * \param[in,out] input Action wrapper for "then" action in ordering * * \return true if the ordering creates a loop, otherwise false */ bool pcmk__graph_has_loop(const pcmk_action_t *init_action, const pcmk_action_t *action, pcmk__related_action_t *input) { bool has_loop = false; if (pcmk_is_set(input->action->flags, pcmk_action_detect_loop)) { crm_trace("Breaking tracking loop: %s@%s -> %s@%s (%#.6x)", input->action->uuid, input->action->node? input->action->node->details->uname : "", action->uuid, action->node? action->node->details->uname : "", input->type); return false; } // Don't need to check inputs that won't be used if (!should_add_input_to_graph(action, input)) { return false; } if (input->action == init_action) { crm_debug("Input loop found in %s@%s ->...-> %s@%s", action->uuid, action->node? action->node->details->uname : "", init_action->uuid, init_action->node? init_action->node->details->uname : ""); return true; } pcmk__set_action_flags(input->action, pcmk_action_detect_loop); crm_trace("Checking inputs of action %s@%s input %s@%s (%#.6x)" "for graph loop with %s@%s ", action->uuid, action->node? action->node->details->uname : "", input->action->uuid, input->action->node? input->action->node->details->uname : "", input->type, init_action->uuid, init_action->node? init_action->node->details->uname : ""); // Recursively check input itself for loops for (GList *iter = input->action->actions_before; iter != NULL; iter = iter->next) { if (pcmk__graph_has_loop(init_action, input->action, (pcmk__related_action_t *) iter->data)) { // Recursive call already logged a debug message has_loop = true; break; } } pcmk__clear_action_flags(input->action, pcmk_action_detect_loop); if (!has_loop) { crm_trace("No input loop found in %s@%s -> %s@%s (%#.6x)", input->action->uuid, input->action->node? input->action->node->details->uname : "", action->uuid, action->node? action->node->details->uname : "", input->type); } return has_loop; } /*! * \internal * \brief Create a synapse XML element for a transition graph * * \param[in] action Action that synapse is for * \param[in,out] scheduler Scheduler data containing graph * * \return Newly added XML element for new graph synapse */ static xmlNode * create_graph_synapse(const pcmk_action_t *action, pcmk_scheduler_t *scheduler) { int synapse_priority = 0; xmlNode *syn = pcmk__xe_create(scheduler->graph, "synapse"); crm_xml_add_int(syn, PCMK_XA_ID, scheduler->num_synapse); scheduler->num_synapse++; if (action->rsc != NULL) { synapse_priority = action->rsc->private->priority; } if (action->priority > synapse_priority) { synapse_priority = action->priority; } if (synapse_priority > 0) { crm_xml_add_int(syn, PCMK__XA_PRIORITY, synapse_priority); } return syn; } /*! * \internal * \brief Add an action to the transition graph XML if appropriate * * \param[in,out] data Action to possibly add * \param[in,out] user_data Scheduler data * * \note This will de-duplicate the action inputs, meaning that the * pcmk__related_action_t:type flags can no longer be relied on to retain * their original settings. That means this MUST be called after * pcmk__apply_orderings() is complete, and nothing after this should rely * on those type flags. (For example, some code looks for type equal to * some flag rather than whether the flag is set, and some code looks for * particular combinations of flags -- such code must be done before * pcmk__create_graph().) */ static void add_action_to_graph(gpointer data, gpointer user_data) { pcmk_action_t *action = (pcmk_action_t *) data; pcmk_scheduler_t *scheduler = (pcmk_scheduler_t *) user_data; xmlNode *syn = NULL; xmlNode *set = NULL; xmlNode *in = NULL; /* If we haven't already, de-duplicate inputs (even if we won't be adding * the action to the graph, so that crm_simulate's dot graphs don't have * duplicates). */ if (!pcmk_is_set(action->flags, pcmk_action_inputs_deduplicated)) { pcmk__deduplicate_action_inputs(action); pcmk__set_action_flags(action, pcmk_action_inputs_deduplicated); } if (pcmk_is_set(action->flags, pcmk_action_added_to_graph) || !should_add_action_to_graph(action)) { return; // Already added, or shouldn't be } pcmk__set_action_flags(action, pcmk_action_added_to_graph); crm_trace("Adding action %d (%s%s%s) to graph", action->id, action->uuid, ((action->node == NULL)? "" : " on "), ((action->node == NULL)? "" : action->node->details->uname)); syn = create_graph_synapse(action, scheduler); set = pcmk__xe_create(syn, "action_set"); in = pcmk__xe_create(syn, "inputs"); create_graph_action(set, action, false, scheduler); for (GList *lpc = action->actions_before; lpc != NULL; lpc = lpc->next) { pcmk__related_action_t *input = lpc->data; if (should_add_input_to_graph(action, input)) { xmlNode *input_xml = pcmk__xe_create(in, "trigger"); input->state = pe_link_dumped; create_graph_action(input_xml, input->action, true, scheduler); } } } static int transition_id = -1; /*! * \internal * \brief Log a message after calculating a transition * * \param[in] filename Where transition input is stored */ void pcmk__log_transition_summary(const char *filename) { if (was_processing_error || crm_config_error) { crm_err("Calculated transition %d (with errors)%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } else if (was_processing_warning || crm_config_warning) { crm_warn("Calculated transition %d (with warnings)%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } else { crm_notice("Calculated transition %d%s%s", transition_id, (filename == NULL)? "" : ", saving inputs in ", (filename == NULL)? "" : filename); } if (crm_config_error) { crm_notice("Configuration errors found during scheduler processing," " please run \"crm_verify -L\" to identify issues"); } } /*! * \internal * \brief Add a resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void pcmk__add_rsc_actions_to_graph(pcmk_resource_t *rsc) { GList *iter = NULL; CRM_ASSERT(rsc != NULL); pcmk__rsc_trace(rsc, "Adding actions for %s to graph", rsc->id); // First add the resource's own actions g_list_foreach(rsc->private->actions, add_action_to_graph, rsc->private->scheduler); // Then recursively add its children's actions (appropriate to variant) for (iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) iter->data; child_rsc->private->cmds->add_actions_to_graph(child_rsc); } } /*! * \internal * \brief Create a transition graph with all cluster actions needed * * \param[in,out] scheduler Scheduler data */ void pcmk__create_graph(pcmk_scheduler_t *scheduler) { GList *iter = NULL; const char *value = NULL; long long limit = 0LL; GHashTable *config_hash = scheduler->config_hash; transition_id++; crm_trace("Creating transition graph %d", transition_id); scheduler->graph = pcmk__xe_create(NULL, PCMK__XE_TRANSITION_GRAPH); value = pcmk__cluster_option(config_hash, PCMK_OPT_CLUSTER_DELAY); crm_xml_add(scheduler->graph, PCMK_OPT_CLUSTER_DELAY, value); value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT); crm_xml_add(scheduler->graph, PCMK_OPT_STONITH_TIMEOUT, value); crm_xml_add(scheduler->graph, "failed-stop-offset", "INFINITY"); if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) { crm_xml_add(scheduler->graph, "failed-start-offset", "INFINITY"); } else { crm_xml_add(scheduler->graph, "failed-start-offset", "1"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_BATCH_LIMIT); crm_xml_add(scheduler->graph, PCMK_OPT_BATCH_LIMIT, value); crm_xml_add_int(scheduler->graph, "transition_id", transition_id); value = pcmk__cluster_option(config_hash, PCMK_OPT_MIGRATION_LIMIT); if ((pcmk__scan_ll(value, &limit, 0LL) == pcmk_rc_ok) && (limit > 0)) { crm_xml_add(scheduler->graph, PCMK_OPT_MIGRATION_LIMIT, value); } if (scheduler->recheck_by > 0) { char *recheck_epoch = NULL; recheck_epoch = crm_strdup_printf("%llu", (long long) scheduler->recheck_by); crm_xml_add(scheduler->graph, "recheck-by", recheck_epoch); free(recheck_epoch); } /* The following code will de-duplicate action inputs, so nothing past this * should rely on the action input type flags retaining their original * values. */ // Add resource actions to graph for (iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; pcmk__rsc_trace(rsc, "Processing actions for %s", rsc->id); rsc->private->cmds->add_actions_to_graph(rsc); } // Add pseudo-action for list of nodes with maintenance state update add_maintenance_update(scheduler); // Add non-resource (node) actions for (iter = scheduler->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = (pcmk_action_t *) iter->data; if ((action->rsc != NULL) && (action->node != NULL) && action->node->details->shutdown && !pcmk_is_set(action->rsc->flags, pcmk__rsc_maintenance) && !pcmk_any_flags_set(action->flags, pcmk_action_optional|pcmk_action_runnable) && pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_none)) { /* Eventually we should just ignore the 'fence' case, but for now * it's the best way to detect (in CTS) when CIB resource updates * are being lost. */ if (pcmk_is_set(scheduler->flags, pcmk_sched_quorate) || (scheduler->no_quorum_policy == pcmk_no_quorum_ignore)) { const bool managed = pcmk_is_set(action->rsc->flags, pcmk__rsc_managed); const bool failed = pcmk_is_set(action->rsc->flags, pcmk__rsc_failed); crm_crit("Cannot %s %s because of %s:%s%s (%s)", action->node->details->unclean? "fence" : "shut down", pcmk__node_name(action->node), action->rsc->id, (managed? " blocked" : " unmanaged"), (failed? " failed" : ""), action->uuid); } } add_action_to_graph((gpointer) action, (gpointer) scheduler); } crm_log_xml_trace(scheduler->graph, "graph"); } diff --git a/lib/pacemaker/pcmk_sched_bundle.c b/lib/pacemaker/pcmk_sched_bundle.c index 1195d2b511..90bdbf6df9 100644 --- a/lib/pacemaker/pcmk_sched_bundle.c +++ b/lib/pacemaker/pcmk_sched_bundle.c @@ -1,1065 +1,1065 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "libpacemaker_private.h" struct assign_data { const pcmk_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(pcmk__bundle_replica_t *replica, void *user_data) { pcmk_node_t *container_host = NULL; struct assign_data *assign_data = user_data; const pcmk_node_t *prefer = assign_data->prefer; bool stop_if_fail = assign_data->stop_if_fail; const pcmk_resource_t *bundle = pe__const_top_resource(replica->container, true); if (replica->ip != NULL) { pcmk__rsc_trace(bundle, "Assigning bundle %s IP %s", bundle->id, replica->ip->id); replica->ip->private->cmds->assign(replica->ip, prefer, stop_if_fail); } - container_host = replica->container->allocated_to; + container_host = replica->container->private->assigned_node; if (replica->remote != NULL) { if (pcmk__is_pacemaker_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, PCMK_SCORE_INFINITY, replica->remote, container_host->details->remote_rsc, NULL, NULL, pcmk__coloc_influence); } pcmk__rsc_trace(bundle, "Assigning bundle %s connection %s", bundle->id, replica->remote->id); replica->remote->private->cmds->assign(replica->remote, prefer, stop_if_fail); } if (replica->child != NULL) { pcmk_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 (!pcmk__same_node(node, replica->node)) { node->weight = -PCMK_SCORE_INFINITY; } else if (!pcmk__threshold_reached(replica->child, node, NULL)) { node->weight = PCMK_SCORE_INFINITY; } } pcmk__set_rsc_flags(replica->child->private->parent, pcmk__rsc_assigning); pcmk__rsc_trace(bundle, "Assigning bundle %s replica child %s", bundle->id, replica->child->id); replica->child->private->cmds->assign(replica->child, replica->node, stop_if_fail); pcmk__clear_rsc_flags(replica->child->private->parent, pcmk__rsc_assigning); } 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. */ pcmk_node_t * pcmk__bundle_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { GList *containers = NULL; pcmk_resource_t *bundled_resource = NULL; struct assign_data assign_data = { prefer, stop_if_fail }; CRM_ASSERT(pcmk__is_bundle(rsc)); pcmk__rsc_trace(rsc, "Assigning bundle %s", rsc->id); pcmk__set_rsc_flags(rsc, pcmk__rsc_assigning); pe__show_node_scores(!pcmk_is_set(rsc->private->scheduler->flags, pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->private->scheduler); // 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->private->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) { pcmk_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 = -PCMK_SCORE_INFINITY; } } bundled_resource->private->cmds->assign(bundled_resource, prefer, stop_if_fail); } pcmk__clear_rsc_flags(rsc, pcmk__rsc_assigning|pcmk__rsc_unassigned); 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(pcmk__bundle_replica_t *replica, void *user_data) { if (replica->ip != NULL) { replica->ip->private->cmds->create_actions(replica->ip); } if (replica->container != NULL) { replica->container->private->cmds->create_actions(replica->container); } if (replica->remote != NULL) { replica->remote->private->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(pcmk_resource_t *rsc) { pcmk_action_t *action = NULL; GList *containers = NULL; pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT(pcmk__is_bundle(rsc)); 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->private->cmds->create_actions(bundled_resource); if (pcmk_is_set(bundled_resource->flags, pcmk__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 = PCMK_SCORE_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 = PCMK_SCORE_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(pcmk__bundle_replica_t *replica, void *user_data) { pcmk_resource_t *bundle = user_data; replica->container->private->cmds->internal_constraints(replica->container); // Start bundle -> start replica container pcmk__order_starts(bundle, replica->container, pcmk__ar_unrunnable_first_blocks |pcmk__ar_then_implies_first_graphed); // Stop bundle -> stop replica child and container if (replica->child != NULL) { pcmk__order_stops(bundle, replica->child, pcmk__ar_then_implies_first_graphed); } pcmk__order_stops(bundle, replica->container, pcmk__ar_then_implies_first_graphed); // Start replica container -> bundle is started pcmk__order_resource_actions(replica->container, PCMK_ACTION_START, bundle, PCMK_ACTION_RUNNING, pcmk__ar_first_implies_then_graphed); // Stop replica container -> bundle is stopped pcmk__order_resource_actions(replica->container, PCMK_ACTION_STOP, bundle, PCMK_ACTION_STOPPED, pcmk__ar_first_implies_then_graphed); if (replica->ip != NULL) { replica->ip->private->cmds->internal_constraints(replica->ip); // Replica IP address -> replica container (symmetric) pcmk__order_starts(replica->ip, replica->container, pcmk__ar_unrunnable_first_blocks |pcmk__ar_guest_allowed); pcmk__order_stops(replica->container, replica->ip, pcmk__ar_then_implies_first|pcmk__ar_guest_allowed); pcmk__new_colocation("#ip-with-container", NULL, PCMK_SCORE_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->private->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(pcmk_resource_t *rsc) { pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT(pcmk__is_bundle(rsc)); 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, pcmk__ar_then_implies_first_graphed); // Bundled clone is started -> bundle is started pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_RUNNING, rsc, PCMK_ACTION_RUNNING, pcmk__ar_first_implies_then_graphed); // Stop bundle -> stop bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, bundled_resource, PCMK_ACTION_STOP, pcmk__ar_then_implies_first_graphed); // Bundled clone is stopped -> bundle is stopped pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_STOPPED, rsc, PCMK_ACTION_STOPPED, pcmk__ar_first_implies_then_graphed); bundled_resource->private->cmds->internal_constraints(bundled_resource); if (!pcmk_is_set(bundled_resource->flags, pcmk__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, pcmk__ar_then_implies_first_graphed); // Bundled clone is demoted -> bundle is demoted pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_DEMOTED, rsc, PCMK_ACTION_DEMOTED, pcmk__ar_first_implies_then_graphed); // Promote bundle -> promote bundled clone pcmk__order_resource_actions(rsc, PCMK_ACTION_PROMOTE, bundled_resource, PCMK_ACTION_PROMOTE, pcmk__ar_then_implies_first_graphed); // Bundled clone is promoted -> bundle is promoted pcmk__order_resource_actions(bundled_resource, PCMK_ACTION_PROMOTED, rsc, PCMK_ACTION_PROMOTED, pcmk__ar_first_implies_then_graphed); } struct match_data { const pcmk_node_t *node; // Node to compare against replica pcmk_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 pcmk__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 pcmk_node_t * get_bundle_node_host(const pcmk_node_t *node) { if (pcmk__is_bundle_node(node)) { const pcmk_resource_t *container = node->details->remote_rsc->container; return container->private->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 pcmk_resource_t * compatible_container(const pcmk_resource_t *dependent, const pcmk_resource_t *bundle) { GList *scratch = NULL; struct match_data match_data = { NULL, NULL }; // If dependent is assigned, only check there match_data.node = dependent->private->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; pcmk_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 pcmk__bundle_replica_t *replica, void *user_data) { struct coloc_data *coloc_data = user_data; pcmk_node_t *chosen = NULL; pcmk_resource_t *container = replica->container; if (coloc_data->colocation->score < PCMK_SCORE_INFINITY) { container->private->cmds->apply_coloc_score(coloc_data->dependent, container, coloc_data->colocation, false); return true; } chosen = container->private->fns->location(container, NULL, 0); if ((chosen == NULL) || is_set_recursive(container, pcmk__rsc_blocked, true)) { return true; } if ((coloc_data->colocation->primary_role >= pcmk_role_promoted) && ((replica->child == NULL) || (replica->child->next_role < pcmk_role_promoted))) { return true; } pcmk__rsc_trace(pe__const_top_resource(container, true), "Allowing mandatory colocation %s using %s @%d", coloc_data->colocation->id, pcmk__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(pcmk_resource_t *dependent, const pcmk_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(pcmk__is_bundle(primary) && pcmk__is_primitive(dependent) && (colocation != NULL) && !for_dependent); if (pcmk_is_set(primary->flags, pcmk__rsc_unassigned)) { pcmk__rsc_trace(primary, "Skipping applying colocation %s " "because %s is still provisional", colocation->id, primary->id); return; } pcmk__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->private->variant > pcmk__rsc_variant_group) { const pcmk_resource_t *primary_container = NULL; primary_container = compatible_container(dependent, primary); if (primary_container != NULL) { // Success, we found one pcmk__rsc_debug(primary, "Pairing %s with %s", dependent->id, primary_container->id); dependent->private->cmds->apply_coloc_score(dependent, primary_container, colocation, true); } else if (colocation->score >= PCMK_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 pcmk__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 >= PCMK_SCORE_INFINITY) { pcmk__colocation_intersect_nodes(dependent, primary, colocation, coloc_data.container_hosts, false); } g_list_free(coloc_data.container_hosts); } // Bundle implementation of pcmk__assignment_methods_t:with_this_colocations() void pcmk__with_bundle_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *bundled_rsc = NULL; CRM_ASSERT(pcmk__is_bundle(rsc) && (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, pcmk__rsc_replica_container)) { pcmk__add_with_this_list(list, rsc->private->with_this_colocations, 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, pcmk__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, pcmk__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->private->with_this_colocations, orig_rsc); } } else if (!pcmk_is_set(orig_rsc->flags, pcmk__rsc_unassigned)) { /* 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->private->with_this_colocations, orig_rsc); } } // Bundle implementation of pcmk__assignment_methods_t:this_with_colocations() void pcmk__bundle_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *bundled_rsc = NULL; CRM_ASSERT(pcmk__is_bundle(rsc) && (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, pcmk__rsc_replica_container)) { pcmk__add_this_with_list(list, rsc->private->this_with_colocations, 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, pcmk__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, pcmk__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->private->this_with_colocations, orig_rsc); } } else if (!pcmk_is_set(orig_rsc->flags, pcmk__rsc_unassigned)) { /* 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->private->this_with_colocations, 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(pcmk_action_t *action, const pcmk_node_t *node) { GList *containers = NULL; uint32_t flags = 0; pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((action != NULL) && pcmk__is_bundle(action->rsc)); 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(pcmk__bundle_replica_t *replica, void *user_data) { pcmk__location_t *location = user_data; replica->container->private->cmds->apply_location(replica->container, location); if (replica->ip != NULL) { replica->ip->private->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(pcmk_resource_t *rsc, pcmk__location_t *location) { pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT((location != NULL) && pcmk__is_bundle(rsc)); 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->private->cmds->apply_location(bundled_resource, location); bundled_resource->private->location_constraints = g_list_prepend(bundled_resource->private->location_constraints, location); } } #define XPATH_REMOTE "//nvpair[@name='" PCMK_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(pcmk__bundle_replica_t *replica, void *user_data) { if ((replica->remote != 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->private->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, nvpair, PCMK_XA_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 = NULL; params = pe_rsc_params(replica->remote, NULL, replica->remote->private->scheduler); pcmk__insert_dup(params, PCMK_REMOTE_RA_ADDR, calculated_addr); } else { pcmk_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). */ pcmk__rsc_info(bundle, "Unable to determine address for bundle %s " "remote connection", bundle->id); } } if (replica->ip != NULL) { replica->ip->private->cmds->add_actions_to_graph(replica->ip); } replica->container->private->cmds->add_actions_to_graph(replica->container); if (replica->remote != NULL) { replica->remote->private->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(pcmk_resource_t *rsc) { pcmk_resource_t *bundled_resource = NULL; CRM_ASSERT(pcmk__is_bundle(rsc)); bundled_resource = pe__bundled_resource(rsc); if (bundled_resource != NULL) { bundled_resource->private->cmds->add_actions_to_graph(bundled_resource); } pe__foreach_bundle_replica(rsc, add_replica_actions_to_graph, rsc); } struct probe_data { pcmk_resource_t *bundle; // Bundle being probed pcmk_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(pcmk__bundle_replica_t *replica, void *user_data) { pcmk__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, pcmk__ar_ordered|pcmk__ar_if_on_same_node, replica->container->private->scheduler); 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(pcmk__bundle_replica_t *replica, void *user_data) { struct probe_data *probe_data = user_data; pcmk_resource_t *bundle = probe_data->bundle; if ((replica->ip != NULL) && replica->ip->private->cmds->create_probe(replica->ip, probe_data->node)) { probe_data->any_created = true; } if ((replica->child != NULL) && pcmk__same_node(probe_data->node, replica->node) && replica->child->private->cmds->create_probe(replica->child, probe_data->node)) { probe_data->any_created = true; } if (replica->container->private->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 (bundle->private->fns->max_per_node(bundle) == 1) { pe__foreach_bundle_replica(bundle, order_replica_start_after, replica); } } if ((replica->remote != NULL) && replica->remote->private->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); pcmk_action_t *probe = NULL; probe = find_first_action(replica->remote->private->actions, probe_uuid, NULL, probe_data->node); free(probe_uuid); if (probe != NULL) { probe_data->any_created = true; pcmk__rsc_trace(bundle, "Ordering %s probe on %s", replica->remote->id, pcmk__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, pcmk__ar_nested_remote_probe, bundle->private->scheduler); } } 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(pcmk_resource_t *rsc, pcmk_node_t *node) { struct probe_data probe_data = { rsc, node, false }; CRM_ASSERT(pcmk__is_bundle(rsc)); 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(pcmk__bundle_replica_t *replica, void *user_data) { if (replica->ip != NULL) { replica->ip->private->cmds->output_actions(replica->ip); } replica->container->private->cmds->output_actions(replica->container); if (replica->remote != NULL) { replica->remote->private->cmds->output_actions(replica->remote); } if (replica->child != NULL) { replica->child->private->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(pcmk_resource_t *rsc) { CRM_ASSERT(pcmk__is_bundle(rsc)); pe__foreach_bundle_replica(rsc, output_replica_actions, NULL); } // Bundle implementation of pcmk__assignment_methods_t:add_utilization() void pcmk__bundle_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { pcmk_resource_t *container = NULL; CRM_ASSERT(pcmk__is_bundle(rsc)); if (!pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) { 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->private->cmds->add_utilization(container, orig_rsc, all_rscs, utilization); } } // Bundle implementation of pcmk__assignment_methods_t:shutdown_lock() void pcmk__bundle_shutdown_lock(pcmk_resource_t *rsc) { CRM_ASSERT(pcmk__is_bundle(rsc)); // Bundles currently don't support shutdown locks } diff --git a/lib/pacemaker/pcmk_sched_colocation.c b/lib/pacemaker/pcmk_sched_colocation.c index 9d9fb80a9d..f6822edd04 100644 --- a/lib/pacemaker/pcmk_sched_colocation.c +++ b/lib/pacemaker/pcmk_sched_colocation.c @@ -1,1946 +1,1951 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include "crm/common/util.h" #include "crm/common/xml_internal.h" #include "crm/common/xml.h" #include "libpacemaker_private.h" // Used to temporarily mark a node as unusable #define INFINITY_HACK (PCMK_SCORE_INFINITY * -100) /*! * \internal * \brief Compare two colocations according to priority * * Compare two colocations according to the order in which they should be * considered, based on either their dependent resources or their primary * resources -- preferring (in order): * * Colocation that is not \c NULL * * Colocation whose resource has higher priority * * Colocation whose resource is of a higher-level variant * (bundle > clone > group > primitive) * * Colocation whose resource is promotable, if both are clones * * Colocation whose resource has lower ID in lexicographic order * * \param[in] colocation1 First colocation to compare * \param[in] colocation2 Second colocation to compare * \param[in] dependent If \c true, compare colocations by dependent * priority; otherwise compare them by primary priority * * \return A negative number if \p colocation1 should be considered first, * a positive number if \p colocation2 should be considered first, * or 0 if order doesn't matter */ static gint cmp_colocation_priority(const pcmk__colocation_t *colocation1, const pcmk__colocation_t *colocation2, bool dependent) { const pcmk_resource_t *rsc1 = NULL; const pcmk_resource_t *rsc2 = NULL; if (colocation1 == NULL) { return 1; } if (colocation2 == NULL) { return -1; } if (dependent) { rsc1 = colocation1->dependent; rsc2 = colocation2->dependent; CRM_ASSERT(colocation1->primary != NULL); } else { rsc1 = colocation1->primary; rsc2 = colocation2->primary; CRM_ASSERT(colocation1->dependent != NULL); } CRM_ASSERT((rsc1 != NULL) && (rsc2 != NULL)); if (rsc1->private->priority > rsc2->private->priority) { return -1; } if (rsc1->private->priority < rsc2->private->priority) { return 1; } // Process clones before primitives and groups if (rsc1->private->variant > rsc2->private->variant) { return -1; } if (rsc1->private->variant < rsc2->private->variant) { return 1; } /* @COMPAT scheduler <2.0.0: Process promotable clones before nonpromotable * clones (probably unnecessary, but avoids having to update regression * tests) */ if (pcmk__is_clone(rsc1)) { if (pcmk_is_set(rsc1->flags, pcmk__rsc_promotable) && !pcmk_is_set(rsc2->flags, pcmk__rsc_promotable)) { return -1; } if (!pcmk_is_set(rsc1->flags, pcmk__rsc_promotable) && pcmk_is_set(rsc2->flags, pcmk__rsc_promotable)) { return 1; } } return strcmp(rsc1->id, rsc2->id); } /*! * \internal * \brief Compare two colocations according to priority based on dependents * * Compare two colocations according to the order in which they should be * considered, based on their dependent resources -- preferring (in order): * * Colocation that is not \c NULL * * Colocation whose resource has higher priority * * Colocation whose resource is of a higher-level variant * (bundle > clone > group > primitive) * * Colocation whose resource is promotable, if both are clones * * Colocation whose resource has lower ID in lexicographic order * * \param[in] a First colocation to compare * \param[in] b Second colocation to compare * * \return A negative number if \p a should be considered first, * a positive number if \p b should be considered first, * or 0 if order doesn't matter */ static gint cmp_dependent_priority(gconstpointer a, gconstpointer b) { return cmp_colocation_priority(a, b, true); } /*! * \internal * \brief Compare two colocations according to priority based on primaries * * Compare two colocations according to the order in which they should be * considered, based on their primary resources -- preferring (in order): * * Colocation that is not \c NULL * * Colocation whose primary has higher priority * * Colocation whose primary is of a higher-level variant * (bundle > clone > group > primitive) * * Colocation whose primary is promotable, if both are clones * * Colocation whose primary has lower ID in lexicographic order * * \param[in] a First colocation to compare * \param[in] b Second colocation to compare * * \return A negative number if \p a should be considered first, * a positive number if \p b should be considered first, * or 0 if order doesn't matter */ static gint cmp_primary_priority(gconstpointer a, gconstpointer b) { return cmp_colocation_priority(a, b, false); } /*! * \internal * \brief Add a "this with" colocation constraint to a sorted list * * \param[in,out] list List of constraints to add \p colocation to * \param[in] colocation Colocation constraint to add to \p list * \param[in] rsc Resource whose colocations we're getting (for * logging only) * * \note The list will be sorted using cmp_primary_priority(). */ void pcmk__add_this_with(GList **list, const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc) { CRM_ASSERT((list != NULL) && (colocation != NULL) && (rsc != NULL)); pcmk__rsc_trace(rsc, "Adding colocation %s (%s with %s using %s @%s) to " "'this with' list for %s", colocation->id, colocation->dependent->id, colocation->primary->id, colocation->node_attribute, pcmk_readable_score(colocation->score), rsc->id); *list = g_list_insert_sorted(*list, (gpointer) colocation, cmp_primary_priority); } /*! * \internal * \brief Add a list of "this with" colocation constraints to a list * * \param[in,out] list List of constraints to add \p addition to * \param[in] addition List of colocation constraints to add to \p list * \param[in] rsc Resource whose colocations we're getting (for * logging only) * * \note The lists must be pre-sorted by cmp_primary_priority(). */ void pcmk__add_this_with_list(GList **list, GList *addition, const pcmk_resource_t *rsc) { CRM_ASSERT((list != NULL) && (rsc != NULL)); pcmk__if_tracing( {}, // Always add each colocation individually if tracing { if (*list == NULL) { // Trivial case for efficiency if not tracing *list = g_list_copy(addition); return; } } ); for (const GList *iter = addition; iter != NULL; iter = iter->next) { pcmk__add_this_with(list, addition->data, rsc); } } /*! * \internal * \brief Add a "with this" colocation constraint to a sorted list * * \param[in,out] list List of constraints to add \p colocation to * \param[in] colocation Colocation constraint to add to \p list * \param[in] rsc Resource whose colocations we're getting (for * logging only) * * \note The list will be sorted using cmp_dependent_priority(). */ void pcmk__add_with_this(GList **list, const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc) { CRM_ASSERT((list != NULL) && (colocation != NULL) && (rsc != NULL)); pcmk__rsc_trace(rsc, "Adding colocation %s (%s with %s using %s @%s) to " "'with this' list for %s", colocation->id, colocation->dependent->id, colocation->primary->id, colocation->node_attribute, pcmk_readable_score(colocation->score), rsc->id); *list = g_list_insert_sorted(*list, (gpointer) colocation, cmp_dependent_priority); } /*! * \internal * \brief Add a list of "with this" colocation constraints to a list * * \param[in,out] list List of constraints to add \p addition to * \param[in] addition List of colocation constraints to add to \p list * \param[in] rsc Resource whose colocations we're getting (for * logging only) * * \note The lists must be pre-sorted by cmp_dependent_priority(). */ void pcmk__add_with_this_list(GList **list, GList *addition, const pcmk_resource_t *rsc) { CRM_ASSERT((list != NULL) && (rsc != NULL)); pcmk__if_tracing( {}, // Always add each colocation individually if tracing { if (*list == NULL) { // Trivial case for efficiency if not tracing *list = g_list_copy(addition); return; } } ); for (const GList *iter = addition; iter != NULL; iter = iter->next) { pcmk__add_with_this(list, addition->data, rsc); } } /*! * \internal * \brief Add orderings necessary for an anti-colocation constraint * * \param[in,out] first_rsc One resource in an anti-colocation * \param[in] first_role Anti-colocation role of \p first_rsc * \param[in] then_rsc Other resource in the anti-colocation * \param[in] then_role Anti-colocation role of \p then_rsc */ static void anti_colocation_order(pcmk_resource_t *first_rsc, int first_role, pcmk_resource_t *then_rsc, int then_role) { const char *first_tasks[] = { NULL, NULL }; const char *then_tasks[] = { NULL, NULL }; /* Actions to make first_rsc lose first_role */ if (first_role == pcmk_role_promoted) { first_tasks[0] = PCMK_ACTION_DEMOTE; } else { first_tasks[0] = PCMK_ACTION_STOP; if (first_role == pcmk_role_unpromoted) { first_tasks[1] = PCMK_ACTION_PROMOTE; } } /* Actions to make then_rsc gain then_role */ if (then_role == pcmk_role_promoted) { then_tasks[0] = PCMK_ACTION_PROMOTE; } else { then_tasks[0] = PCMK_ACTION_START; if (then_role == pcmk_role_unpromoted) { then_tasks[1] = PCMK_ACTION_DEMOTE; } } for (int first_lpc = 0; (first_lpc <= 1) && (first_tasks[first_lpc] != NULL); first_lpc++) { for (int then_lpc = 0; (then_lpc <= 1) && (then_tasks[then_lpc] != NULL); then_lpc++) { pcmk__order_resource_actions(first_rsc, first_tasks[first_lpc], then_rsc, then_tasks[then_lpc], pcmk__ar_if_required_on_same_node); } } } /*! * \internal * \brief Add a new colocation constraint to scheduler data * * \param[in] id XML ID for this constraint * \param[in] node_attr Colocate by this attribute (NULL for #uname) * \param[in] score Constraint score * \param[in,out] dependent Resource to be colocated * \param[in,out] primary Resource to colocate \p dependent with * \param[in] dependent_role Current role of \p dependent * \param[in] primary_role Current role of \p primary * \param[in] flags Group of enum pcmk__coloc_flags */ void pcmk__new_colocation(const char *id, const char *node_attr, int score, pcmk_resource_t *dependent, pcmk_resource_t *primary, const char *dependent_role, const char *primary_role, uint32_t flags) { pcmk__colocation_t *new_con = NULL; CRM_CHECK(id != NULL, return); if ((dependent == NULL) || (primary == NULL)) { pcmk__config_err("Ignoring colocation '%s' because resource " "does not exist", id); return; } if (score == 0) { pcmk__rsc_trace(dependent, "Ignoring colocation '%s' (%s with %s) because score is 0", id, dependent->id, primary->id); return; } new_con = pcmk__assert_alloc(1, sizeof(pcmk__colocation_t)); if (pcmk__str_eq(dependent_role, PCMK_ROLE_STARTED, pcmk__str_null_matches|pcmk__str_casei)) { dependent_role = PCMK__ROLE_UNKNOWN; } if (pcmk__str_eq(primary_role, PCMK_ROLE_STARTED, pcmk__str_null_matches|pcmk__str_casei)) { primary_role = PCMK__ROLE_UNKNOWN; } new_con->id = id; new_con->dependent = dependent; new_con->primary = primary; new_con->score = score; new_con->dependent_role = pcmk_parse_role(dependent_role); new_con->primary_role = pcmk_parse_role(primary_role); new_con->node_attribute = pcmk__s(node_attr, CRM_ATTR_UNAME); new_con->flags = flags; pcmk__add_this_with(&(dependent->private->this_with_colocations), new_con, dependent); pcmk__add_with_this(&(primary->private->with_this_colocations), new_con, primary); dependent->private->scheduler->colocation_constraints = g_list_prepend(dependent->private->scheduler->colocation_constraints, new_con); if (score <= -PCMK_SCORE_INFINITY) { anti_colocation_order(dependent, new_con->dependent_role, primary, new_con->primary_role); anti_colocation_order(primary, new_con->primary_role, dependent, new_con->dependent_role); } } /*! * \internal * \brief Return the boolean influence corresponding to configuration * * \param[in] coloc_id Colocation XML ID (for error logging) * \param[in] rsc Resource involved in constraint (for default) * \param[in] influence_s String value of \c PCMK_XA_INFLUENCE option * * \return \c pcmk__coloc_influence if string evaluates true, or string is * \c NULL or invalid and resource's \c PCMK_META_CRITICAL option * evaluates true, otherwise \c pcmk__coloc_none */ static uint32_t unpack_influence(const char *coloc_id, const pcmk_resource_t *rsc, const char *influence_s) { if (influence_s != NULL) { int influence_i = 0; if (crm_str_to_boolean(influence_s, &influence_i) < 0) { pcmk__config_err("Constraint '%s' has invalid value for " PCMK_XA_INFLUENCE " (using default)", coloc_id); } else { return (influence_i == 0)? pcmk__coloc_none : pcmk__coloc_influence; } } if (pcmk_is_set(rsc->flags, pcmk__rsc_critical)) { return pcmk__coloc_influence; } return pcmk__coloc_none; } static void unpack_colocation_set(xmlNode *set, int score, const char *coloc_id, const char *influence_s, pcmk_scheduler_t *scheduler) { xmlNode *xml_rsc = NULL; pcmk_resource_t *other = NULL; pcmk_resource_t *resource = NULL; const char *set_id = pcmk__xe_id(set); const char *role = crm_element_value(set, PCMK_XA_ROLE); bool with_previous = false; int local_score = score; bool sequential = false; uint32_t flags = pcmk__coloc_none; const char *xml_rsc_id = NULL; const char *score_s = crm_element_value(set, PCMK_XA_SCORE); if (score_s) { local_score = char2score(score_s); } if (local_score == 0) { crm_trace("Ignoring colocation '%s' for set '%s' because score is 0", coloc_id, set_id); return; } /* @COMPAT The deprecated PCMK__XA_ORDERING attribute specifies whether * resources in a positive-score set are colocated with the previous or next * resource. */ if (pcmk__str_eq(crm_element_value(set, PCMK__XA_ORDERING), PCMK__VALUE_GROUP, pcmk__str_null_matches|pcmk__str_casei)) { with_previous = true; } else { pcmk__warn_once(pcmk__wo_set_ordering, "Support for '" PCMK__XA_ORDERING "' other than" " '" PCMK__VALUE_GROUP "' in " PCMK_XE_RESOURCE_SET " (such as %s) is deprecated and will be removed in a" " future release", set_id); } if ((pcmk__xe_get_bool_attr(set, PCMK_XA_SEQUENTIAL, &sequential) == pcmk_rc_ok) && !sequential) { return; } if (local_score > 0) { for (xml_rsc = pcmk__xe_first_child(set, PCMK_XE_RESOURCE_REF, NULL, NULL); xml_rsc != NULL; xml_rsc = pcmk__xe_next_same(xml_rsc)) { xml_rsc_id = pcmk__xe_id(xml_rsc); resource = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (resource == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring %s and later resources in set %s: " "No such resource", xml_rsc_id, set_id); return; } if (other != NULL) { flags = pcmk__coloc_explicit | unpack_influence(coloc_id, resource, influence_s); if (with_previous) { pcmk__rsc_trace(resource, "Colocating %s with %s in set %s", resource->id, other->id, set_id); pcmk__new_colocation(set_id, NULL, local_score, resource, other, role, role, flags); } else { pcmk__rsc_trace(resource, "Colocating %s with %s in set %s", other->id, resource->id, set_id); pcmk__new_colocation(set_id, NULL, local_score, other, resource, role, role, flags); } } other = resource; } } else { /* Anti-colocating with every prior resource is * the only way to ensure the intuitive result * (i.e. that no one in the set can run with anyone else in the set) */ for (xml_rsc = pcmk__xe_first_child(set, PCMK_XE_RESOURCE_REF, NULL, NULL); xml_rsc != NULL; xml_rsc = pcmk__xe_next_same(xml_rsc)) { xmlNode *xml_rsc_with = NULL; xml_rsc_id = pcmk__xe_id(xml_rsc); resource = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (resource == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring %s and later resources in set %s: " "No such resource", xml_rsc_id, set_id); return; } flags = pcmk__coloc_explicit | unpack_influence(coloc_id, resource, influence_s); for (xml_rsc_with = pcmk__xe_first_child(set, PCMK_XE_RESOURCE_REF, NULL, NULL); xml_rsc_with != NULL; xml_rsc_with = pcmk__xe_next_same(xml_rsc_with)) { xml_rsc_id = pcmk__xe_id(xml_rsc_with); if (pcmk__str_eq(resource->id, xml_rsc_id, pcmk__str_none)) { break; } other = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); CRM_ASSERT(other != NULL); // We already processed it pcmk__new_colocation(set_id, NULL, local_score, resource, other, role, role, flags); } } } } /*! * \internal * \brief Colocate two resource sets relative to each other * * \param[in] id Colocation XML ID * \param[in] set1 Dependent set * \param[in] set2 Primary set * \param[in] score Colocation score * \param[in] influence_s Value of colocation's \c PCMK_XA_INFLUENCE * attribute * \param[in,out] scheduler Scheduler data */ static void colocate_rsc_sets(const char *id, const xmlNode *set1, const xmlNode *set2, int score, const char *influence_s, pcmk_scheduler_t *scheduler) { xmlNode *xml_rsc = NULL; pcmk_resource_t *rsc_1 = NULL; pcmk_resource_t *rsc_2 = NULL; const char *xml_rsc_id = NULL; const char *role_1 = crm_element_value(set1, PCMK_XA_ROLE); const char *role_2 = crm_element_value(set2, PCMK_XA_ROLE); int rc = pcmk_rc_ok; bool sequential = false; uint32_t flags = pcmk__coloc_none; if (score == 0) { crm_trace("Ignoring colocation '%s' between sets %s and %s " "because score is 0", id, pcmk__xe_id(set1), pcmk__xe_id(set2)); return; } rc = pcmk__xe_get_bool_attr(set1, PCMK_XA_SEQUENTIAL, &sequential); if ((rc != pcmk_rc_ok) || sequential) { // Get the first one xml_rsc = pcmk__xe_first_child(set1, PCMK_XE_RESOURCE_REF, NULL, NULL); if (xml_rsc != NULL) { xml_rsc_id = pcmk__xe_id(xml_rsc); rsc_1 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_1 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring colocation of set %s with set %s " "because first resource %s not found", pcmk__xe_id(set1), pcmk__xe_id(set2), xml_rsc_id); return; } } } rc = pcmk__xe_get_bool_attr(set2, PCMK_XA_SEQUENTIAL, &sequential); if ((rc != pcmk_rc_ok) || sequential) { // Get the last one for (xml_rsc = pcmk__xe_first_child(set2, PCMK_XE_RESOURCE_REF, NULL, NULL); xml_rsc != NULL; xml_rsc = pcmk__xe_next_same(xml_rsc)) { xml_rsc_id = pcmk__xe_id(xml_rsc); } rsc_2 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_2 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring colocation of set %s with set %s " "because last resource %s not found", pcmk__xe_id(set1), pcmk__xe_id(set2), xml_rsc_id); return; } } if ((rsc_1 != NULL) && (rsc_2 != NULL)) { // Both sets are sequential flags = pcmk__coloc_explicit | unpack_influence(id, rsc_1, influence_s); pcmk__new_colocation(id, NULL, score, rsc_1, rsc_2, role_1, role_2, flags); } else if (rsc_1 != NULL) { // Only set1 is sequential flags = pcmk__coloc_explicit | unpack_influence(id, rsc_1, influence_s); for (xml_rsc = pcmk__xe_first_child(set2, PCMK_XE_RESOURCE_REF, NULL, NULL); xml_rsc != NULL; xml_rsc = pcmk__xe_next_same(xml_rsc)) { xml_rsc_id = pcmk__xe_id(xml_rsc); rsc_2 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_2 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring set %s colocation with resource %s " "in set %s: No such resource", pcmk__xe_id(set1), xml_rsc_id, pcmk__xe_id(set2)); continue; } pcmk__new_colocation(id, NULL, score, rsc_1, rsc_2, role_1, role_2, flags); } } else if (rsc_2 != NULL) { // Only set2 is sequential for (xml_rsc = pcmk__xe_first_child(set1, PCMK_XE_RESOURCE_REF, NULL, NULL); xml_rsc != NULL; xml_rsc = pcmk__xe_next_same(xml_rsc)) { xml_rsc_id = pcmk__xe_id(xml_rsc); rsc_1 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_1 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring colocation of set %s resource %s " "with set %s: No such resource", pcmk__xe_id(set1), xml_rsc_id, pcmk__xe_id(set2)); continue; } flags = pcmk__coloc_explicit | unpack_influence(id, rsc_1, influence_s); pcmk__new_colocation(id, NULL, score, rsc_1, rsc_2, role_1, role_2, flags); } } else { // Neither set is sequential for (xml_rsc = pcmk__xe_first_child(set1, PCMK_XE_RESOURCE_REF, NULL, NULL); xml_rsc != NULL; xml_rsc = pcmk__xe_next_same(xml_rsc)) { xmlNode *xml_rsc_2 = NULL; xml_rsc_id = pcmk__xe_id(xml_rsc); rsc_1 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_1 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring colocation of set %s resource %s " "with set %s: No such resource", pcmk__xe_id(set1), xml_rsc_id, pcmk__xe_id(set2)); continue; } flags = pcmk__coloc_explicit | unpack_influence(id, rsc_1, influence_s); for (xml_rsc_2 = pcmk__xe_first_child(set2, PCMK_XE_RESOURCE_REF, NULL, NULL); xml_rsc_2 != NULL; xml_rsc_2 = pcmk__xe_next_same(xml_rsc_2)) { xml_rsc_id = pcmk__xe_id(xml_rsc_2); rsc_2 = pcmk__find_constraint_resource(scheduler->resources, xml_rsc_id); if (rsc_2 == NULL) { // Should be possible only with validation disabled pcmk__config_err("Ignoring colocation of set %s resource " "%s with set %s resource %s: No such " "resource", pcmk__xe_id(set1), pcmk__xe_id(xml_rsc), pcmk__xe_id(set2), xml_rsc_id); continue; } pcmk__new_colocation(id, NULL, score, rsc_1, rsc_2, role_1, role_2, flags); } } } } static void unpack_simple_colocation(xmlNode *xml_obj, const char *id, const char *influence_s, pcmk_scheduler_t *scheduler) { int score_i = 0; uint32_t flags = pcmk__coloc_none; const char *score = crm_element_value(xml_obj, PCMK_XA_SCORE); const char *dependent_id = crm_element_value(xml_obj, PCMK_XA_RSC); const char *primary_id = crm_element_value(xml_obj, PCMK_XA_WITH_RSC); const char *dependent_role = crm_element_value(xml_obj, PCMK_XA_RSC_ROLE); const char *primary_role = crm_element_value(xml_obj, PCMK_XA_WITH_RSC_ROLE); const char *attr = crm_element_value(xml_obj, PCMK_XA_NODE_ATTRIBUTE); const char *primary_instance = NULL; const char *dependent_instance = NULL; pcmk_resource_t *primary = NULL; pcmk_resource_t *dependent = NULL; primary = pcmk__find_constraint_resource(scheduler->resources, primary_id); dependent = pcmk__find_constraint_resource(scheduler->resources, dependent_id); // @COMPAT: Deprecated since 2.1.5 primary_instance = crm_element_value(xml_obj, PCMK__XA_WITH_RSC_INSTANCE); dependent_instance = crm_element_value(xml_obj, PCMK__XA_RSC_INSTANCE); if (dependent_instance != NULL) { pcmk__warn_once(pcmk__wo_coloc_inst, "Support for " PCMK__XA_RSC_INSTANCE " is deprecated " "and will be removed in a future release"); } if (primary_instance != NULL) { pcmk__warn_once(pcmk__wo_coloc_inst, "Support for " PCMK__XA_WITH_RSC_INSTANCE " is " "deprecated and will be removed in a future release"); } if (dependent == NULL) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "does not exist", id, dependent_id); return; } else if (primary == NULL) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "does not exist", id, primary_id); return; } else if ((dependent_instance != NULL) && !pcmk__is_clone(dependent)) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "is not a clone but instance '%s' was requested", id, dependent_id, dependent_instance); return; } else if ((primary_instance != NULL) && !pcmk__is_clone(primary)) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "is not a clone but instance '%s' was requested", id, primary_id, primary_instance); return; } if (dependent_instance != NULL) { dependent = find_clone_instance(dependent, dependent_instance); if (dependent == NULL) { pcmk__config_warn("Ignoring constraint '%s' because resource '%s' " "does not have an instance '%s'", id, dependent_id, dependent_instance); return; } } if (primary_instance != NULL) { primary = find_clone_instance(primary, primary_instance); if (primary == NULL) { pcmk__config_warn("Ignoring constraint '%s' because resource '%s' " "does not have an instance '%s'", "'%s'", id, primary_id, primary_instance); return; } } if (pcmk__xe_attr_is_true(xml_obj, PCMK_XA_SYMMETRICAL)) { pcmk__config_warn("The colocation constraint " "'" PCMK_XA_SYMMETRICAL "' attribute has been " "removed"); } if (score) { score_i = char2score(score); } flags = pcmk__coloc_explicit | unpack_influence(id, dependent, influence_s); pcmk__new_colocation(id, attr, score_i, dependent, primary, dependent_role, primary_role, flags); } // \return Standard Pacemaker return code static int unpack_colocation_tags(xmlNode *xml_obj, xmlNode **expanded_xml, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *dependent_id = NULL; const char *primary_id = NULL; const char *dependent_role = NULL; const char *primary_role = NULL; pcmk_resource_t *dependent = NULL; pcmk_resource_t *primary = NULL; pcmk_tag_t *dependent_tag = NULL; pcmk_tag_t *primary_tag = NULL; xmlNode *dependent_set = NULL; xmlNode *primary_set = NULL; bool any_sets = false; *expanded_xml = NULL; CRM_CHECK(xml_obj != NULL, return EINVAL); id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("Ignoring <%s> constraint without " PCMK_XA_ID, xml_obj->name); return pcmk_rc_unpack_error; } // Check whether there are any resource sets with template or tag references *expanded_xml = pcmk__expand_tags_in_sets(xml_obj, scheduler); if (*expanded_xml != NULL) { crm_log_xml_trace(*expanded_xml, "Expanded " PCMK_XE_RSC_COLOCATION); return pcmk_rc_ok; } dependent_id = crm_element_value(xml_obj, PCMK_XA_RSC); primary_id = crm_element_value(xml_obj, PCMK_XA_WITH_RSC); if ((dependent_id == NULL) || (primary_id == NULL)) { return pcmk_rc_ok; } if (!pcmk__valid_resource_or_tag(scheduler, dependent_id, &dependent, &dependent_tag)) { pcmk__config_err("Ignoring constraint '%s' because '%s' is not a " "valid resource or tag", id, dependent_id); return pcmk_rc_unpack_error; } if (!pcmk__valid_resource_or_tag(scheduler, primary_id, &primary, &primary_tag)) { pcmk__config_err("Ignoring constraint '%s' because '%s' is not a " "valid resource or tag", id, primary_id); return pcmk_rc_unpack_error; } if ((dependent != NULL) && (primary != NULL)) { /* Neither side references any template/tag. */ return pcmk_rc_ok; } if ((dependent_tag != NULL) && (primary_tag != NULL)) { // A colocation constraint between two templates/tags makes no sense pcmk__config_err("Ignoring constraint '%s' because two templates or " "tags cannot be colocated", id); return pcmk_rc_unpack_error; } dependent_role = crm_element_value(xml_obj, PCMK_XA_RSC_ROLE); primary_role = crm_element_value(xml_obj, PCMK_XA_WITH_RSC_ROLE); *expanded_xml = pcmk__xml_copy(NULL, xml_obj); /* Convert dependent's template/tag reference into constraint * PCMK_XE_RESOURCE_SET */ if (!pcmk__tag_to_set(*expanded_xml, &dependent_set, PCMK_XA_RSC, true, scheduler)) { pcmk__xml_free(*expanded_xml); *expanded_xml = NULL; return pcmk_rc_unpack_error; } if (dependent_set != NULL) { if (dependent_role != NULL) { /* Move PCMK_XA_RSC_ROLE into converted PCMK_XE_RESOURCE_SET as * PCMK_XA_ROLE */ crm_xml_add(dependent_set, PCMK_XA_ROLE, dependent_role); pcmk__xe_remove_attr(*expanded_xml, PCMK_XA_RSC_ROLE); } any_sets = true; } /* Convert primary's template/tag reference into constraint * PCMK_XE_RESOURCE_SET */ if (!pcmk__tag_to_set(*expanded_xml, &primary_set, PCMK_XA_WITH_RSC, true, scheduler)) { pcmk__xml_free(*expanded_xml); *expanded_xml = NULL; return pcmk_rc_unpack_error; } if (primary_set != NULL) { if (primary_role != NULL) { /* Move PCMK_XA_WITH_RSC_ROLE into converted PCMK_XE_RESOURCE_SET as * PCMK_XA_ROLE */ crm_xml_add(primary_set, PCMK_XA_ROLE, primary_role); pcmk__xe_remove_attr(*expanded_xml, PCMK_XA_WITH_RSC_ROLE); } any_sets = true; } if (any_sets) { crm_log_xml_trace(*expanded_xml, "Expanded " PCMK_XE_RSC_COLOCATION); } else { pcmk__xml_free(*expanded_xml); *expanded_xml = NULL; } return pcmk_rc_ok; } /*! * \internal * \brief Parse a colocation constraint from XML into scheduler data * * \param[in,out] xml_obj Colocation constraint XML to unpack * \param[in,out] scheduler Scheduler data to add constraint to */ void pcmk__unpack_colocation(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { int score_i = 0; xmlNode *set = NULL; xmlNode *last = NULL; xmlNode *orig_xml = NULL; xmlNode *expanded_xml = NULL; const char *id = crm_element_value(xml_obj, PCMK_XA_ID); const char *score = NULL; const char *influence_s = NULL; if (pcmk__str_empty(id)) { pcmk__config_err("Ignoring " PCMK_XE_RSC_COLOCATION " without " CRM_ATTR_ID); return; } if (unpack_colocation_tags(xml_obj, &expanded_xml, scheduler) != pcmk_rc_ok) { return; } if (expanded_xml != NULL) { orig_xml = xml_obj; xml_obj = expanded_xml; } score = crm_element_value(xml_obj, PCMK_XA_SCORE); if (score != NULL) { score_i = char2score(score); } influence_s = crm_element_value(xml_obj, PCMK_XA_INFLUENCE); for (set = pcmk__xe_first_child(xml_obj, PCMK_XE_RESOURCE_SET, NULL, NULL); set != NULL; set = pcmk__xe_next_same(set)) { set = pcmk__xe_resolve_idref(set, scheduler->input); if (set == NULL) { // Configuration error, message already logged if (expanded_xml != NULL) { pcmk__xml_free(expanded_xml); } return; } if (pcmk__str_empty(pcmk__xe_id(set))) { pcmk__config_err("Ignoring " PCMK_XE_RESOURCE_SET " without " CRM_ATTR_ID); continue; } unpack_colocation_set(set, score_i, id, influence_s, scheduler); if (last != NULL) { colocate_rsc_sets(id, last, set, score_i, influence_s, scheduler); } last = set; } if (expanded_xml) { pcmk__xml_free(expanded_xml); xml_obj = orig_xml; } if (last == NULL) { unpack_simple_colocation(xml_obj, id, influence_s, scheduler); } } /*! * \internal * \brief Make actions of a given type unrunnable for a given resource * * \param[in,out] rsc Resource whose actions should be blocked * \param[in] task Name of action to block * \param[in] reason Unrunnable start action causing the block */ static void mark_action_blocked(pcmk_resource_t *rsc, const char *task, const pcmk_resource_t *reason) { GList *iter = NULL; char *reason_text = crm_strdup_printf("colocation with %s", reason->id); for (iter = rsc->private->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = iter->data; if (pcmk_is_set(action->flags, pcmk_action_runnable) && pcmk__str_eq(action->task, task, pcmk__str_none)) { pcmk__clear_action_flags(action, pcmk_action_runnable); pe_action_set_reason(action, reason_text, false); pcmk__block_colocation_dependents(action); pcmk__update_action_for_orderings(action, rsc->private->scheduler); } } // If parent resource can't perform an action, neither can any children for (iter = rsc->children; iter != NULL; iter = iter->next) { mark_action_blocked((pcmk_resource_t *) (iter->data), task, reason); } free(reason_text); } /*! * \internal * \brief If an action is unrunnable, block any relevant dependent actions * * If a given action is an unrunnable start or promote, block the start or * promote actions of resources colocated with it, as appropriate to the * colocations' configured roles. * * \param[in,out] action Action to check */ void pcmk__block_colocation_dependents(pcmk_action_t *action) { GList *iter = NULL; GList *colocations = NULL; pcmk_resource_t *rsc = NULL; bool is_start = false; if (pcmk_is_set(action->flags, pcmk_action_runnable)) { return; // Only unrunnable actions block dependents } is_start = pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_none); if (!is_start && !pcmk__str_eq(action->task, PCMK_ACTION_PROMOTE, pcmk__str_none)) { return; // Only unrunnable starts and promotes block dependents } CRM_ASSERT(action->rsc != NULL); // Start and promote are resource actions /* If this resource is part of a collective resource, dependents are blocked * only if all instances of the collective are unrunnable, so check the * collective resource. */ rsc = uber_parent(action->rsc); if (rsc->private->parent != NULL) { rsc = rsc->private->parent; // Bundle } // Colocation fails only if entire primary can't reach desired role for (iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = iter->data; pcmk_action_t *child_action = NULL; child_action = find_first_action(child->private->actions, NULL, action->task, NULL); if ((child_action == NULL) || pcmk_is_set(child_action->flags, pcmk_action_runnable)) { crm_trace("Not blocking %s colocation dependents because " "at least %s has runnable %s", rsc->id, child->id, action->task); return; // At least one child can reach desired role } } crm_trace("Blocking %s colocation dependents due to unrunnable %s %s", rsc->id, action->rsc->id, action->task); // Check each colocation where this resource is primary colocations = pcmk__with_this_colocations(rsc); for (iter = colocations; iter != NULL; iter = iter->next) { pcmk__colocation_t *colocation = iter->data; if (colocation->score < PCMK_SCORE_INFINITY) { continue; // Only mandatory colocations block dependent } /* If the primary can't start, the dependent can't reach its colocated * role, regardless of what the primary or dependent colocation role is. * * If the primary can't be promoted, the dependent can't reach its * colocated role if the primary's colocation role is promoted. */ if (!is_start && (colocation->primary_role != pcmk_role_promoted)) { continue; } // Block the dependent from reaching its colocated role if (colocation->dependent_role == pcmk_role_promoted) { mark_action_blocked(colocation->dependent, PCMK_ACTION_PROMOTE, action->rsc); } else { mark_action_blocked(colocation->dependent, PCMK_ACTION_START, action->rsc); } } g_list_free(colocations); } /*! * \internal * \brief Get the resource to use for role comparisons * * A bundle replica includes a container and possibly an instance of the bundled * resource. The dependent in a "with bundle" colocation is colocated with a * particular bundle container. However, if the colocation includes a role, then * the role must be checked on the bundled resource instance inside the * container. The container itself will never be promoted; the bundled resource * may be. * * If the given resource is a bundle replica container, return the resource * inside it, if any. Otherwise, return the resource itself. * * \param[in] rsc Resource to check * * \return Resource to use for role comparisons */ static const pcmk_resource_t * get_resource_for_role(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk__rsc_replica_container)) { const pcmk_resource_t *child = pe__get_rsc_in_container(rsc); if (child != NULL) { return child; } } return rsc; } /*! * \internal * \brief Determine how a colocation constraint should affect a resource * * Colocation constraints have different effects at different points in the * scheduler sequence. Initially, they affect a resource's location; once that * is determined, then for promotable clones they can affect a resource * instance's role; after both are determined, the constraints no longer matter. * Given a specific colocation constraint, check what has been done so far to * determine what should be affected at the current point in the scheduler. * * \param[in] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint * \param[in] preview If true, pretend resources have already been assigned * * \return How colocation constraint should be applied at this point */ enum pcmk__coloc_affects pcmk__colocation_affects(const pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool preview) { const pcmk_resource_t *dependent_role_rsc = NULL; const pcmk_resource_t *primary_role_rsc = NULL; CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if (!preview && pcmk_is_set(primary->flags, pcmk__rsc_unassigned)) { // Primary resource has not been assigned yet, so we can't do anything return pcmk__coloc_affects_nothing; } dependent_role_rsc = get_resource_for_role(dependent); primary_role_rsc = get_resource_for_role(primary); if ((colocation->dependent_role >= pcmk_role_unpromoted) && (dependent_role_rsc->private->parent != NULL) && pcmk_is_set(dependent_role_rsc->private->parent->flags, pcmk__rsc_promotable) && !pcmk_is_set(dependent_role_rsc->flags, pcmk__rsc_unassigned)) { /* This is a colocation by role, and the dependent is a promotable clone * that has already been assigned, so the colocation should now affect * the role. */ return pcmk__coloc_affects_role; } if (!preview && !pcmk_is_set(dependent->flags, pcmk__rsc_unassigned)) { /* The dependent resource has already been through assignment, so the * constraint no longer has any effect. Log an error if a mandatory * colocation constraint has been violated. */ - const pcmk_node_t *primary_node = primary->allocated_to; + const pcmk_node_t *primary_node = primary->private->assigned_node; + const pcmk_node_t *dependent_node = dependent->private->assigned_node; - if (dependent->allocated_to == NULL) { + if (dependent_node == NULL) { crm_trace("Skipping colocation '%s': %s will not run anywhere", colocation->id, dependent->id); } else if (colocation->score >= PCMK_SCORE_INFINITY) { // Dependent resource must colocate with primary resource - if (!pcmk__same_node(primary_node, dependent->allocated_to)) { + if (!pcmk__same_node(primary_node, dependent_node)) { pcmk__sched_err("%s must be colocated with %s but is not " "(%s vs. %s)", dependent->id, primary->id, - pcmk__node_name(dependent->allocated_to), + pcmk__node_name(dependent_node), pcmk__node_name(primary_node)); } } else if (colocation->score <= -PCMK_SCORE_INFINITY) { // Dependent resource must anti-colocate with primary resource - if (pcmk__same_node(dependent->allocated_to, primary_node)) { + if (pcmk__same_node(dependent_node, primary_node)) { pcmk__sched_err("%s and %s must be anti-colocated but are " "assigned to the same node (%s)", dependent->id, primary->id, pcmk__node_name(primary_node)); } } return pcmk__coloc_affects_nothing; } if ((colocation->dependent_role != pcmk_role_unknown) && (colocation->dependent_role != dependent_role_rsc->next_role)) { crm_trace("Skipping %scolocation '%s': dependent limited to %s role " "but %s next role is %s", ((colocation->score < 0)? "anti-" : ""), colocation->id, pcmk_role_text(colocation->dependent_role), dependent_role_rsc->id, pcmk_role_text(dependent_role_rsc->next_role)); return pcmk__coloc_affects_nothing; } if ((colocation->primary_role != pcmk_role_unknown) && (colocation->primary_role != primary_role_rsc->next_role)) { crm_trace("Skipping %scolocation '%s': primary limited to %s role " "but %s next role is %s", ((colocation->score < 0)? "anti-" : ""), colocation->id, pcmk_role_text(colocation->primary_role), primary_role_rsc->id, pcmk_role_text(primary_role_rsc->next_role)); return pcmk__coloc_affects_nothing; } return pcmk__coloc_affects_location; } /*! * \internal * \brief Apply colocation to dependent for assignment purposes * * Update the allowed node scores of the dependent resource in a colocation, * for the purposes of assigning it to a node. * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint */ void pcmk__apply_coloc_to_scores(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation) { const char *attr = colocation->node_attribute; const char *value = NULL; GHashTable *work = NULL; GHashTableIter iter; pcmk_node_t *node = NULL; - if (primary->allocated_to != NULL) { - value = pcmk__colocation_node_attr(primary->allocated_to, attr, - primary); + if (primary->private->assigned_node != NULL) { + value = pcmk__colocation_node_attr(primary->private->assigned_node, + attr, primary); } else if (colocation->score < 0) { // Nothing to do (anti-colocation with something that is not running) return; } work = pcmk__copy_node_table(dependent->allowed_nodes); g_hash_table_iter_init(&iter, work); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { - if (primary->allocated_to == NULL) { + if (primary->private->assigned_node == NULL) { node->weight = pcmk__add_scores(-colocation->score, node->weight); pcmk__rsc_trace(dependent, "Applied %s to %s score on %s (now %s after " "subtracting %s because primary %s inactive)", colocation->id, dependent->id, pcmk__node_name(node), pcmk_readable_score(node->weight), pcmk_readable_score(colocation->score), primary->id); continue; } if (pcmk__str_eq(pcmk__colocation_node_attr(node, attr, dependent), value, pcmk__str_casei)) { /* Add colocation score only if optional (or minus infinity). A * mandatory colocation is a requirement rather than a preference, * so we don't need to consider it for relative assignment purposes. * The resource will simply be forbidden from running on the node if * the primary isn't active there (via the condition above). */ if (colocation->score < PCMK_SCORE_INFINITY) { node->weight = pcmk__add_scores(colocation->score, node->weight); pcmk__rsc_trace(dependent, "Applied %s to %s score on %s (now %s after " "adding %s)", colocation->id, dependent->id, pcmk__node_name(node), pcmk_readable_score(node->weight), pcmk_readable_score(colocation->score)); } continue; } if (colocation->score >= PCMK_SCORE_INFINITY) { /* Only mandatory colocations are relevant when the colocation * attribute doesn't match, because an attribute not matching is not * a negative preference -- the colocation is simply relevant only * where it matches. */ node->weight = -PCMK_SCORE_INFINITY; pcmk__rsc_trace(dependent, "Banned %s from %s because colocation %s attribute %s " "does not match", dependent->id, pcmk__node_name(node), colocation->id, attr); } } if ((colocation->score <= -PCMK_SCORE_INFINITY) || (colocation->score >= PCMK_SCORE_INFINITY) || pcmk__any_node_available(work)) { g_hash_table_destroy(dependent->allowed_nodes); dependent->allowed_nodes = work; work = NULL; } else { pcmk__rsc_info(dependent, "%s: Rolling back scores from %s (no available nodes)", dependent->id, primary->id); } if (work != NULL) { g_hash_table_destroy(work); } } /*! * \internal * \brief Apply colocation to dependent for role purposes * * Update the priority of the dependent resource in a colocation, for the * purposes of selecting its role * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint */ void pcmk__apply_coloc_to_priority(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation) { const char *dependent_value = NULL; const char *primary_value = NULL; const char *attr = colocation->node_attribute; int score_multiplier = 1; + const pcmk_node_t *primary_node = NULL; + const pcmk_node_t *dependent_node = NULL; const pcmk_resource_t *primary_role_rsc = NULL; CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); - if ((primary->allocated_to == NULL) || (dependent->allocated_to == NULL)) { + primary_node = primary->private->assigned_node; + dependent_node = dependent->private->assigned_node; + + if ((primary_node == NULL) || (dependent_node == NULL)) { return; } - dependent_value = pcmk__colocation_node_attr(dependent->allocated_to, attr, + dependent_value = pcmk__colocation_node_attr(dependent_node, attr, dependent); - primary_value = pcmk__colocation_node_attr(primary->allocated_to, attr, - primary); + primary_value = pcmk__colocation_node_attr(primary_node, attr, primary); primary_role_rsc = get_resource_for_role(primary); if (!pcmk__str_eq(dependent_value, primary_value, pcmk__str_casei)) { if ((colocation->score == PCMK_SCORE_INFINITY) && (colocation->dependent_role == pcmk_role_promoted)) { dependent->private->priority = -PCMK_SCORE_INFINITY; } return; } if ((colocation->primary_role != pcmk_role_unknown) && (colocation->primary_role != primary_role_rsc->next_role)) { return; } if (colocation->dependent_role == pcmk_role_unpromoted) { score_multiplier = -1; } dependent->private->priority = pcmk__add_scores(score_multiplier * colocation->score, dependent->private->priority); pcmk__rsc_trace(dependent, "Applied %s to %s promotion priority (now %s after %s %s)", colocation->id, dependent->id, pcmk_readable_score(dependent->private->priority), ((score_multiplier == 1)? "adding" : "subtracting"), pcmk_readable_score(colocation->score)); } /*! * \internal * \brief Find score of highest-scored node that matches colocation attribute * * \param[in] rsc Resource whose allowed nodes should be searched * \param[in] attr Colocation attribute name (must not be NULL) * \param[in] value Colocation attribute value to require */ static int best_node_score_matching_attr(const pcmk_resource_t *rsc, const char *attr, const char *value) { GHashTableIter iter; pcmk_node_t *node = NULL; int best_score = -PCMK_SCORE_INFINITY; const char *best_node = NULL; // Find best allowed node with matching attribute g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if ((node->weight > best_score) && pcmk__node_available(node, false, false) && pcmk__str_eq(value, pcmk__colocation_node_attr(node, attr, rsc), pcmk__str_casei)) { best_score = node->weight; best_node = node->details->uname; } } if (!pcmk__str_eq(attr, CRM_ATTR_UNAME, pcmk__str_none)) { if (best_node == NULL) { crm_info("No allowed node for %s matches node attribute %s=%s", rsc->id, attr, value); } else { crm_info("Allowed node %s for %s had best score (%d) " "of those matching node attribute %s=%s", best_node, rsc->id, best_score, attr, value); } } return best_score; } /*! * \internal * \brief Check whether a resource is allowed only on a single node * * \param[in] rsc Resource to check * * \return \c true if \p rsc is allowed only on one node, otherwise \c false */ static bool allowed_on_one(const pcmk_resource_t *rsc) { GHashTableIter iter; pcmk_node_t *allowed_node = NULL; int allowed_nodes = 0; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &allowed_node)) { if ((allowed_node->weight >= 0) && (++allowed_nodes > 1)) { pcmk__rsc_trace(rsc, "%s is allowed on multiple nodes", rsc->id); return false; } } pcmk__rsc_trace(rsc, "%s is allowed %s", rsc->id, ((allowed_nodes == 1)? "on a single node" : "nowhere")); return (allowed_nodes == 1); } /*! * \internal * \brief Add resource's colocation matches to current node assignment scores * * For each node in a given table, if any of a given resource's allowed nodes * have a matching value for the colocation attribute, add the highest of those * nodes' scores to the node's score. * * \param[in,out] nodes Table of nodes with assignment scores so far * \param[in] source_rsc Resource whose node scores to add * \param[in] target_rsc Resource on whose behalf to update \p nodes * \param[in] colocation Original colocation constraint (used to get * configured primary resource's stickiness, and * to get colocation node attribute; pass NULL to * ignore stickiness and use default attribute) * \param[in] factor Factor by which to multiply scores being added * \param[in] only_positive Whether to add only positive scores */ static void add_node_scores_matching_attr(GHashTable *nodes, const pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const pcmk__colocation_t *colocation, float factor, bool only_positive) { GHashTableIter iter; pcmk_node_t *node = NULL; const char *attr = colocation->node_attribute; // Iterate through each node g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { float delta_f = 0; int delta = 0; int score = 0; int new_score = 0; const char *value = pcmk__colocation_node_attr(node, attr, target_rsc); score = best_node_score_matching_attr(source_rsc, attr, value); if ((factor < 0) && (score < 0)) { /* If the dependent is anti-colocated, we generally don't want the * primary to prefer nodes that the dependent avoids. That could * lead to unnecessary shuffling of the primary when the dependent * hits its migration threshold somewhere, for example. * * However, there are cases when it is desirable. If the dependent * can't run anywhere but where the primary is, it would be * worthwhile to move the primary for the sake of keeping the * dependent active. * * We can't know that exactly at this point since we don't know * where the primary will be assigned, but we can limit considering * the preference to when the dependent is allowed only on one node. * This is less than ideal for multiple reasons: * * - the dependent could be allowed on more than one node but have * anti-colocation primaries on each; * - the dependent could be a clone or bundle with multiple * instances, and the dependent as a whole is allowed on multiple * nodes but some instance still can't run * - the dependent has considered node-specific criteria such as * location constraints and stickiness by this point, but might * have other factors that end up disallowing a node * * but the alternative is making the primary move when it doesn't * need to. * * We also consider the primary's stickiness and influence, so the * user has some say in the matter. (This is the configured primary, * not a particular instance of the primary, but that doesn't matter * unless stickiness uses a rule to vary by node, and that seems * acceptable to ignore.) */ if ((colocation->primary->private->stickiness >= -score) || !pcmk__colocation_has_influence(colocation, NULL) || !allowed_on_one(colocation->dependent)) { crm_trace("%s: Filtering %d + %f * %d " "(double negative disallowed)", pcmk__node_name(node), node->weight, factor, score); continue; } } if (node->weight == INFINITY_HACK) { crm_trace("%s: Filtering %d + %f * %d (node was marked unusable)", pcmk__node_name(node), node->weight, factor, score); continue; } delta_f = factor * score; // Round the number; see http://c-faq.com/fp/round.html delta = (int) ((delta_f < 0)? (delta_f - 0.5) : (delta_f + 0.5)); /* Small factors can obliterate the small scores that are often actually * used in configurations. If the score and factor are nonzero, ensure * that the result is nonzero as well. */ if ((delta == 0) && (score != 0)) { if (factor > 0.0) { delta = 1; } else if (factor < 0.0) { delta = -1; } } new_score = pcmk__add_scores(delta, node->weight); if (only_positive && (new_score < 0) && (node->weight > 0)) { crm_trace("%s: Filtering %d + %f * %d = %d " "(negative disallowed, marking node unusable)", pcmk__node_name(node), node->weight, factor, score, new_score); node->weight = INFINITY_HACK; continue; } if (only_positive && (new_score < 0) && (node->weight == 0)) { crm_trace("%s: Filtering %d + %f * %d = %d (negative disallowed)", pcmk__node_name(node), node->weight, factor, score, new_score); continue; } crm_trace("%s: %d + %f * %d = %d", pcmk__node_name(node), node->weight, factor, score, new_score); node->weight = new_score; } } /*! * \internal * \brief Update nodes with scores of colocated resources' nodes * * Given a table of nodes and a resource, update the nodes' scores with the * scores of the best nodes matching the attribute used for each of the * resource's relevant colocations. * * \param[in,out] source_rsc Resource whose node scores to add * \param[in] target_rsc Resource on whose behalf to update \p *nodes * \param[in] log_id Resource ID for logs (if \c NULL, use * \p source_rsc ID) * \param[in,out] nodes Nodes to update (set initial contents to \c NULL * to copy allowed nodes from \p source_rsc) * \param[in] colocation Original colocation constraint (used to get * configured primary resource's stickiness, and * to get colocation node attribute; if \c NULL, * source_rsc's own matching node scores * will not be added, and \p *nodes must be \c NULL * as well) * \param[in] factor Incorporate scores multiplied by this factor * \param[in] flags Bitmask of enum pcmk__coloc_select values * * \note \c NULL \p target_rsc, \c NULL \p *nodes, \c NULL \p colocation, and * the \c pcmk__coloc_select_this_with flag are used together (and only by * \c cmp_resources()). * \note The caller remains responsible for freeing \p *nodes. * \note This is the shared implementation of * \c pcmk__assignment_methods_t:add_colocated_node_scores(). */ void pcmk__add_colocated_node_scores(pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags) { GHashTable *work = NULL; CRM_ASSERT((source_rsc != NULL) && (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, pcmk__rsc_updating_nodes)) { pcmk__rsc_info(source_rsc, "%s: Breaking dependency loop at %s", log_id, source_rsc->id); return; } pcmk__set_rsc_flags(source_rsc, pcmk__rsc_updating_nodes); if (*nodes == NULL) { work = pcmk__copy_node_table(source_rsc->allowed_nodes); target_rsc = source_rsc; } else { const bool pos = pcmk_is_set(flags, pcmk__coloc_select_nonnegative); pcmk__rsc_trace(source_rsc, "%s: Merging %s scores from %s (at %.6f)", log_id, (pos? "positive" : "all"), source_rsc->id, factor); work = pcmk__copy_node_table(*nodes); add_node_scores_matching_attr(work, source_rsc, target_rsc, colocation, factor, pos); } if (work == NULL) { pcmk__clear_rsc_flags(source_rsc, pcmk__rsc_updating_nodes); return; } if (pcmk__any_node_available(work)) { GList *colocations = NULL; if (pcmk_is_set(flags, pcmk__coloc_select_this_with)) { colocations = pcmk__this_with_colocations(source_rsc); pcmk__rsc_trace(source_rsc, "Checking additional %d optional '%s with' " "constraints", g_list_length(colocations), source_rsc->id); } else { colocations = pcmk__with_this_colocations(source_rsc); pcmk__rsc_trace(source_rsc, "Checking additional %d optional 'with %s' " "constraints", g_list_length(colocations), source_rsc->id); } flags |= pcmk__coloc_select_active; for (GList *iter = colocations; iter != NULL; iter = iter->next) { pcmk__colocation_t *constraint = iter->data; pcmk_resource_t *other = NULL; float other_factor = factor * constraint->score / (float) PCMK_SCORE_INFINITY; if (pcmk_is_set(flags, pcmk__coloc_select_this_with)) { other = constraint->primary; } else if (!pcmk__colocation_has_influence(constraint, NULL)) { continue; } else { other = constraint->dependent; } pcmk__rsc_trace(source_rsc, "Optionally merging score of '%s' constraint " "(%s with %s)", constraint->id, constraint->dependent->id, constraint->primary->id); other->private->cmds->add_colocated_node_scores(other, target_rsc, log_id, &work, constraint, other_factor, flags); pe__show_node_scores(true, NULL, log_id, work, source_rsc->private->scheduler); } g_list_free(colocations); } else if (pcmk_is_set(flags, pcmk__coloc_select_active)) { pcmk__rsc_info(source_rsc, "%s: Rolling back optional scores from %s", log_id, source_rsc->id); g_hash_table_destroy(work); pcmk__clear_rsc_flags(source_rsc, pcmk__rsc_updating_nodes); return; } if (pcmk_is_set(flags, pcmk__coloc_select_nonnegative)) { pcmk_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, work); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if (node->weight == INFINITY_HACK) { node->weight = 1; } } } if (*nodes != NULL) { g_hash_table_destroy(*nodes); } *nodes = work; pcmk__clear_rsc_flags(source_rsc, pcmk__rsc_updating_nodes); } /*! * \internal * \brief Apply a "with this" colocation to a resource's allowed node scores * * \param[in,out] data Colocation to apply * \param[in,out] user_data Resource being assigned */ void pcmk__add_dependent_scores(gpointer data, gpointer user_data) { pcmk__colocation_t *colocation = data; pcmk_resource_t *primary = user_data; pcmk_resource_t *dependent = colocation->dependent; const float factor = colocation->score / (float) PCMK_SCORE_INFINITY; uint32_t flags = pcmk__coloc_select_active; if (!pcmk__colocation_has_influence(colocation, NULL)) { return; } if (pcmk__is_clone(primary)) { flags |= pcmk__coloc_select_nonnegative; } pcmk__rsc_trace(primary, "%s: Incorporating attenuated %s assignment scores due " "to colocation %s", primary->id, dependent->id, colocation->id); dependent->private->cmds->add_colocated_node_scores(dependent, primary, dependent->id, &primary->allowed_nodes, colocation, factor, flags); } /*! * \internal * \brief Exclude nodes from a dependent's node table if not in a given list * * Given a dependent resource in a colocation and a list of nodes where the * primary resource will run, set a node's score to \c -INFINITY in the * dependent's node table if not found in the primary nodes list. * * \param[in,out] dependent Dependent resource * \param[in] primary Primary resource (for logging only) * \param[in] colocation Colocation constraint (for logging only) * \param[in] primary_nodes List of nodes where the primary will have * unblocked instances in a suitable role * \param[in] merge_scores If \c true and a node is found in both \p table * and \p list, add the node's score in \p list to * the node's score in \p table */ void pcmk__colocation_intersect_nodes(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, const GList *primary_nodes, bool merge_scores) { GHashTableIter iter; pcmk_node_t *dependent_node = NULL; CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); g_hash_table_iter_init(&iter, dependent->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &dependent_node)) { const pcmk_node_t *primary_node = NULL; primary_node = pe_find_node_id(primary_nodes, dependent_node->details->id); if (primary_node == NULL) { dependent_node->weight = -PCMK_SCORE_INFINITY; pcmk__rsc_trace(dependent, "Banning %s from %s (no primary instance) for %s", dependent->id, pcmk__node_name(dependent_node), colocation->id); } else if (merge_scores) { dependent_node->weight = pcmk__add_scores(dependent_node->weight, primary_node->weight); pcmk__rsc_trace(dependent, "Added %s's score %s to %s's score for %s (now %s) " "for colocation %s", primary->id, pcmk_readable_score(primary_node->weight), dependent->id, pcmk__node_name(dependent_node), pcmk_readable_score(dependent_node->weight), colocation->id); } } } /*! * \internal * \brief Get all colocations affecting a resource as the primary * * \param[in] rsc Resource to get colocations for * * \return Newly allocated list of colocations affecting \p rsc as primary * * \note This is a convenience wrapper for the with_this_colocations() method. */ GList * pcmk__with_this_colocations(const pcmk_resource_t *rsc) { GList *list = NULL; rsc->private->cmds->with_this_colocations(rsc, rsc, &list); return list; } /*! * \internal * \brief Get all colocations affecting a resource as the dependent * * \param[in] rsc Resource to get colocations for * * \return Newly allocated list of colocations affecting \p rsc as dependent * * \note This is a convenience wrapper for the this_with_colocations() method. */ GList * pcmk__this_with_colocations(const pcmk_resource_t *rsc) { GList *list = NULL; rsc->private->cmds->this_with_colocations(rsc, rsc, &list); return list; } diff --git a/lib/pacemaker/pcmk_sched_group.c b/lib/pacemaker/pcmk_sched_group.c index b4be5b7ac4..7ad47753a4 100644 --- a/lib/pacemaker/pcmk_sched_group.c +++ b/lib/pacemaker/pcmk_sched_group.c @@ -1,973 +1,973 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "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. */ pcmk_node_t * pcmk__group_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { pcmk_node_t *first_assigned_node = NULL; pcmk_resource_t *first_member = NULL; CRM_ASSERT(pcmk__is_group(rsc)); if (!pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) { - return rsc->allocated_to; // Assignment already done + return rsc->private->assigned_node; // Assignment already done } if (pcmk_is_set(rsc->flags, pcmk__rsc_assigning)) { pcmk__rsc_debug(rsc, "Assignment dependency loop detected involving %s", rsc->id); return NULL; } if (rsc->children == NULL) { // No members to assign pcmk__clear_rsc_flags(rsc, pcmk__rsc_unassigned); return NULL; } pcmk__set_rsc_flags(rsc, pcmk__rsc_assigning); first_member = (pcmk_resource_t *) rsc->children->data; rsc->role = first_member->role; pe__show_node_scores(!pcmk_is_set(rsc->private->scheduler->flags, pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, rsc->private->scheduler); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; pcmk_node_t *node = NULL; pcmk__rsc_trace(rsc, "Assigning group %s member %s", rsc->id, member->id); node = member->private->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"); pcmk__clear_rsc_flags(rsc, pcmk__rsc_assigning|pcmk__rsc_unassigned); if (!pe__group_flag_is_set(rsc, pcmk__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 pcmk_action_t * create_group_pseudo_op(pcmk_resource_t *group, const char *action) { pcmk_action_t *op = custom_action(group, pcmk__op_key(group->id, action, 0), action, NULL, TRUE, group->private->scheduler); pcmk__set_action_flags(op, pcmk_action_pseudo|pcmk_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(pcmk_resource_t *rsc) { CRM_ASSERT(pcmk__is_group(rsc)); pcmk__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) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; member->private->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, PCMK_META_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; pcmk_resource_t *last_active; pcmk_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) { pcmk_resource_t *member = (pcmk_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 = pcmk__ar_then_implies_first_graphed; // For ordering demote vs demoted or stop vs stopped uint32_t post_down_flags = pcmk__ar_first_implies_then_graphed; // Create the individual member's implicit constraints member->private->cmds->internal_constraints(member); if (member_data->previous_member == NULL) { // This is first member if (member_data->ordered) { pcmk__set_relation_flags(down_flags, pcmk__ar_ordered); post_down_flags = pcmk__ar_first_implies_then; } } else if (member_data->colocated) { uint32_t flags = pcmk__coloc_none; if (pcmk_is_set(member->flags, pcmk__rsc_critical)) { flags |= pcmk__coloc_influence; } // Colocate this member with the previous one pcmk__new_colocation("#group-members", NULL, PCMK_SCORE_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->private->parent, PCMK_ACTION_DEMOTE, member, PCMK_ACTION_DEMOTE, down_flags); pcmk__order_resource_actions(member, PCMK_ACTION_DEMOTE, member->private->parent, PCMK_ACTION_DEMOTED, post_down_flags); // Promote group -> promote member -> group is promoted pcmk__order_resource_actions(member, PCMK_ACTION_PROMOTE, member->private->parent, PCMK_ACTION_PROMOTED, pcmk__ar_unrunnable_first_blocks |pcmk__ar_first_implies_then |pcmk__ar_first_implies_then_graphed); pcmk__order_resource_actions(member->private->parent, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pcmk__ar_then_implies_first_graphed); } // Stop group -> stop member -> group is stopped pcmk__order_stops(member->private->parent, member, down_flags); pcmk__order_resource_actions(member, PCMK_ACTION_STOP, member->private->parent, PCMK_ACTION_STOPPED, post_down_flags); // Start group -> start member -> group is started pcmk__order_starts(member->private->parent, member, pcmk__ar_then_implies_first_graphed); pcmk__order_resource_actions(member, PCMK_ACTION_START, member->private->parent, PCMK_ACTION_RUNNING, pcmk__ar_unrunnable_first_blocks |pcmk__ar_first_implies_then |pcmk__ar_first_implies_then_graphed); if (!member_data->ordered) { pcmk__order_starts(member->private->parent, member, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks |pcmk__ar_then_implies_first_graphed); if (member_data->promotable) { pcmk__order_resource_actions(member->private->parent, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks |pcmk__ar_then_implies_first_graphed); } } else if (member_data->previous_member == NULL) { pcmk__order_starts(member->private->parent, member, pcmk__ar_none); if (member_data->promotable) { pcmk__order_resource_actions(member->private->parent, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pcmk__ar_none); } } else { // Order this member relative to the previous one pcmk__order_starts(member_data->previous_member, member, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks); pcmk__order_stops(member, member_data->previous_member, pcmk__ar_ordered|pcmk__ar_intermediate_stop); /* 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, pcmk__ar_then_implies_first |pcmk__ar_unrunnable_first_blocks); } if (member_data->promotable) { pcmk__order_resource_actions(member_data->previous_member, PCMK_ACTION_PROMOTE, member, PCMK_ACTION_PROMOTE, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks); pcmk__order_resource_actions(member, PCMK_ACTION_DEMOTE, member_data->previous_member, PCMK_ACTION_DEMOTE, pcmk__ar_ordered); } } // 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, pcmk__ar_ordered); } 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(pcmk_resource_t *rsc) { struct member_data member_data = { false, }; const pcmk_resource_t *top = NULL; CRM_ASSERT(pcmk__is_group(rsc)); /* 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, pcmk__ar_unrunnable_first_blocks); pcmk__order_resource_actions(rsc, PCMK_ACTION_STOPPED, rsc, PCMK_ACTION_START, pcmk__ar_ordered); pcmk__order_resource_actions(rsc, PCMK_ACTION_START, rsc, PCMK_ACTION_RUNNING, pcmk__ar_unrunnable_first_blocks); top = pe__const_top_resource(rsc, false); member_data.ordered = pe__group_flag_is_set(rsc, pcmk__group_ordered); member_data.colocated = pe__group_flag_is_set(rsc, pcmk__group_colocated); member_data.promotable = pcmk_is_set(top->flags, pcmk__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(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation) { pcmk_resource_t *member = NULL; if (dependent->children == NULL) { return; } pcmk__rsc_trace(primary, "Processing %s (group %s with %s) for dependent", colocation->id, dependent->id, primary->id); if (pe__group_flag_is_set(dependent, pcmk__group_colocated)) { // Colocate first member (internal colocations will handle the rest) member = (pcmk_resource_t *) dependent->children->data; member->private->cmds->apply_coloc_score(member, primary, colocation, true); return; } if (colocation->score >= PCMK_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 = (pcmk_resource_t *) iter->data; member->private->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(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation) { const pcmk_resource_t *member = NULL; pcmk__rsc_trace(primary, "Processing colocation %s (%s with group %s) for primary", colocation->id, dependent->id, primary->id); if (pcmk_is_set(primary->flags, pcmk__rsc_unassigned)) { return; } if (pe__group_flag_is_set(primary, pcmk__group_colocated)) { if (colocation->score >= PCMK_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 = (pcmk_resource_t *) primary->children->data; } if (member == NULL) { return; // Nothing to colocate with } member->private->cmds->apply_coloc_score(dependent, member, colocation, false); return; } if (colocation->score >= PCMK_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->private->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(pcmk_resource_t *dependent, const pcmk_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(pcmk__is_primitive(dependent)); 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(pcmk_action_t *action, const pcmk_node_t *node) { // Default flags for a group action uint32_t flags = pcmk_action_optional |pcmk_action_runnable |pcmk_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) { pcmk_resource_t *member = (pcmk_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 = pcmk_action_text(task); pcmk_action_t *member_action = NULL; member_action = find_first_action(member->private->actions, NULL, task_s, node); if (member_action != NULL) { uint32_t member_flags = 0U; member_flags = member->private->cmds->action_flags(member_action, node); // Group action is mandatory if any member action is if (pcmk_is_set(flags, pcmk_action_optional) && !pcmk_is_set(member_flags, pcmk_action_optional)) { pcmk__rsc_trace(action->rsc, "%s is mandatory because %s is", action->uuid, member_action->uuid); pcmk__clear_raw_action_flags(flags, "group action", pcmk_action_optional); pcmk__clear_action_flags(action, pcmk_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, pcmk_action_runnable) && !pcmk_is_set(member_flags, pcmk_action_runnable)) { pcmk__rsc_trace(action->rsc, "%s is unrunnable because %s is", action->uuid, member_action->uuid); pcmk__clear_raw_action_flags(flags, "group action", pcmk_action_runnable); pcmk__clear_action_flags(action, pcmk_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)) { pcmk__rsc_trace(action->rsc, "%s is not runnable because %s will not %s", action->uuid, member->id, task_s); pcmk__clear_raw_action_flags(flags, "group action", pcmk_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 pcmk_action_optional to affect only * mandatory actions, and pcmk_action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t pcmk__group_update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler) { 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) && (scheduler != NULL)); // Update the actions for the group itself changed |= pcmk__update_ordered_actions(first, then, node, flags, filter, type, scheduler); // Update the actions for each group member for (GList *iter = then->rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; pcmk_action_t *member_action = NULL; member_action = find_first_action(member->private->actions, NULL, then->task, node); if (member_action == NULL) { continue; } changed |= member->private->cmds->update_ordered_actions(first, member_action, node, flags, filter, type, scheduler); } 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(pcmk_resource_t *rsc, pcmk__location_t *location) { GList *node_list_orig = NULL; GList *node_list_copy = NULL; bool reset_scores = true; CRM_ASSERT(pcmk__is_group(rsc) && (location != NULL)); node_list_orig = location->nodes; node_list_copy = pcmk__copy_node_list(node_list_orig, true); reset_scores = pe__group_flag_is_set(rsc, pcmk__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) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; member->private->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->nodes = node_list_copy; } } location->nodes = node_list_orig; g_list_free_full(node_list_copy, free); } // Group implementation of pcmk__assignment_methods_t:colocated_resources() GList * pcmk__group_colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs) { CRM_ASSERT(pcmk__is_group(rsc)); if (orig_rsc == NULL) { orig_rsc = rsc; } if (pe__group_flag_is_set(rsc, pcmk__group_colocated) || pcmk__is_clone(rsc->private->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) { const pcmk_resource_t *member = iter->data; colocated_rscs = member->private->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 pcmk__assignment_methods_t:with_this_colocations() void pcmk__with_group_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *parent = NULL; CRM_ASSERT((orig_rsc != NULL) && (list != NULL) && pcmk__is_group(rsc)); parent = rsc->private->parent; // 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; } pcmk__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->private->with_this_colocations, orig_rsc); // If cloned, add any relevant colocations with the clone if (parent != NULL) { parent->private->cmds->with_this_colocations(parent, orig_rsc, list); } if (!pe__group_flag_is_set(rsc, pcmk__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 pcmk_resource_t *member = iter->data; if (member == orig_rsc) { continue; } member->private->cmds->with_this_colocations(member, orig_rsc, list); } } // Group implementation of pcmk__assignment_methods_t:this_with_colocations() void pcmk__group_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *parent = NULL; const pcmk_resource_t *member = NULL; CRM_ASSERT((orig_rsc != NULL) && (list != NULL) && pcmk__is_group(rsc)); parent = rsc->private->parent; // 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 pcmk_resource_t *) rsc->children->data)) { pcmk__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->private->this_with_colocations, orig_rsc); // If cloned, add any relevant colocations involving the clone if (parent != NULL) { parent->private->cmds->this_with_colocations(parent, orig_rsc, list); } if (!pe__group_flag_is_set(rsc, pcmk__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) { continue; } member->private->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, pcmk__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->private->this_with_colocations; 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 == PCMK_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 pcmk__assignment_methods_t:add_colocated_node_scores(). */ void pcmk__group_add_colocated_node_scores(pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags) { pcmk_resource_t *member = NULL; CRM_ASSERT(pcmk__is_group(source_rsc) && (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, pcmk__rsc_updating_nodes)) { pcmk__rsc_info(source_rsc, "%s: Breaking dependency loop at %s", log_id, source_rsc->id); return; } pcmk__set_rsc_flags(source_rsc, pcmk__rsc_updating_nodes); // 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; } pcmk__rsc_trace(source_rsc, "%s: Merging scores from group %s using member %s " "(at %.6f)", log_id, source_rsc->id, member->id, factor); member->private->cmds->add_colocated_node_scores(member, target_rsc, log_id, nodes, colocation, factor, flags); pcmk__clear_rsc_flags(source_rsc, pcmk__rsc_updating_nodes); } // Group implementation of pcmk__assignment_methods_t:add_utilization() void pcmk__group_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { pcmk_resource_t *member = NULL; CRM_ASSERT((orig_rsc != NULL) && (utilization != NULL) && pcmk__is_group(rsc)); if (!pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) { return; } pcmk__rsc_trace(orig_rsc, "%s: Adding group %s as colocated utilization", orig_rsc->id, rsc->id); if (pe__group_flag_is_set(rsc, pcmk__group_colocated) || pcmk__is_clone(rsc->private->parent)) { // Every group member will be on same node, so sum all members for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = (pcmk_resource_t *) iter->data; if (pcmk_is_set(member->flags, pcmk__rsc_unassigned) && (g_list_find(all_rscs, member) == NULL)) { member->private->cmds->add_utilization(member, orig_rsc, all_rscs, utilization); } } } else if (rsc->children != NULL) { // Just add first member's utilization member = (pcmk_resource_t *) rsc->children->data; if ((member != NULL) && pcmk_is_set(member->flags, pcmk__rsc_unassigned) && (g_list_find(all_rscs, member) == NULL)) { member->private->cmds->add_utilization(member, orig_rsc, all_rscs, utilization); } } } void pcmk__group_shutdown_lock(pcmk_resource_t *rsc) { CRM_ASSERT(pcmk__is_group(rsc)); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *member = (pcmk_resource_t *) iter->data; member->private->cmds->shutdown_lock(member); } } diff --git a/lib/pacemaker/pcmk_sched_migration.c b/lib/pacemaker/pcmk_sched_migration.c index 95e9a3a2f6..349cae97d5 100644 --- a/lib/pacemaker/pcmk_sched_migration.c +++ b/lib/pacemaker/pcmk_sched_migration.c @@ -1,406 +1,409 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Add migration source and target meta-attributes to an action * * \param[in,out] action Action to add meta-attributes to * \param[in] source Node to add as migration source * \param[in] target Node to add as migration target */ static void add_migration_meta(pcmk_action_t *action, const pcmk_node_t *source, const pcmk_node_t *target) { pcmk__insert_meta(action, PCMK__META_MIGRATE_SOURCE, source->details->uname); pcmk__insert_meta(action, PCMK__META_MIGRATE_TARGET, target->details->uname); } /*! * \internal * \brief Create internal migration actions for a migrateable resource * * \param[in,out] rsc Resource to create migration actions for * \param[in] current Node that resource is originally active on */ void pcmk__create_migration_actions(pcmk_resource_t *rsc, const pcmk_node_t *current) { pcmk_action_t *migrate_to = NULL; pcmk_action_t *migrate_from = NULL; pcmk_action_t *start = NULL; pcmk_action_t *stop = NULL; pcmk__rsc_trace(rsc, "Creating actions to %smigrate %s from %s to %s", ((rsc->partial_migration_target == NULL)? "" : "partially "), rsc->id, pcmk__node_name(current), - pcmk__node_name(rsc->allocated_to)); - start = start_action(rsc, rsc->allocated_to, TRUE); + pcmk__node_name(rsc->private->assigned_node)); + start = start_action(rsc, rsc->private->assigned_node, TRUE); stop = stop_action(rsc, current, TRUE); if (rsc->partial_migration_target == NULL) { migrate_to = custom_action(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_TO, 0), PCMK_ACTION_MIGRATE_TO, current, TRUE, rsc->private->scheduler); } migrate_from = custom_action(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_FROM, 0), - PCMK_ACTION_MIGRATE_FROM, rsc->allocated_to, - TRUE, rsc->private->scheduler); + PCMK_ACTION_MIGRATE_FROM, + rsc->private->assigned_node, TRUE, + rsc->private->scheduler); pcmk__set_action_flags(start, pcmk_action_migratable); pcmk__set_action_flags(stop, pcmk_action_migratable); // This is easier than trying to delete it from the graph pcmk__set_action_flags(start, pcmk_action_pseudo); if (rsc->partial_migration_target == NULL) { pcmk__set_action_flags(migrate_from, pcmk_action_migratable); pcmk__set_action_flags(migrate_to, pcmk_action_migratable); migrate_to->needs = start->needs; // Probe -> migrate_to -> migrate_from pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_TO, 0), NULL, pcmk__ar_ordered, rsc->private->scheduler); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_TO, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_FROM, 0), NULL, pcmk__ar_ordered|pcmk__ar_unmigratable_then_blocks, rsc->private->scheduler); } else { pcmk__set_action_flags(migrate_from, pcmk_action_migratable); migrate_from->needs = start->needs; // Probe -> migrate_from (migrate_to already completed) pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_FROM, 0), NULL, pcmk__ar_ordered, rsc->private->scheduler); } // migrate_from before stop or start pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_FROM, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, pcmk__ar_ordered|pcmk__ar_unmigratable_then_blocks, rsc->private->scheduler); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_FROM, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered |pcmk__ar_unmigratable_then_blocks |pcmk__ar_first_else_then, rsc->private->scheduler); if (migrate_to != NULL) { - add_migration_meta(migrate_to, current, rsc->allocated_to); + add_migration_meta(migrate_to, current, rsc->private->assigned_node); if (!pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { /* migrate_to takes place on the source node, but can affect the * target node depending on how the agent is written. Because of * this, pending migrate_to actions must be recorded in the CIB, * in case the source node loses membership while the migrate_to * action is still in flight. * * However we know Pacemaker Remote connection resources don't * require this, so we skip this for them. (Although it wouldn't * hurt, and now that PCMK_META_RECORD_PENDING defaults to true, * skipping it matters even less.) */ pcmk__insert_meta(migrate_to, PCMK_META_RECORD_PENDING, PCMK_VALUE_TRUE); } } - add_migration_meta(migrate_from, current, rsc->allocated_to); + add_migration_meta(migrate_from, current, rsc->private->assigned_node); } /*! * \internal * \brief Abort a dangling migration by scheduling a stop (and possibly cleanup) * * \param[in] data Source node of dangling migration * \param[in,out] user_data Resource involved in dangling migration */ void pcmk__abort_dangling_migration(void *data, void *user_data) { const pcmk_node_t *dangling_source = (const pcmk_node_t *) data; pcmk_resource_t *rsc = (pcmk_resource_t *) user_data; pcmk_action_t *stop = NULL; bool cleanup = pcmk_is_set(rsc->private->scheduler->flags, pcmk_sched_remove_after_stop); pcmk__rsc_trace(rsc, "Scheduling stop%s for %s on %s due to dangling migration", (cleanup? " and cleanup" : ""), rsc->id, pcmk__node_name(dangling_source)); stop = stop_action(rsc, dangling_source, FALSE); pcmk__set_action_flags(stop, pcmk_action_migration_abort); if (cleanup) { pcmk__schedule_cleanup(rsc, dangling_source, false); } } /*! * \internal * \brief Check whether a resource can migrate * * \param[in] rsc Resource to check * \param[in] node Resource's current node * * \return true if \p rsc can migrate, otherwise false */ bool pcmk__rsc_can_migrate(const pcmk_resource_t *rsc, const pcmk_node_t *current) { CRM_CHECK(rsc != NULL, return false); if (!pcmk_is_set(rsc->flags, pcmk__rsc_migratable)) { pcmk__rsc_trace(rsc, "%s cannot migrate because " "the configuration does not allow it", rsc->id); return false; } if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__rsc_trace(rsc, "%s cannot migrate because it is not managed", rsc->id); return false; } if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { pcmk__rsc_trace(rsc, "%s cannot migrate because it is failed", rsc->id); return false; } if (pcmk_is_set(rsc->flags, pcmk__rsc_start_pending)) { pcmk__rsc_trace(rsc, "%s cannot migrate because it has a start pending", rsc->id); return false; } if ((current == NULL) || current->details->unclean) { pcmk__rsc_trace(rsc, "%s cannot migrate because " "its current node (%s) is unclean", rsc->id, pcmk__node_name(current)); return false; } - if ((rsc->allocated_to == NULL) || rsc->allocated_to->details->unclean) { + if ((rsc->private->assigned_node == NULL) + || rsc->private->assigned_node->details->unclean) { + pcmk__rsc_trace(rsc, "%s cannot migrate because " "its next node (%s) is unclean", - rsc->id, pcmk__node_name(rsc->allocated_to)); + rsc->id, pcmk__node_name(rsc->private->assigned_node)); return false; } return true; } /*! * \internal * \brief Get an action name from an action or operation key * * \param[in] action If not NULL, get action name from here * \param[in] key If not NULL, get action name from here * * \return Newly allocated copy of action name (or NULL if none available) */ static char * task_from_action_or_key(const pcmk_action_t *action, const char *key) { char *res = NULL; if (action != NULL) { res = pcmk__str_copy(action->task); } else if (key != NULL) { parse_op_key(key, NULL, &res, NULL); } return res; } /*! * \internal * \brief Order migration actions equivalent to a given ordering * * Orderings involving start, stop, demote, and promote actions must be honored * during a migration as well, so duplicate any such ordering for the * corresponding migration actions. * * \param[in,out] order Ordering constraint to check */ void pcmk__order_migration_equivalents(pcmk__action_relation_t *order) { char *first_task = NULL; char *then_task = NULL; bool then_migratable; bool first_migratable; // Only orderings between unrelated resources are relevant if ((order->rsc1 == NULL) || (order->rsc2 == NULL) || (order->rsc1 == order->rsc2) || is_parent(order->rsc1, order->rsc2) || is_parent(order->rsc2, order->rsc1)) { return; } // Only orderings involving at least one migratable resource are relevant first_migratable = pcmk_is_set(order->rsc1->flags, pcmk__rsc_migratable); then_migratable = pcmk_is_set(order->rsc2->flags, pcmk__rsc_migratable); if (!first_migratable && !then_migratable) { return; } // Check which actions are involved first_task = task_from_action_or_key(order->action1, order->task1); then_task = task_from_action_or_key(order->action2, order->task2); if (pcmk__str_eq(first_task, PCMK_ACTION_START, pcmk__str_none) && pcmk__str_eq(then_task, PCMK_ACTION_START, pcmk__str_none)) { uint32_t flags = pcmk__ar_ordered; if (first_migratable && then_migratable) { /* A start then B start * -> A migrate_from then B migrate_to */ pcmk__new_ordering(order->rsc1, pcmk__op_key(order->rsc1->id, PCMK_ACTION_MIGRATE_FROM, 0), NULL, order->rsc2, pcmk__op_key(order->rsc2->id, PCMK_ACTION_MIGRATE_TO, 0), NULL, flags, order->rsc1->private->scheduler); } if (then_migratable) { if (first_migratable) { pcmk__set_relation_flags(flags, pcmk__ar_if_first_unmigratable); } /* A start then B start * -> A start then B migrate_to (if start is not part of a * migration) */ pcmk__new_ordering(order->rsc1, pcmk__op_key(order->rsc1->id, PCMK_ACTION_START, 0), NULL, order->rsc2, pcmk__op_key(order->rsc2->id, PCMK_ACTION_MIGRATE_TO, 0), NULL, flags, order->rsc1->private->scheduler); } } else if (then_migratable && pcmk__str_eq(first_task, PCMK_ACTION_STOP, pcmk__str_none) && pcmk__str_eq(then_task, PCMK_ACTION_STOP, pcmk__str_none)) { uint32_t flags = pcmk__ar_ordered; if (first_migratable) { pcmk__set_relation_flags(flags, pcmk__ar_if_first_unmigratable); } /* For an ordering "stop A then stop B", if A is moving via restart, and * B is migrating, enforce that B's migrate_to occurs after A's stop. */ pcmk__new_ordering(order->rsc1, pcmk__op_key(order->rsc1->id, PCMK_ACTION_STOP, 0), NULL, order->rsc2, pcmk__op_key(order->rsc2->id, PCMK_ACTION_MIGRATE_TO, 0), NULL, flags, order->rsc1->private->scheduler); // Also order B's migrate_from after A's stop during partial migrations if (order->rsc2->partial_migration_target != NULL) { pcmk__new_ordering(order->rsc1, pcmk__op_key(order->rsc1->id, PCMK_ACTION_STOP, 0), NULL, order->rsc2, pcmk__op_key(order->rsc2->id, PCMK_ACTION_MIGRATE_FROM, 0), NULL, flags, order->rsc1->private->scheduler); } } else if (pcmk__str_eq(first_task, PCMK_ACTION_PROMOTE, pcmk__str_none) && pcmk__str_eq(then_task, PCMK_ACTION_START, pcmk__str_none)) { uint32_t flags = pcmk__ar_ordered; if (then_migratable) { /* A promote then B start * -> A promote then B migrate_to */ pcmk__new_ordering(order->rsc1, pcmk__op_key(order->rsc1->id, PCMK_ACTION_PROMOTE, 0), NULL, order->rsc2, pcmk__op_key(order->rsc2->id, PCMK_ACTION_MIGRATE_TO, 0), NULL, flags, order->rsc1->private->scheduler); } } else if (pcmk__str_eq(first_task, PCMK_ACTION_DEMOTE, pcmk__str_none) && pcmk__str_eq(then_task, PCMK_ACTION_STOP, pcmk__str_none)) { uint32_t flags = pcmk__ar_ordered; if (then_migratable) { /* A demote then B stop * -> A demote then B migrate_to */ pcmk__new_ordering(order->rsc1, pcmk__op_key(order->rsc1->id, PCMK_ACTION_DEMOTE, 0), NULL, order->rsc2, pcmk__op_key(order->rsc2->id, PCMK_ACTION_MIGRATE_TO, 0), NULL, flags, order->rsc1->private->scheduler); // Order B migrate_from after A demote during partial migrations if (order->rsc2->partial_migration_target != NULL) { pcmk__new_ordering(order->rsc1, pcmk__op_key(order->rsc1->id, PCMK_ACTION_DEMOTE, 0), NULL, order->rsc2, pcmk__op_key(order->rsc2->id, PCMK_ACTION_MIGRATE_FROM, 0), NULL, flags, order->rsc1->private->scheduler); } } } free(first_task); free(then_task); } diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c index d4cb1bcb2c..ee63ded450 100644 --- a/lib/pacemaker/pcmk_sched_primitive.c +++ b/lib/pacemaker/pcmk_sched_primitive.c @@ -1,1699 +1,1700 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include // uint8_t, uint32_t #include #include #include "libpacemaker_private.h" static void stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void start_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); #define RSC_ROLE_MAX (pcmk_role_promoted + 1) static enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = { /* This array lists the immediate next role when transitioning from one role * to a target role. For example, when going from Stopped to Promoted, the * next role is Unpromoted, because the resource must be started before it * can be promoted. The current state then becomes Started, which is fed * into this array again, giving a next role of Promoted. * * Current role Immediate next role Final target role * ------------ ------------------- ----------------- */ /* Unknown */ { pcmk_role_unknown, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_stopped, /* Started */ pcmk_role_stopped, /* Unpromoted */ pcmk_role_stopped, /* Promoted */ }, /* Stopped */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_started, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_unpromoted, /* Promoted */ }, /* Started */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_started, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, /* Unpromoted */ { pcmk_role_stopped, /* Unknown */ pcmk_role_stopped, /* Stopped */ pcmk_role_stopped, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, /* Promoted */ { pcmk_role_stopped, /* Unknown */ pcmk_role_unpromoted, /* Stopped */ pcmk_role_unpromoted, /* Started */ pcmk_role_unpromoted, /* Unpromoted */ pcmk_role_promoted, /* Promoted */ }, }; /*! * \internal * \brief Function to schedule actions needed for a role change * * \param[in,out] rsc Resource whose role is changing * \param[in,out] node Node where resource will be in its next role * \param[in] optional Whether scheduled actions should be optional */ typedef void (*rsc_transition_fn)(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional); static rsc_transition_fn rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = { /* This array lists the function needed to transition directly from one role * to another. NULL indicates that nothing is needed. * * Current role Transition function Next role * ------------ ------------------- ---------- */ /* Unknown */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ assert_role_error, /* Started */ assert_role_error, /* Unpromoted */ assert_role_error, /* Promoted */ }, /* Stopped */ { assert_role_error, /* Unknown */ NULL, /* Stopped */ start_resource, /* Started */ start_resource, /* Unpromoted */ assert_role_error, /* Promoted */ }, /* Started */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ NULL, /* Started */ NULL, /* Unpromoted */ promote_resource, /* Promoted */ }, /* Unpromoted */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ stop_resource, /* Started */ NULL, /* Unpromoted */ promote_resource, /* Promoted */ }, /* Promoted */ { assert_role_error, /* Unknown */ demote_resource, /* Stopped */ demote_resource, /* Started */ demote_resource, /* Unpromoted */ NULL, /* Promoted */ }, }; /*! * \internal * \brief Get a list of a resource's allowed nodes sorted by node score * * \param[in] rsc Resource to check * * \return List of allowed nodes sorted by node score */ static GList * sorted_allowed_nodes(const pcmk_resource_t *rsc) { if (rsc->allowed_nodes != NULL) { GList *nodes = g_hash_table_get_values(rsc->allowed_nodes); if (nodes != NULL) { return pcmk__sort_nodes(nodes, pcmk__current_node(rsc)); } } return NULL; } /*! * \internal * \brief Assign a resource to its best allowed node, if possible * * \param[in,out] rsc Resource to choose a node for * \param[in] prefer If not \c NULL, prefer this node when all else * equal * \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a * node, set next role to stopped and update * existing actions * * \return true if \p rsc could be assigned to a node, otherwise false * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ static bool assign_best_node(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { GList *nodes = NULL; pcmk_node_t *chosen = NULL; pcmk_node_t *best = NULL; const pcmk_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc); if (prefer == NULL) { prefer = most_free_node; } if (!pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) { // We've already finished assignment of resources to nodes - return rsc->allocated_to != NULL; + return rsc->private->assigned_node != NULL; } // Sort allowed nodes by score nodes = sorted_allowed_nodes(rsc); if (nodes != NULL) { best = (pcmk_node_t *) nodes->data; // First node has best score } if ((prefer != NULL) && (nodes != NULL)) { // Get the allowed node version of prefer chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id); if (chosen == NULL) { pcmk__rsc_trace(rsc, "Preferred node %s for %s was unknown", pcmk__node_name(prefer), rsc->id); /* Favor the preferred node as long as its score is at least as good as * the best allowed node's. * * An alternative would be to favor the preferred node even if the best * node is better, when the best node's score is less than INFINITY. */ } else if (chosen->weight < best->weight) { pcmk__rsc_trace(rsc, "Preferred node %s for %s was unsuitable", pcmk__node_name(chosen), rsc->id); chosen = NULL; } else if (!pcmk__node_available(chosen, true, false)) { pcmk__rsc_trace(rsc, "Preferred node %s for %s was unavailable", pcmk__node_name(chosen), rsc->id); chosen = NULL; } else { pcmk__rsc_trace(rsc, "Chose preferred node %s for %s " "(ignoring %d candidates)", pcmk__node_name(chosen), rsc->id, g_list_length(nodes)); } } if ((chosen == NULL) && (best != NULL)) { /* Either there is no preferred node, or the preferred node is not * suitable, but another node is allowed to run the resource. */ chosen = best; if (!pcmk__is_unique_clone(rsc->private->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. */ pcmk_node_t *running = pcmk__current_node(rsc); if (running == NULL) { // Nothing to do } else if (!pcmk__node_available(running, true, false)) { pcmk__rsc_trace(rsc, "Current node for %s (%s) can't run resources", rsc->id, pcmk__node_name(running)); } else { int nodes_with_best_score = 1; for (GList *iter = nodes->next; iter; iter = iter->next) { pcmk_node_t *allowed = (pcmk_node_t *) iter->data; if (allowed->weight != chosen->weight) { // The nodes are sorted by score, so no more are equal break; } if (pcmk__same_node(allowed, running)) { // Scores are equal, so prefer the current node chosen = allowed; } nodes_with_best_score++; } if (nodes_with_best_score > 1) { uint8_t log_level = LOG_INFO; if (chosen->weight >= PCMK_SCORE_INFINITY) { log_level = LOG_WARNING; } do_crm_log(log_level, "Chose %s for %s from %d nodes with score %s", pcmk__node_name(chosen), rsc->id, nodes_with_best_score, pcmk_readable_score(chosen->weight)); } } } pcmk__rsc_trace(rsc, "Chose %s for %s from %d candidates", pcmk__node_name(chosen), rsc->id, g_list_length(nodes)); } pcmk__assign_resource(rsc, chosen, false, stop_if_fail); g_list_free(nodes); - return rsc->allocated_to != NULL; + return rsc->private->assigned_node != NULL; } /*! * \internal * \brief Apply a "this with" colocation to a node's allowed node scores * * \param[in,out] colocation Colocation to apply * \param[in,out] rsc Resource being assigned */ static void apply_this_with(pcmk__colocation_t *colocation, pcmk_resource_t *rsc) { GHashTable *archive = NULL; pcmk_resource_t *other = colocation->primary; // In certain cases, we will need to revert the node scores if ((colocation->dependent_role >= pcmk_role_promoted) || ((colocation->score < 0) && (colocation->score > -PCMK_SCORE_INFINITY))) { archive = pcmk__copy_node_table(rsc->allowed_nodes); } if (pcmk_is_set(other->flags, pcmk__rsc_unassigned)) { pcmk__rsc_trace(rsc, "%s: Assigning colocation %s primary %s first" "(score=%d role=%s)", rsc->id, colocation->id, other->id, colocation->score, pcmk_role_text(colocation->dependent_role)); other->private->cmds->assign(other, NULL, true); } // Apply the colocation score to this resource's allowed node scores rsc->private->cmds->apply_coloc_score(rsc, other, colocation, true); if ((archive != NULL) && !pcmk__any_node_available(rsc->allowed_nodes)) { pcmk__rsc_info(rsc, "%s: Reverting scores from colocation with %s " "because no nodes allowed", rsc->id, other->id); g_hash_table_destroy(rsc->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 pcmk_resource_t *connection) { pcmk_node_t *remote_node = pcmk_find_node(connection->private->scheduler, connection->id); CRM_CHECK(remote_node != NULL, return); - if ((connection->allocated_to != NULL) + if ((connection->private->assigned_node != 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" : ""), + ((connection->private->assigned_node == NULL)? "un" : ""), pcmk_role_text(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. */ pcmk_node_t * pcmk__primitive_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail) { GList *this_with_colocations = NULL; GList *with_this_colocations = NULL; GList *iter = NULL; pcmk_resource_t *parent = NULL; pcmk__colocation_t *colocation = NULL; pcmk_scheduler_t *scheduler = NULL; CRM_ASSERT(pcmk__is_primitive(rsc)); scheduler = rsc->private->scheduler; parent = rsc->private->parent; // Never assign a child without parent being assigned first if ((parent != NULL) && !pcmk_is_set(parent->flags, pcmk__rsc_assigning)) { pcmk__rsc_debug(rsc, "%s: Assigning parent %s first", rsc->id, parent->id); parent->private->cmds->assign(parent, prefer, stop_if_fail); } if (!pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) { // Assignment has already been done const char *node_name = "no node"; - if (rsc->allocated_to != NULL) { - node_name = pcmk__node_name(rsc->allocated_to); + if (rsc->private->assigned_node != NULL) { + node_name = pcmk__node_name(rsc->private->assigned_node); } pcmk__rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name); - return rsc->allocated_to; + return rsc->private->assigned_node; } // Ensure we detect assignment loops if (pcmk_is_set(rsc->flags, pcmk__rsc_assigning)) { pcmk__rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id); return NULL; } pcmk__set_rsc_flags(rsc, pcmk__rsc_assigning); pe__show_node_scores(true, rsc, "Pre-assignment", rsc->allowed_nodes, scheduler); this_with_colocations = pcmk__this_with_colocations(rsc); with_this_colocations = pcmk__with_this_colocations(rsc); // Apply mandatory colocations first, to satisfy as many as possible for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score <= -PCMK_SCORE_INFINITY) || (colocation->score >= PCMK_SCORE_INFINITY)) { apply_this_with(colocation, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score <= -PCMK_SCORE_INFINITY) || (colocation->score >= PCMK_SCORE_INFINITY)) { pcmk__add_dependent_scores(colocation, rsc); } } pe__show_node_scores(true, rsc, "Mandatory-colocations", rsc->allowed_nodes, scheduler); // Then apply optional colocations for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score > -PCMK_SCORE_INFINITY) && (colocation->score < PCMK_SCORE_INFINITY)) { apply_this_with(colocation, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score > -PCMK_SCORE_INFINITY) && (colocation->score < PCMK_SCORE_INFINITY)) { pcmk__add_dependent_scores(colocation, rsc); } } g_list_free(this_with_colocations); g_list_free(with_this_colocations); if (rsc->next_role == pcmk_role_stopped) { pcmk__rsc_trace(rsc, "Banning %s from all nodes because it will be stopped", rsc->id); resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, PCMK_META_TARGET_ROLE, scheduler); } else if ((rsc->next_role > rsc->role) && !pcmk_is_set(scheduler->flags, pcmk_sched_quorate) && (scheduler->no_quorum_policy == pcmk_no_quorum_freeze)) { crm_notice("Resource %s cannot be elevated from %s to %s due to " PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE, rsc->id, pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role)); pe__set_next_role(rsc, rsc->role, PCMK_OPT_NO_QUORUM_POLICY "=" PCMK_VALUE_FREEZE); } pe__show_node_scores(!pcmk_is_set(scheduler->flags, pcmk_sched_output_scores), rsc, __func__, rsc->allowed_nodes, scheduler); // Unmanage resource if fencing is enabled but no device is configured if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled) && !pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); } if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { // Unmanaged resources stay on their current node const char *reason = NULL; pcmk_node_t *assign_to = NULL; pe__set_next_role(rsc, rsc->role, "unmanaged"); assign_to = pcmk__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, pcmk__rsc_failed)) { reason = "failed"; } else { reason = "active"; } pcmk__rsc_info(rsc, "Unmanaged resource %s assigned to %s: %s", rsc->id, (assign_to? assign_to->details->uname : "no node"), reason); pcmk__assign_resource(rsc, assign_to, true, stop_if_fail); } else if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_all)) { // Must stop at some point, but be consistent with stop_if_fail if (stop_if_fail) { pcmk__rsc_debug(rsc, "Forcing %s to stop: " PCMK_OPT_STOP_ALL_RESOURCES, rsc->id); } pcmk__assign_resource(rsc, NULL, true, stop_if_fail); } else if (!assign_best_node(rsc, prefer, stop_if_fail)) { // Assignment failed if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { pcmk__rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id); } else if ((rsc->running_on != NULL) && stop_if_fail) { pcmk__rsc_info(rsc, "Stopping removed resource %s", rsc->id); } } pcmk__clear_rsc_flags(rsc, pcmk__rsc_assigning); if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { remote_connection_assigned(rsc); } - return rsc->allocated_to; + return rsc->private->assigned_node; } /*! * \internal * \brief Schedule actions to bring resource down and back to current role * * \param[in,out] rsc Resource to restart * \param[in,out] current Node that resource should be brought down on * \param[in] need_stop Whether the resource must be stopped * \param[in] need_promote Whether the resource must be promoted * * \return Role that resource would have after scheduled actions are taken */ static void schedule_restart_actions(pcmk_resource_t *rsc, pcmk_node_t *current, bool need_stop, bool need_promote) { enum rsc_role_e role = rsc->role; enum rsc_role_e next_role; rsc_transition_fn fn = NULL; pcmk__set_rsc_flags(rsc, pcmk__rsc_restarting); // Bring resource down to a stop on its current node while (role != pcmk_role_stopped) { next_role = rsc_state_matrix[role][pcmk_role_stopped]; pcmk__rsc_trace(rsc, "Creating %s action to take %s down from %s to %s", (need_stop? "required" : "optional"), rsc->id, pcmk_role_text(role), pcmk_role_text(next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, current, !need_stop); role = next_role; } // Bring resource up to its next role on its next node while ((rsc->role <= rsc->next_role) && (role != rsc->role) && !pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) { bool required = need_stop; next_role = rsc_state_matrix[role][rsc->role]; if ((next_role == pcmk_role_promoted) && need_promote) { required = true; } pcmk__rsc_trace(rsc, "Creating %s action to take %s up from %s to %s", (required? "required" : "optional"), rsc->id, pcmk_role_text(role), pcmk_role_text(next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } - fn(rsc, rsc->allocated_to, !required); + fn(rsc, rsc->private->assigned_node, !required); role = next_role; } pcmk__clear_rsc_flags(rsc, pcmk__rsc_restarting); } /*! * \internal * \brief If a resource's next role is not explicitly specified, set a default * * \param[in,out] rsc Resource to set next role for * * \return "explicit" if next role was explicitly set, otherwise "implicit" */ static const char * set_default_next_role(pcmk_resource_t *rsc) { if (rsc->next_role != pcmk_role_unknown) { return "explicit"; } - if (rsc->allocated_to == NULL) { + if (rsc->private->assigned_node == NULL) { pe__set_next_role(rsc, pcmk_role_stopped, "assignment"); } else { pe__set_next_role(rsc, pcmk_role_started, "assignment"); } return "implicit"; } /*! * \internal * \brief Create an action to represent an already pending start * * \param[in,out] rsc Resource to create start action for */ static void create_pending_start(pcmk_resource_t *rsc) { pcmk_action_t *start = NULL; pcmk__rsc_trace(rsc, "Creating action for %s to represent already pending start", rsc->id); - start = start_action(rsc, rsc->allocated_to, TRUE); + start = start_action(rsc, rsc->private->assigned_node, TRUE); pcmk__set_action_flags(start, pcmk_action_always_in_graph); } /*! * \internal * \brief Schedule actions needed to take a resource to its next role * * \param[in,out] rsc Resource to schedule actions for */ static void schedule_role_transition_actions(pcmk_resource_t *rsc) { enum rsc_role_e role = rsc->role; while (role != rsc->next_role) { enum rsc_role_e next_role = rsc_state_matrix[role][rsc->next_role]; rsc_transition_fn fn = NULL; pcmk__rsc_trace(rsc, "Creating action to take %s from %s to %s " "(ending at %s)", rsc->id, pcmk_role_text(role), pcmk_role_text(next_role), pcmk_role_text(rsc->next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } - fn(rsc, rsc->allocated_to, false); + fn(rsc, rsc->private->assigned_node, false); role = next_role; } } /*! * \internal * \brief Create all actions needed for a given primitive resource * * \param[in,out] rsc Primitive resource to create actions for */ void pcmk__primitive_create_actions(pcmk_resource_t *rsc) { bool need_stop = false; bool need_promote = false; bool is_moving = false; bool allow_migrate = false; bool multiply_active = false; pcmk_node_t *current = NULL; unsigned int num_all_active = 0; unsigned int num_clean_active = 0; const char *next_role_source = NULL; CRM_ASSERT(pcmk__is_primitive(rsc)); next_role_source = set_default_next_role(rsc); pcmk__rsc_trace(rsc, "Creating all actions for %s transition from %s to %s " "(%s) on %s", rsc->id, pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role), next_role_source, - pcmk__node_name(rsc->allocated_to)); + pcmk__node_name(rsc->private->assigned_node)); current = rsc->private->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) - && !pcmk__same_node(current, rsc->allocated_to) + if ((current != NULL) && (rsc->private->assigned_node != NULL) + && !pcmk__same_node(current, rsc->private->assigned_node) && (rsc->next_role >= pcmk_role_started)) { pcmk__rsc_trace(rsc, "Moving %s from %s to %s", rsc->id, pcmk__node_name(current), - pcmk__node_name(rsc->allocated_to)); + pcmk__node_name(rsc->private->assigned_node)); is_moving = true; allow_migrate = pcmk__rsc_can_migrate(rsc, current); // This is needed even if migrating (though I'm not sure why ...) need_stop = true; } // Check whether resource is partially migrated and/or multiply active if ((rsc->partial_migration_source != NULL) && (rsc->partial_migration_target != NULL) && allow_migrate && (num_all_active == 2) && pcmk__same_node(current, rsc->partial_migration_source) - && pcmk__same_node(rsc->allocated_to, rsc->partial_migration_target)) { + && pcmk__same_node(rsc->private->assigned_node, + rsc->partial_migration_target)) { /* A partial migration is in progress, and the migration target remains * the same as when the migration began. */ pcmk__rsc_trace(rsc, "Partial migration of %s from %s to %s will continue", rsc->id, pcmk__node_name(rsc->partial_migration_source), pcmk__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, pcmk__node_name(rsc->partial_migration_source), pcmk__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, pcmk__node_name(rsc->partial_migration_source), pcmk__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, pcmk__rsc_needs_fencing)) { multiply_active = (num_all_active > 1); } else { /* If a resource has PCMK_META_REQUIRES set to PCMK_VALUE_NOTHING or * PCMK_VALUE_QUORUM, don't consider it active on unclean nodes (similar * to how all resources behave when PCMK_OPT_STONITH_ENABLED is false). * We can start such resources elsewhere before fencing completes, and * if we considered the resource active on the failed node, we would * attempt recovery for being active on multiple nodes. */ multiply_active = (num_clean_active > 1); } if (multiply_active) { const char *class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); // Resource was (possibly) incorrectly multiply active pcmk__sched_err("%s resource %s might be active on %u nodes (%s)", pcmk__s(class, "Untyped"), rsc->id, num_all_active, pcmk__multiply_active_text(rsc)); crm_notice("For more information, see \"What are multiply active " "resources?\" at " "https://projects.clusterlabs.org/w/clusterlabs/faq/"); switch (rsc->private->multiply_active_policy) { case pcmk__multiply_active_restart: need_stop = true; break; case pcmk__multiply_active_unexpected: need_stop = true; // stop_resource() will skip expected node pcmk__set_rsc_flags(rsc, pcmk__rsc_stop_unexpected); break; default: break; } } else { pcmk__clear_rsc_flags(rsc, pcmk__rsc_stop_unexpected); } if (pcmk_is_set(rsc->flags, pcmk__rsc_start_pending)) { create_pending_start(rsc); } if (is_moving) { // Remaining tests are only for resources staying where they are } else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { if (pcmk_is_set(rsc->flags, pcmk__rsc_stop_if_failed)) { need_stop = true; pcmk__rsc_trace(rsc, "Recovering %s", rsc->id); } else { pcmk__rsc_trace(rsc, "Recovering %s by demotion", rsc->id); if (rsc->next_role == pcmk_role_promoted) { need_promote = true; } } } else if (pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) { pcmk__rsc_trace(rsc, "Blocking further actions on %s", rsc->id); need_stop = true; } else if ((rsc->role > pcmk_role_started) && (current != NULL) - && (rsc->allocated_to != NULL)) { + && (rsc->private->assigned_node != NULL)) { pcmk_action_t *start = NULL; pcmk__rsc_trace(rsc, "Creating start action for promoted resource %s", rsc->id); - start = start_action(rsc, rsc->allocated_to, TRUE); + start = start_action(rsc, rsc->private->assigned_node, TRUE); if (!pcmk_is_set(start->flags, pcmk_action_optional)) { // Recovery of a promoted resource pcmk__rsc_trace(rsc, "%s restart is required for recovery", rsc->id); need_stop = true; } } // Create any actions needed to bring resource down and back up to same role schedule_restart_actions(rsc, current, need_stop, need_promote); // Create any actions needed to take resource from this role to the next schedule_role_transition_actions(rsc); pcmk__create_recurring_actions(rsc); if (allow_migrate) { pcmk__create_migration_actions(rsc, current); } } /*! * \internal * \brief Ban a resource from any allowed nodes that are Pacemaker Remote nodes * * \param[in] rsc Resource to check */ static void rsc_avoids_remote_nodes(const pcmk_resource_t *rsc) { GHashTableIter iter; pcmk_node_t *node = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (node->details->remote_rsc != NULL) { node->weight = -PCMK_SCORE_INFINITY; } } } /*! * \internal * \brief Return allowed nodes as (possibly sorted) list * * Convert a resource's hash table of allowed nodes to a list. If printing to * stdout, sort the list, to keep action ID numbers consistent for regression * test output (while avoiding the performance hit on a live cluster). * * \param[in] rsc Resource to check for allowed nodes * * \return List of resource's allowed nodes * \note Callers should take care not to rely on the list being sorted. */ static GList * allowed_nodes_as_list(const pcmk_resource_t *rsc) { GList *allowed_nodes = NULL; if (rsc->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(pcmk_resource_t *rsc) { GList *allowed_nodes = NULL; bool check_unfencing = false; bool check_utilization = false; pcmk_scheduler_t *scheduler = NULL; CRM_ASSERT(pcmk__is_primitive(rsc)); scheduler = rsc->private->scheduler; if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__rsc_trace(rsc, "Skipping implicit constraints for unmanaged resource " "%s", rsc->id); return; } // Whether resource requires unfencing check_unfencing = !pcmk_is_set(rsc->flags, pcmk__rsc_fence_device) && pcmk_is_set(scheduler->flags, pcmk_sched_enable_unfencing) && pcmk_is_set(rsc->flags, pcmk__rsc_needs_unfencing); // Whether a non-default placement strategy is used check_utilization = (g_hash_table_size(rsc->utilization) > 0) && !pcmk__str_eq(scheduler->placement_strategy, PCMK_VALUE_DEFAULT, pcmk__str_casei); // Order stops before starts (i.e. restart) pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered |pcmk__ar_first_implies_then |pcmk__ar_intermediate_stop, scheduler); // Promotable ordering: demote before stop, start before promote if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_promotable) || (rsc->role > pcmk_role_unpromoted)) { pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_DEMOTE, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, pcmk__ar_promoted_then_implies_first, scheduler); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_PROMOTE, 0), NULL, pcmk__ar_unrunnable_first_blocks, scheduler); } // Don't clear resource history if probing on same node pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0), NULL, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0), NULL, pcmk__ar_if_on_same_node|pcmk__ar_then_cancels_first, scheduler); // Certain checks need allowed nodes if (check_unfencing || check_utilization || (rsc->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) { pcmk_resource_t *remote_rsc = NULL; if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { // rsc is the implicit remote connection for a guest or bundle node /* Guest resources are not allowed to run on Pacemaker Remote nodes, * to avoid nesting remotes. However, bundles are allowed. */ if (!pcmk_is_set(rsc->flags, pcmk__rsc_remote_nesting_allowed)) { rsc_avoids_remote_nodes(rsc->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, pcmk__ar_ordered); /* 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 (pcmk_is_set(rsc->container->flags, pcmk__rsc_is_remote_connection)) { remote_rsc = rsc->container; } else { remote_rsc = pe__resource_contains_guest_node(scheduler, 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) { pcmk_node_t *node = item->data; if (node->details->remote_rsc != remote_rsc) { node->weight = -PCMK_SCORE_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, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks, scheduler); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL, rsc->container, pcmk__op_key(rsc->container->id, PCMK_ACTION_STOP, 0), NULL, pcmk__ar_then_implies_first, scheduler); if (pcmk_is_set(rsc->flags, pcmk__rsc_remote_nesting_allowed)) { score = 10000; /* Highly preferred but not essential */ } else { score = PCMK_SCORE_INFINITY; // Force to run on same host } pcmk__new_colocation("#resource-with-container", NULL, score, rsc, rsc->container, NULL, NULL, pcmk__coloc_influence); } } if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection) || pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) { /* Remote connections and fencing devices are not allowed to run on * Pacemaker Remote nodes */ rsc_avoids_remote_nodes(rsc); } g_list_free(allowed_nodes); } /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__primitive_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { enum pcmk__coloc_affects filter_results; CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if (for_dependent) { // Always process on behalf of primary resource primary->private->cmds->apply_coloc_score(dependent, primary, colocation, false); return; } filter_results = pcmk__colocation_affects(dependent, primary, colocation, false); pcmk__rsc_trace(dependent, "%s %s with %s (%s, score=%d, filter=%d)", ((colocation->score > 0)? "Colocating" : "Anti-colocating"), dependent->id, primary->id, colocation->id, colocation->score, filter_results); switch (filter_results) { case pcmk__coloc_affects_role: 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 * pcmk__assignment_methods_t:with_this_colocations() */ void pcmk__with_primitive_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *parent = NULL; CRM_ASSERT(pcmk__is_primitive(rsc) && (list != NULL)); parent = rsc->private->parent; if (rsc == orig_rsc) { /* For the resource itself, add all of its own colocations and relevant * colocations from its parent (if any). */ pcmk__add_with_this_list(list, rsc->private->with_this_colocations, orig_rsc); if (parent != NULL) { parent->private->cmds->with_this_colocations(parent, orig_rsc, list); } } else { // For an ancestor, add only explicitly configured constraints for (GList *iter = rsc->private->with_this_colocations; iter != NULL; iter = iter->next) { pcmk__colocation_t *colocation = iter->data; if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) { pcmk__add_with_this(list, colocation, orig_rsc); } } } } /* Primitive implementation of * pcmk__assignment_methods_t:this_with_colocations() */ void pcmk__primitive_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { const pcmk_resource_t *parent = NULL; CRM_ASSERT(pcmk__is_primitive(rsc) && (list != NULL)); parent = rsc->private->parent; if (rsc == orig_rsc) { /* For the resource itself, add all of its own colocations and relevant * colocations from its parent (if any). */ pcmk__add_this_with_list(list, rsc->private->this_with_colocations, orig_rsc); if (parent != NULL) { parent->private->cmds->this_with_colocations(parent, orig_rsc, list); } } else { // For an ancestor, add only explicitly configured constraints for (GList *iter = rsc->private->this_with_colocations; iter != NULL; iter = iter->next) { pcmk__colocation_t *colocation = iter->data; if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) { pcmk__add_this_with(list, colocation, orig_rsc); } } } } /*! * \internal * \brief Return action flags for a given primitive resource action * * \param[in,out] action Action to get flags for * \param[in] node If not NULL, limit effects to this node (ignored) * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__primitive_action_flags(pcmk_action_t *action, const pcmk_node_t *node) { 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 \c true if \p rsc is multiply active with * \c PCMK_META_MULTIPLE_ACTIVE set to \c PCMK_VALUE_STOP_UNEXPECTED, * and \p node is the node where it will remain active * \note This assumes that the resource's next role cannot be changed to stopped * after this is called, which should be reasonable if status has already * been unpacked and resources have been assigned to nodes. */ static bool is_expected_node(const pcmk_resource_t *rsc, const pcmk_node_t *node) { return pcmk_all_flags_set(rsc->flags, pcmk__rsc_stop_unexpected|pcmk__rsc_restarting) && (rsc->next_role > pcmk_role_stopped) - && pcmk__same_node(rsc->allocated_to, node); + && pcmk__same_node(rsc->private->assigned_node, node); } /*! * \internal * \brief Schedule actions needed to stop a resource wherever it is active * * \param[in,out] rsc Resource being stopped * \param[in] node Node where resource is being stopped (ignored) * \param[in] optional Whether actions should be optional */ static void stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *current = (pcmk_node_t *) iter->data; pcmk_action_t *stop = NULL; if (is_expected_node(rsc, current)) { /* We are scheduling restart actions for a multiply active resource * with PCMK_META_MULTIPLE_ACTIVE=PCMK_VALUE_STOP_UNEXPECTED, and * this is where it should not be stopped. */ pcmk__rsc_trace(rsc, "Skipping stop of multiply active resource %s " "on expected node %s", rsc->id, pcmk__node_name(current)); continue; } if (rsc->partial_migration_target != NULL) { // Continue migration if node originally was and remains target if (pcmk__same_node(current, rsc->partial_migration_target) - && pcmk__same_node(current, rsc->allocated_to)) { + && pcmk__same_node(current, rsc->private->assigned_node)) { pcmk__rsc_trace(rsc, "Skipping stop of %s on %s " "because partial migration there will continue", rsc->id, pcmk__node_name(current)); continue; } else { pcmk__rsc_trace(rsc, "Forcing stop of %s on %s " "because migration target changed", rsc->id, pcmk__node_name(current)); optional = false; } } pcmk__rsc_trace(rsc, "Scheduling stop of %s on %s", rsc->id, pcmk__node_name(current)); stop = stop_action(rsc, current, optional); - if (rsc->allocated_to == NULL) { + if (rsc->private->assigned_node == NULL) { pe_action_set_reason(stop, "node availability", true); } else if (pcmk_all_flags_set(rsc->flags, pcmk__rsc_restarting |pcmk__rsc_stop_unexpected)) { /* We are stopping a multiply active resource on a node that is * not its expected node, and we are still scheduling restart * actions, so the stop is for being multiply active. */ pe_action_set_reason(stop, "being multiply active", true); } if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__clear_action_flags(stop, pcmk_action_runnable); } if (pcmk_is_set(rsc->private->scheduler->flags, pcmk_sched_remove_after_stop)) { pcmk__schedule_cleanup(rsc, current, optional); } if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_unfencing)) { pcmk_action_t *unfence = pe_fence_op(current, PCMK_ACTION_ON, true, NULL, false, rsc->private->scheduler); order_actions(stop, unfence, pcmk__ar_then_implies_first); if (!pcmk__node_unfenced(current)) { pcmk__sched_err("Stopping %s until %s can be unfenced", rsc->id, pcmk__node_name(current)); } } } } /*! * \internal * \brief Schedule actions needed to start a resource on a node * * \param[in,out] rsc Resource being started * \param[in,out] node Node where resource should be started * \param[in] optional Whether actions should be optional */ static void start_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { pcmk_action_t *start = NULL; CRM_ASSERT(node != NULL); pcmk__rsc_trace(rsc, "Scheduling %s start of %s on %s (score %d)", (optional? "optional" : "required"), rsc->id, pcmk__node_name(node), node->weight); start = start_action(rsc, node, TRUE); pcmk__order_vs_unfence(rsc, node, start, pcmk__ar_first_implies_then); if (pcmk_is_set(start->flags, pcmk_action_runnable) && !optional) { pcmk__clear_action_flags(start, pcmk_action_optional); } if (is_expected_node(rsc, node)) { /* This could be a problem if the start becomes necessary for other * reasons later. */ pcmk__rsc_trace(rsc, "Start of multiply active resouce %s " "on expected node %s will be a pseudo-action", rsc->id, pcmk__node_name(node)); pcmk__set_action_flags(start, pcmk_action_pseudo); } } /*! * \internal * \brief Schedule actions needed to promote a resource on a node * * \param[in,out] rsc Resource being promoted * \param[in] node Node where resource should be promoted * \param[in] optional Whether actions should be optional */ static void promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { GList *iter = NULL; GList *action_list = NULL; bool runnable = true; 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) { pcmk_action_t *start = (pcmk_action_t *) iter->data; if (!pcmk_is_set(start->flags, pcmk_action_runnable)) { runnable = false; } } g_list_free(action_list); if (runnable) { pcmk_action_t *promote = promote_action(rsc, node, optional); pcmk__rsc_trace(rsc, "Scheduling %s promotion of %s on %s", (optional? "optional" : "required"), rsc->id, pcmk__node_name(node)); if (is_expected_node(rsc, node)) { /* This could be a problem if the promote becomes necessary for * other reasons later. */ pcmk__rsc_trace(rsc, "Promotion of multiply active resouce %s " "on expected node %s will be a pseudo-action", rsc->id, pcmk__node_name(node)); pcmk__set_action_flags(promote, pcmk_action_pseudo); } } else { pcmk__rsc_trace(rsc, "Not promoting %s on %s: start unrunnable", rsc->id, pcmk__node_name(node)); action_list = pe__resource_actions(rsc, node, PCMK_ACTION_PROMOTE, true); for (iter = action_list; iter != NULL; iter = iter->next) { pcmk_action_t *promote = (pcmk_action_t *) iter->data; pcmk__clear_action_flags(promote, pcmk_action_runnable); } g_list_free(action_list); } } /*! * \internal * \brief Schedule actions needed to demote a resource wherever it is active * * \param[in,out] rsc Resource being demoted * \param[in] node Node where resource should be demoted (ignored) * \param[in] optional Whether actions should be optional */ static void demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { /* Since this will only be called for a primitive (possibly as an instance * of a collective resource), the resource is multiply active if it is * running on more than one node, so we want to demote on all of them as * part of recovery, regardless of which one is the desired node. */ for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *current = (pcmk_node_t *) iter->data; if (is_expected_node(rsc, current)) { pcmk__rsc_trace(rsc, "Skipping demote of multiply active resource %s " "on expected node %s", rsc->id, pcmk__node_name(current)); } else { pcmk__rsc_trace(rsc, "Scheduling %s demotion of %s on %s", (optional? "optional" : "required"), rsc->id, pcmk__node_name(current)); demote_action(rsc, current, optional); } } } static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { 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(pcmk_resource_t *rsc, const pcmk_node_t *node, bool optional) { /* If the cleanup is required, its orderings are optional, because they're * relevant only if both actions are required. Conversely, if the cleanup is * optional, the orderings make the then action required if the first action * becomes required. */ uint32_t flag = optional? pcmk__ar_first_implies_then : pcmk__ar_ordered; CRM_CHECK((rsc != NULL) && (node != NULL), return); if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: resource failed", rsc->id, pcmk__node_name(node)); return; } if (node->details->unclean || !node->details->online) { pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: node unavailable", rsc->id, pcmk__node_name(node)); return; } crm_notice("Scheduling clean-up of %s on %s", rsc->id, pcmk__node_name(node)); delete_action(rsc, node, optional); // stop -> clean-up -> start pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP, rsc, PCMK_ACTION_DELETE, flag); pcmk__order_resource_actions(rsc, PCMK_ACTION_DELETE, rsc, PCMK_ACTION_START, flag); } /*! * \internal * \brief Add primitive meta-attributes relevant to graph actions to XML * * \param[in] rsc Primitive resource whose meta-attributes should be added * \param[in,out] xml Transition graph action attributes XML to add to */ void pcmk__primitive_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml) { char *name = NULL; char *value = NULL; const pcmk_resource_t *parent = NULL; CRM_ASSERT(pcmk__is_primitive(rsc) && (xml != NULL)); /* Clone instance numbers get set internally as meta-attributes, and are * needed in the transition graph (for example, to tell unique clone * instances apart). */ value = g_hash_table_lookup(rsc->meta, PCMK__META_CLONE); if (value != NULL) { name = crm_meta_name(PCMK__META_CLONE); crm_xml_add(xml, name, value); free(name); } // Not sure if this one is really needed ... value = g_hash_table_lookup(rsc->meta, PCMK_META_REMOTE_NODE); if (value != NULL) { name = crm_meta_name(PCMK_META_REMOTE_NODE); crm_xml_add(xml, name, value); free(name); } /* The 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->private->parent) { if (parent->container != NULL) { crm_xml_add(xml, CRM_META "_" PCMK__META_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 pcmk__assignment_methods_t:add_utilization() void pcmk__primitive_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { CRM_ASSERT(pcmk__is_primitive(rsc) && (orig_rsc != NULL) && (utilization != NULL)); if (!pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) { return; } pcmk__rsc_trace(orig_rsc, "%s: Adding primitive %s as colocated utilization", orig_rsc->id, rsc->id); pcmk__release_node_capacity(utilization, rsc); } /*! * \internal * \brief Get epoch time of node's shutdown attribute (or now if none) * * \param[in,out] node Node to check * * \return Epoch time corresponding to shutdown attribute if set or now if not */ static time_t shutdown_time(pcmk_node_t *node) { const char *shutdown = pcmk__node_attr(node, PCMK__NODE_ATTR_SHUTDOWN, NULL, pcmk__rsc_node_current); time_t result = 0; if (shutdown != NULL) { long long result_ll; 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 pcmk_node_t *node = (const pcmk_node_t *) data; pcmk_resource_t *rsc = (pcmk_resource_t *) user_data; if (strcmp(node->details->uname, rsc->lock_node->details->uname) != 0) { resource_location(rsc, node, -PCMK_SCORE_INFINITY, PCMK_OPT_SHUTDOWN_LOCK, rsc->private->scheduler); } } // Primitive implementation of pcmk__assignment_methods_t:shutdown_lock() void pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc) { const char *class = NULL; pcmk_scheduler_t *scheduler = NULL; CRM_ASSERT(pcmk__is_primitive(rsc)); scheduler = rsc->private->scheduler; class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); // Fence devices and remote connections can't be locked if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches) || pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { 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. */ pcmk__rsc_info(rsc, "Cancelling shutdown lock " "because %s is already active", rsc->id); pe__clear_resource_history(rsc, rsc->lock_node); 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)) { pcmk_node_t *node = rsc->running_on->data; if (node->details->shutdown) { if (node->details->unclean) { pcmk__rsc_debug(rsc, "Not locking %s to unclean %s for shutdown", rsc->id, pcmk__node_name(node)); } else { rsc->lock_node = node; rsc->lock_time = shutdown_time(node); } } } if (rsc->lock_node == NULL) { // No lock needed return; } if (scheduler->shutdown_lock > 0) { time_t lock_expiration = rsc->lock_time + scheduler->shutdown_lock; pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)", rsc->id, pcmk__node_name(rsc->lock_node), (long long) lock_expiration); pe__update_recheck_time(++lock_expiration, scheduler, "shutdown lock expiration"); } else { pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown", rsc->id, pcmk__node_name(rsc->lock_node)); } // If resource is locked to one node, ban it from all other nodes g_list_foreach(scheduler->nodes, ban_if_not_locked, rsc); } diff --git a/lib/pacemaker/pcmk_sched_probes.c b/lib/pacemaker/pcmk_sched_probes.c index f0e2f6f9b3..d9d7f71622 100644 --- a/lib/pacemaker/pcmk_sched_probes.c +++ b/lib/pacemaker/pcmk_sched_probes.c @@ -1,912 +1,914 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Add the expected result to a newly created probe * * \param[in,out] probe Probe action to add expected result to * \param[in] rsc Resource that probe is for * \param[in] node Node that probe will run on */ static void add_expected_result(pcmk_action_t *probe, const pcmk_resource_t *rsc, const pcmk_node_t *node) { // Check whether resource is currently active on node pcmk_node_t *running = pe_find_node_id(rsc->running_on, node->details->id); // The expected result is what we think the resource's current state is if (running == NULL) { pe__add_action_expected_result(probe, CRM_EX_NOT_RUNNING); } else if (rsc->role == pcmk_role_promoted) { pe__add_action_expected_result(probe, CRM_EX_PROMOTED); } } /*! * \internal * \brief Create any needed robes on a node for a list of resources * * \param[in,out] rscs List of resources to create probes for * \param[in,out] node Node to create probes on * * \return true if any probe was created, otherwise false */ bool pcmk__probe_resource_list(GList *rscs, pcmk_node_t *node) { bool any_created = false; for (GList *iter = rscs; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; if (rsc->private->cmds->create_probe(rsc, node)) { any_created = true; } } return any_created; } /*! * \internal * \brief Order one resource's start after another's start-up probe * * \param[in,out] rsc1 Resource that might get start-up probe * \param[in] rsc2 Resource that might be started */ static void probe_then_start(pcmk_resource_t *rsc1, pcmk_resource_t *rsc2) { - if ((rsc1->allocated_to != NULL) + const pcmk_node_t *rsc1_node = rsc1->private->assigned_node; + + if ((rsc1_node != NULL) && (g_hash_table_lookup(rsc1->known_on, - rsc1->allocated_to->details->id) == NULL)) { + rsc1_node->details->id) == NULL)) { pcmk__new_ordering(rsc1, pcmk__op_key(rsc1->id, PCMK_ACTION_MONITOR, 0), NULL, rsc2, pcmk__op_key(rsc2->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered, rsc1->private->scheduler); } } /*! * \internal * \brief Check whether a guest resource will stop * * \param[in] node Guest node to check * * \return true if guest resource will likely stop, otherwise false */ static bool guest_resource_will_stop(const pcmk_node_t *node) { const pcmk_resource_t *guest_rsc = node->details->remote_rsc->container; + const pcmk_node_t *guest_node = guest_rsc->private->assigned_node; /* Ideally, we'd check whether the guest has a required stop, but that * information doesn't exist yet, so approximate it ... */ return node->details->remote_requires_reset || node->details->unclean || pcmk_is_set(guest_rsc->flags, pcmk__rsc_failed) || (guest_rsc->next_role == pcmk_role_stopped) // Guest is moving - || ((guest_rsc->role > pcmk_role_stopped) - && (guest_rsc->allocated_to != NULL) - && (pcmk__find_node_in_list(guest_rsc->running_on, - guest_rsc->allocated_to->details->uname) == NULL)); + || ((guest_rsc->role > pcmk_role_stopped) && (guest_node != NULL) + && pcmk__find_node_in_list(guest_rsc->running_on, + guest_node->details->uname) == NULL); } /*! * \internal * \brief Create a probe action 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 Newly created probe action */ static pcmk_action_t * probe_action(pcmk_resource_t *rsc, pcmk_node_t *node) { pcmk_action_t *probe = NULL; char *key = pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0); crm_debug("Scheduling probe of %s %s on %s", pcmk_role_text(rsc->role), rsc->id, pcmk__node_name(node)); probe = custom_action(rsc, key, PCMK_ACTION_MONITOR, node, FALSE, rsc->private->scheduler); pcmk__clear_action_flags(probe, pcmk_action_optional); pcmk__order_vs_unfence(rsc, node, probe, pcmk__ar_ordered); add_expected_result(probe, rsc, node); return probe; } /*! * \internal * \brief Create probes for a resource on a node, if needed * * \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__probe_rsc_on_node(pcmk_resource_t *rsc, pcmk_node_t *node) { uint32_t flags = pcmk__ar_ordered; pcmk_action_t *probe = NULL; pcmk_node_t *allowed = NULL; pcmk_resource_t *top = uber_parent(rsc); const char *reason = NULL; CRM_ASSERT((rsc != NULL) && (node != NULL)); if (!pcmk_is_set(rsc->private->scheduler->flags, pcmk_sched_probe_resources)) { reason = "start-up probes are disabled"; goto no_probe; } if (pcmk__is_pacemaker_remote_node(node)) { const char *class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_none)) { reason = "Pacemaker Remote nodes cannot run stonith agents"; goto no_probe; } else if (pcmk__is_guest_or_bundle_node(node) && pe__resource_contains_guest_node(rsc->private->scheduler, rsc)) { reason = "guest nodes cannot run resources containing guest nodes"; goto no_probe; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { reason = "Pacemaker Remote nodes cannot host remote connections"; goto no_probe; } } // If this is a collective resource, probes are created for its children if (rsc->children != NULL) { return pcmk__probe_resource_list(rsc->children, node); } if ((rsc->container != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { reason = "resource is inside a container"; goto no_probe; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { reason = "resource is orphaned"; goto no_probe; } else if (g_hash_table_lookup(rsc->known_on, node->details->id) != NULL) { reason = "resource state is already known"; goto no_probe; } allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (pcmk_is_set(rsc->flags, pcmk__rsc_exclusive_probes) || pcmk_is_set(top->flags, pcmk__rsc_exclusive_probes)) { // Exclusive discovery is enabled ... if (allowed == NULL) { // ... but this node is not allowed to run the resource reason = "resource has exclusive discovery but is not allowed " "on node"; goto no_probe; } else if (allowed->rsc_discover_mode != pcmk_probe_exclusive) { // ... but no constraint marks this node for discovery of resource reason = "resource has exclusive discovery but is not enabled " "on node"; goto no_probe; } } if (allowed == NULL) { allowed = node; } if (allowed->rsc_discover_mode == pcmk_probe_never) { reason = "node has discovery disabled"; goto no_probe; } if (pcmk__is_guest_or_bundle_node(node)) { pcmk_resource_t *guest = node->details->remote_rsc->container; if (guest->role == pcmk_role_stopped) { // The guest is stopped, so we know no resource is active there reason = "node's guest is stopped"; probe_then_start(guest, top); goto no_probe; } else if (guest_resource_will_stop(node)) { reason = "node's guest will stop"; // Order resource start after guest stop (in case it's restarting) pcmk__new_ordering(guest, pcmk__op_key(guest->id, PCMK_ACTION_STOP, 0), NULL, top, pcmk__op_key(top->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered, rsc->private->scheduler); goto no_probe; } } // We've eliminated all cases where a probe is not needed, so now it is probe = probe_action(rsc, node); /* Below, we will order the probe relative to start or reload. If this is a * clone instance, the start or reload is for the entire clone rather than * just the instance. Otherwise, the start or reload is for the resource * itself. */ if (!pcmk__is_clone(top)) { top = rsc; } /* Prevent a start if the resource can't be probed, but don't cause the * resource or entire clone to stop if already active. */ if (!pcmk_is_set(probe->flags, pcmk_action_runnable) && (top->running_on == NULL)) { pcmk__set_relation_flags(flags, pcmk__ar_unrunnable_first_blocks); } // Start or reload after probing the resource pcmk__new_ordering(rsc, NULL, probe, top, pcmk__op_key(top->id, PCMK_ACTION_START, 0), NULL, flags, rsc->private->scheduler); pcmk__new_ordering(rsc, NULL, probe, top, reload_key(rsc), NULL, pcmk__ar_ordered, rsc->private->scheduler); return true; no_probe: pcmk__rsc_trace(rsc, "Skipping probe for %s on %s because %s", rsc->id, node->details->id, reason); return false; } /*! * \internal * \brief Check whether a probe should be ordered before another action * * \param[in] probe Probe action to check * \param[in] then Other action to check * * \return true if \p probe should be ordered before \p then, otherwise false */ static bool probe_needed_before_action(const pcmk_action_t *probe, const pcmk_action_t *then) { // Probes on a node are performed after unfencing it, not before if (pcmk__str_eq(then->task, PCMK_ACTION_STONITH, pcmk__str_none) && pcmk__same_node(probe->node, then->node)) { const char *op = g_hash_table_lookup(then->meta, PCMK__META_STONITH_ACTION); if (pcmk__str_eq(op, PCMK_ACTION_ON, pcmk__str_casei)) { return false; } } // Probes should be done on a node before shutting it down if (pcmk__str_eq(then->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none) && (probe->node != NULL) && (then->node != NULL) && !pcmk__same_node(probe->node, then->node)) { return false; } // Otherwise probes should always be done before any other action return true; } /*! * \internal * \brief Add implicit "probe then X" orderings for "stop then X" orderings * * If the state of a resource is not known yet, a probe will be scheduled, * expecting a "not running" result. If the probe fails, a stop will not be * scheduled until the next transition. Thus, if there are ordering constraints * like "stop this resource then do something else that's not for the same * resource", add implicit "probe this resource then do something" equivalents * so the relation is upheld until we know whether a stop is needed. * * \param[in,out] scheduler Scheduler data */ static void add_probe_orderings_for_stops(pcmk_scheduler_t *scheduler) { for (GList *iter = scheduler->ordering_constraints; iter != NULL; iter = iter->next) { pcmk__action_relation_t *order = iter->data; uint32_t order_flags = pcmk__ar_ordered; GList *probes = NULL; GList *then_actions = NULL; pcmk_action_t *first = NULL; pcmk_action_t *then = NULL; // Skip disabled orderings if (order->flags == pcmk__ar_none) { continue; } // Skip non-resource orderings, and orderings for the same resource if ((order->rsc1 == NULL) || (order->rsc1 == order->rsc2)) { continue; } // Skip invalid orderings (shouldn't be possible) first = order->action1; then = order->action2; if (((first == NULL) && (order->task1 == NULL)) || ((then == NULL) && (order->task2 == NULL))) { continue; } // Skip orderings for first actions other than stop if ((first != NULL) && !pcmk__str_eq(first->task, PCMK_ACTION_STOP, pcmk__str_none)) { continue; } else if ((first == NULL) && !pcmk__ends_with(order->task1, "_" PCMK_ACTION_STOP "_0")) { continue; } /* Do not imply a probe ordering for a resource inside of a stopping * container. Otherwise, it might introduce a transition loop, since a * probe could be scheduled after the container starts again. */ if ((order->rsc2 != NULL) && (order->rsc1->container == order->rsc2)) { if ((then != NULL) && pcmk__str_eq(then->task, PCMK_ACTION_STOP, pcmk__str_none)) { continue; } else if ((then == NULL) && pcmk__ends_with(order->task2, "_" PCMK_ACTION_STOP "_0")) { continue; } } // Preserve certain order options for future filtering if (pcmk_is_set(order->flags, pcmk__ar_if_first_unmigratable)) { pcmk__set_relation_flags(order_flags, pcmk__ar_if_first_unmigratable); } if (pcmk_is_set(order->flags, pcmk__ar_if_on_same_node)) { pcmk__set_relation_flags(order_flags, pcmk__ar_if_on_same_node); } // Preserve certain order types for future filtering if ((order->flags == pcmk__ar_if_required_on_same_node) || (order->flags == pcmk__ar_if_on_same_node_or_target)) { order_flags = order->flags; } // List all scheduled probes for the first resource probes = pe__resource_actions(order->rsc1, NULL, PCMK_ACTION_MONITOR, FALSE); if (probes == NULL) { // There aren't any continue; } // List all relevant "then" actions if (then != NULL) { then_actions = g_list_prepend(NULL, then); } else if (order->rsc2 != NULL) { then_actions = find_actions(order->rsc2->private->actions, order->task2, NULL); if (then_actions == NULL) { // There aren't any g_list_free(probes); continue; } } crm_trace("Implying 'probe then' orderings for '%s then %s' " "(id=%d, type=%.6x)", ((first == NULL)? order->task1 : first->uuid), ((then == NULL)? order->task2 : then->uuid), order->id, order->flags); for (GList *probe_iter = probes; probe_iter != NULL; probe_iter = probe_iter->next) { pcmk_action_t *probe = (pcmk_action_t *) probe_iter->data; for (GList *then_iter = then_actions; then_iter != NULL; then_iter = then_iter->next) { pcmk_action_t *then = (pcmk_action_t *) then_iter->data; if (probe_needed_before_action(probe, then)) { order_actions(probe, then, order_flags); } } } g_list_free(then_actions); g_list_free(probes); } } /*! * \internal * \brief Add necessary orderings between probe and starts of clone instances * * , in additon to the ordering with the parent resource added upon creating * the probe. * * \param[in,out] probe Probe as 'first' action in an ordering * \param[in,out] after 'then' action wrapper in the ordering */ static void add_start_orderings_for_probe(pcmk_action_t *probe, pcmk__related_action_t *after) { uint32_t flags = pcmk__ar_ordered|pcmk__ar_unrunnable_first_blocks; /* Although the ordering between the probe of the clone instance and the * start of its parent has been added in pcmk__probe_rsc_on_node(), we * avoided enforcing `pcmk__ar_unrunnable_first_blocks` order type for that * as long as any of the clone instances are running to prevent them from * being unexpectedly stopped. * * On the other hand, we still need to prevent any inactive instances from * starting unless the probe is runnable so that we don't risk starting too * many instances before we know the state on all nodes. */ if ((after->action->rsc->private->variant <= pcmk__rsc_variant_group) || pcmk_is_set(probe->flags, pcmk_action_runnable) // The order type is already enforced for its parent. || pcmk_is_set(after->type, pcmk__ar_unrunnable_first_blocks) || (pe__const_top_resource(probe->rsc, false) != after->action->rsc) || !pcmk__str_eq(after->action->task, PCMK_ACTION_START, pcmk__str_none)) { return; } crm_trace("Adding probe start orderings for 'unrunnable %s@%s " "then instances of %s@%s'", probe->uuid, pcmk__node_name(probe->node), after->action->uuid, pcmk__node_name(after->action->node)); for (GList *then_iter = after->action->actions_after; then_iter != NULL; then_iter = then_iter->next) { pcmk__related_action_t *then = then_iter->data; if (then->action->rsc->running_on || (pe__const_top_resource(then->action->rsc, false) != after->action->rsc) || !pcmk__str_eq(then->action->task, PCMK_ACTION_START, pcmk__str_none)) { continue; } crm_trace("Adding probe start ordering for 'unrunnable %s@%s " "then %s@%s' (type=%#.6x)", probe->uuid, pcmk__node_name(probe->node), then->action->uuid, pcmk__node_name(then->action->node), flags); /* Prevent the instance from starting if the instance can't, but don't * cause any other intances to stop if already active. */ order_actions(probe, then->action, flags); } return; } /*! * \internal * \brief Order probes before restarts and re-promotes * * If a given ordering is a "probe then start" or "probe then promote" ordering, * add an implicit "probe then stop/demote" ordering in case the action is part * of a restart/re-promote, and do the same recursively for all actions ordered * after the "then" action. * * \param[in,out] probe Probe as 'first' action in an ordering * \param[in,out] after 'then' action in the ordering */ static void add_restart_orderings_for_probe(pcmk_action_t *probe, pcmk_action_t *after) { GList *iter = NULL; bool interleave = false; pcmk_resource_t *compatible_rsc = NULL; // Validate that this is a resource probe followed by some action if ((after == NULL) || (probe == NULL) || !pcmk__is_primitive(probe->rsc) || !pcmk__str_eq(probe->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { return; } // Avoid running into any possible loop if (pcmk_is_set(after->flags, pcmk_action_detect_loop)) { return; } pcmk__set_action_flags(after, pcmk_action_detect_loop); crm_trace("Adding probe restart orderings for '%s@%s then %s@%s'", probe->uuid, pcmk__node_name(probe->node), after->uuid, pcmk__node_name(after->node)); /* Add restart orderings if "then" is for a different primitive. * Orderings for collective resources will be added later. */ if (pcmk__is_primitive(after->rsc) && (probe->rsc != after->rsc)) { GList *then_actions = NULL; if (pcmk__str_eq(after->task, PCMK_ACTION_START, pcmk__str_none)) { then_actions = pe__resource_actions(after->rsc, NULL, PCMK_ACTION_STOP, FALSE); } else if (pcmk__str_eq(after->task, PCMK_ACTION_PROMOTE, pcmk__str_none)) { then_actions = pe__resource_actions(after->rsc, NULL, PCMK_ACTION_DEMOTE, FALSE); } for (iter = then_actions; iter != NULL; iter = iter->next) { pcmk_action_t *then = (pcmk_action_t *) iter->data; // Skip pseudo-actions (for example, those implied by fencing) if (!pcmk_is_set(then->flags, pcmk_action_pseudo)) { order_actions(probe, then, pcmk__ar_ordered); } } g_list_free(then_actions); } /* Detect whether "then" is an interleaved clone action. For these, we want * to add orderings only for the relevant instance. */ if ((after->rsc != NULL) && (after->rsc->private->variant > pcmk__rsc_variant_group)) { const char *interleave_s = g_hash_table_lookup(after->rsc->meta, PCMK_META_INTERLEAVE); interleave = crm_is_true(interleave_s); if (interleave) { compatible_rsc = pcmk__find_compatible_instance(probe->rsc, after->rsc, pcmk_role_unknown, false); } } /* Now recursively do the same for all actions ordered after "then". This * also handles collective resources since the collective action will be * ordered before its individual instances' actions. */ for (iter = after->actions_after; iter != NULL; iter = iter->next) { pcmk__related_action_t *after_wrapper = iter->data; const pcmk_resource_t *chained_rsc = NULL; /* pcmk__ar_first_implies_then is the reason why a required A.start * implies/enforces B.start to be required too, which is the cause of * B.restart/re-promote. * * Not sure about pcmk__ar_first_implies_same_node_then though. It's now * only used for unfencing case, which tends to introduce transition * loops... */ if (!pcmk_is_set(after_wrapper->type, pcmk__ar_first_implies_then)) { /* The order type between a group/clone and its child such as * B.start-> B_child.start is: * pcmk__ar_then_implies_first_graphed * |pcmk__ar_unrunnable_first_blocks * * Proceed through the ordering chain and build dependencies with * its children. */ if ((after->rsc == NULL) || (after->rsc->private->variant < pcmk__rsc_variant_group) || (probe->rsc->private->parent == after->rsc) || (after_wrapper->action->rsc == NULL)) { continue; } chained_rsc = after_wrapper->action->rsc; if ((chained_rsc->private->variant > pcmk__rsc_variant_group) || (after->rsc != chained_rsc->private->parent)) { continue; } /* Proceed to the children of a group or a non-interleaved clone. * For an interleaved clone, proceed only to the relevant child. */ if ((after->rsc->private->variant > pcmk__rsc_variant_group) && interleave && ((compatible_rsc == NULL) || (compatible_rsc != chained_rsc))) { continue; } } crm_trace("Recursively adding probe restart orderings for " "'%s@%s then %s@%s' (type=%#.6x)", after->uuid, pcmk__node_name(after->node), after_wrapper->action->uuid, pcmk__node_name(after_wrapper->action->node), after_wrapper->type); add_restart_orderings_for_probe(probe, after_wrapper->action); } } /*! * \internal * \brief Clear the tracking flag on all scheduled actions * * \param[in,out] scheduler Scheduler data */ static void clear_actions_tracking_flag(pcmk_scheduler_t *scheduler) { for (GList *iter = scheduler->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = iter->data; pcmk__clear_action_flags(action, pcmk_action_detect_loop); } } /*! * \internal * \brief Add start and restart orderings for probes scheduled for a resource * * \param[in,out] data Resource whose probes should be ordered * \param[in] user_data Unused */ static void add_start_restart_orderings_for_rsc(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; GList *probes = NULL; // For collective resources, order each instance recursively if (!pcmk__is_primitive(rsc)) { g_list_foreach(rsc->children, add_start_restart_orderings_for_rsc, NULL); return; } // Find all probes for given resource probes = pe__resource_actions(rsc, NULL, PCMK_ACTION_MONITOR, FALSE); // Add probe restart orderings for each probe found for (GList *iter = probes; iter != NULL; iter = iter->next) { pcmk_action_t *probe = (pcmk_action_t *) iter->data; for (GList *then_iter = probe->actions_after; then_iter != NULL; then_iter = then_iter->next) { pcmk__related_action_t *then = then_iter->data; add_start_orderings_for_probe(probe, then); add_restart_orderings_for_probe(probe, then->action); clear_actions_tracking_flag(rsc->private->scheduler); } } g_list_free(probes); } /*! * \internal * \brief Add "A then probe B" orderings for "A then B" orderings * * \param[in,out] scheduler Scheduler data * * \note This function is currently disabled (see next comment). */ static void order_then_probes(pcmk_scheduler_t *scheduler) { #if 0 /* Given an ordering "A then B", we would prefer to wait for A to be started * before probing B. * * For example, if A is a filesystem which B can't even run without, it * would be helpful if the author of B's agent could assume that A is * running before B.monitor will be called. * * However, we can't _only_ probe after A is running, otherwise we wouldn't * detect the state of B if A could not be started. We can't even do an * opportunistic version of this, because B may be moving: * * A.stop -> A.start -> B.probe -> B.stop -> B.start * * and if we add B.stop -> A.stop here, we get a loop: * * A.stop -> A.start -> B.probe -> B.stop -> A.stop * * We could kill the "B.probe -> B.stop" dependency, but that could mean * stopping B "too" soon, because B.start must wait for the probe, and * we don't want to stop B if we can't start it. * * We could add the ordering only if A is an anonymous clone with * clone-max == node-max (since we'll never be moving it). However, we could * still be stopping one instance at the same time as starting another. * * The complexity of checking for allowed conditions combined with the ever * narrowing use case suggests that this code should remain disabled until * someone gets smarter. */ for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; pcmk_action_t *start = NULL; GList *actions = NULL; GList *probes = NULL; actions = pe__resource_actions(rsc, NULL, PCMK_ACTION_START, FALSE); if (actions) { start = actions->data; g_list_free(actions); } if (start == NULL) { crm_debug("No start action for %s", rsc->id); continue; } probes = pe__resource_actions(rsc, NULL, PCMK_ACTION_MONITOR, FALSE); for (actions = start->actions_before; actions != NULL; actions = actions->next) { pcmk__related_action_t *before = actions->data; pcmk_action_t *first = before->action; pcmk_resource_t *first_rsc = first->rsc; if (first->required_runnable_before) { for (GList *clone_actions = first->actions_before; clone_actions != NULL; clone_actions = clone_actions->next) { before = clone_actions->data; crm_trace("Testing '%s then %s' for %s", first->uuid, before->action->uuid, start->uuid); CRM_ASSERT(before->action->rsc != NULL); first_rsc = before->action->rsc; break; } } else if (!pcmk__str_eq(first->task, PCMK_ACTION_START, pcmk__str_none)) { crm_trace("Not a start op %s for %s", first->uuid, start->uuid); } if (first_rsc == NULL) { continue; } else if (pe__const_top_resource(first_rsc, false) == pe__const_top_resource(start->rsc, false)) { crm_trace("Same parent %s for %s", first_rsc->id, start->uuid); continue; } else if (!pcmk__is_clone(pe__const_top_resource(first_rsc, false))) { crm_trace("Not a clone %s for %s", first_rsc->id, start->uuid); continue; } crm_debug("Applying %s before %s", first->uuid, start->uuid); for (GList *probe_iter = probes; probe_iter != NULL; probe_iter = probe_iter->next) { pcmk_action_t *probe = (pcmk_action_t *) probe_iter->data; crm_debug("Ordering %s before %s", first->uuid, probe->uuid); order_actions(first, probe, pcmk__ar_ordered); } } } #endif } void pcmk__order_probes(pcmk_scheduler_t *scheduler) { // Add orderings for "probe then X" g_list_foreach(scheduler->resources, add_start_restart_orderings_for_rsc, NULL); add_probe_orderings_for_stops(scheduler); order_then_probes(scheduler); } /*! * \internal * \brief Schedule any probes needed * * \param[in,out] scheduler Scheduler data * * \note This may also schedule fencing of failed remote nodes. */ void pcmk__schedule_probes(pcmk_scheduler_t *scheduler) { // Schedule probes on each node in the cluster as needed for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; const char *probed = NULL; if (!node->details->online) { // Don't probe offline nodes if (pcmk__is_failed_remote_node(node)) { pe_fence_node(scheduler, node, "the connection is unrecoverable", FALSE); } continue; } else if (node->details->unclean) { // ... or nodes that need fencing continue; } else if (!node->details->rsc_discovery_enabled) { // The user requested that probes not be done on this node continue; } /* This is no longer needed for live clusters, since the probe_complete * node attribute will never be in the CIB. However this is still useful * for processing old saved CIBs (< 1.1.14), including the * reprobe-target_rc regression test. */ probed = pcmk__node_attr(node, CRM_OP_PROBED, NULL, pcmk__rsc_node_current); if (probed != NULL && crm_is_true(probed) == FALSE) { pcmk_action_t *probe_op = NULL; probe_op = custom_action(NULL, crm_strdup_printf("%s-%s", CRM_OP_REPROBE, node->details->uname), CRM_OP_REPROBE, node, FALSE, scheduler); pcmk__insert_meta(probe_op, PCMK__META_OP_NO_WAIT, PCMK_VALUE_TRUE); continue; } // Probe each resource in the cluster on this node, as needed pcmk__probe_resource_list(scheduler->resources, node); } } diff --git a/lib/pacemaker/pcmk_sched_recurring.c b/lib/pacemaker/pcmk_sched_recurring.c index 14db1916c3..33620bba99 100644 --- a/lib/pacemaker/pcmk_sched_recurring.c +++ b/lib/pacemaker/pcmk_sched_recurring.c @@ -1,752 +1,754 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include "libpacemaker_private.h" // Information parsed from an operation history entry in the CIB struct op_history { // XML attributes const char *id; // ID of history entry const char *name; // Action name // Parsed information char *key; // Operation key for action enum rsc_role_e role; // Action role (or pcmk_role_unknown for default) guint interval_ms; // Action interval }; /*! * \internal * \brief Parse an interval from XML * * \param[in] xml XML containing an interval attribute * * \return Interval parsed from XML (or 0 as default) */ static guint xe_interval(const xmlNode *xml) { guint interval_ms = 0U; pcmk_parse_interval_spec(crm_element_value(xml, PCMK_META_INTERVAL), &interval_ms); return interval_ms; } /*! * \internal * \brief Check whether an operation exists multiple times in resource history * * \param[in] rsc Resource with history to search * \param[in] name Name of action to search for * \param[in] interval_ms Interval (in milliseconds) of action to search for * * \return true if an operation with \p name and \p interval_ms exists more than * once in the operation history of \p rsc, otherwise false */ static bool is_op_dup(const pcmk_resource_t *rsc, const char *name, guint interval_ms) { const char *id = NULL; for (xmlNode *op = pcmk__xe_first_child(rsc->private->ops_xml, PCMK_XE_OP, NULL, NULL); op != NULL; op = pcmk__xe_next_same(op)) { // Check whether action name and interval match if (!pcmk__str_eq(crm_element_value(op, PCMK_XA_NAME), name, pcmk__str_none) || (xe_interval(op) != interval_ms)) { continue; } if (pcmk__xe_id(op) == NULL) { continue; // Shouldn't be possible } if (id == NULL) { id = pcmk__xe_id(op); // First matching op } else { pcmk__config_err("Operation %s is duplicate of %s (do not use " "same name and interval combination more " "than once per resource)", pcmk__xe_id(op), id); return true; } } return false; } /*! * \internal * \brief Check whether an action name is one that can be recurring * * \param[in] name Action name to check * * \return true if \p name is an action known to be unsuitable as a recurring * operation, otherwise false * * \note Pacemaker's current philosophy is to allow users to configure recurring * operations except for a short list of actions known not to be suitable * for that (as opposed to allowing only actions known to be suitable, * which includes only monitor). Among other things, this approach allows * users to define their own custom operations and make them recurring, * though that use case is not well tested. */ static bool op_cannot_recur(const char *name) { return pcmk__str_any_of(name, PCMK_ACTION_STOP, PCMK_ACTION_START, PCMK_ACTION_DEMOTE, PCMK_ACTION_PROMOTE, PCMK_ACTION_RELOAD_AGENT, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL); } /*! * \internal * \brief Check whether a resource history entry is for a recurring action * * \param[in] rsc Resource that history entry is for * \param[in] xml XML of resource history entry to check * \param[out] op Where to store parsed info if recurring * * \return true if \p xml is for a recurring action, otherwise false */ static bool is_recurring_history(const pcmk_resource_t *rsc, const xmlNode *xml, struct op_history *op) { const char *role = NULL; op->interval_ms = xe_interval(xml); if (op->interval_ms == 0) { return false; // Not recurring } op->id = pcmk__xe_id(xml); if (pcmk__str_empty(op->id)) { pcmk__config_err("Ignoring resource history entry without ID"); return false; // Shouldn't be possible (unless CIB was manually edited) } op->name = crm_element_value(xml, PCMK_XA_NAME); if (op_cannot_recur(op->name)) { pcmk__config_err("Ignoring %s because %s action cannot be recurring", op->id, pcmk__s(op->name, "unnamed")); return false; } // There should only be one recurring operation per action/interval if (is_op_dup(rsc, op->name, op->interval_ms)) { return false; } // Ensure role is valid if specified role = crm_element_value(xml, PCMK_XA_ROLE); if (role == NULL) { op->role = pcmk_role_unknown; } else { op->role = pcmk_parse_role(role); if (op->role == pcmk_role_unknown) { pcmk__config_err("Ignoring %s role because %s is not a valid role", op->id, role); return false; } } // Only actions that are still configured and enabled matter if (pcmk__find_action_config(rsc, op->name, op->interval_ms, false) == NULL) { pcmk__rsc_trace(rsc, "Ignoring %s (%s-interval %s for %s) because it is " "disabled or no longer in configuration", op->id, pcmk__readable_interval(op->interval_ms), op->name, rsc->id); return false; } op->key = pcmk__op_key(rsc->id, op->name, op->interval_ms); return true; } /*! * \internal * \brief Check whether a recurring action for an active role should be optional * * \param[in] rsc Resource that recurring action is for * \param[in] node Node that \p rsc will be active on (if any) * \param[in] key Operation key for recurring action to check * \param[in,out] start Start action for \p rsc * * \return true if recurring action should be optional, otherwise false */ static bool active_recurring_should_be_optional(const pcmk_resource_t *rsc, const pcmk_node_t *node, const char *key, pcmk_action_t *start) { GList *possible_matches = NULL; if (node == NULL) { // Should only be possible if unmanaged and stopped pcmk__rsc_trace(rsc, "%s will be mandatory because resource is unmanaged", key); return false; } if (!pcmk_is_set(rsc->private->cmds->action_flags(start, NULL), pcmk_action_optional)) { pcmk__rsc_trace(rsc, "%s will be mandatory because %s is", key, start->uuid); return false; } possible_matches = find_actions_exact(rsc->private->actions, key, node); if (possible_matches == NULL) { pcmk__rsc_trace(rsc, "%s will be mandatory because it is not active on %s", key, pcmk__node_name(node)); return false; } for (const GList *iter = possible_matches; iter != NULL; iter = iter->next) { const pcmk_action_t *op = (const pcmk_action_t *) iter->data; if (pcmk_is_set(op->flags, pcmk_action_reschedule)) { pcmk__rsc_trace(rsc, "%s will be mandatory because " "it needs to be rescheduled", key); g_list_free(possible_matches); return false; } } g_list_free(possible_matches); return true; } /*! * \internal * \brief Create recurring action from resource history entry for an active role * * \param[in,out] rsc Resource that resource history is for * \param[in,out] start Start action for \p rsc on \p node * \param[in] node Node that resource will be active on (if any) * \param[in] op Resource history entry */ static void recurring_op_for_active(pcmk_resource_t *rsc, pcmk_action_t *start, const pcmk_node_t *node, const struct op_history *op) { pcmk_action_t *mon = NULL; bool is_optional = true; bool role_match = false; enum rsc_role_e monitor_role = op->role; // We're only interested in recurring actions for active roles if (monitor_role == pcmk_role_stopped) { return; } is_optional = active_recurring_should_be_optional(rsc, node, op->key, start); // Check whether monitor's role matches role resource will have if (monitor_role == pcmk_role_unknown) { monitor_role = pcmk_role_unpromoted; role_match = (rsc->next_role != pcmk_role_promoted); } else { role_match = (rsc->next_role == monitor_role); } if (!role_match) { if (is_optional) { // It's running, so cancel it char *after_key = NULL; pcmk_action_t *cancel_op = pcmk__new_cancel_action(rsc, op->name, op->interval_ms, node); switch (rsc->role) { case pcmk_role_unpromoted: case pcmk_role_started: if (rsc->next_role == pcmk_role_promoted) { after_key = promote_key(rsc); } else if (rsc->next_role == pcmk_role_stopped) { after_key = stop_key(rsc); } break; case pcmk_role_promoted: after_key = demote_key(rsc); break; default: break; } if (after_key) { pcmk__new_ordering(rsc, NULL, cancel_op, rsc, after_key, NULL, pcmk__ar_unrunnable_first_blocks, rsc->private->scheduler); } } do_crm_log((is_optional? LOG_INFO : LOG_TRACE), "%s recurring action %s because %s configured for %s role " "(not %s)", (is_optional? "Cancelling" : "Ignoring"), op->key, op->id, pcmk_role_text(monitor_role), pcmk_role_text(rsc->next_role)); return; } pcmk__rsc_trace(rsc, "Creating %s recurring action %s for %s (%s %s on %s)", (is_optional? "optional" : "mandatory"), op->key, op->id, rsc->id, pcmk_role_text(rsc->next_role), pcmk__node_name(node)); mon = custom_action(rsc, strdup(op->key), op->name, node, is_optional, rsc->private->scheduler); if (!pcmk_is_set(start->flags, pcmk_action_runnable)) { pcmk__rsc_trace(rsc, "%s is unrunnable because start is", mon->uuid); pcmk__clear_action_flags(mon, pcmk_action_runnable); } else if ((node == NULL) || !node->details->online || node->details->unclean) { pcmk__rsc_trace(rsc, "%s is unrunnable because no node is available", mon->uuid); pcmk__clear_action_flags(mon, pcmk_action_runnable); } else if (!pcmk_is_set(mon->flags, pcmk_action_optional)) { pcmk__rsc_info(rsc, "Start %s-interval %s for %s on %s", pcmk__readable_interval(op->interval_ms), mon->task, rsc->id, pcmk__node_name(node)); } if (rsc->next_role == pcmk_role_promoted) { pe__add_action_expected_result(mon, CRM_EX_PROMOTED); } // Order monitor relative to other actions if ((node == NULL) || pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__new_ordering(rsc, start_key(rsc), NULL, NULL, strdup(mon->uuid), mon, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks, rsc->private->scheduler); pcmk__new_ordering(rsc, reload_key(rsc), NULL, NULL, strdup(mon->uuid), mon, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks, rsc->private->scheduler); if (rsc->next_role == pcmk_role_promoted) { pcmk__new_ordering(rsc, promote_key(rsc), NULL, rsc, NULL, mon, pcmk__ar_ordered |pcmk__ar_unrunnable_first_blocks, rsc->private->scheduler); } else if (rsc->role == pcmk_role_promoted) { pcmk__new_ordering(rsc, demote_key(rsc), NULL, rsc, NULL, mon, pcmk__ar_ordered |pcmk__ar_unrunnable_first_blocks, rsc->private->scheduler); } } } /*! * \internal * \brief Cancel a recurring action if running on a node * * \param[in,out] rsc Resource that action is for * \param[in] node Node to cancel action on * \param[in] key Operation key for action * \param[in] name Action name * \param[in] interval_ms Action interval (in milliseconds) */ static void cancel_if_running(pcmk_resource_t *rsc, const pcmk_node_t *node, const char *key, const char *name, guint interval_ms) { GList *possible_matches = find_actions_exact(rsc->private->actions, key, node); pcmk_action_t *cancel_op = NULL; if (possible_matches == NULL) { return; // Recurring action isn't running on this node } g_list_free(possible_matches); cancel_op = pcmk__new_cancel_action(rsc, name, interval_ms, node); switch (rsc->next_role) { case pcmk_role_started: case pcmk_role_unpromoted: /* Order starts after cancel. If the current role is * stopped, this cancels the monitor before the resource * starts; if the current role is started, then this cancels * the monitor on a migration target before starting there. */ pcmk__new_ordering(rsc, NULL, cancel_op, rsc, start_key(rsc), NULL, pcmk__ar_unrunnable_first_blocks, rsc->private->scheduler); break; default: break; } pcmk__rsc_info(rsc, "Cancelling %s-interval %s action for %s on %s because " "configured for " PCMK_ROLE_STOPPED " role (not %s)", pcmk__readable_interval(interval_ms), name, rsc->id, pcmk__node_name(node), pcmk_role_text(rsc->next_role)); } /*! * \internal * \brief Order an action after all probes of a resource on a node * * \param[in,out] rsc Resource to check for probes * \param[in] node Node to check for probes of \p rsc * \param[in,out] action Action to order after probes of \p rsc on \p node */ static void order_after_probes(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_action_t *action) { GList *probes = pe__resource_actions(rsc, node, PCMK_ACTION_MONITOR, FALSE); for (GList *iter = probes; iter != NULL; iter = iter->next) { order_actions((pcmk_action_t *) iter->data, action, pcmk__ar_unrunnable_first_blocks); } g_list_free(probes); } /*! * \internal * \brief Order an action after all stops of a resource on a node * * \param[in,out] rsc Resource to check for stops * \param[in] node Node to check for stops of \p rsc * \param[in,out] action Action to order after stops of \p rsc on \p node */ static void order_after_stops(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_action_t *action) { GList *stop_ops = pe__resource_actions(rsc, node, PCMK_ACTION_STOP, TRUE); for (GList *iter = stop_ops; iter != NULL; iter = iter->next) { pcmk_action_t *stop = (pcmk_action_t *) iter->data; if (!pcmk_is_set(stop->flags, pcmk_action_optional) && !pcmk_is_set(action->flags, pcmk_action_optional) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__rsc_trace(rsc, "%s optional on %s: unmanaged", action->uuid, pcmk__node_name(node)); pcmk__set_action_flags(action, pcmk_action_optional); } if (!pcmk_is_set(stop->flags, pcmk_action_runnable)) { crm_debug("%s unrunnable on %s: stop is unrunnable", action->uuid, pcmk__node_name(node)); pcmk__clear_action_flags(action, pcmk_action_runnable); } if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__new_ordering(rsc, stop_key(rsc), stop, NULL, NULL, action, pcmk__ar_first_implies_then |pcmk__ar_unrunnable_first_blocks, rsc->private->scheduler); } } g_list_free(stop_ops); } /*! * \internal * \brief Create recurring action from resource history entry for inactive role * * \param[in,out] rsc Resource that resource history is for * \param[in] node Node that resource will be active on (if any) * \param[in] op Resource history entry */ static void recurring_op_for_inactive(pcmk_resource_t *rsc, const pcmk_node_t *node, const struct op_history *op) { GList *possible_matches = NULL; // We're only interested in recurring actions for the inactive role if (op->role != pcmk_role_stopped) { return; } if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { crm_notice("Ignoring %s (recurring monitors for " PCMK_ROLE_STOPPED " role are not supported for anonymous clones)", op->id); return; // @TODO add support } pcmk__rsc_trace(rsc, "Creating recurring action %s for %s on nodes " "where it should not be running", op->id, rsc->id); for (GList *iter = rsc->private->scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *stop_node = (pcmk_node_t *) iter->data; bool is_optional = true; pcmk_action_t *stopped_mon = NULL; // Cancel action on node where resource will be active if ((node != NULL) && pcmk__str_eq(stop_node->details->uname, node->details->uname, pcmk__str_casei)) { cancel_if_running(rsc, node, op->key, op->name, op->interval_ms); continue; } // Recurring action on this node is optional if it's already active here possible_matches = find_actions_exact(rsc->private->actions, op->key, stop_node); is_optional = (possible_matches != NULL); g_list_free(possible_matches); pcmk__rsc_trace(rsc, "Creating %s recurring action %s for %s (%s " PCMK_ROLE_STOPPED " on %s)", (is_optional? "optional" : "mandatory"), op->key, op->id, rsc->id, pcmk__node_name(stop_node)); stopped_mon = custom_action(rsc, strdup(op->key), op->name, stop_node, is_optional, rsc->private->scheduler); pe__add_action_expected_result(stopped_mon, CRM_EX_NOT_RUNNING); if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { order_after_probes(rsc, stop_node, stopped_mon); } /* The recurring action is for the inactive role, so it shouldn't be * performed until the resource is inactive. */ order_after_stops(rsc, stop_node, stopped_mon); if (!stop_node->details->online || stop_node->details->unclean) { pcmk__rsc_debug(rsc, "%s unrunnable on %s: node unavailable)", stopped_mon->uuid, pcmk__node_name(stop_node)); pcmk__clear_action_flags(stopped_mon, pcmk_action_runnable); } if (pcmk_is_set(stopped_mon->flags, pcmk_action_runnable) && !pcmk_is_set(stopped_mon->flags, pcmk_action_optional)) { crm_notice("Start recurring %s-interval %s for " PCMK_ROLE_STOPPED " %s on %s", pcmk__readable_interval(op->interval_ms), stopped_mon->task, rsc->id, pcmk__node_name(stop_node)); } } } /*! * \internal * \brief Create recurring actions for a resource * * \param[in,out] rsc Resource to create recurring actions for */ void pcmk__create_recurring_actions(pcmk_resource_t *rsc) { pcmk_action_t *start = NULL; if (pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) { pcmk__rsc_trace(rsc, "Skipping recurring actions for blocked resource %s", rsc->id); return; } if (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance)) { pcmk__rsc_trace(rsc, "Skipping recurring actions for %s " "in maintenance mode", rsc->id); return; } - if (rsc->allocated_to == NULL) { + if (rsc->private->assigned_node == NULL) { // Recurring actions for active roles not needed - } else if (rsc->allocated_to->details->maintenance) { + } else if (rsc->private->assigned_node->details->maintenance) { pcmk__rsc_trace(rsc, "Skipping recurring actions for %s on %s " "in maintenance mode", - rsc->id, pcmk__node_name(rsc->allocated_to)); + rsc->id, pcmk__node_name(rsc->private->assigned_node)); } else if ((rsc->next_role != pcmk_role_stopped) || !pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { // Recurring actions for active roles needed - start = start_action(rsc, rsc->allocated_to, TRUE); + start = start_action(rsc, rsc->private->assigned_node, TRUE); } pcmk__rsc_trace(rsc, "Creating any recurring actions needed for %s", rsc->id); for (xmlNode *op = pcmk__xe_first_child(rsc->private->ops_xml, PCMK_XE_OP, NULL, NULL); op != NULL; op = pcmk__xe_next_same(op)) { struct op_history op_history = { NULL, }; if (!is_recurring_history(rsc, op, &op_history)) { continue; } if (start != NULL) { - recurring_op_for_active(rsc, start, rsc->allocated_to, &op_history); + recurring_op_for_active(rsc, start, rsc->private->assigned_node, + &op_history); } - recurring_op_for_inactive(rsc, rsc->allocated_to, &op_history); + recurring_op_for_inactive(rsc, rsc->private->assigned_node, + &op_history); free(op_history.key); } } /*! * \internal * \brief Create an executor cancel action * * \param[in,out] rsc Resource of action to cancel * \param[in] task Name of action to cancel * \param[in] interval_ms Interval of action to cancel * \param[in] node Node of action to cancel * * \return Created op */ pcmk_action_t * pcmk__new_cancel_action(pcmk_resource_t *rsc, const char *task, guint interval_ms, const pcmk_node_t *node) { pcmk_action_t *cancel_op = NULL; char *key = NULL; char *interval_ms_s = NULL; CRM_ASSERT((rsc != NULL) && (task != NULL) && (node != NULL)); key = pcmk__op_key(rsc->id, task, interval_ms); /* This finds an existing action by key, so custom_action() does not change * cancel_op->task. */ cancel_op = custom_action(rsc, key, PCMK_ACTION_CANCEL, node, FALSE, rsc->private->scheduler); pcmk__str_update(&(cancel_op->task), PCMK_ACTION_CANCEL); pcmk__str_update(&(cancel_op->cancel_task), task); interval_ms_s = crm_strdup_printf("%u", interval_ms); pcmk__insert_meta(cancel_op, PCMK_XA_OPERATION, task); pcmk__insert_meta(cancel_op, PCMK_META_INTERVAL, interval_ms_s); free(interval_ms_s); return cancel_op; } /*! * \internal * \brief Schedule cancellation of a recurring action * * \param[in,out] rsc Resource that action is for * \param[in] call_id Action's call ID from history * \param[in] task Action name * \param[in] interval_ms Action interval * \param[in] node Node that history entry is for * \param[in] reason Short description of why action is cancelled */ void pcmk__schedule_cancel(pcmk_resource_t *rsc, const char *call_id, const char *task, guint interval_ms, const pcmk_node_t *node, const char *reason) { pcmk_action_t *cancel = NULL; CRM_CHECK((rsc != NULL) && (task != NULL) && (node != NULL) && (reason != NULL), return); crm_info("Recurring %s-interval %s for %s will be stopped on %s: %s", pcmk__readable_interval(interval_ms), task, rsc->id, pcmk__node_name(node), reason); cancel = pcmk__new_cancel_action(rsc, task, interval_ms, node); pcmk__insert_meta(cancel, PCMK__XA_CALL_ID, call_id); // Cancellations happen after stops pcmk__new_ordering(rsc, stop_key(rsc), NULL, rsc, NULL, cancel, pcmk__ar_ordered, rsc->private->scheduler); } /*! * \internal * \brief Create a recurring action marked as needing rescheduling if active * * \param[in,out] rsc Resource that action is for * \param[in] task Name of action being rescheduled * \param[in] interval_ms Action interval (in milliseconds) * \param[in,out] node Node where action should be rescheduled */ void pcmk__reschedule_recurring(pcmk_resource_t *rsc, const char *task, guint interval_ms, pcmk_node_t *node) { pcmk_action_t *op = NULL; trigger_unfencing(rsc, node, "Device parameters changed (reschedule)", NULL, rsc->private->scheduler); op = custom_action(rsc, pcmk__op_key(rsc->id, task, interval_ms), task, node, TRUE, rsc->private->scheduler); pcmk__set_action_flags(op, pcmk_action_reschedule); } /*! * \internal * \brief Check whether an action is recurring * * \param[in] action Action to check * * \return true if \p action has a nonzero interval, otherwise false */ bool pcmk__action_is_recurring(const pcmk_action_t *action) { guint interval_ms = 0; if (pcmk__guint_from_hash(action->meta, PCMK_META_INTERVAL, 0, &interval_ms) != pcmk_rc_ok) { return false; } return (interval_ms > 0); } diff --git a/lib/pacemaker/pcmk_sched_remote.c b/lib/pacemaker/pcmk_sched_remote.c index 1f61d17f4e..134517c30d 100644 --- a/lib/pacemaker/pcmk_sched_remote.c +++ b/lib/pacemaker/pcmk_sched_remote.c @@ -1,736 +1,738 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include "libpacemaker_private.h" enum remote_connection_state { remote_state_unknown = 0, remote_state_alive = 1, remote_state_resting = 2, remote_state_failed = 3, remote_state_stopped = 4 }; static const char * state2text(enum remote_connection_state state) { switch (state) { case remote_state_unknown: return "unknown"; case remote_state_alive: return "alive"; case remote_state_resting: return "resting"; case remote_state_failed: return "failed"; case remote_state_stopped: return "stopped"; } return "impossible"; } /* We always use pcmk__ar_guest_allowed with these convenience functions to * exempt internally generated constraints from the prohibition of user * constraints involving remote connection resources. * * The start ordering additionally uses pcmk__ar_unrunnable_first_blocks so that * the specified action is not runnable if the start is not runnable. */ static inline void order_start_then_action(pcmk_resource_t *first_rsc, pcmk_action_t *then_action, uint32_t extra) { if ((first_rsc != NULL) && (then_action != NULL)) { pcmk__new_ordering(first_rsc, start_key(first_rsc), NULL, then_action->rsc, NULL, then_action, pcmk__ar_guest_allowed |pcmk__ar_unrunnable_first_blocks |extra, first_rsc->private->scheduler); } } static inline void order_action_then_stop(pcmk_action_t *first_action, pcmk_resource_t *then_rsc, uint32_t extra) { if ((first_action != NULL) && (then_rsc != NULL)) { pcmk__new_ordering(first_action->rsc, NULL, first_action, then_rsc, stop_key(then_rsc), NULL, pcmk__ar_guest_allowed|extra, then_rsc->private->scheduler); } } static enum remote_connection_state get_remote_node_state(const pcmk_node_t *node) { const pcmk_resource_t *remote_rsc = NULL; const pcmk_node_t *cluster_node = NULL; CRM_ASSERT(node != NULL); remote_rsc = node->details->remote_rsc; CRM_ASSERT(remote_rsc != NULL); cluster_node = pcmk__current_node(remote_rsc); /* If the cluster node the remote connection resource resides on * is unclean or went offline, we can't process any operations * on that remote node until after it starts elsewhere. */ if ((remote_rsc->next_role == pcmk_role_stopped) - || (remote_rsc->allocated_to == NULL)) { + || (remote_rsc->private->assigned_node == NULL)) { // The connection resource is not going to run anywhere if ((cluster_node != NULL) && cluster_node->details->unclean) { /* The remote connection is failed because its resource is on a * failed node and can't be recovered elsewhere, so we must fence. */ return remote_state_failed; } if (!pcmk_is_set(remote_rsc->flags, pcmk__rsc_failed)) { /* Connection resource is cleanly stopped */ return remote_state_stopped; } /* Connection resource is failed */ if ((remote_rsc->next_role == pcmk_role_stopped) && (remote_rsc->private->remote_reconnect_ms > 0U) && node->details->remote_was_fenced && !pe__shutdown_requested(node)) { /* We won't know whether the connection is recoverable until the * reconnect interval expires and we reattempt connection. */ return remote_state_unknown; } /* The remote connection is in a failed state. If there are any * resources known to be active on it (stop) or in an unknown state * (probe), we must assume the worst and fence it. */ return remote_state_failed; } else if (cluster_node == NULL) { /* Connection is recoverable but not currently running anywhere, so see * if we can recover it first */ return remote_state_unknown; } else if (cluster_node->details->unclean || !(cluster_node->details->online)) { // Connection is running on a dead node, see if we can recover it first return remote_state_resting; } else if (pcmk__list_of_multiple(remote_rsc->running_on) && (remote_rsc->partial_migration_source != NULL) && (remote_rsc->partial_migration_target != NULL)) { /* We're in the middle of migrating a connection resource, so wait until * after the migration completes before performing any actions. */ return remote_state_resting; } return remote_state_alive; } /*! * \internal * \brief Order actions on remote node relative to actions for the connection * * \param[in,out] action An action scheduled on a Pacemaker Remote node */ static void apply_remote_ordering(pcmk_action_t *action) { pcmk_resource_t *remote_rsc = NULL; enum action_tasks task = pcmk_parse_action(action->task); enum remote_connection_state state = get_remote_node_state(action->node); uint32_t order_opts = pcmk__ar_none; if (action->rsc == NULL) { return; } CRM_ASSERT(pcmk__is_pacemaker_remote_node(action->node)); remote_rsc = action->node->details->remote_rsc; CRM_ASSERT(remote_rsc != NULL); crm_trace("Order %s action %s relative to %s%s (state: %s)", action->task, action->uuid, pcmk_is_set(remote_rsc->flags, pcmk__rsc_failed)? "failed " : "", remote_rsc->id, state2text(state)); if (pcmk__strcase_any_of(action->task, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { /* Migration ops map to pcmk_action_unspecified, but we need to apply * the same ordering as for stop or demote (see get_router_node()). */ task = pcmk_action_stop; } switch (task) { case pcmk_action_start: case pcmk_action_promote: order_opts = pcmk__ar_none; if (state == remote_state_failed) { /* Force recovery, by making this action required */ pcmk__set_relation_flags(order_opts, pcmk__ar_first_implies_then); } /* Ensure connection is up before running this action */ order_start_then_action(remote_rsc, action, order_opts); break; case pcmk_action_stop: if (state == remote_state_alive) { order_action_then_stop(action, remote_rsc, pcmk__ar_then_implies_first); } else if (state == remote_state_failed) { /* The resource is active on the node, but since we don't have a * valid connection, the only way to stop the resource is by * fencing the node. There is no need to order the stop relative * to the remote connection, since the stop will become implied * by the fencing. */ pe_fence_node(remote_rsc->private->scheduler, action->node, "resources are active but " "connection is unrecoverable", FALSE); } else if (remote_rsc->next_role == pcmk_role_stopped) { /* State must be remote_state_unknown or remote_state_stopped. * Since the connection is not coming back up in this * transition, stop this resource first. */ order_action_then_stop(action, remote_rsc, pcmk__ar_then_implies_first); } else { /* The connection is going to be started somewhere else, so * stop this resource after that completes. */ order_start_then_action(remote_rsc, action, pcmk__ar_none); } break; case pcmk_action_demote: /* Only order this demote relative to the connection start if the * connection isn't being torn down. Otherwise, the demote would be * blocked because the connection start would not be allowed. */ if ((state == remote_state_resting) || (state == remote_state_unknown)) { order_start_then_action(remote_rsc, action, pcmk__ar_none); } /* Otherwise we can rely on the stop ordering */ break; default: /* Wait for the connection resource to be up */ if (pcmk__action_is_recurring(action)) { /* In case we ever get the recovery logic wrong, force * recurring monitors to be restarted, even if just * the connection was re-established */ order_start_then_action(remote_rsc, action, pcmk__ar_first_implies_then); } else { pcmk_node_t *cluster_node = pcmk__current_node(remote_rsc); if ((task == pcmk_action_monitor) && (state == remote_state_failed)) { /* We would only be here if we do not know the state of the * resource on the remote node. Since we have no way to find * out, it is necessary to fence the node. */ pe_fence_node(remote_rsc->private->scheduler, action->node, "resources are in unknown state " "and connection is unrecoverable", FALSE); } if ((cluster_node != NULL) && (state == remote_state_stopped)) { /* The connection is currently up, but is going down * permanently. Make sure we check services are actually * stopped _before_ we let the connection get closed. */ order_action_then_stop(action, remote_rsc, pcmk__ar_unrunnable_first_blocks); } else { order_start_then_action(remote_rsc, action, pcmk__ar_none); } } break; } } static void apply_container_ordering(pcmk_action_t *action) { /* VMs are also classified as containers for these purposes... in * that they both involve a 'thing' running on a real or remote * cluster node. * * This allows us to be smarter about the type and extent of * recovery actions required in various scenarios */ pcmk_resource_t *remote_rsc = NULL; pcmk_resource_t *container = NULL; enum action_tasks task = pcmk_parse_action(action->task); CRM_ASSERT(action->rsc != NULL); CRM_ASSERT(pcmk__is_pacemaker_remote_node(action->node)); remote_rsc = action->node->details->remote_rsc; CRM_ASSERT(remote_rsc != NULL); container = remote_rsc->container; CRM_ASSERT(container != NULL); if (pcmk_is_set(container->flags, pcmk__rsc_failed)) { pe_fence_node(action->rsc->private->scheduler, action->node, "container failed", FALSE); } crm_trace("Order %s action %s relative to %s%s for %s%s", action->task, action->uuid, pcmk_is_set(remote_rsc->flags, pcmk__rsc_failed)? "failed " : "", remote_rsc->id, pcmk_is_set(container->flags, pcmk__rsc_failed)? "failed " : "", container->id); if (pcmk__strcase_any_of(action->task, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { /* Migration ops map to pcmk_action_unspecified, but we need to apply * the same ordering as for stop or demote (see get_router_node()). */ task = pcmk_action_stop; } switch (task) { case pcmk_action_start: case pcmk_action_promote: // Force resource recovery if the container is recovered order_start_then_action(container, action, pcmk__ar_first_implies_then); // Wait for the connection resource to be up, too order_start_then_action(remote_rsc, action, pcmk__ar_none); break; case pcmk_action_stop: case pcmk_action_demote: if (pcmk_is_set(container->flags, pcmk__rsc_failed)) { /* When the container representing a guest node fails, any stop * or demote actions for resources running on the guest node * are implied by the container stopping. This is similar to * how fencing operations work for cluster nodes and remote * nodes. */ } else { /* Ensure the operation happens before the connection is brought * down. * * If we really wanted to, we could order these after the * connection start, IFF the container's current role was * stopped (otherwise we re-introduce an ordering loop when the * connection is restarting). */ order_action_then_stop(action, remote_rsc, pcmk__ar_none); } break; default: /* Wait for the connection resource to be up */ if (pcmk__action_is_recurring(action)) { /* In case we ever get the recovery logic wrong, force * recurring monitors to be restarted, even if just * the connection was re-established */ if (task != pcmk_action_unspecified) { order_start_then_action(remote_rsc, action, pcmk__ar_first_implies_then); } } else { order_start_then_action(remote_rsc, action, pcmk__ar_none); } break; } } /*! * \internal * \brief Order all relevant actions relative to remote connection actions * * \param[in,out] scheduler Scheduler data */ void pcmk__order_remote_connection_actions(pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_have_remote_nodes)) { return; } crm_trace("Creating remote connection orderings"); for (GList *iter = scheduler->actions; iter != NULL; iter = iter->next) { pcmk_action_t *action = iter->data; pcmk_resource_t *remote = NULL; // We are only interested in resource actions if (action->rsc == NULL) { continue; } /* Special case: If we are clearing the failcount of an actual * remote connection resource, then make sure this happens before * any start of the resource in this transition. */ if (pcmk_is_set(action->rsc->flags, pcmk__rsc_is_remote_connection) && pcmk__str_eq(action->task, PCMK_ACTION_CLEAR_FAILCOUNT, pcmk__str_none)) { pcmk__new_ordering(action->rsc, NULL, action, action->rsc, pcmk__op_key(action->rsc->id, PCMK_ACTION_START, 0), NULL, pcmk__ar_ordered, scheduler); continue; } // We are only interested in actions assigned to a node if (action->node == NULL) { continue; } if (!pcmk__is_pacemaker_remote_node(action->node)) { continue; } /* We are only interested in real actions. * * @TODO This is probably wrong; pseudo-actions might be converted to * real actions and vice versa later in update_actions() at the end of * pcmk__apply_orderings(). */ if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { continue; } remote = action->node->details->remote_rsc; if (remote == NULL) { // Orphaned continue; } /* Another special case: if a resource is moving to a Pacemaker Remote * node, order the stop on the original node after any start of the * remote connection. This ensures that if the connection fails to * start, we leave the resource running on the original node. */ if (pcmk__str_eq(action->task, PCMK_ACTION_START, pcmk__str_none)) { for (GList *item = action->rsc->private->actions; item != NULL; item = item->next) { pcmk_action_t *rsc_action = item->data; if (!pcmk__same_node(rsc_action->node, action->node) && pcmk__str_eq(rsc_action->task, PCMK_ACTION_STOP, pcmk__str_none)) { pcmk__new_ordering(remote, start_key(remote), NULL, action->rsc, NULL, rsc_action, pcmk__ar_ordered, scheduler); } } } /* The action occurs across a remote connection, so create * ordering constraints that guarantee the action occurs while the node * is active (after start, before stop ... things like that). * * This is somewhat brittle in that we need to make sure the results of * this ordering are compatible with the result of get_router_node(). * It would probably be better to add PCMK__XA_ROUTER_NODE as part of * this logic rather than create_graph_action(). */ if (remote->container) { crm_trace("Container ordering for %s", action->uuid); apply_container_ordering(action); } else { crm_trace("Remote ordering for %s", action->uuid); apply_remote_ordering(action); } } } /*! * \internal * \brief Check whether a node is a failed remote node * * \param[in] node Node to check * * \return true if \p node is a failed remote node, false otherwise */ bool pcmk__is_failed_remote_node(const pcmk_node_t *node) { return pcmk__is_remote_node(node) && (node->details->remote_rsc != NULL) && (get_remote_node_state(node) == remote_state_failed); } /*! * \internal * \brief Check whether a given resource corresponds to a given node as guest * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p node is a guest node and \p rsc is its containing * resource, otherwise false */ bool pcmk__rsc_corresponds_to_guest(const pcmk_resource_t *rsc, const pcmk_node_t *node) { return (rsc != NULL) && (rsc->fillers != NULL) && (node != NULL) && (node->details->remote_rsc != NULL) && (node->details->remote_rsc->container == rsc); } /*! * \internal * \brief Get proper connection host that a remote action must be routed through * * A remote connection resource might be starting, stopping, or migrating in the * same transition that an action needs to be executed on its Pacemaker Remote * node. Determine the proper node that the remote action should be routed * through. * * \param[in] action (Potentially remote) action to route * * \return Connection host that action should be routed through if remote, * otherwise NULL */ pcmk_node_t * pcmk__connection_host_for_action(const pcmk_action_t *action) { pcmk_node_t *began_on = NULL; pcmk_node_t *ended_on = NULL; bool partial_migration = false; const char *task = action->task; if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_none) || !pcmk__is_pacemaker_remote_node(action->node)) { return NULL; } CRM_ASSERT(action->node->details->remote_rsc != NULL); began_on = pcmk__current_node(action->node->details->remote_rsc); - ended_on = action->node->details->remote_rsc->allocated_to; + ended_on = action->node->details->remote_rsc->private->assigned_node; if (action->node->details->remote_rsc && (action->node->details->remote_rsc->container == NULL) && action->node->details->remote_rsc->partial_migration_target) { partial_migration = true; } if (began_on == NULL) { crm_trace("Routing %s for %s through remote connection's " "next node %s (starting)%s", action->task, (action->rsc? action->rsc->id : "no resource"), (ended_on? ended_on->details->uname : "none"), partial_migration? " (partial migration)" : ""); return ended_on; } if (ended_on == NULL) { crm_trace("Routing %s for %s through remote connection's " "current node %s (stopping)%s", action->task, (action->rsc? action->rsc->id : "no resource"), (began_on? began_on->details->uname : "none"), partial_migration? " (partial migration)" : ""); return began_on; } if (pcmk__same_node(began_on, ended_on)) { crm_trace("Routing %s for %s through remote connection's " "current node %s (not moving)%s", action->task, (action->rsc? action->rsc->id : "no resource"), (began_on? began_on->details->uname : "none"), partial_migration? " (partial migration)" : ""); return began_on; } /* If we get here, the remote connection is moving during this transition. * This means some actions for resources behind the connection will get * routed through the cluster node the connection resource is currently on, * and others are routed through the cluster node the connection will end up * on. */ if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)) { task = g_hash_table_lookup(action->meta, "notify_operation"); } /* * Stop, demote, and migration actions must occur before the connection can * move (these actions are required before the remote resource can stop). In * this case, we know these actions have to be routed through the initial * cluster node the connection resource lived on before the move takes * place. * * The exception is a partial migration of a (non-guest) remote connection * resource; in that case, all actions (even these) will be ordered after * the connection's pseudo-start on the migration target, so the target is * the router node. */ if (pcmk__strcase_any_of(task, PCMK_ACTION_CANCEL, PCMK_ACTION_STOP, PCMK_ACTION_DEMOTE, PCMK_ACTION_MIGRATE_FROM, PCMK_ACTION_MIGRATE_TO, NULL) && !partial_migration) { crm_trace("Routing %s for %s through remote connection's " "current node %s (moving)%s", action->task, (action->rsc? action->rsc->id : "no resource"), (began_on? began_on->details->uname : "none"), partial_migration? " (partial migration)" : ""); return began_on; } /* Everything else (start, promote, monitor, probe, refresh, * clear failcount, delete, ...) must occur after the connection starts on * the node it is moving to. */ crm_trace("Routing %s for %s through remote connection's " "next node %s (moving)%s", action->task, (action->rsc? action->rsc->id : "no resource"), (ended_on? ended_on->details->uname : "none"), partial_migration? " (partial migration)" : ""); return ended_on; } /*! * \internal * \brief Replace remote connection's addr="#uname" with actual address * * REMOTE_CONTAINER_HACK: If a given resource is a remote connection resource * with its "addr" parameter set to "#uname", pull the actual value from the * parameters evaluated without a node (which was put there earlier in * pcmk__create_graph() when the bundle's expand() method was called). * * \param[in,out] rsc Resource to check * \param[in,out] params Resource parameters evaluated per node */ void pcmk__substitute_remote_addr(pcmk_resource_t *rsc, GHashTable *params) { const char *remote_addr = g_hash_table_lookup(params, PCMK_REMOTE_RA_ADDR); if (pcmk__str_eq(remote_addr, "#uname", pcmk__str_none)) { GHashTable *base = pe_rsc_params(rsc, NULL, rsc->private->scheduler); remote_addr = g_hash_table_lookup(base, PCMK_REMOTE_RA_ADDR); if (remote_addr != NULL) { pcmk__insert_dup(params, PCMK_REMOTE_RA_ADDR, remote_addr); } } } /*! * \brief Add special guest node meta-attributes to XML * * If a given action will be executed on a guest node, add the following as XML * attributes (using meta-attribute naming): * * The resource's \c PCMK_META_CONTAINER_ATTRIBUTE_TARGET meta-attribute * (usually set only for bundles), as \c PCMK_META_CONTAINER_ATTRIBUTE_TARGET * * The guest's physical host (current host for "down" actions, next host for * "up" actions), as \c PCMK__META_PHYSICAL_HOST * * If the guest node has no physical host, then don't add either attribute. * * \param[in,out] args_xml XML to add attributes to * \param[in] action Action to check */ void pcmk__add_guest_meta_to_xml(xmlNode *args_xml, const pcmk_action_t *action) { const pcmk_node_t *guest = action->node; const pcmk_node_t *host = NULL; + const pcmk_resource_t *container = NULL; enum action_tasks task; if (!pcmk__is_guest_or_bundle_node(guest)) { return; } + container = guest->details->remote_rsc->container; task = pcmk_parse_action(action->task); if ((task == pcmk_action_notify) || (task == pcmk_action_notified)) { task = pcmk_parse_action(g_hash_table_lookup(action->meta, "notify_operation")); } switch (task) { case pcmk_action_stop: case pcmk_action_stopped: case pcmk_action_demote: case pcmk_action_demoted: // "Down" actions take place on guest's current host - host = pcmk__current_node(guest->details->remote_rsc->container); + host = pcmk__current_node(container); break; case pcmk_action_start: case pcmk_action_started: case pcmk_action_monitor: case pcmk_action_promote: case pcmk_action_promoted: // "Up" actions take place on guest's next host - host = guest->details->remote_rsc->container->allocated_to; + host = container->private->assigned_node; break; default: break; } if (host != NULL) { gpointer target = g_hash_table_lookup(action->rsc->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); hash2metafield((gpointer) PCMK_META_CONTAINER_ATTRIBUTE_TARGET, target, (gpointer) args_xml); hash2metafield((gpointer) PCMK__META_PHYSICAL_HOST, (gpointer) host->details->uname, (gpointer) args_xml); } } diff --git a/lib/pacemaker/pcmk_sched_resource.c b/lib/pacemaker/pcmk_sched_resource.c index 1ecb0d4d7b..61c1e5bd51 100644 --- a/lib/pacemaker/pcmk_sched_resource.c +++ b/lib/pacemaker/pcmk_sched_resource.c @@ -1,780 +1,780 @@ /* * Copyright 2014-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include "libpacemaker_private.h" // Resource assignment methods by resource variant static pcmk__assignment_methods_t assignment_methods[] = { { pcmk__primitive_assign, pcmk__primitive_create_actions, pcmk__probe_rsc_on_node, pcmk__primitive_internal_constraints, pcmk__primitive_apply_coloc_score, pcmk__colocated_resources, pcmk__with_primitive_colocations, pcmk__primitive_with_colocations, pcmk__add_colocated_node_scores, pcmk__apply_location, pcmk__primitive_action_flags, pcmk__update_ordered_actions, pcmk__output_resource_actions, pcmk__add_rsc_actions_to_graph, pcmk__primitive_add_graph_meta, pcmk__primitive_add_utilization, pcmk__primitive_shutdown_lock, }, { pcmk__group_assign, pcmk__group_create_actions, pcmk__probe_rsc_on_node, pcmk__group_internal_constraints, pcmk__group_apply_coloc_score, pcmk__group_colocated_resources, pcmk__with_group_colocations, pcmk__group_with_colocations, pcmk__group_add_colocated_node_scores, pcmk__group_apply_location, pcmk__group_action_flags, pcmk__group_update_ordered_actions, pcmk__output_resource_actions, pcmk__add_rsc_actions_to_graph, pcmk__noop_add_graph_meta, pcmk__group_add_utilization, pcmk__group_shutdown_lock, }, { pcmk__clone_assign, pcmk__clone_create_actions, pcmk__clone_create_probe, pcmk__clone_internal_constraints, pcmk__clone_apply_coloc_score, pcmk__colocated_resources, pcmk__with_clone_colocations, pcmk__clone_with_colocations, pcmk__add_colocated_node_scores, pcmk__clone_apply_location, pcmk__clone_action_flags, pcmk__instance_update_ordered_actions, pcmk__output_resource_actions, pcmk__clone_add_actions_to_graph, pcmk__clone_add_graph_meta, pcmk__clone_add_utilization, pcmk__clone_shutdown_lock, }, { pcmk__bundle_assign, pcmk__bundle_create_actions, pcmk__bundle_create_probe, pcmk__bundle_internal_constraints, pcmk__bundle_apply_coloc_score, pcmk__colocated_resources, pcmk__with_bundle_colocations, pcmk__bundle_with_colocations, pcmk__add_colocated_node_scores, pcmk__bundle_apply_location, pcmk__bundle_action_flags, pcmk__instance_update_ordered_actions, pcmk__output_bundle_actions, pcmk__bundle_add_actions_to_graph, pcmk__noop_add_graph_meta, pcmk__bundle_add_utilization, pcmk__bundle_shutdown_lock, } }; /*! * \internal * \brief Check whether a resource's agent standard, provider, or type changed * * \param[in,out] rsc Resource to check * \param[in,out] node Node needing unfencing if agent changed * \param[in] rsc_entry XML with previously known agent information * \param[in] active_on_node Whether \p rsc is active on \p node * * \return true if agent for \p rsc changed, otherwise false */ bool pcmk__rsc_agent_changed(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *rsc_entry, bool active_on_node) { bool changed = false; const char *attr_list[] = { PCMK_XA_TYPE, PCMK_XA_CLASS, PCMK_XA_PROVIDER, }; for (int i = 0; i < PCMK__NELEM(attr_list); i++) { const char *value = crm_element_value(rsc->private->xml, attr_list[i]); const char *old_value = crm_element_value(rsc_entry, attr_list[i]); if (!pcmk__str_eq(value, old_value, pcmk__str_none)) { changed = true; trigger_unfencing(rsc, node, "Device definition changed", NULL, rsc->private->scheduler); if (active_on_node) { crm_notice("Forcing restart of %s on %s " "because %s changed from '%s' to '%s'", rsc->id, pcmk__node_name(node), attr_list[i], pcmk__s(old_value, ""), pcmk__s(value, "")); } } } if (changed && active_on_node) { // Make sure the resource is restarted custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, node, FALSE, rsc->private->scheduler); pcmk__set_rsc_flags(rsc, pcmk__rsc_start_pending); } return changed; } /*! * \internal * \brief Add resource (and any matching children) to list if it matches ID * * \param[in] result List to add resource to * \param[in] rsc Resource to check * \param[in] id ID to match * * \return (Possibly new) head of list */ static GList * add_rsc_if_matching(GList *result, pcmk_resource_t *rsc, const char *id) { if (pcmk__str_eq(id, rsc->id, pcmk__str_none) || pcmk__str_eq(id, rsc->private->history_id, pcmk__str_none)) { result = g_list_prepend(result, rsc); } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) iter->data; result = add_rsc_if_matching(result, child, id); } return result; } /*! * \internal * \brief Find all resources matching a given ID by either ID or clone name * * \param[in] id Resource ID to check * \param[in] scheduler Scheduler data * * \return List of all resources that match \p id * \note The caller is responsible for freeing the return value with * g_list_free(). */ GList * pcmk__rscs_matching_id(const char *id, const pcmk_scheduler_t *scheduler) { GList *result = NULL; CRM_CHECK((id != NULL) && (scheduler != NULL), return NULL); for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { result = add_rsc_if_matching(result, (pcmk_resource_t *) iter->data, id); } return result; } /*! * \internal * \brief Set the variant-appropriate assignment methods for a resource * * \param[in,out] data Resource to set assignment methods for * \param[in] user_data Ignored */ static void set_assignment_methods_for_rsc(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; rsc->private->cmds = &assignment_methods[rsc->private->variant]; g_list_foreach(rsc->children, set_assignment_methods_for_rsc, NULL); } /*! * \internal * \brief Set the variant-appropriate assignment methods for all resources * * \param[in,out] scheduler Scheduler data */ void pcmk__set_assignment_methods(pcmk_scheduler_t *scheduler) { g_list_foreach(scheduler->resources, set_assignment_methods_for_rsc, NULL); } /*! * \internal * \brief Wrapper for colocated_resources() method for readability * * \param[in] rsc Resource to add to colocated list * \param[in] orig_rsc Resource originally requested * \param[in,out] list Pointer to list to add to * * \return (Possibly new) head of list */ static inline void add_colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { *list = rsc->private->cmds->colocated_resources(rsc, orig_rsc, *list); } // Shared implementation of pcmk__assignment_methods_t:colocated_resources() GList * pcmk__colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs) { const GList *iter = NULL; GList *colocations = NULL; if (orig_rsc == NULL) { orig_rsc = rsc; } if ((rsc == NULL) || (g_list_find(colocated_rscs, rsc) != NULL)) { return colocated_rscs; } pcmk__rsc_trace(orig_rsc, "%s is in colocation chain with %s", rsc->id, orig_rsc->id); colocated_rscs = g_list_prepend(colocated_rscs, (gpointer) rsc); // Follow colocations where this resource is the dependent resource colocations = pcmk__this_with_colocations(rsc); for (iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *constraint = iter->data; const pcmk_resource_t *primary = constraint->primary; if (primary == orig_rsc) { continue; // Break colocation loop } if ((constraint->score == PCMK_SCORE_INFINITY) && (pcmk__colocation_affects(rsc, primary, constraint, true) == pcmk__coloc_affects_location)) { add_colocated_resources(primary, orig_rsc, &colocated_rscs); } } g_list_free(colocations); // Follow colocations where this resource is the primary resource colocations = pcmk__with_this_colocations(rsc); for (iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *constraint = iter->data; const pcmk_resource_t *dependent = constraint->dependent; if (dependent == orig_rsc) { continue; // Break colocation loop } if (pcmk__is_clone(rsc) && !pcmk__is_clone(dependent)) { continue; // We can't be sure whether dependent will be colocated } if ((constraint->score == PCMK_SCORE_INFINITY) && (pcmk__colocation_affects(dependent, rsc, constraint, true) == pcmk__coloc_affects_location)) { add_colocated_resources(dependent, orig_rsc, &colocated_rscs); } } g_list_free(colocations); return colocated_rscs; } // No-op function for variants that don't need to implement add_graph_meta() void pcmk__noop_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml) { } /*! * \internal * \brief Output a summary of scheduled actions for a resource * * \param[in,out] rsc Resource to output actions for */ void pcmk__output_resource_actions(pcmk_resource_t *rsc) { pcmk_node_t *next = NULL; pcmk_node_t *current = NULL; pcmk__output_t *out = NULL; CRM_ASSERT(rsc != NULL); out = rsc->private->scheduler->priv; if (rsc->children != NULL) { for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) iter->data; child->private->cmds->output_actions(child); } return; } - next = rsc->allocated_to; + next = rsc->private->assigned_node; if (rsc->running_on) { current = pcmk__current_node(rsc); if (rsc->role == pcmk_role_stopped) { /* This can occur when resources are being recovered because * the current role can change in pcmk__primitive_create_actions() */ rsc->role = pcmk_role_started; } } if ((current == NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { /* Don't log stopped orphans */ return; } out->message(out, "rsc-action", rsc, current, next); } /*! * \internal * \brief Add a resource to a node's list of assigned resources * * \param[in,out] node Node to add resource to * \param[in] rsc Resource to add */ static inline void add_assigned_resource(pcmk_node_t *node, pcmk_resource_t *rsc) { node->details->allocated_rsc = g_list_prepend(node->details->allocated_rsc, rsc); } /*! * \internal * \brief Assign a specified resource (of any variant) to a node * * Assign a specified resource and its children (if any) to a specified node, if * the node can run the resource (or unconditionally, if \p force is true). Mark * the resources as no longer provisional. * * If a resource can't be assigned (or \p node is \c NULL), unassign any * previous assignment. If \p stop_if_fail is \c true, set next role to stopped * and update any existing actions scheduled for the resource. * * \param[in,out] rsc Resource to assign * \param[in,out] node Node to assign \p rsc to * \param[in] force If true, assign to \p node even if unavailable * \param[in] stop_if_fail If \c true and either \p rsc can't be assigned * or \p chosen is \c NULL, set next role to * stopped and update existing actions (if \p rsc * is not a primitive, this applies to its * primitive descendants instead) * * \return \c true if the assignment of \p rsc changed, or \c false otherwise * * \note Assigning a resource to the NULL node using this function is different * from calling pcmk__unassign_resource(), in that it may also update any * actions created for the resource. * \note The \c pcmk__assignment_methods_t:assign() method is preferred, unless * a resource should be assigned to the \c NULL node or every resource in * a tree should be assigned to the same 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. */ bool pcmk__assign_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool force, bool stop_if_fail) { bool changed = false; pcmk_scheduler_t *scheduler = NULL; CRM_ASSERT(rsc != NULL); scheduler = rsc->private->scheduler; if (rsc->children != NULL) { for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child_rsc = iter->data; changed |= pcmk__assign_resource(child_rsc, node, force, stop_if_fail); } return changed; } // Assigning a primitive if (!force && (node != NULL) && ((node->weight < 0) // Allow graph to assume that guest node connections will come up || (!pcmk__node_available(node, true, false) && !pcmk__is_guest_or_bundle_node(node)))) { pcmk__rsc_debug(rsc, "All nodes for resource %s are unavailable, unclean or " "shutting down (%s can%s run resources, with score %s)", rsc->id, pcmk__node_name(node), (pcmk__node_available(node, true, false)? "" : "not"), pcmk_readable_score(node->weight)); if (stop_if_fail) { pe__set_next_role(rsc, pcmk_role_stopped, "node availability"); } node = NULL; } - if (rsc->allocated_to != NULL) { - changed = !pcmk__same_node(rsc->allocated_to, node); + if (rsc->private->assigned_node != NULL) { + changed = !pcmk__same_node(rsc->private->assigned_node, node); } else { changed = (node != NULL); } pcmk__unassign_resource(rsc); pcmk__clear_rsc_flags(rsc, pcmk__rsc_unassigned); if (node == NULL) { char *rc_stopped = NULL; pcmk__rsc_debug(rsc, "Could not assign %s to a node", rsc->id); if (!stop_if_fail) { return changed; } pe__set_next_role(rsc, pcmk_role_stopped, "unable to assign"); for (GList *iter = rsc->private->actions; iter != NULL; iter = iter->next) { pcmk_action_t *op = (pcmk_action_t *) iter->data; pcmk__rsc_debug(rsc, "Updating %s for %s assignment failure", op->uuid, rsc->id); if (pcmk__str_eq(op->task, PCMK_ACTION_STOP, pcmk__str_none)) { pcmk__clear_action_flags(op, pcmk_action_optional); } else if (pcmk__str_eq(op->task, PCMK_ACTION_START, pcmk__str_none)) { pcmk__clear_action_flags(op, pcmk_action_runnable); } else { // Cancel recurring actions, unless for stopped state const char *interval_ms_s = NULL; const char *target_rc_s = NULL; interval_ms_s = g_hash_table_lookup(op->meta, PCMK_META_INTERVAL); target_rc_s = g_hash_table_lookup(op->meta, PCMK__META_OP_TARGET_RC); if (rc_stopped == NULL) { rc_stopped = pcmk__itoa(PCMK_OCF_NOT_RUNNING); } if (!pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches) && !pcmk__str_eq(rc_stopped, target_rc_s, pcmk__str_none)) { pcmk__clear_action_flags(op, pcmk_action_runnable); } } } free(rc_stopped); return changed; } pcmk__rsc_debug(rsc, "Assigning %s to %s", rsc->id, pcmk__node_name(node)); - rsc->allocated_to = pe__copy_node(node); + rsc->private->assigned_node = pe__copy_node(node); add_assigned_resource(node, rsc); node->details->num_resources++; node->count++; pcmk__consume_node_capacity(node->details->utilization, rsc); if (pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) { pcmk__output_t *out = scheduler->priv; out->message(out, "resource-util", rsc, node, __func__); } return changed; } /*! * \internal * \brief Remove any node assignment from a specified resource and its children * * If a specified resource has been assigned to a node, remove that assignment * and mark the resource as provisional again. * * \param[in,out] rsc Resource to unassign * * \note This function is called recursively on \p rsc and its children. */ void pcmk__unassign_resource(pcmk_resource_t *rsc) { - pcmk_node_t *old = rsc->allocated_to; + pcmk_node_t *old = rsc->private->assigned_node; if (old == NULL) { crm_info("Unassigning %s", rsc->id); } else { crm_info("Unassigning %s from %s", rsc->id, pcmk__node_name(old)); } pcmk__set_rsc_flags(rsc, pcmk__rsc_unassigned); if (rsc->children == NULL) { if (old == NULL) { return; } - rsc->allocated_to = NULL; + rsc->private->assigned_node = NULL; /* We're going to free the pcmk_node_t, but its details member is shared * and will remain, so update that appropriately first. */ old->details->allocated_rsc = g_list_remove(old->details->allocated_rsc, rsc); old->details->num_resources--; pcmk__release_node_capacity(old->details->utilization, rsc); free(old); return; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk__unassign_resource((pcmk_resource_t *) iter->data); } } /*! * \internal * \brief Check whether a resource has reached its migration threshold on a node * * \param[in,out] rsc Resource to check * \param[in] node Node to check * \param[out] failed If threshold has been reached, this will be set to * resource that failed (possibly a parent of \p rsc) * * \return true if the migration threshold has been reached, false otherwise */ bool pcmk__threshold_reached(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_resource_t **failed) { int fail_count, remaining_tries; pcmk_resource_t *rsc_to_ban = rsc; // Migration threshold of 0 means never force away if (rsc->private->ban_after_failures == 0) { return false; } // If we're ignoring failures, also ignore the migration threshold if (pcmk_is_set(rsc->flags, pcmk__rsc_ignore_failure)) { return false; } // If there are no failures, there's no need to force away fail_count = pe_get_failcount(node, rsc, NULL, pcmk__fc_effective|pcmk__fc_fillers, NULL); if (fail_count <= 0) { return false; } // If failed resource is anonymous clone instance, we'll force clone away if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { rsc_to_ban = uber_parent(rsc); } // How many more times recovery will be tried on this node remaining_tries = rsc->private->ban_after_failures - fail_count; if (remaining_tries <= 0) { pcmk__sched_warn("%s cannot run on %s due to reaching migration " "threshold (clean up resource to allow again)" QB_XS " failures=%d " PCMK_META_MIGRATION_THRESHOLD "=%d", rsc_to_ban->id, pcmk__node_name(node), fail_count, rsc->private->ban_after_failures); if (failed != NULL) { *failed = rsc_to_ban; } return true; } crm_info("%s can fail %d more time%s on " "%s before reaching migration threshold (%d)", rsc_to_ban->id, remaining_tries, pcmk__plural_s(remaining_tries), pcmk__node_name(node), rsc->private->ban_after_failures); return false; } /*! * \internal * \brief Get a node's score * * \param[in] node Node with ID to check * \param[in] nodes List of nodes to look for \p node score in * * \return Node's score, or -INFINITY if not found */ static int get_node_score(const pcmk_node_t *node, GHashTable *nodes) { pcmk_node_t *found_node = NULL; if ((node != NULL) && (nodes != NULL)) { found_node = g_hash_table_lookup(nodes, node->details->id); } return (found_node == NULL)? -PCMK_SCORE_INFINITY : found_node->weight; } /*! * \internal * \brief Compare two resources according to which should be assigned first * * \param[in] a First resource to compare * \param[in] b Second resource to compare * \param[in] data Sorted list of all nodes in cluster * * \return -1 if \p a should be assigned before \b, 0 if they are equal, * or +1 if \p a should be assigned after \b */ static gint cmp_resources(gconstpointer a, gconstpointer b, gpointer data) { /* GLib insists that this function require gconstpointer arguments, but we * make a small, temporary change to each argument (setting the * pe_rsc_merging flag) during comparison */ pcmk_resource_t *resource1 = (pcmk_resource_t *) a; pcmk_resource_t *resource2 = (pcmk_resource_t *) b; const GList *nodes = data; int rc = 0; int r1_score = -PCMK_SCORE_INFINITY; int r2_score = -PCMK_SCORE_INFINITY; pcmk_node_t *r1_node = NULL; pcmk_node_t *r2_node = NULL; GHashTable *r1_nodes = NULL; GHashTable *r2_nodes = NULL; const char *reason = NULL; // Resources with highest priority should be assigned first reason = "priority"; r1_score = resource1->private->priority; r2_score = resource2->private->priority; if (r1_score > r2_score) { rc = -1; goto done; } if (r1_score < r2_score) { rc = 1; goto done; } // We need nodes to make any other useful comparisons reason = "no node list"; if (nodes == NULL) { goto done; } // Calculate and log node scores resource1->private->cmds->add_colocated_node_scores(resource1, NULL, resource1->id, &r1_nodes, NULL, 1, pcmk__coloc_select_this_with); resource2->private->cmds->add_colocated_node_scores(resource2, NULL, resource2->id, &r2_nodes, NULL, 1, pcmk__coloc_select_this_with); pe__show_node_scores(true, NULL, resource1->id, r1_nodes, resource1->private->scheduler); pe__show_node_scores(true, NULL, resource2->id, r2_nodes, resource2->private->scheduler); // The resource with highest score on its current node goes first reason = "current location"; if (resource1->running_on != NULL) { r1_node = pcmk__current_node(resource1); } if (resource2->running_on != NULL) { r2_node = pcmk__current_node(resource2); } r1_score = get_node_score(r1_node, r1_nodes); r2_score = get_node_score(r2_node, r2_nodes); if (r1_score > r2_score) { rc = -1; goto done; } if (r1_score < r2_score) { rc = 1; goto done; } // Otherwise a higher score on any node will do reason = "score"; for (const GList *iter = nodes; iter != NULL; iter = iter->next) { const pcmk_node_t *node = (const pcmk_node_t *) iter->data; r1_score = get_node_score(node, r1_nodes); r2_score = get_node_score(node, r2_nodes); if (r1_score > r2_score) { rc = -1; goto done; } if (r1_score < r2_score) { rc = 1; goto done; } } done: crm_trace("%s (%d)%s%s %c %s (%d)%s%s: %s", resource1->id, r1_score, ((r1_node == NULL)? "" : " on "), ((r1_node == NULL)? "" : r1_node->details->id), ((rc < 0)? '>' : ((rc > 0)? '<' : '=')), resource2->id, r2_score, ((r2_node == NULL)? "" : " on "), ((r2_node == NULL)? "" : r2_node->details->id), reason); if (r1_nodes != NULL) { g_hash_table_destroy(r1_nodes); } if (r2_nodes != NULL) { g_hash_table_destroy(r2_nodes); } return rc; } /*! * \internal * \brief Sort resources in the order they should be assigned to nodes * * \param[in,out] scheduler Scheduler data */ void pcmk__sort_resources(pcmk_scheduler_t *scheduler) { GList *nodes = g_list_copy(scheduler->nodes); nodes = pcmk__sort_nodes(nodes, NULL); scheduler->resources = g_list_sort_with_data(scheduler->resources, cmp_resources, nodes); g_list_free(nodes); } diff --git a/lib/pengine/bundle.c b/lib/pengine/bundle.c index 29e06c902c..dd25cfbf69 100644 --- a/lib/pengine/bundle.c +++ b/lib/pengine/bundle.c @@ -1,2139 +1,2139 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include enum pe__bundle_mount_flags { pe__bundle_mount_none = 0x00, // mount instance-specific subdirectory rather than source directly pe__bundle_mount_subdir = 0x01 }; typedef struct { char *source; char *target; char *options; uint32_t flags; // bitmask of pe__bundle_mount_flags } pe__bundle_mount_t; typedef struct { char *source; char *target; } pe__bundle_port_t; enum pe__container_agent { PE__CONTAINER_AGENT_UNKNOWN, PE__CONTAINER_AGENT_DOCKER, PE__CONTAINER_AGENT_RKT, PE__CONTAINER_AGENT_PODMAN, }; #define PE__CONTAINER_AGENT_UNKNOWN_S "unknown" #define PE__CONTAINER_AGENT_DOCKER_S "docker" #define PE__CONTAINER_AGENT_RKT_S "rkt" #define PE__CONTAINER_AGENT_PODMAN_S "podman" typedef struct pe__bundle_variant_data_s { int promoted_max; int nreplicas; int nreplicas_per_host; char *prefix; char *image; const char *ip_last; char *host_network; char *host_netmask; char *control_port; char *container_network; char *ip_range_start; gboolean add_host; gchar *container_host_options; char *container_command; char *launcher_options; const char *attribute_target; pcmk_resource_t *child; GList *replicas; // pcmk__bundle_replica_t * GList *ports; // pe__bundle_port_t * GList *mounts; // pe__bundle_mount_t * enum pe__container_agent agent_type; } pe__bundle_variant_data_t; #define get_bundle_variant_data(data, rsc) do { \ CRM_ASSERT(pcmk__is_bundle(rsc)); \ data = rsc->private->variant_opaque; \ } while (0) /*! * \internal * \brief Get maximum number of bundle replicas allowed to run * * \param[in] rsc Bundle or bundled resource to check * * \return Maximum replicas for bundle corresponding to \p rsc */ int pe__bundle_max(const pcmk_resource_t *rsc) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, pe__const_top_resource(rsc, true)); return bundle_data->nreplicas; } /*! * \internal * \brief Get the resource inside a bundle * * \param[in] bundle Bundle to check * * \return Resource inside \p bundle if any, otherwise NULL */ pcmk_resource_t * pe__bundled_resource(const pcmk_resource_t *rsc) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, pe__const_top_resource(rsc, true)); return bundle_data->child; } /*! * \internal * \brief Get containerized resource corresponding to a given bundle container * * \param[in] instance Collective instance that might be a bundle container * * \return Bundled resource instance inside \p instance if it is a bundle * container instance, otherwise NULL */ const pcmk_resource_t * pe__get_rsc_in_container(const pcmk_resource_t *instance) { const pe__bundle_variant_data_t *data = NULL; const pcmk_resource_t *top = pe__const_top_resource(instance, true); if (!pcmk__is_bundle(top)) { return NULL; } get_bundle_variant_data(data, top); for (const GList *iter = data->replicas; iter != NULL; iter = iter->next) { const pcmk__bundle_replica_t *replica = iter->data; if (instance == replica->container) { return replica->child; } } return NULL; } /*! * \internal * \brief Check whether a given node is created by a bundle * * \param[in] bundle Bundle resource to check * \param[in] node Node to check * * \return true if \p node is an instance of \p bundle, otherwise false */ bool pe__node_is_bundle_instance(const pcmk_resource_t *bundle, const pcmk_node_t *node) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; if (pcmk__same_node(node, replica->node)) { return true; } } return false; } /*! * \internal * \brief Get the container of a bundle's first replica * * \param[in] bundle Bundle resource to get container for * * \return Container resource from first replica of \p bundle if any, * otherwise NULL */ pcmk_resource_t * pe__first_container(const pcmk_resource_t *bundle) { const pe__bundle_variant_data_t *bundle_data = NULL; const pcmk__bundle_replica_t *replica = NULL; get_bundle_variant_data(bundle_data, bundle); if (bundle_data->replicas == NULL) { return NULL; } replica = bundle_data->replicas->data; return replica->container; } /*! * \internal * \brief Iterate over bundle replicas * * \param[in,out] bundle Bundle to iterate over * \param[in] fn Function to call for each replica (its return value * indicates whether to continue iterating) * \param[in,out] user_data Pointer to pass to \p fn */ void pe__foreach_bundle_replica(pcmk_resource_t *bundle, bool (*fn)(pcmk__bundle_replica_t *, void *), void *user_data) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { if (!fn((pcmk__bundle_replica_t *) iter->data, user_data)) { break; } } } /*! * \internal * \brief Iterate over const bundle replicas * * \param[in] bundle Bundle to iterate over * \param[in] fn Function to call for each replica (its return value * indicates whether to continue iterating) * \param[in,out] user_data Pointer to pass to \p fn */ void pe__foreach_const_bundle_replica(const pcmk_resource_t *bundle, bool (*fn)(const pcmk__bundle_replica_t *, void *), void *user_data) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (const GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { if (!fn((const pcmk__bundle_replica_t *) iter->data, user_data)) { break; } } } static char * next_ip(const char *last_ip) { unsigned int oct1 = 0; unsigned int oct2 = 0; unsigned int oct3 = 0; unsigned int oct4 = 0; int rc = sscanf(last_ip, "%u.%u.%u.%u", &oct1, &oct2, &oct3, &oct4); if (rc != 4) { /*@ TODO check for IPv6 */ return NULL; } else if (oct3 > 253) { return NULL; } else if (oct4 > 253) { ++oct3; oct4 = 1; } else { ++oct4; } return crm_strdup_printf("%u.%u.%u.%u", oct1, oct2, oct3, oct4); } static void allocate_ip(pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica, GString *buffer) { if(data->ip_range_start == NULL) { return; } else if(data->ip_last) { replica->ipaddr = next_ip(data->ip_last); } else { replica->ipaddr = strdup(data->ip_range_start); } data->ip_last = replica->ipaddr; switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: if (data->add_host) { g_string_append_printf(buffer, " --add-host=%s-%d:%s", data->prefix, replica->offset, replica->ipaddr); } else { g_string_append_printf(buffer, " --hosts-entry=%s=%s-%d", replica->ipaddr, data->prefix, replica->offset); } break; case PE__CONTAINER_AGENT_RKT: g_string_append_printf(buffer, " --hosts-entry=%s=%s-%d", replica->ipaddr, data->prefix, replica->offset); break; default: // PE__CONTAINER_AGENT_UNKNOWN break; } } static xmlNode * create_resource(const char *name, const char *provider, const char *kind) { xmlNode *rsc = pcmk__xe_create(NULL, PCMK_XE_PRIMITIVE); crm_xml_add(rsc, PCMK_XA_ID, name); crm_xml_add(rsc, PCMK_XA_CLASS, PCMK_RESOURCE_CLASS_OCF); crm_xml_add(rsc, PCMK_XA_PROVIDER, provider); crm_xml_add(rsc, PCMK_XA_TYPE, kind); return rsc; } /*! * \internal * \brief Check whether cluster can manage resource inside container * * \param[in,out] data Container variant data * * \return TRUE if networking configuration is acceptable, FALSE otherwise * * \note The resource is manageable if an IP range or control port has been * specified. If a control port is used without an IP range, replicas per * host must be 1. */ static bool valid_network(pe__bundle_variant_data_t *data) { if(data->ip_range_start) { return TRUE; } if(data->control_port) { if(data->nreplicas_per_host > 1) { pcmk__config_err("Specifying the '" PCMK_XA_CONTROL_PORT "' for %s " "requires '" PCMK_XA_REPLICAS_PER_HOST "=1'", data->prefix); data->nreplicas_per_host = 1; // @TODO to be sure: // pcmk__clear_rsc_flags(rsc, pcmk__rsc_unique); } return TRUE; } return FALSE; } static int create_ip_resource(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { if(data->ip_range_start) { char *id = NULL; xmlNode *xml_ip = NULL; xmlNode *xml_obj = NULL; id = crm_strdup_printf("%s-ip-%s", data->prefix, replica->ipaddr); pcmk__xml_sanitize_id(id); xml_ip = create_resource(id, "heartbeat", "IPaddr2"); free(id); xml_obj = pcmk__xe_create(xml_ip, PCMK_XE_INSTANCE_ATTRIBUTES); pcmk__xe_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "ip", replica->ipaddr); if(data->host_network) { crm_create_nvpair_xml(xml_obj, NULL, "nic", data->host_network); } if(data->host_netmask) { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", data->host_netmask); } else { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", "32"); } xml_obj = pcmk__xe_create(xml_ip, PCMK_XE_OPERATIONS); crm_create_op_xml(xml_obj, pcmk__xe_id(xml_ip), PCMK_ACTION_MONITOR, "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (pe__unpack_resource(xml_ip, &replica->ip, parent, parent->private->scheduler) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } parent->children = g_list_append(parent->children, replica->ip); } return pcmk_rc_ok; } static const char* container_agent_str(enum pe__container_agent t) { switch (t) { case PE__CONTAINER_AGENT_DOCKER: return PE__CONTAINER_AGENT_DOCKER_S; case PE__CONTAINER_AGENT_RKT: return PE__CONTAINER_AGENT_RKT_S; case PE__CONTAINER_AGENT_PODMAN: return PE__CONTAINER_AGENT_PODMAN_S; default: // PE__CONTAINER_AGENT_UNKNOWN break; } return PE__CONTAINER_AGENT_UNKNOWN_S; } static int create_container_resource(pcmk_resource_t *parent, const pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { char *id = NULL; xmlNode *xml_container = NULL; xmlNode *xml_obj = NULL; // Agent-specific const char *hostname_opt = NULL; const char *env_opt = NULL; const char *agent_str = NULL; int volid = 0; // rkt-only GString *buffer = NULL; GString *dbuffer = NULL; // Where syntax differences are drop-in replacements, set them now switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: hostname_opt = "-h "; env_opt = "-e "; break; case PE__CONTAINER_AGENT_RKT: hostname_opt = "--hostname="; env_opt = "--environment="; break; default: // PE__CONTAINER_AGENT_UNKNOWN return pcmk_rc_unpack_error; } agent_str = container_agent_str(data->agent_type); buffer = g_string_sized_new(4096); id = crm_strdup_printf("%s-%s-%d", data->prefix, agent_str, replica->offset); pcmk__xml_sanitize_id(id); xml_container = create_resource(id, "heartbeat", agent_str); free(id); xml_obj = pcmk__xe_create(xml_container, PCMK_XE_INSTANCE_ATTRIBUTES); pcmk__xe_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "image", data->image); crm_create_nvpair_xml(xml_obj, NULL, "allow_pull", PCMK_VALUE_TRUE); crm_create_nvpair_xml(xml_obj, NULL, "force_kill", PCMK_VALUE_FALSE); crm_create_nvpair_xml(xml_obj, NULL, "reuse", PCMK_VALUE_FALSE); if (data->agent_type == PE__CONTAINER_AGENT_DOCKER) { g_string_append(buffer, " --restart=no"); } /* Set a container hostname only if we have an IP to map it to. The user can * set -h or --uts=host themselves if they want a nicer name for logs, but * this makes applications happy who need their hostname to match the IP * they bind to. */ if (data->ip_range_start != NULL) { g_string_append_printf(buffer, " %s%s-%d", hostname_opt, data->prefix, replica->offset); } pcmk__g_strcat(buffer, " ", env_opt, "PCMK_stderr=1", NULL); if (data->container_network != NULL) { pcmk__g_strcat(buffer, " --net=", data->container_network, NULL); } if (data->control_port != NULL) { pcmk__g_strcat(buffer, " ", env_opt, "PCMK_" PCMK__ENV_REMOTE_PORT "=", data->control_port, NULL); } else { g_string_append_printf(buffer, " %sPCMK_" PCMK__ENV_REMOTE_PORT "=%d", env_opt, DEFAULT_REMOTE_PORT); } for (GList *iter = data->mounts; iter != NULL; iter = iter->next) { pe__bundle_mount_t *mount = (pe__bundle_mount_t *) iter->data; char *source = NULL; if (pcmk_is_set(mount->flags, pe__bundle_mount_subdir)) { source = crm_strdup_printf("%s/%s-%d", mount->source, data->prefix, replica->offset); pcmk__add_separated_word(&dbuffer, 1024, source, ","); } switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: pcmk__g_strcat(buffer, " -v ", pcmk__s(source, mount->source), ":", mount->target, NULL); if (mount->options != NULL) { pcmk__g_strcat(buffer, ":", mount->options, NULL); } break; case PE__CONTAINER_AGENT_RKT: g_string_append_printf(buffer, " --volume vol%d,kind=host," "source=%s%s%s " "--mount volume=vol%d,target=%s", volid, pcmk__s(source, mount->source), (mount->options != NULL)? "," : "", pcmk__s(mount->options, ""), volid, mount->target); volid++; break; default: break; } free(source); } for (GList *iter = data->ports; iter != NULL; iter = iter->next) { pe__bundle_port_t *port = (pe__bundle_port_t *) iter->data; switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: if (replica->ipaddr != NULL) { pcmk__g_strcat(buffer, " -p ", replica->ipaddr, ":", port->source, ":", port->target, NULL); } else if (!pcmk__str_eq(data->container_network, PCMK_VALUE_HOST, pcmk__str_none)) { // No need to do port mapping if net == host pcmk__g_strcat(buffer, " -p ", port->source, ":", port->target, NULL); } break; case PE__CONTAINER_AGENT_RKT: if (replica->ipaddr != NULL) { pcmk__g_strcat(buffer, " --port=", port->target, ":", replica->ipaddr, ":", port->source, NULL); } else { pcmk__g_strcat(buffer, " --port=", port->target, ":", port->source, NULL); } break; default: break; } } /* @COMPAT: We should use pcmk__add_word() here, but we can't yet, because * it would cause restarts during rolling upgrades. * * In a previous version of the container resource creation logic, if * data->launcher_options is not NULL, we append * (" %s", data->launcher_options) even if data->launcher_options is an * empty string. Likewise for data->container_host_options. Using * * pcmk__add_word(buffer, 0, data->launcher_options) * * removes that extra trailing space, causing a resource definition change. */ if (data->launcher_options != NULL) { pcmk__g_strcat(buffer, " ", data->launcher_options, NULL); } if (data->container_host_options != NULL) { pcmk__g_strcat(buffer, " ", data->container_host_options, NULL); } crm_create_nvpair_xml(xml_obj, NULL, "run_opts", (const char *) buffer->str); g_string_free(buffer, TRUE); crm_create_nvpair_xml(xml_obj, NULL, "mount_points", (dbuffer != NULL)? (const char *) dbuffer->str : ""); if (dbuffer != NULL) { g_string_free(dbuffer, TRUE); } if (replica->child != NULL) { if (data->container_command != NULL) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } else { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", SBIN_DIR "/pacemaker-remoted"); } /* TODO: Allow users to specify their own? * * We just want to know if the container is alive; we'll monitor the * child independently. */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); #if 0 /* @TODO Consider supporting the use case where we can start and stop * resources, but not proxy local commands (such as setting node * attributes), by running the local executor in stand-alone mode. * However, this would probably be better done via ACLs as with other * Pacemaker Remote nodes. */ } else if ((child != NULL) && data->untrusted) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", CRM_DAEMON_DIR "/pacemaker-execd"); crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", CRM_DAEMON_DIR "/pacemaker/cts-exec-helper -c poke"); #endif } else { if (data->container_command != NULL) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } /* TODO: Allow users to specify their own? * * We don't know what's in the container, so we just want to know if it * is alive. */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); } xml_obj = pcmk__xe_create(xml_container, PCMK_XE_OPERATIONS); crm_create_op_xml(xml_obj, pcmk__xe_id(xml_container), PCMK_ACTION_MONITOR, "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (pe__unpack_resource(xml_container, &replica->container, parent, parent->private->scheduler) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } pcmk__set_rsc_flags(replica->container, pcmk__rsc_replica_container); parent->children = g_list_append(parent->children, replica->container); return pcmk_rc_ok; } /*! * \brief Ban a node from a resource's (and its children's) allowed nodes list * * \param[in,out] rsc Resource to modify * \param[in] uname Name of node to ban */ static void disallow_node(pcmk_resource_t *rsc, const char *uname) { gpointer match = g_hash_table_lookup(rsc->allowed_nodes, uname); if (match) { ((pcmk_node_t *) match)->weight = -PCMK_SCORE_INFINITY; ((pcmk_node_t *) match)->rsc_discover_mode = pcmk_probe_never; } if (rsc->children) { g_list_foreach(rsc->children, (GFunc) disallow_node, (gpointer) uname); } } static int create_remote_resource(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { if (replica->child && valid_network(data)) { GHashTableIter gIter; pcmk_node_t *node = NULL; xmlNode *xml_remote = NULL; char *id = crm_strdup_printf("%s-%d", data->prefix, replica->offset); char *port_s = NULL; const char *uname = NULL; const char *connect_name = NULL; if (pe_find_resource(parent->private->scheduler->resources, id) != NULL) { free(id); // The biggest hammer we have id = crm_strdup_printf("pcmk-internal-%s-remote-%d", replica->child->id, replica->offset); //@TODO return error instead of asserting? CRM_ASSERT(pe_find_resource(parent->private->scheduler->resources, id) == NULL); } /* REMOTE_CONTAINER_HACK: Using "#uname" as the server name when the * connection does not have its own IP is a magic string that we use to * support nested remotes (i.e. a bundle running on a remote node). */ connect_name = (replica->ipaddr? replica->ipaddr : "#uname"); if (data->control_port == NULL) { port_s = pcmk__itoa(DEFAULT_REMOTE_PORT); } /* This sets replica->container as replica->remote's container, which is * similar to what happens with guest nodes. This is how the scheduler * knows that the bundle node is fenced by recovering the container, and * that remote should be ordered relative to the container. */ xml_remote = pe_create_remote_xml(NULL, id, replica->container->id, NULL, NULL, NULL, connect_name, (data->control_port? data->control_port : port_s)); free(port_s); /* Abandon our created ID, and pull the copy from the XML, because we * need something that will get freed during scheduler data cleanup to * use as the node ID and uname. */ free(id); id = NULL; uname = pcmk__xe_id(xml_remote); /* Ensure a node has been created for the guest (it may have already * been, if it has a permanent node attribute), and ensure its weight is * -INFINITY so no other resources can run on it. */ node = pcmk_find_node(parent->private->scheduler, uname); if (node == NULL) { node = pe_create_node(uname, uname, PCMK_VALUE_REMOTE, PCMK_VALUE_MINUS_INFINITY, parent->private->scheduler); } else { node->weight = -PCMK_SCORE_INFINITY; } node->rsc_discover_mode = pcmk_probe_never; /* unpack_remote_nodes() ensures that each remote node and guest node * has a pcmk_node_t entry. Ideally, it would do the same for bundle * nodes. Unfortunately, a bundle has to be mostly unpacked before it's * obvious what nodes will be needed, so we do it just above. * * Worse, that means that the node may have been utilized while * unpacking other resources, without our weight correction. The most * likely place for this to happen is when pe__unpack_resource() calls * resource_location() to set a default score in symmetric clusters. * This adds a node *copy* to each resource's allowed nodes, and these * copies will have the wrong weight. * * As a hacky workaround, fix those copies here. * * @TODO Possible alternative: ensure bundles are unpacked before other * resources, so the weight is correct before any copies are made. */ g_list_foreach(parent->private->scheduler->resources, (GFunc) disallow_node, (gpointer) uname); replica->node = pe__copy_node(node); replica->node->weight = 500; replica->node->rsc_discover_mode = pcmk_probe_exclusive; /* Ensure the node shows up as allowed and with the correct discovery set */ if (replica->child->allowed_nodes != NULL) { g_hash_table_destroy(replica->child->allowed_nodes); } replica->child->allowed_nodes = pcmk__strkey_table(NULL, free); g_hash_table_insert(replica->child->allowed_nodes, (gpointer) replica->node->details->id, pe__copy_node(replica->node)); { pcmk_node_t *copy = pe__copy_node(replica->node); copy->weight = -PCMK_SCORE_INFINITY; g_hash_table_insert(replica->child->private->parent->allowed_nodes, (gpointer) replica->node->details->id, copy); } if (pe__unpack_resource(xml_remote, &replica->remote, parent, parent->private->scheduler) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } g_hash_table_iter_init(&gIter, replica->remote->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&node)) { if (pcmk__is_pacemaker_remote_node(node)) { /* Remote resources can only run on 'normal' cluster node */ node->weight = -PCMK_SCORE_INFINITY; } } replica->node->details->remote_rsc = replica->remote; // Ensure pcmk__is_guest_or_bundle_node() functions correctly replica->remote->container = replica->container; /* A bundle's #kind is closer to "container" (guest node) than the * "remote" set by pe_create_node(). */ pcmk__insert_dup(replica->node->details->attrs, CRM_ATTR_KIND, "container"); /* One effect of this is that setup_container() will add * replica->remote to replica->container's fillers, which will make * pe__resource_contains_guest_node() true for replica->container. * * replica->child does NOT get added to replica->container's fillers. * The only noticeable effect if it did would be for its fail count to * be taken into account when checking replica->container's migration * threshold. */ parent->children = g_list_append(parent->children, replica->remote); } return pcmk_rc_ok; } static int create_replica_resources(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { int rc = pcmk_rc_ok; rc = create_container_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } rc = create_ip_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } rc = create_remote_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } if ((replica->child != NULL) && (replica->ipaddr != NULL)) { pcmk__insert_meta(replica->child, "external-ip", replica->ipaddr); } if (replica->remote != NULL) { /* * Allow the remote connection resource to be allocated to a * different node than the one on which the container is active. * * This makes it possible to have Pacemaker Remote nodes running * containers with pacemaker-remoted inside in order to start * services inside those containers. */ pcmk__set_rsc_flags(replica->remote, pcmk__rsc_remote_nesting_allowed); } return rc; } static void mount_add(pe__bundle_variant_data_t *bundle_data, const char *source, const char *target, const char *options, uint32_t flags) { pe__bundle_mount_t *mount = pcmk__assert_alloc(1, sizeof(pe__bundle_mount_t)); mount->source = pcmk__str_copy(source); mount->target = pcmk__str_copy(target); mount->options = pcmk__str_copy(options); mount->flags = flags; bundle_data->mounts = g_list_append(bundle_data->mounts, mount); } static void mount_free(pe__bundle_mount_t *mount) { free(mount->source); free(mount->target); free(mount->options); free(mount); } static void port_free(pe__bundle_port_t *port) { free(port->source); free(port->target); free(port); } static pcmk__bundle_replica_t * replica_for_remote(pcmk_resource_t *remote) { pcmk_resource_t *top = remote; pe__bundle_variant_data_t *bundle_data = NULL; if (top == NULL) { return NULL; } while (top->private->parent != NULL) { top = top->private->parent; } get_bundle_variant_data(bundle_data, top); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; if (replica->remote == remote) { return replica; } } CRM_LOG_ASSERT(FALSE); return NULL; } bool pe__bundle_needs_remote_name(pcmk_resource_t *rsc) { const char *value; GHashTable *params = NULL; if (rsc == NULL) { return false; } // Use NULL node since pcmk__bundle_expand() uses that to set value params = pe_rsc_params(rsc, NULL, rsc->private->scheduler); value = g_hash_table_lookup(params, PCMK_REMOTE_RA_ADDR); return pcmk__str_eq(value, "#uname", pcmk__str_casei) && xml_contains_remote_node(rsc->private->xml); } const char * pe__add_bundle_remote_name(pcmk_resource_t *rsc, xmlNode *xml, const char *field) { // REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside pcmk_node_t *node = NULL; pcmk__bundle_replica_t *replica = NULL; if (!pe__bundle_needs_remote_name(rsc)) { return NULL; } replica = replica_for_remote(rsc); if (replica == NULL) { return NULL; } - node = replica->container->allocated_to; + node = replica->container->private->assigned_node; if (node == NULL) { /* If it won't be running anywhere after the * transition, go with where it's running now. */ node = pcmk__current_node(replica->container); } if(node == NULL) { crm_trace("Cannot determine address for bundle connection %s", rsc->id); return NULL; } crm_trace("Setting address for bundle connection %s to bundle host %s", rsc->id, pcmk__node_name(node)); if(xml != NULL && field != NULL) { crm_xml_add(xml, field, node->details->uname); } return node->details->uname; } #define pe__set_bundle_mount_flags(mount_xml, flags, flags_to_set) do { \ flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Bundle mount", pcmk__xe_id(mount_xml), \ flags, (flags_to_set), #flags_to_set); \ } while (0) gboolean pe__unpack_bundle(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { const char *value = NULL; xmlNode *xml_obj = NULL; const xmlNode *xml_child = NULL; xmlNode *xml_resource = NULL; pe__bundle_variant_data_t *bundle_data = NULL; bool need_log_mount = TRUE; CRM_ASSERT(rsc != NULL); pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id); bundle_data = pcmk__assert_alloc(1, sizeof(pe__bundle_variant_data_t)); rsc->private->variant_opaque = bundle_data; bundle_data->prefix = strdup(rsc->id); xml_obj = pcmk__xe_first_child(rsc->private->xml, PCMK_XE_DOCKER, NULL, NULL); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_DOCKER; } else { xml_obj = pcmk__xe_first_child(rsc->private->xml, PCMK__XE_RKT, NULL, NULL); if (xml_obj != NULL) { pcmk__warn_once(pcmk__wo_rkt, "Support for " PCMK__XE_RKT " in bundles " "(such as %s) is deprecated and will be " "removed in a future release", rsc->id); bundle_data->agent_type = PE__CONTAINER_AGENT_RKT; } else { xml_obj = pcmk__xe_first_child(rsc->private->xml, PCMK_XE_PODMAN, NULL, NULL); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_PODMAN; } else { return FALSE; } } } // Use 0 for default, minimum, and invalid PCMK_XA_PROMOTED_MAX value = crm_element_value(xml_obj, PCMK_XA_PROMOTED_MAX); if (value == NULL) { // @COMPAT deprecated since 2.0.0 value = crm_element_value(xml_obj, PCMK__XA_PROMOTED_MAX_LEGACY); if (value != NULL) { pcmk__warn_once(pcmk__wo_bundle_master, "Support for the " PCMK__XA_PROMOTED_MAX_LEGACY " attribute (such as in %s) is deprecated and " "will be removed in a future release. Use " PCMK_XA_PROMOTED_MAX " instead.", rsc->id); } } pcmk__scan_min_int(value, &bundle_data->promoted_max, 0); /* Default replicas to PCMK_XA_PROMOTED_MAX if it was specified and 1 * otherwise */ value = crm_element_value(xml_obj, PCMK_XA_REPLICAS); if ((value == NULL) && (bundle_data->promoted_max > 0)) { bundle_data->nreplicas = bundle_data->promoted_max; } else { pcmk__scan_min_int(value, &bundle_data->nreplicas, 1); } /* * Communication between containers on the same host via the * floating IPs only works if the container is started with: * --userland-proxy=false --ip-masq=false */ value = crm_element_value(xml_obj, PCMK_XA_REPLICAS_PER_HOST); pcmk__scan_min_int(value, &bundle_data->nreplicas_per_host, 1); if (bundle_data->nreplicas_per_host == 1) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_unique); } bundle_data->container_command = crm_element_value_copy(xml_obj, PCMK_XA_RUN_COMMAND); bundle_data->launcher_options = crm_element_value_copy(xml_obj, PCMK_XA_OPTIONS); bundle_data->image = crm_element_value_copy(xml_obj, PCMK_XA_IMAGE); bundle_data->container_network = crm_element_value_copy(xml_obj, PCMK_XA_NETWORK); xml_obj = pcmk__xe_first_child(rsc->private->xml, PCMK_XE_NETWORK, NULL, NULL); if(xml_obj) { bundle_data->ip_range_start = crm_element_value_copy(xml_obj, PCMK_XA_IP_RANGE_START); bundle_data->host_netmask = crm_element_value_copy(xml_obj, PCMK_XA_HOST_NETMASK); bundle_data->host_network = crm_element_value_copy(xml_obj, PCMK_XA_HOST_INTERFACE); bundle_data->control_port = crm_element_value_copy(xml_obj, PCMK_XA_CONTROL_PORT); value = crm_element_value(xml_obj, PCMK_XA_ADD_HOST); if (crm_str_to_boolean(value, &bundle_data->add_host) != 1) { bundle_data->add_host = TRUE; } for (xml_child = pcmk__xe_first_child(xml_obj, PCMK_XE_PORT_MAPPING, NULL, NULL); xml_child != NULL; xml_child = pcmk__xe_next_same(xml_child)) { pe__bundle_port_t *port = pcmk__assert_alloc(1, sizeof(pe__bundle_port_t)); port->source = crm_element_value_copy(xml_child, PCMK_XA_PORT); if(port->source == NULL) { port->source = crm_element_value_copy(xml_child, PCMK_XA_RANGE); } else { port->target = crm_element_value_copy(xml_child, PCMK_XA_INTERNAL_PORT); } if(port->source != NULL && strlen(port->source) > 0) { if(port->target == NULL) { port->target = strdup(port->source); } bundle_data->ports = g_list_append(bundle_data->ports, port); } else { pcmk__config_err("Invalid " PCMK_XA_PORT " directive %s", pcmk__xe_id(xml_child)); port_free(port); } } } xml_obj = pcmk__xe_first_child(rsc->private->xml, PCMK_XE_STORAGE, NULL, NULL); for (xml_child = pcmk__xe_first_child(xml_obj, PCMK_XE_STORAGE_MAPPING, NULL, NULL); xml_child != NULL; xml_child = pcmk__xe_next_same(xml_child)) { const char *source = crm_element_value(xml_child, PCMK_XA_SOURCE_DIR); const char *target = crm_element_value(xml_child, PCMK_XA_TARGET_DIR); const char *options = crm_element_value(xml_child, PCMK_XA_OPTIONS); int flags = pe__bundle_mount_none; if (source == NULL) { source = crm_element_value(xml_child, PCMK_XA_SOURCE_DIR_ROOT); pe__set_bundle_mount_flags(xml_child, flags, pe__bundle_mount_subdir); } if (source && target) { mount_add(bundle_data, source, target, options, flags); if (strcmp(target, "/var/log") == 0) { need_log_mount = FALSE; } } else { pcmk__config_err("Invalid mount directive %s", pcmk__xe_id(xml_child)); } } xml_obj = pcmk__xe_first_child(rsc->private->xml, PCMK_XE_PRIMITIVE, NULL, NULL); if (xml_obj && valid_network(bundle_data)) { const char *suffix = NULL; char *value = NULL; xmlNode *xml_set = NULL; xml_resource = pcmk__xe_create(NULL, PCMK_XE_CLONE); /* @COMPAT We no longer use the tag, but we need to keep it as * part of the resource name, so that bundles don't restart in a rolling * upgrade. (It also avoids needing to change regression tests.) */ suffix = (const char *) xml_resource->name; if (bundle_data->promoted_max > 0) { suffix = "master"; } pcmk__xe_set_id(xml_resource, "%s-%s", bundle_data->prefix, suffix); xml_set = pcmk__xe_create(xml_resource, PCMK_XE_META_ATTRIBUTES); pcmk__xe_set_id(xml_set, "%s-%s-meta", bundle_data->prefix, xml_resource->name); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_ORDERED, PCMK_VALUE_TRUE); value = pcmk__itoa(bundle_data->nreplicas); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_CLONE_MAX, value); free(value); value = pcmk__itoa(bundle_data->nreplicas_per_host); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_CLONE_NODE_MAX, value); free(value); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_GLOBALLY_UNIQUE, pcmk__btoa(bundle_data->nreplicas_per_host > 1)); if (bundle_data->promoted_max) { crm_create_nvpair_xml(xml_set, NULL, PCMK_META_PROMOTABLE, PCMK_VALUE_TRUE); value = pcmk__itoa(bundle_data->promoted_max); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_PROMOTED_MAX, value); free(value); } //crm_xml_add(xml_obj, PCMK_XA_ID, bundle_data->prefix); pcmk__xml_copy(xml_resource, xml_obj); } else if(xml_obj) { pcmk__config_err("Cannot control %s inside %s without either " PCMK_XA_IP_RANGE_START " or " PCMK_XA_CONTROL_PORT, rsc->id, pcmk__xe_id(xml_obj)); return FALSE; } if(xml_resource) { int lpc = 0; GList *childIter = NULL; pe__bundle_port_t *port = NULL; GString *buffer = NULL; if (pe__unpack_resource(xml_resource, &(bundle_data->child), rsc, scheduler) != pcmk_rc_ok) { return FALSE; } /* Currently, we always map the default authentication key location * into the same location inside the container. * * Ideally, we would respect the host's PCMK_authkey_location, but: * - it may be different on different nodes; * - the actual connection will do extra checking to make sure the key * file exists and is readable, that we can't do here on the DC * - tools such as crm_resource and crm_simulate may not have the same * environment variables as the cluster, causing operation digests to * differ * * Always using the default location inside the container is fine, * because we control the pacemaker_remote environment, and it avoids * having to pass another environment variable to the container. * * @TODO A better solution may be to have only pacemaker_remote use the * environment variable, and have the cluster nodes use a new * cluster option for key location. This would introduce the limitation * of the location being the same on all cluster nodes, but that's * reasonable. */ mount_add(bundle_data, DEFAULT_REMOTE_KEY_LOCATION, DEFAULT_REMOTE_KEY_LOCATION, NULL, pe__bundle_mount_none); if (need_log_mount) { mount_add(bundle_data, CRM_BUNDLE_DIR, "/var/log", NULL, pe__bundle_mount_subdir); } port = pcmk__assert_alloc(1, sizeof(pe__bundle_port_t)); if(bundle_data->control_port) { port->source = strdup(bundle_data->control_port); } else { /* If we wanted to respect PCMK_remote_port, we could use * crm_default_remote_port() here and elsewhere in this file instead * of DEFAULT_REMOTE_PORT. * * However, it gains nothing, since we control both the container * environment and the connection resource parameters, and the user * can use a different port if desired by setting * PCMK_XA_CONTROL_PORT. */ port->source = pcmk__itoa(DEFAULT_REMOTE_PORT); } port->target = strdup(port->source); bundle_data->ports = g_list_append(bundle_data->ports, port); buffer = g_string_sized_new(1024); for (childIter = bundle_data->child->children; childIter != NULL; childIter = childIter->next) { pcmk__bundle_replica_t *replica = NULL; replica = pcmk__assert_alloc(1, sizeof(pcmk__bundle_replica_t)); replica->child = childIter->data; pcmk__set_rsc_flags(replica->child, pcmk__rsc_exclusive_probes); replica->offset = lpc++; // Ensure the child's notify gets set based on the underlying primitive's value if (pcmk_is_set(replica->child->flags, pcmk__rsc_notify)) { pcmk__set_rsc_flags(bundle_data->child, pcmk__rsc_notify); } allocate_ip(bundle_data, replica, buffer); bundle_data->replicas = g_list_append(bundle_data->replicas, replica); bundle_data->attribute_target = g_hash_table_lookup(replica->child->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); } bundle_data->container_host_options = g_string_free(buffer, FALSE); if (bundle_data->attribute_target) { pcmk__insert_dup(rsc->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET, bundle_data->attribute_target); pcmk__insert_dup(bundle_data->child->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET, bundle_data->attribute_target); } } else { // Just a naked container, no pacemaker-remote GString *buffer = g_string_sized_new(1024); for (int lpc = 0; lpc < bundle_data->nreplicas; lpc++) { pcmk__bundle_replica_t *replica = NULL; replica = pcmk__assert_alloc(1, sizeof(pcmk__bundle_replica_t)); replica->offset = lpc; allocate_ip(bundle_data, replica, buffer); bundle_data->replicas = g_list_append(bundle_data->replicas, replica); } bundle_data->container_host_options = g_string_free(buffer, FALSE); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; if (create_replica_resources(rsc, bundle_data, replica) != pcmk_rc_ok) { pcmk__config_err("Failed unpacking resource %s", rsc->id); rsc->private->fns->free(rsc); return FALSE; } /* Utilization needs special handling for bundles. It makes no sense for * the inner primitive to have utilization, because it is tied * one-to-one to the guest node created by the container resource -- and * there's no way to set capacities for that guest node anyway. * * What the user really wants is to configure utilization for the * container. However, the schema only allows utilization for * primitives, and the container resource is implicit anyway, so the * user can *only* configure utilization for the inner primitive. If * they do, move the primitive's utilization values to the container. * * @TODO This means that bundles without an inner primitive can't have * utilization. An alternative might be to allow utilization values in * the top-level bundle XML in the schema, and copy those to each * container. */ if (replica->child != NULL) { GHashTable *empty = replica->container->utilization; replica->container->utilization = replica->child->utilization; replica->child->utilization = empty; } } if (bundle_data->child) { rsc->children = g_list_append(rsc->children, bundle_data->child); } return TRUE; } static int replica_resource_active(pcmk_resource_t *rsc, gboolean all) { if (rsc) { gboolean child_active = rsc->private->fns->active(rsc, all); if (child_active && !all) { return TRUE; } else if (!child_active && all) { return FALSE; } } return -1; } gboolean pe__bundle_active(pcmk_resource_t *rsc, gboolean all) { pe__bundle_variant_data_t *bundle_data = NULL; GList *iter = NULL; get_bundle_variant_data(bundle_data, rsc); for (iter = bundle_data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; int rsc_active; rsc_active = replica_resource_active(replica->ip, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->child, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->container, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->remote, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } } /* If "all" is TRUE, we've already checked that no resources were inactive, * so return TRUE; if "all" is FALSE, we didn't find any active resources, * so return FALSE. */ return all; } /*! * \internal * \brief Find the bundle replica corresponding to a given node * * \param[in] bundle Top-level bundle resource * \param[in] node Node to search for * * \return Bundle replica if found, NULL otherwise */ pcmk_resource_t * pe__find_bundle_replica(const pcmk_resource_t *bundle, const pcmk_node_t *node) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_ASSERT(bundle && node); get_bundle_variant_data(bundle_data, bundle); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica && replica->node); if (pcmk__same_node(replica->node, node)) { return replica->child; } } return NULL; } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_xml(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean printed_header = FALSE; gboolean print_everything = TRUE; const char *desc = NULL; CRM_ASSERT(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; pcmk_resource_t *ip = replica->ip; pcmk_resource_t *child = replica->child; pcmk_resource_t *container = replica->container; pcmk_resource_t *remote = replica->remote; char *id = NULL; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(container, only_node)) { continue; } print_ip = (ip != NULL) && !ip->private->fns->is_filtered(ip, only_rsc, print_everything); print_child = (child != NULL) && !child->private->fns->is_filtered(child, only_rsc, print_everything); print_ctnr = !container->private->fns->is_filtered(container, only_rsc, print_everything); print_remote = (remote != NULL) && !remote->private->fns->is_filtered(remote, only_rsc, print_everything); if (!print_everything && !print_ip && !print_child && !print_ctnr && !print_remote) { continue; } if (!printed_header) { const char *type = container_agent_str(bundle_data->agent_type); const char *unique = pcmk__flag_text(rsc->flags, pcmk__rsc_unique); const char *maintenance = pcmk__flag_text(rsc->flags, pcmk__rsc_maintenance); const char *managed = pcmk__flag_text(rsc->flags, pcmk__rsc_managed); const char *failed = pcmk__flag_text(rsc->flags, pcmk__rsc_failed); printed_header = TRUE; desc = pe__resource_description(rsc, show_opts); rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_BUNDLE, PCMK_XA_ID, rsc->id, PCMK_XA_TYPE, type, PCMK_XA_IMAGE, bundle_data->image, PCMK_XA_UNIQUE, unique, PCMK_XA_MAINTENANCE, maintenance, PCMK_XA_MANAGED, managed, PCMK_XA_FAILED, failed, PCMK_XA_DESCRIPTION, desc, NULL); CRM_ASSERT(rc == pcmk_rc_ok); } id = pcmk__itoa(replica->offset); rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_REPLICA, PCMK_XA_ID, id, NULL); free(id); CRM_ASSERT(rc == pcmk_rc_ok); if (print_ip) { out->message(out, (const char *) ip->private->xml->name, show_opts, ip, only_node, only_rsc); } if (print_child) { out->message(out, (const char *) child->private->xml->name, show_opts, child, only_node, only_rsc); } if (print_ctnr) { out->message(out, (const char *) container->private->xml->name, show_opts, container, only_node, only_rsc); } if (print_remote) { out->message(out, (const char *) remote->private->xml->name, show_opts, remote, only_node, only_rsc); } pcmk__output_xml_pop_parent(out); // replica } if (printed_header) { pcmk__output_xml_pop_parent(out); // bundle } return rc; } static void pe__bundle_replica_output_html(pcmk__output_t *out, pcmk__bundle_replica_t *replica, pcmk_node_t *node, uint32_t show_opts) { pcmk_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } pe__common_output_html(out, rsc, buffer, node, show_opts); } /*! * \internal * \brief Get a string describing a resource's unmanaged state or lack thereof * * \param[in] rsc Resource to describe * * \return A string indicating that a resource is in maintenance mode or * otherwise unmanaged, or an empty string otherwise */ static const char * get_unmanaged_str(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance)) { return " (maintenance)"; } if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { return " (unmanaged)"; } return ""; } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_html(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const char *desc = NULL; pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; CRM_ASSERT(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); desc = pe__resource_description(rsc, show_opts); if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; pcmk_resource_t *ip = replica->ip; pcmk_resource_t *child = replica->child; pcmk_resource_t *container = replica->container; pcmk_resource_t *remote = replica->remote; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(container, only_node)) { continue; } print_ip = (ip != NULL) && !ip->private->fns->is_filtered(ip, only_rsc, print_everything); print_child = (child != NULL) && !child->private->fns->is_filtered(child, only_rsc, print_everything); print_ctnr = !container->private->fns->is_filtered(container, only_rsc, print_everything); print_remote = (remote != NULL) && !remote->private->fns->is_filtered(remote, only_rsc, print_everything); if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) || (print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) { /* The text output messages used below require pe_print_implicit to * be set to do anything. */ uint32_t new_show_opts = show_opts | pcmk_show_implicit_rscs; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); if (pcmk__list_of_multiple(bundle_data->replicas)) { out->begin_list(out, NULL, NULL, "Replica[%d]", replica->offset); } if (print_ip) { out->message(out, (const char *) ip->private->xml->name, new_show_opts, ip, only_node, only_rsc); } if (print_child) { out->message(out, (const char *) child->private->xml->name, new_show_opts, child, only_node, only_rsc); } if (print_ctnr) { out->message(out, (const char *) container->private->xml->name, new_show_opts, container, only_node, only_rsc); } if (print_remote) { out->message(out, (const char *) remote->private->xml->name, new_show_opts, remote, only_node, only_rsc); } if (pcmk__list_of_multiple(bundle_data->replicas)) { out->end_list(out); } } else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) { continue; } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); pe__bundle_replica_output_html(out, replica, pcmk__current_node(container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } static void pe__bundle_replica_output_text(pcmk__output_t *out, pcmk__bundle_replica_t *replica, pcmk_node_t *node, uint32_t show_opts) { const pcmk_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } pe__common_output_text(out, rsc, buffer, node, show_opts); } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_text(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const char *desc = NULL; pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; desc = pe__resource_description(rsc, show_opts); CRM_ASSERT(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; pcmk_resource_t *ip = replica->ip; pcmk_resource_t *child = replica->child; pcmk_resource_t *container = replica->container; pcmk_resource_t *remote = replica->remote; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(container, only_node)) { continue; } print_ip = (ip != NULL) && !ip->private->fns->is_filtered(ip, only_rsc, print_everything); print_child = (child != NULL) && !child->private->fns->is_filtered(child, only_rsc, print_everything); print_ctnr = !container->private->fns->is_filtered(container, only_rsc, print_everything); print_remote = (remote != NULL) && !remote->private->fns->is_filtered(remote, only_rsc, print_everything); if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) || (print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) { /* The text output messages used below require pe_print_implicit to * be set to do anything. */ uint32_t new_show_opts = show_opts | pcmk_show_implicit_rscs; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); if (pcmk__list_of_multiple(bundle_data->replicas)) { out->list_item(out, NULL, "Replica[%d]", replica->offset); } out->begin_list(out, NULL, NULL, NULL); if (print_ip) { out->message(out, (const char *) ip->private->xml->name, new_show_opts, ip, only_node, only_rsc); } if (print_child) { out->message(out, (const char *) child->private->xml->name, new_show_opts, child, only_node, only_rsc); } if (print_ctnr) { out->message(out, (const char *) container->private->xml->name, new_show_opts, container, only_node, only_rsc); } if (print_remote) { out->message(out, (const char *) remote->private->xml->name, new_show_opts, remote, only_node, only_rsc); } out->end_list(out); } else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) { continue; } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); pe__bundle_replica_output_text(out, replica, pcmk__current_node(container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } static void free_bundle_replica(pcmk__bundle_replica_t *replica) { if (replica == NULL) { return; } if (replica->node) { free(replica->node); replica->node = NULL; } if (replica->ip) { pcmk__xml_free(replica->ip->private->xml); replica->ip->private->xml = NULL; replica->ip->private->fns->free(replica->ip); } if (replica->container) { pcmk__xml_free(replica->container->private->xml); replica->container->private->xml = NULL; replica->container->private->fns->free(replica->container); } if (replica->remote) { pcmk__xml_free(replica->remote->private->xml); replica->remote->private->xml = NULL; replica->remote->private->fns->free(replica->remote); } free(replica->ipaddr); free(replica); } void pe__free_bundle(pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); pcmk__rsc_trace(rsc, "Freeing %s", rsc->id); free(bundle_data->prefix); free(bundle_data->image); free(bundle_data->control_port); free(bundle_data->host_network); free(bundle_data->host_netmask); free(bundle_data->ip_range_start); free(bundle_data->container_network); free(bundle_data->launcher_options); free(bundle_data->container_command); g_free(bundle_data->container_host_options); g_list_free_full(bundle_data->replicas, (GDestroyNotify) free_bundle_replica); g_list_free_full(bundle_data->mounts, (GDestroyNotify)mount_free); g_list_free_full(bundle_data->ports, (GDestroyNotify)port_free); g_list_free(rsc->children); if(bundle_data->child) { pcmk__xml_free(bundle_data->child->private->xml); bundle_data->child->private->xml = NULL; bundle_data->child->private->fns->free(bundle_data->child); } common_free(rsc); } enum rsc_role_e pe__bundle_resource_state(const pcmk_resource_t *rsc, gboolean current) { enum rsc_role_e container_role = pcmk_role_unknown; return container_role; } /*! * \brief Get the number of configured replicas in a bundle * * \param[in] rsc Bundle resource * * \return Number of configured replicas, or 0 on error */ int pe_bundle_replicas(const pcmk_resource_t *rsc) { if (pcmk__is_bundle(rsc)) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); return bundle_data->nreplicas; } return 0; } void pe__count_bundle(pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); for (GList *item = bundle_data->replicas; item != NULL; item = item->next) { pcmk__bundle_replica_t *replica = item->data; if (replica->ip) { replica->ip->private->fns->count(replica->ip); } if (replica->child) { replica->child->private->fns->count(replica->child); } if (replica->container) { replica->container->private->fns->count(replica->container); } if (replica->remote) { replica->remote->private->fns->count(replica->remote); } } } gboolean pe__bundle_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent) { gboolean passes = FALSE; pe__bundle_variant_data_t *bundle_data = NULL; if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches)) { passes = TRUE; } else { get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; pcmk_resource_t *ip = replica->ip; pcmk_resource_t *child = replica->child; pcmk_resource_t *container = replica->container; pcmk_resource_t *remote = replica->remote; if ((ip != NULL) && !ip->private->fns->is_filtered(ip, only_rsc, FALSE)) { passes = TRUE; break; } if ((child != NULL) && !child->private->fns->is_filtered(child, only_rsc, FALSE)) { passes = TRUE; break; } if (!container->private->fns->is_filtered(container, only_rsc, FALSE)) { passes = TRUE; break; } if ((remote != NULL) && !remote->private->fns->is_filtered(remote, only_rsc, FALSE)) { passes = TRUE; break; } } } return !passes; } /*! * \internal * \brief Get a list of a bundle's containers * * \param[in] bundle Bundle resource * * \return Newly created list of \p bundle's containers * \note It is the caller's responsibility to free the result with * g_list_free(). */ GList * pe__bundle_containers(const pcmk_resource_t *bundle) { GList *containers = NULL; const pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, bundle); for (GList *iter = data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; containers = g_list_append(containers, replica->container); } return containers; } // Bundle implementation of pcmk__rsc_methods_t:active_node() pcmk_node_t * pe__bundle_active_node(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean) { pcmk_node_t *active = NULL; pcmk_node_t *node = NULL; pcmk_resource_t *container = NULL; GList *containers = NULL; GList *iter = NULL; GHashTable *nodes = NULL; const pe__bundle_variant_data_t *data = NULL; if (count_all != NULL) { *count_all = 0; } if (count_clean != NULL) { *count_clean = 0; } if (rsc == NULL) { return NULL; } /* For the purposes of this method, we only care about where the bundle's * containers are active, so build a list of active containers. */ get_bundle_variant_data(data, rsc); for (iter = data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; if (replica->container->running_on != NULL) { containers = g_list_append(containers, replica->container); } } if (containers == NULL) { return NULL; } /* If the bundle has only a single active container, just use that * container's method. If live migration is ever supported for bundle * containers, this will allow us to prefer the migration source when there * is only one container and it is migrating. For now, this just lets us * avoid creating the nodes table. */ if (pcmk__list_of_1(containers)) { container = containers->data; node = container->private->fns->active_node(container, count_all, count_clean); g_list_free(containers); return node; } // Add all containers' active nodes to a hash table (for uniqueness) nodes = g_hash_table_new(NULL, NULL); for (iter = containers; iter != NULL; iter = iter->next) { container = iter->data; for (GList *node_iter = container->running_on; node_iter != NULL; node_iter = node_iter->next) { node = node_iter->data; // If insert returns true, we haven't counted this node yet if (g_hash_table_insert(nodes, (gpointer) node->details, (gpointer) node) && !pe__count_active_node(rsc, node, &active, count_all, count_clean)) { goto done; } } } done: g_list_free(containers); g_hash_table_destroy(nodes); return active; } /*! * \internal * \brief Get maximum bundle resource instances per node * * \param[in] rsc Bundle resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int pe__bundle_max_per_node(const pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); CRM_ASSERT(bundle_data->nreplicas_per_host >= 0); return (unsigned int) bundle_data->nreplicas_per_host; } diff --git a/lib/pengine/complex.c b/lib/pengine/complex.c index f4426398a6..ca64f9377e 100644 --- a/lib/pengine/complex.c +++ b/lib/pengine/complex.c @@ -1,1268 +1,1268 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include "pe_status_private.h" void populate_hash(xmlNode * nvpair_list, GHashTable * hash, const char **attrs, int attrs_length); static pcmk_node_t *active_node(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean); static pcmk__rsc_methods_t resource_class_functions[] = { { native_unpack, native_find_rsc, native_parameter, native_active, native_resource_state, native_location, native_free, pe__count_common, pe__native_is_filtered, active_node, pe__primitive_max_per_node, }, { group_unpack, native_find_rsc, native_parameter, group_active, group_resource_state, native_location, group_free, pe__count_common, pe__group_is_filtered, active_node, pe__group_max_per_node, }, { clone_unpack, native_find_rsc, native_parameter, clone_active, clone_resource_state, native_location, clone_free, pe__count_common, pe__clone_is_filtered, active_node, pe__clone_max_per_node, }, { pe__unpack_bundle, native_find_rsc, native_parameter, pe__bundle_active, pe__bundle_resource_state, native_location, pe__free_bundle, pe__count_bundle, pe__bundle_is_filtered, pe__bundle_active_node, pe__bundle_max_per_node, } }; static enum pcmk__rsc_variant get_resource_type(const char *name) { if (pcmk__str_eq(name, PCMK_XE_PRIMITIVE, pcmk__str_casei)) { return pcmk__rsc_variant_primitive; } else if (pcmk__str_eq(name, PCMK_XE_GROUP, pcmk__str_casei)) { return pcmk__rsc_variant_group; } else if (pcmk__str_eq(name, PCMK_XE_CLONE, pcmk__str_casei)) { return pcmk__rsc_variant_clone; } else if (pcmk__str_eq(name, PCMK__XE_PROMOTABLE_LEGACY, pcmk__str_casei)) { // @COMPAT deprecated since 2.0.0 return pcmk__rsc_variant_clone; } else if (pcmk__str_eq(name, PCMK_XE_BUNDLE, pcmk__str_casei)) { return pcmk__rsc_variant_bundle; } return pcmk__rsc_variant_unknown; } /*! * \internal * \brief Insert a meta-attribute if not already present * * \param[in] key Meta-attribute name * \param[in] value Meta-attribute value to add if not already present * \param[in,out] table Meta-attribute hash table to insert into * * \note This is like pcmk__insert_meta() except it won't overwrite existing * values. */ static void dup_attr(gpointer key, gpointer value, gpointer user_data) { GHashTable *table = user_data; CRM_CHECK((key != NULL) && (table != NULL), return); if (pcmk__str_eq((const char *) value, "#default", pcmk__str_casei)) { // @COMPAT Deprecated since 2.1.8 pcmk__config_warn("Support for setting meta-attributes (such as %s) to " "the explicit value '#default' is deprecated and " "will be removed in a future release", (const char *) key); } else if ((value != NULL) && (g_hash_table_lookup(table, key) == NULL)) { pcmk__insert_dup(table, (const char *) key, (const char *) value); } } static void expand_parents_fixed_nvpairs(pcmk_resource_t *rsc, pe_rule_eval_data_t *rule_data, GHashTable *meta_hash, pcmk_scheduler_t *scheduler) { GHashTable *parent_orig_meta = pcmk__strkey_table(free, free); pcmk_resource_t *p = rsc->private->parent; if (p == NULL) { return ; } /* Search all parent resources, get the fixed value of * PCMK_XE_META_ATTRIBUTES set only in the original xml, and stack it in the * hash table. The fixed value of the lower parent resource takes precedence * and is not overwritten. */ while(p != NULL) { /* A hash table for comparison is generated, including the id-ref. */ pe__unpack_dataset_nvpairs(p->private->xml, PCMK_XE_META_ATTRIBUTES, rule_data, parent_orig_meta, NULL, FALSE, scheduler); p = p->private->parent; } if (parent_orig_meta != NULL) { // This will not overwrite any values already existing for child g_hash_table_foreach(parent_orig_meta, dup_attr, meta_hash); } if (parent_orig_meta != NULL) { g_hash_table_destroy(parent_orig_meta); } return ; } void get_meta_attributes(GHashTable * meta_hash, pcmk_resource_t * rsc, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pe_rsc_eval_data_t rsc_rule_data = { .standard = crm_element_value(rsc->private->xml, PCMK_XA_CLASS), .provider = crm_element_value(rsc->private->xml, PCMK_XA_PROVIDER), .agent = crm_element_value(rsc->private->xml, PCMK_XA_TYPE) }; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->now, .match_data = NULL, .rsc_data = &rsc_rule_data, .op_data = NULL }; if (node) { /* @COMPAT Support for node attribute expressions in rules for * meta-attributes is deprecated. When we can break behavioral backward * compatibility, drop this block. */ rule_data.node_hash = node->details->attrs; } for (xmlAttrPtr a = pcmk__xe_first_attr(rsc->private->xml); a != NULL; a = a->next) { if (a->children != NULL) { dup_attr((gpointer) a->name, (gpointer) a->children->content, meta_hash); } } pe__unpack_dataset_nvpairs(rsc->private->xml, PCMK_XE_META_ATTRIBUTES, &rule_data, meta_hash, NULL, FALSE, scheduler); /* Set the PCMK_XE_META_ATTRIBUTES explicitly set in the parent resource to * the hash table of the child resource. If it is already explicitly set as * a child, it will not be overwritten. */ if (rsc->private->parent != NULL) { expand_parents_fixed_nvpairs(rsc, &rule_data, meta_hash, scheduler); } /* check the defaults */ pe__unpack_dataset_nvpairs(scheduler->rsc_defaults, PCMK_XE_META_ATTRIBUTES, &rule_data, meta_hash, NULL, FALSE, scheduler); /* If there is PCMK_XE_META_ATTRIBUTES that the parent resource has not * explicitly set, set a value that is not set from PCMK_XE_RSC_DEFAULTS * either. The values already set up to this point will not be overwritten. */ if (rsc->private->parent != NULL) { g_hash_table_foreach(rsc->private->parent->meta, dup_attr, meta_hash); } } void get_rsc_attributes(GHashTable *meta_hash, const pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; if (node) { rule_data.node_hash = node->details->attrs; } pe__unpack_dataset_nvpairs(rsc->private->xml, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, meta_hash, NULL, FALSE, scheduler); /* set anything else based on the parent */ if (rsc->private->parent != NULL) { get_rsc_attributes(meta_hash, rsc->private->parent, node, scheduler); } else { if (pcmk__xe_first_child(scheduler->rsc_defaults, PCMK_XE_INSTANCE_ATTRIBUTES, NULL, NULL) != NULL) { /* Not possible with schema validation enabled * * @COMPAT Drop support when we can break behavioral * backward compatibility */ pcmk__warn_once(pcmk__wo_instance_defaults, "Support for " PCMK_XE_INSTANCE_ATTRIBUTES " in " PCMK_XE_RSC_DEFAULTS " is deprecated and will be " "removed in a future release"); } /* and finally check the defaults */ pe__unpack_dataset_nvpairs(scheduler->rsc_defaults, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, meta_hash, NULL, FALSE, scheduler); } } static char * template_op_key(xmlNode * op) { const char *name = crm_element_value(op, PCMK_XA_NAME); const char *role = crm_element_value(op, PCMK_XA_ROLE); char *key = NULL; if ((role == NULL) || pcmk__strcase_any_of(role, PCMK_ROLE_STARTED, PCMK_ROLE_UNPROMOTED, PCMK__ROLE_UNPROMOTED_LEGACY, NULL)) { role = PCMK__ROLE_UNKNOWN; } key = crm_strdup_printf("%s-%s", name, role); return key; } static gboolean unpack_template(xmlNode *xml_obj, xmlNode **expanded_xml, pcmk_scheduler_t *scheduler) { xmlNode *cib_resources = NULL; xmlNode *template = NULL; xmlNode *new_xml = NULL; xmlNode *child_xml = NULL; xmlNode *rsc_ops = NULL; xmlNode *template_ops = NULL; const char *template_ref = NULL; const char *id = NULL; if (xml_obj == NULL) { pcmk__config_err("No resource object for template unpacking"); return FALSE; } template_ref = crm_element_value(xml_obj, PCMK_XA_TEMPLATE); if (template_ref == NULL) { return TRUE; } id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("'%s' object must have a id", xml_obj->name); return FALSE; } if (pcmk__str_eq(template_ref, id, pcmk__str_none)) { pcmk__config_err("The resource object '%s' should not reference itself", id); return FALSE; } cib_resources = get_xpath_object("//" PCMK_XE_RESOURCES, scheduler->input, LOG_TRACE); if (cib_resources == NULL) { pcmk__config_err("No resources configured"); return FALSE; } template = pcmk__xe_first_child(cib_resources, PCMK_XE_TEMPLATE, PCMK_XA_ID, template_ref); if (template == NULL) { pcmk__config_err("No template named '%s'", template_ref); return FALSE; } new_xml = pcmk__xml_copy(NULL, template); xmlNodeSetName(new_xml, xml_obj->name); crm_xml_add(new_xml, PCMK_XA_ID, id); crm_xml_add(new_xml, PCMK__META_CLONE, crm_element_value(xml_obj, PCMK__META_CLONE)); template_ops = pcmk__xe_first_child(new_xml, PCMK_XE_OPERATIONS, NULL, NULL); for (child_xml = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL); child_xml != NULL; child_xml = pcmk__xe_next(child_xml)) { xmlNode *new_child = pcmk__xml_copy(new_xml, child_xml); if (pcmk__xe_is(new_child, PCMK_XE_OPERATIONS)) { rsc_ops = new_child; } } if (template_ops && rsc_ops) { xmlNode *op = NULL; GHashTable *rsc_ops_hash = pcmk__strkey_table(free, NULL); for (op = pcmk__xe_first_child(rsc_ops, NULL, NULL, NULL); op != NULL; op = pcmk__xe_next(op)) { char *key = template_op_key(op); g_hash_table_insert(rsc_ops_hash, key, op); } for (op = pcmk__xe_first_child(template_ops, NULL, NULL, NULL); op != NULL; op = pcmk__xe_next(op)) { char *key = template_op_key(op); if (g_hash_table_lookup(rsc_ops_hash, key) == NULL) { pcmk__xml_copy(rsc_ops, op); } free(key); } if (rsc_ops_hash) { g_hash_table_destroy(rsc_ops_hash); } pcmk__xml_free(template_ops); } /*pcmk__xml_free(*expanded_xml); */ *expanded_xml = new_xml; #if 0 /* Disable multi-level templates for now */ if (!unpack_template(new_xml, expanded_xml, scheduler)) { pcmk__xml_free(*expanded_xml); *expanded_xml = NULL; return FALSE; } #endif return TRUE; } static gboolean add_template_rsc(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { const char *template_ref = NULL; const char *id = NULL; if (xml_obj == NULL) { pcmk__config_err("No resource object for processing resource list " "of template"); return FALSE; } template_ref = crm_element_value(xml_obj, PCMK_XA_TEMPLATE); if (template_ref == NULL) { return TRUE; } id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("'%s' object must have a id", xml_obj->name); return FALSE; } if (pcmk__str_eq(template_ref, id, pcmk__str_none)) { pcmk__config_err("The resource object '%s' should not reference itself", id); return FALSE; } if (add_tag_ref(scheduler->template_rsc_sets, template_ref, id) == FALSE) { return FALSE; } return TRUE; } static bool detect_promotable(pcmk_resource_t *rsc) { const char *promotable = g_hash_table_lookup(rsc->meta, PCMK_META_PROMOTABLE); if (crm_is_true(promotable)) { return TRUE; } // @COMPAT deprecated since 2.0.0 if (pcmk__xe_is(rsc->private->xml, PCMK__XE_PROMOTABLE_LEGACY)) { pcmk__warn_once(pcmk__wo_master_element, "Support for <" PCMK__XE_PROMOTABLE_LEGACY "> (such " "as in %s) is deprecated and will be removed in a " "future release. Use <" PCMK_XE_CLONE "> with a " PCMK_META_PROMOTABLE " meta-attribute instead.", rsc->id); pcmk__insert_dup(rsc->meta, PCMK_META_PROMOTABLE, PCMK_VALUE_TRUE); return TRUE; } return FALSE; } static void free_params_table(gpointer data) { g_hash_table_destroy((GHashTable *) data); } /*! * \brief Get a table of resource parameters * * \param[in,out] rsc Resource to query * \param[in] node Node for evaluating rules (NULL for defaults) * \param[in,out] scheduler Scheduler data * * \return Hash table containing resource parameter names and values * (or NULL if \p rsc or \p scheduler is NULL) * \note The returned table will be destroyed when the resource is freed, so * callers should not destroy it. */ GHashTable * pe_rsc_params(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { GHashTable *params_on_node = NULL; /* A NULL node is used to request the resource's default parameters * (not evaluated for node), but we always want something non-NULL * as a hash table key. */ const char *node_name = ""; // Sanity check if ((rsc == NULL) || (scheduler == NULL)) { return NULL; } if ((node != NULL) && (node->details->uname != NULL)) { node_name = node->details->uname; } // Find the parameter table for given node if (rsc->parameter_cache == NULL) { rsc->parameter_cache = pcmk__strikey_table(free, free_params_table); } else { params_on_node = g_hash_table_lookup(rsc->parameter_cache, node_name); } // If none exists yet, create one with parameters evaluated for node if (params_on_node == NULL) { params_on_node = pcmk__strkey_table(free, free); get_rsc_attributes(params_on_node, rsc, node, scheduler); g_hash_table_insert(rsc->parameter_cache, strdup(node_name), params_on_node); } return params_on_node; } /*! * \internal * \brief Unpack a resource's \c PCMK_META_REQUIRES meta-attribute * * \param[in,out] rsc Resource being unpacked * \param[in] value Value of \c PCMK_META_REQUIRES meta-attribute * \param[in] is_default Whether \p value was selected by default */ static void unpack_requires(pcmk_resource_t *rsc, const char *value, bool is_default) { const pcmk_scheduler_t *scheduler = rsc->private->scheduler; if (pcmk__str_eq(value, PCMK_VALUE_NOTHING, pcmk__str_casei)) { } else if (pcmk__str_eq(value, PCMK_VALUE_QUORUM, pcmk__str_casei)) { pcmk__set_rsc_flags(rsc, pcmk__rsc_needs_quorum); } else if (pcmk__str_eq(value, PCMK_VALUE_FENCING, pcmk__str_casei)) { pcmk__set_rsc_flags(rsc, pcmk__rsc_needs_fencing); if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("%s requires fencing but fencing is disabled", rsc->id); } } else if (pcmk__str_eq(value, PCMK_VALUE_UNFENCING, pcmk__str_casei)) { if (pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) { pcmk__config_warn("Resetting \"" PCMK_META_REQUIRES "\" for %s " "to \"" PCMK_VALUE_QUORUM "\" because fencing " "devices cannot require unfencing", rsc->id); unpack_requires(rsc, PCMK_VALUE_QUORUM, true); return; } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("Resetting \"" PCMK_META_REQUIRES "\" for %s " "to \"" PCMK_VALUE_QUORUM "\" because fencing is " "disabled", rsc->id); unpack_requires(rsc, PCMK_VALUE_QUORUM, true); return; } else { pcmk__set_rsc_flags(rsc, pcmk__rsc_needs_fencing |pcmk__rsc_needs_unfencing); } } else { const char *orig_value = value; if (pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) { value = PCMK_VALUE_QUORUM; } else if (pcmk__is_primitive(rsc) && xml_contains_remote_node(rsc->private->xml)) { value = PCMK_VALUE_QUORUM; } else if (pcmk_is_set(scheduler->flags, pcmk_sched_enable_unfencing)) { value = PCMK_VALUE_UNFENCING; } else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { value = PCMK_VALUE_FENCING; } else if (scheduler->no_quorum_policy == pcmk_no_quorum_ignore) { value = PCMK_VALUE_NOTHING; } else { value = PCMK_VALUE_QUORUM; } if (orig_value != NULL) { pcmk__config_err("Resetting '" PCMK_META_REQUIRES "' for %s " "to '%s' because '%s' is not valid", rsc->id, value, orig_value); } unpack_requires(rsc, value, true); return; } pcmk__rsc_trace(rsc, "\tRequired to start: %s%s", value, (is_default? " (default)" : "")); } static void warn_about_deprecated_classes(pcmk_resource_t *rsc) { const char *std = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); if (pcmk__str_eq(std, PCMK_RESOURCE_CLASS_UPSTART, pcmk__str_none)) { pcmk__warn_once(pcmk__wo_upstart, "Support for Upstart resources (such as %s) is " "deprecated and will be removed in a future release", rsc->id); } else if (pcmk__str_eq(std, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_none)) { pcmk__warn_once(pcmk__wo_nagios, "Support for Nagios resources (such as %s) is " "deprecated and will be removed in a future release", rsc->id); } } /*! * \internal * \brief Unpack configuration XML for a given resource * * Unpack the XML object containing a resource's configuration into a new * \c pcmk_resource_t object. * * \param[in] xml_obj XML node containing the resource's configuration * \param[out] rsc Where to store the unpacked resource information * \param[in] parent Resource's parent, if any * \param[in,out] scheduler Scheduler data * * \return Standard Pacemaker return code * \note If pcmk_rc_ok is returned, \p *rsc is guaranteed to be non-NULL, and * the caller is responsible for freeing it using its variant-specific * free() method. Otherwise, \p *rsc is guaranteed to be NULL. */ int pe__unpack_resource(xmlNode *xml_obj, pcmk_resource_t **rsc, pcmk_resource_t *parent, pcmk_scheduler_t *scheduler) { xmlNode *expanded_xml = NULL; xmlNode *ops = NULL; const char *value = NULL; const char *id = NULL; bool guest_node = false; bool remote_node = false; pcmk__resource_private_t *rsc_private = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = NULL, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; CRM_CHECK(rsc != NULL, return EINVAL); CRM_CHECK((xml_obj != NULL) && (scheduler != NULL), *rsc = NULL; return EINVAL); rule_data.now = scheduler->now; crm_log_xml_trace(xml_obj, "[raw XML]"); id = crm_element_value(xml_obj, PCMK_XA_ID); if (id == NULL) { pcmk__config_err("Ignoring <%s> configuration without " PCMK_XA_ID, xml_obj->name); return pcmk_rc_unpack_error; } if (unpack_template(xml_obj, &expanded_xml, scheduler) == FALSE) { return pcmk_rc_unpack_error; } *rsc = calloc(1, sizeof(pcmk_resource_t)); if (*rsc == NULL) { pcmk__sched_err("Unable to allocate memory for resource '%s'", id); return ENOMEM; } (*rsc)->private = calloc(1, sizeof(pcmk__resource_private_t)); if ((*rsc)->private == NULL) { pcmk__sched_err("Unable to allocate memory for resource '%s'", id); free(*rsc); return ENOMEM; } rsc_private = (*rsc)->private; rsc_private->scheduler = scheduler; if (expanded_xml) { crm_log_xml_trace(expanded_xml, "[expanded XML]"); rsc_private->xml = expanded_xml; rsc_private->orig_xml = xml_obj; } else { rsc_private->xml = xml_obj; rsc_private->orig_xml = NULL; } /* Do not use xml_obj from here on, use (*rsc)->xml in case templates are involved */ rsc_private->parent = parent; ops = pcmk__xe_first_child(rsc_private->xml, PCMK_XE_OPERATIONS, NULL, NULL); rsc_private->ops_xml = pcmk__xe_resolve_idref(ops, scheduler->input); rsc_private->variant = get_resource_type((const char *) rsc_private->xml->name); if (rsc_private->variant == pcmk__rsc_variant_unknown) { pcmk__config_err("Ignoring resource '%s' of unknown type '%s'", id, rsc_private->xml->name); common_free(*rsc); *rsc = NULL; return pcmk_rc_unpack_error; } (*rsc)->meta = pcmk__strkey_table(free, free); (*rsc)->allowed_nodes = pcmk__strkey_table(NULL, free); (*rsc)->known_on = pcmk__strkey_table(NULL, free); value = crm_element_value(rsc_private->xml, PCMK__META_CLONE); if (value) { (*rsc)->id = crm_strdup_printf("%s:%s", id, value); pcmk__insert_meta(*rsc, PCMK__META_CLONE, value); } else { (*rsc)->id = strdup(id); } warn_about_deprecated_classes(*rsc); rsc_private->fns = &resource_class_functions[rsc_private->variant]; get_meta_attributes((*rsc)->meta, *rsc, NULL, scheduler); (*rsc)->parameters = pe_rsc_params(*rsc, NULL, scheduler); // \deprecated (*rsc)->flags = 0; pcmk__set_rsc_flags(*rsc, pcmk__rsc_unassigned); if (!pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_managed); } (*rsc)->role = pcmk_role_stopped; (*rsc)->next_role = pcmk_role_unknown; rsc_private->ban_after_failures = PCMK_SCORE_INFINITY; value = g_hash_table_lookup((*rsc)->meta, PCMK_META_PRIORITY); rsc_private->priority = char2score(value); value = g_hash_table_lookup((*rsc)->meta, PCMK_META_CRITICAL); if ((value == NULL) || crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_critical); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_NOTIFY); if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_notify); } if (xml_contains_remote_node(rsc_private->xml)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_is_remote_connection); if (g_hash_table_lookup((*rsc)->meta, PCMK__META_CONTAINER)) { guest_node = true; } else { remote_node = true; } } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_ALLOW_MIGRATE); if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_migratable); } else if ((value == NULL) && remote_node) { /* By default, we want remote nodes to be able * to float around the cluster without having to stop all the * resources within the remote-node before moving. Allowing * migration support enables this feature. If this ever causes * problems, migration support can be explicitly turned off with * PCMK_META_ALLOW_MIGRATE=false. */ pcmk__set_rsc_flags(*rsc, pcmk__rsc_migratable); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_IS_MANAGED); if (value != NULL) { if (pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) { // @COMPAT Deprecated since 2.1.8 pcmk__config_warn("Support for setting " PCMK_META_IS_MANAGED " to the explicit value '" PCMK_VALUE_DEFAULT "' is deprecated and will be removed in a " "future release (just leave it unset)"); } else if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_managed); } else { pcmk__clear_rsc_flags(*rsc, pcmk__rsc_managed); } } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_MAINTENANCE); if (crm_is_true(value)) { pcmk__clear_rsc_flags(*rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(*rsc, pcmk__rsc_maintenance); } if (pcmk_is_set(scheduler->flags, pcmk_sched_in_maintenance)) { pcmk__clear_rsc_flags(*rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(*rsc, pcmk__rsc_maintenance); } if (pcmk__is_clone(pe__const_top_resource(*rsc, false))) { value = g_hash_table_lookup((*rsc)->meta, PCMK_META_GLOBALLY_UNIQUE); if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_unique); } if (detect_promotable(*rsc)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_promotable); } } else { pcmk__set_rsc_flags(*rsc, pcmk__rsc_unique); } // @COMPAT Deprecated meta-attribute value = g_hash_table_lookup((*rsc)->meta, PCMK__META_RESTART_TYPE); if (pcmk__str_eq(value, PCMK_VALUE_RESTART, pcmk__str_casei)) { rsc_private->restart_type = pcmk__restart_restart; pcmk__rsc_trace(*rsc, "%s dependency restart handling: restart", (*rsc)->id); pcmk__warn_once(pcmk__wo_restart_type, "Support for " PCMK__META_RESTART_TYPE " is deprecated " "and will be removed in a future release"); } else { rsc_private->restart_type = pcmk__restart_ignore; pcmk__rsc_trace(*rsc, "%s dependency restart handling: ignore", (*rsc)->id); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_MULTIPLE_ACTIVE); if (pcmk__str_eq(value, PCMK_VALUE_STOP_ONLY, pcmk__str_casei)) { rsc_private->multiply_active_policy = pcmk__multiply_active_stop; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: stop only", (*rsc)->id); } else if (pcmk__str_eq(value, PCMK_VALUE_BLOCK, pcmk__str_casei)) { rsc_private->multiply_active_policy = pcmk__multiply_active_block; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: block", (*rsc)->id); } else if (pcmk__str_eq(value, PCMK_VALUE_STOP_UNEXPECTED, pcmk__str_casei)) { rsc_private->multiply_active_policy = pcmk__multiply_active_unexpected; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: " "stop unexpected instances", (*rsc)->id); } else { // PCMK_VALUE_STOP_START if (!pcmk__str_eq(value, PCMK_VALUE_STOP_START, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_warn("%s is not a valid value for " PCMK_META_MULTIPLE_ACTIVE ", using default of " "\"" PCMK_VALUE_STOP_START "\"", value); } rsc_private->multiply_active_policy = pcmk__multiply_active_restart; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: stop/start", (*rsc)->id); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_RESOURCE_STICKINESS); if (value != NULL) { if (pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) { // @COMPAT Deprecated since 2.1.8 pcmk__config_warn("Support for setting " PCMK_META_RESOURCE_STICKINESS " to the explicit value '" PCMK_VALUE_DEFAULT "' is deprecated and will be removed in a " "future release (just leave it unset)"); } else { rsc_private->stickiness = char2score(value); } } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_MIGRATION_THRESHOLD); if (value != NULL) { if (pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) { // @COMPAT Deprecated since 2.1.8 pcmk__config_warn("Support for setting " PCMK_META_MIGRATION_THRESHOLD " to the explicit value '" PCMK_VALUE_DEFAULT "' is deprecated and will be removed in a " "future release (just leave it unset)"); } else { rsc_private->ban_after_failures = char2score(value); if (rsc_private->ban_after_failures < 0) { /* @COMPAT We use 1 here to preserve previous behavior, but this * should probably use the default (INFINITY) or 0 (to disable) * instead. */ pcmk__warn_once(pcmk__wo_neg_threshold, PCMK_META_MIGRATION_THRESHOLD " must be non-negative, using 1 instead"); rsc_private->ban_after_failures = 1; } } } if (pcmk__str_eq(crm_element_value(rsc_private->xml, PCMK_XA_CLASS), PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_fencing); pcmk__set_rsc_flags(*rsc, pcmk__rsc_fence_device); } value = g_hash_table_lookup((*rsc)->meta, PCMK_META_REQUIRES); unpack_requires(*rsc, value, false); value = g_hash_table_lookup((*rsc)->meta, PCMK_META_FAILURE_TIMEOUT); if (value != NULL) { pcmk_parse_interval_spec(value, &(rsc_private->failure_expiration_ms)); } if (remote_node) { GHashTable *params = pe_rsc_params(*rsc, NULL, scheduler); /* Grabbing the value now means that any rules based on node attributes * will evaluate to false, so such rules should not be used with * PCMK_REMOTE_RA_RECONNECT_INTERVAL. * * @TODO Evaluate per node before using */ value = g_hash_table_lookup(params, PCMK_REMOTE_RA_RECONNECT_INTERVAL); if (value) { /* reconnect delay works by setting failure_timeout and preventing the * connection from starting until the failure is cleared. */ pcmk_parse_interval_spec(value, &(rsc_private->remote_reconnect_ms)); /* We want to override any default failure_timeout in use when remote * PCMK_REMOTE_RA_RECONNECT_INTERVAL is in use. */ rsc_private->failure_expiration_ms = rsc_private->remote_reconnect_ms; } } get_target_role(*rsc, &((*rsc)->next_role)); pcmk__rsc_trace(*rsc, "%s desired next state: %s", (*rsc)->id, ((*rsc)->next_role == pcmk_role_unknown)? "default" : pcmk_role_text((*rsc)->next_role)); if (rsc_private->fns->unpack(*rsc, scheduler) == FALSE) { rsc_private->fns->free(*rsc); *rsc = NULL; return pcmk_rc_unpack_error; } if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) { // This tag must stay exactly the same because it is tested elsewhere resource_location(*rsc, NULL, 0, "symmetric_default", scheduler); } else if (guest_node) { /* remote resources tied to a container resource must always be allowed * to opt-in to the cluster. Whether the connection resource is actually * allowed to be placed on a node is dependent on the container resource */ resource_location(*rsc, NULL, 0, "remote_connection_default", scheduler); } pcmk__rsc_trace(*rsc, "%s action notification: %s", (*rsc)->id, pcmk_is_set((*rsc)->flags, pcmk__rsc_notify)? "required" : "not required"); (*rsc)->utilization = pcmk__strkey_table(free, free); pe__unpack_dataset_nvpairs(rsc_private->xml, PCMK_XE_UTILIZATION, &rule_data, (*rsc)->utilization, NULL, FALSE, scheduler); if (expanded_xml) { if (add_template_rsc(xml_obj, scheduler) == FALSE) { rsc_private->fns->free(*rsc); *rsc = NULL; return pcmk_rc_unpack_error; } } return pcmk_rc_ok; } gboolean is_parent(pcmk_resource_t *child, pcmk_resource_t *rsc) { pcmk_resource_t *parent = child; if (parent == NULL || rsc == NULL) { return FALSE; } while (parent->private->parent != NULL) { if (parent->private->parent == rsc) { return TRUE; } parent = parent->private->parent; } return FALSE; } pcmk_resource_t * uber_parent(pcmk_resource_t *rsc) { pcmk_resource_t *parent = rsc; if (parent == NULL) { return NULL; } while ((parent->private->parent != NULL) && !pcmk__is_bundle(parent->private->parent)) { parent = parent->private->parent; } return parent; } /*! * \internal * \brief Get the topmost parent of a resource as a const pointer * * \param[in] rsc Resource to check * \param[in] include_bundle If true, go all the way to bundle * * \return \p NULL if \p rsc is NULL, \p rsc if \p rsc has no parent, * the bundle if \p rsc is bundled and \p include_bundle is true, * otherwise the topmost parent of \p rsc up to a clone */ const pcmk_resource_t * pe__const_top_resource(const pcmk_resource_t *rsc, bool include_bundle) { const pcmk_resource_t *parent = rsc; if (parent == NULL) { return NULL; } while (parent->private->parent != NULL) { if (!include_bundle && pcmk__is_bundle(parent->private->parent)) { break; } parent = parent->private->parent; } return parent; } void common_free(pcmk_resource_t * rsc) { if (rsc == NULL) { return; } pcmk__rsc_trace(rsc, "Freeing %s %s", (const char *) rsc->private->xml->name, rsc->id); g_list_free(rsc->dangling_migrations); if (rsc->parameter_cache != NULL) { g_hash_table_destroy(rsc->parameter_cache); } if (rsc->meta != NULL) { g_hash_table_destroy(rsc->meta); } if (rsc->utilization != NULL) { g_hash_table_destroy(rsc->utilization); } if ((rsc->private->parent == NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { pcmk__xml_free(rsc->private->xml); rsc->private->xml = NULL; pcmk__xml_free(rsc->private->orig_xml); rsc->private->orig_xml = NULL; } else if (rsc->private->orig_xml != NULL) { // rsc->private->xml was expanded from a template pcmk__xml_free(rsc->private->xml); rsc->private->xml = NULL; } if (rsc->running_on) { g_list_free(rsc->running_on); rsc->running_on = NULL; } if (rsc->known_on) { g_hash_table_destroy(rsc->known_on); rsc->known_on = NULL; } if (rsc->allowed_nodes) { g_hash_table_destroy(rsc->allowed_nodes); rsc->allowed_nodes = NULL; } g_list_free(rsc->fillers); free(rsc->id); - free(rsc->allocated_to); free(rsc->private->variant_opaque); free(rsc->private->history_id); free(rsc->private->pending_action); + free(rsc->private->assigned_node); g_list_free(rsc->private->actions); g_list_free(rsc->private->with_this_colocations); g_list_free(rsc->private->this_with_colocations); g_list_free(rsc->private->location_constraints); g_list_free(rsc->private->ticket_constraints); free(rsc->private); free(rsc); } /*! * \internal * \brief Count a node and update most preferred to it as appropriate * * \param[in] rsc An active resource * \param[in] node A node that \p rsc is active on * \param[in,out] active This will be set to \p node if \p node is more * preferred than the current value * \param[in,out] count_all If not NULL, this will be incremented * \param[in,out] count_clean If not NULL, this will be incremented if \p node * is online and clean * * \return true if the count should continue, or false if sufficiently known */ bool pe__count_active_node(const pcmk_resource_t *rsc, pcmk_node_t *node, pcmk_node_t **active, unsigned int *count_all, unsigned int *count_clean) { bool keep_looking = false; bool is_happy = false; CRM_CHECK((rsc != NULL) && (node != NULL) && (active != NULL), return false); is_happy = node->details->online && !node->details->unclean; if (count_all != NULL) { ++*count_all; } if ((count_clean != NULL) && is_happy) { ++*count_clean; } if ((count_all != NULL) || (count_clean != NULL)) { keep_looking = true; // We're counting, so go through entire list } if (rsc->partial_migration_source != NULL) { if (pcmk__same_node(node, rsc->partial_migration_source)) { *active = node; // This is the migration source } else { keep_looking = true; } } else if (!pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)) { if (is_happy && ((*active == NULL) || !(*active)->details->online || (*active)->details->unclean)) { *active = node; // This is the first clean node } else { keep_looking = true; } } if (*active == NULL) { *active = node; // This is the first node checked } return keep_looking; } // Shared implementation of pcmk__rsc_methods_t:active_node() static pcmk_node_t * active_node(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean) { pcmk_node_t *active = NULL; if (count_all != NULL) { *count_all = 0; } if (count_clean != NULL) { *count_clean = 0; } if (rsc == NULL) { return NULL; } for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { if (!pe__count_active_node(rsc, (pcmk_node_t *) iter->data, &active, count_all, count_clean)) { break; // Don't waste time iterating if we don't have to } } return active; } /*! * \brief * \internal Find and count active nodes according to \c PCMK_META_REQUIRES * * \param[in] rsc Resource to check * \param[out] count If not NULL, will be set to count of active nodes * * \return An active node (or NULL if resource is not active anywhere) * * \note This is a convenience wrapper for active_node() where the count of all * active nodes or only clean active nodes is desired according to the * \c PCMK_META_REQUIRES meta-attribute. */ pcmk_node_t * pe__find_active_requires(const pcmk_resource_t *rsc, unsigned int *count) { if (rsc == NULL) { if (count != NULL) { *count = 0; } return NULL; } if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)) { return rsc->private->fns->active_node(rsc, count, NULL); } else { return rsc->private->fns->active_node(rsc, NULL, count); } } void pe__count_common(pcmk_resource_t *rsc) { if (rsc->children != NULL) { for (GList *item = rsc->children; item != NULL; item = item->next) { pcmk_resource_t *child = item->data; child->private->fns->count(item->data); } } else if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed) || (rsc->role > pcmk_role_stopped)) { rsc->private->scheduler->ninstances++; if (pe__resource_is_disabled(rsc)) { rsc->private->scheduler->disabled_resources++; } if (pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) { rsc->private->scheduler->blocked_resources++; } } } /*! * \internal * \brief Update a resource's next role * * \param[in,out] rsc Resource to be updated * \param[in] role Resource's new next role * \param[in] why Human-friendly reason why role is changing (for logs) */ void pe__set_next_role(pcmk_resource_t *rsc, enum rsc_role_e role, const char *why) { CRM_ASSERT((rsc != NULL) && (why != NULL)); if (rsc->next_role != role) { pcmk__rsc_trace(rsc, "Resetting next role for %s from %s to %s (%s)", rsc->id, pcmk_role_text(rsc->next_role), pcmk_role_text(role), why); rsc->next_role = role; } } diff --git a/lib/pengine/native.c b/lib/pengine/native.c index 12a7c2b7ff..b8c9da7c2b 100644 --- a/lib/pengine/native.c +++ b/lib/pengine/native.c @@ -1,1170 +1,1170 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include /*! * \internal * \brief Check whether a resource is active on multiple nodes */ static bool is_multiply_active(const pcmk_resource_t *rsc) { unsigned int count = 0; if (pcmk__is_primitive(rsc)) { pe__find_active_requires(rsc, &count); } return count > 1; } static void native_priority_to_node(pcmk_resource_t *rsc, pcmk_node_t *node, gboolean failed) { int priority = 0; if ((rsc->private->priority == 0) || failed) { return; } if (rsc->role == pcmk_role_promoted) { // Promoted instance takes base priority + 1 priority = rsc->private->priority + 1; } else { priority = rsc->private->priority; } node->details->priority += priority; pcmk__rsc_trace(rsc, "%s now has priority %d with %s'%s' (priority: %d%s)", pcmk__node_name(node), node->details->priority, (rsc->role == pcmk_role_promoted)? "promoted " : "", rsc->id, rsc->private->priority, (rsc->role == pcmk_role_promoted)? " + 1" : ""); /* Priority of a resource running on a guest node is added to the cluster * node as well. */ if (node->details->remote_rsc && node->details->remote_rsc->container) { GList *gIter = node->details->remote_rsc->container->running_on; for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *a_node = gIter->data; a_node->details->priority += priority; pcmk__rsc_trace(rsc, "%s now has priority %d with %s'%s' " "(priority: %d%s) from guest node %s", pcmk__node_name(a_node), a_node->details->priority, (rsc->role == pcmk_role_promoted)? "promoted " : "", rsc->id, rsc->private->priority, (rsc->role == pcmk_role_promoted)? " + 1" : "", pcmk__node_name(node)); } } } void native_add_running(pcmk_resource_t *rsc, pcmk_node_t *node, pcmk_scheduler_t *scheduler, gboolean failed) { pcmk_resource_t *parent = rsc->private->parent; GList *gIter = rsc->running_on; CRM_CHECK(node != NULL, return); for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *a_node = (pcmk_node_t *) gIter->data; CRM_CHECK(a_node != NULL, return); if (pcmk__str_eq(a_node->details->id, node->details->id, pcmk__str_casei)) { return; } } pcmk__rsc_trace(rsc, "Adding %s to %s %s", rsc->id, pcmk__node_name(node), pcmk_is_set(rsc->flags, pcmk__rsc_managed)? "" : "(unmanaged)"); rsc->running_on = g_list_append(rsc->running_on, node); if (pcmk__is_primitive(rsc)) { node->details->running_rsc = g_list_append(node->details->running_rsc, rsc); native_priority_to_node(rsc, node, failed); if (node->details->maintenance) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(rsc, pcmk__rsc_maintenance); } } if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk_resource_t *p = parent; pcmk__rsc_info(rsc, "resource %s isn't managed", rsc->id); resource_location(rsc, node, PCMK_SCORE_INFINITY, "not_managed_default", scheduler); while(p && node->details->online) { /* add without the additional location constraint */ p->running_on = g_list_append(p->running_on, node); p = p->private->parent; } return; } if (is_multiply_active(rsc)) { switch (rsc->private->multiply_active_policy) { case pcmk__multiply_active_stop: { GHashTableIter gIter; pcmk_node_t *local_node = NULL; /* make sure it doesn't come up again */ if (rsc->allowed_nodes != NULL) { g_hash_table_destroy(rsc->allowed_nodes); } rsc->allowed_nodes = pe__node_list2table(scheduler->nodes); g_hash_table_iter_init(&gIter, rsc->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&local_node)) { local_node->weight = -PCMK_SCORE_INFINITY; } } break; case pcmk__multiply_active_block: pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(rsc, pcmk__rsc_blocked); /* If the resource belongs to a group or bundle configured with * PCMK_META_MULTIPLE_ACTIVE=PCMK_VALUE_BLOCK, block the entire * entity. */ if ((pcmk__is_group(parent) || pcmk__is_bundle(parent)) && (parent->private->multiply_active_policy == pcmk__multiply_active_block)) { for (GList *gIter = parent->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = gIter->data; pcmk__clear_rsc_flags(child, pcmk__rsc_managed); pcmk__set_rsc_flags(child, pcmk__rsc_blocked); } } break; // pcmk__multiply_active_restart, pcmk__multiply_active_unexpected default: /* The scheduler will do the right thing because the relevant * variables and flags are set when unpacking the history. */ break; } crm_debug("%s is active on multiple nodes including %s: %s", rsc->id, pcmk__node_name(node), pcmk__multiply_active_text(rsc)); } else { pcmk__rsc_trace(rsc, "Resource %s is active on %s", rsc->id, pcmk__node_name(node)); } if (parent != NULL) { native_add_running(parent, node, scheduler, FALSE); } } static void recursive_clear_unique(pcmk_resource_t *rsc, gpointer user_data) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_unique); pcmk__insert_meta(rsc, PCMK_META_GLOBALLY_UNIQUE, PCMK_VALUE_FALSE); g_list_foreach(rsc->children, (GFunc) recursive_clear_unique, NULL); } gboolean native_unpack(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { pcmk_resource_t *parent = uber_parent(rsc); const char *standard = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); uint32_t ra_caps = pcmk_get_ra_caps(standard); pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id); // Only some agent standards support unique and promotable clones if (!pcmk_is_set(ra_caps, pcmk_ra_cap_unique) && pcmk_is_set(rsc->flags, pcmk__rsc_unique) && pcmk__is_clone(parent)) { /* @COMPAT We should probably reject this situation as an error (as we * do for promotable below) rather than warn and convert, but that would * be a backward-incompatible change that we should probably do with a * transform at a schema major version bump. */ pe__force_anon(standard, parent, rsc->id, scheduler); /* Clear PCMK_META_GLOBALLY_UNIQUE on the parent and all its descendants * unpacked so far (clearing the parent should make any future children * unpacking correct). We have to clear this resource explicitly because * it isn't hooked into the parent's children yet. */ recursive_clear_unique(parent, NULL); recursive_clear_unique(rsc, NULL); } if (!pcmk_is_set(ra_caps, pcmk_ra_cap_promotable) && pcmk_is_set(parent->flags, pcmk__rsc_promotable)) { pcmk__config_err("Resource %s is of type %s and therefore " "cannot be used as a promotable clone resource", rsc->id, standard); return FALSE; } return TRUE; } static bool rsc_is_on_node(pcmk_resource_t *rsc, const pcmk_node_t *node, int flags) { pcmk__rsc_trace(rsc, "Checking whether %s is on %s", rsc->id, pcmk__node_name(node)); if (pcmk_is_set(flags, pcmk_rsc_match_current_node) && (rsc->running_on != NULL)) { for (GList *iter = rsc->running_on; iter; iter = iter->next) { if (pcmk__same_node((pcmk_node_t *) iter->data, node)) { return true; } } } else if (pcmk_is_set(flags, pe_find_inactive) // @COMPAT deprecated && (rsc->running_on == NULL)) { return true; } else if (!pcmk_is_set(flags, pcmk_rsc_match_current_node) - && (rsc->allocated_to != NULL) - && pcmk__same_node(rsc->allocated_to, node)) { + && (rsc->private->assigned_node != NULL) + && pcmk__same_node(rsc->private->assigned_node, node)) { return true; } return false; } pcmk_resource_t * native_find_rsc(pcmk_resource_t *rsc, const char *id, const pcmk_node_t *on_node, int flags) { bool match = false; pcmk_resource_t *result = NULL; CRM_CHECK(id && rsc && rsc->id, return NULL); if (pcmk_is_set(flags, pcmk_rsc_match_clone_only)) { const char *rid = pcmk__xe_id(rsc->private->xml); if (!pcmk__is_clone(pe__const_top_resource(rsc, false))) { match = false; } else if (!strcmp(id, rsc->id) || pcmk__str_eq(id, rid, pcmk__str_none)) { match = true; } } else if (!strcmp(id, rsc->id)) { match = true; } else if (pcmk_is_set(flags, pcmk_rsc_match_history) && pcmk__str_eq(rsc->private->history_id, id, pcmk__str_none)) { match = true; } else if (pcmk_is_set(flags, pcmk_rsc_match_basename) || (pcmk_is_set(flags, pcmk_rsc_match_anon_basename) && !pcmk_is_set(rsc->flags, pcmk__rsc_unique))) { match = pe_base_name_eq(rsc, id); } if (match && on_node) { if (!rsc_is_on_node(rsc, on_node, flags)) { match = false; } } if (match) { return rsc; } for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = (pcmk_resource_t *) gIter->data; result = rsc->private->fns->find_rsc(child, id, on_node, flags); if (result) { return result; } } return NULL; } // create is ignored char * native_parameter(pcmk_resource_t *rsc, pcmk_node_t *node, gboolean create, const char *name, pcmk_scheduler_t *scheduler) { const char *value = NULL; GHashTable *params = NULL; CRM_CHECK(rsc != NULL, return NULL); CRM_CHECK(name != NULL && strlen(name) != 0, return NULL); pcmk__rsc_trace(rsc, "Looking up %s in %s", name, rsc->id); params = pe_rsc_params(rsc, node, scheduler); value = g_hash_table_lookup(params, name); if (value == NULL) { /* try meta attributes instead */ value = g_hash_table_lookup(rsc->meta, name); } return pcmk__str_copy(value); } gboolean native_active(pcmk_resource_t * rsc, gboolean all) { for (GList *gIter = rsc->running_on; gIter != NULL; gIter = gIter->next) { pcmk_node_t *a_node = (pcmk_node_t *) gIter->data; if (a_node->details->unclean) { pcmk__rsc_trace(rsc, "Resource %s: %s is unclean", rsc->id, pcmk__node_name(a_node)); return TRUE; } else if (!a_node->details->online && pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__rsc_trace(rsc, "Resource %s: %s is offline", rsc->id, pcmk__node_name(a_node)); } else { pcmk__rsc_trace(rsc, "Resource %s active on %s", rsc->id, pcmk__node_name(a_node)); return TRUE; } } return FALSE; } struct print_data_s { long options; void *print_data; }; static const char * native_pending_state(const pcmk_resource_t *rsc) { const char *pending_state = NULL; if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_START, pcmk__str_none)) { pending_state = "Starting"; } else if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_STOP, pcmk__str_none)) { pending_state = "Stopping"; } else if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)) { pending_state = "Migrating"; } else if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_MIGRATE_FROM, pcmk__str_none)) { /* Work might be done in here. */ pending_state = "Migrating"; } else if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_PROMOTE, pcmk__str_none)) { pending_state = "Promoting"; } else if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_DEMOTE, pcmk__str_none)) { pending_state = "Demoting"; } return pending_state; } static const char * native_pending_action(const pcmk_resource_t *rsc) { const char *pending_action = NULL; if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_MONITOR, pcmk__str_none)) { pending_action = "Monitoring"; /* Pending probes are not printed, even if pending * operations are requested. If someone ever requests that * behavior, uncomment this and the corresponding part of * unpack.c:unpack_rsc_op(). */ #if 0 } else if (pcmk__str_eq(rsc->private->pending_action, "probe", pcmk__str_none)) { pending_action = "Checking"; #endif } return pending_action; } static enum rsc_role_e native_displayable_role(const pcmk_resource_t *rsc) { enum rsc_role_e role = rsc->role; if ((role == pcmk_role_started) && pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_promotable)) { role = pcmk_role_unpromoted; } return role; } static const char * native_displayable_state(const pcmk_resource_t *rsc, bool print_pending) { const char *rsc_state = NULL; if (print_pending) { rsc_state = native_pending_state(rsc); } if (rsc_state == NULL) { rsc_state = pcmk_role_text(native_displayable_role(rsc)); } return rsc_state; } // Append a flag to resource description string's flags list static bool add_output_flag(GString *s, const char *flag_desc, bool have_flags) { g_string_append(s, (have_flags? ", " : " (")); g_string_append(s, flag_desc); return true; } // Append a node name to resource description string's node list static bool add_output_node(GString *s, const char *node, bool have_nodes) { g_string_append(s, (have_nodes? " " : " [ ")); g_string_append(s, node); return true; } /*! * \internal * \brief Create a string description of a resource * * \param[in] rsc Resource to describe * \param[in] name Desired identifier for the resource * \param[in] node If not NULL, node that resource is "on" * \param[in] show_opts Bitmask of pcmk_show_opt_e. * \param[in] target_role Resource's target role * \param[in] show_nodes Whether to display nodes when multiply active * * \return Newly allocated string description of resource * \note Caller must free the result with g_free(). */ gchar * pcmk__native_output_string(const pcmk_resource_t *rsc, const char *name, const pcmk_node_t *node, uint32_t show_opts, const char *target_role, bool show_nodes) { const char *class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); const char *provider = NULL; const char *kind = crm_element_value(rsc->private->xml, PCMK_XA_TYPE); GString *outstr = NULL; bool have_flags = false; if (!pcmk__is_primitive(rsc)) { return NULL; } CRM_CHECK(name != NULL, name = "unknown"); CRM_CHECK(kind != NULL, kind = "unknown"); CRM_CHECK(class != NULL, class = "unknown"); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) { provider = crm_element_value(rsc->private->xml, PCMK_XA_PROVIDER); } if ((node == NULL) && (rsc->lock_node != NULL)) { node = rsc->lock_node; } if (pcmk_any_flags_set(show_opts, pcmk_show_rsc_only) || pcmk__list_of_multiple(rsc->running_on)) { node = NULL; } outstr = g_string_sized_new(128); // Resource name and agent pcmk__g_strcat(outstr, name, "\t(", class, ((provider == NULL)? "" : ":"), pcmk__s(provider, ""), ":", kind, "):\t", NULL); // State on node if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { g_string_append(outstr, " ORPHANED"); } if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { enum rsc_role_e role = native_displayable_role(rsc); g_string_append(outstr, " FAILED"); if (role > pcmk_role_unpromoted) { pcmk__add_word(&outstr, 0, pcmk_role_text(role)); } } else { bool show_pending = pcmk_is_set(show_opts, pcmk_show_pending); pcmk__add_word(&outstr, 0, native_displayable_state(rsc, show_pending)); } if (node) { pcmk__add_word(&outstr, 0, pcmk__node_name(node)); } // Failed probe operation if (native_displayable_role(rsc) == pcmk_role_stopped) { xmlNode *probe_op = pe__failed_probe_for_rsc(rsc, node ? node->details->uname : NULL); if (probe_op != NULL) { int rc; pcmk__scan_min_int(crm_element_value(probe_op, PCMK__XA_RC_CODE), &rc, 0); pcmk__g_strcat(outstr, " (", services_ocf_exitcode_str(rc), ") ", NULL); } } // Flags, as: ( [...]) if (node && !(node->details->online) && node->details->unclean) { have_flags = add_output_flag(outstr, "UNCLEAN", have_flags); } if (node && (node == rsc->lock_node)) { have_flags = add_output_flag(outstr, "LOCKED", have_flags); } if (pcmk_is_set(show_opts, pcmk_show_pending)) { const char *pending_action = native_pending_action(rsc); if (pending_action != NULL) { have_flags = add_output_flag(outstr, pending_action, have_flags); } } if (target_role != NULL) { switch (pcmk_parse_role(target_role)) { case pcmk_role_unknown: pcmk__config_err("Invalid " PCMK_META_TARGET_ROLE " %s for resource %s", target_role, rsc->id); break; case pcmk_role_stopped: have_flags = add_output_flag(outstr, "disabled", have_flags); break; case pcmk_role_unpromoted: if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_promotable)) { have_flags = add_output_flag(outstr, PCMK_META_TARGET_ROLE ":", have_flags); g_string_append(outstr, target_role); } break; default: /* Only show target role if it limits our abilities (i.e. ignore * Started, as it is the default anyways, and doesn't prevent * the resource from becoming promoted). */ break; } } // Blocked or maintenance implies unmanaged if (pcmk_any_flags_set(rsc->flags, pcmk__rsc_blocked|pcmk__rsc_maintenance)) { if (pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) { have_flags = add_output_flag(outstr, "blocked", have_flags); } else if (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance)) { have_flags = add_output_flag(outstr, "maintenance", have_flags); } } else if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { have_flags = add_output_flag(outstr, "unmanaged", have_flags); } if (pcmk_is_set(rsc->flags, pcmk__rsc_ignore_failure)) { have_flags = add_output_flag(outstr, "failure ignored", have_flags); } if (have_flags) { g_string_append_c(outstr, ')'); } // User-supplied description if (pcmk_any_flags_set(show_opts, pcmk_show_rsc_only|pcmk_show_description) || pcmk__list_of_multiple(rsc->running_on)) { const char *desc = crm_element_value(rsc->private->xml, PCMK_XA_DESCRIPTION); if (desc) { g_string_append(outstr, " ("); g_string_append(outstr, desc); g_string_append(outstr, ")"); } } if (show_nodes && !pcmk_is_set(show_opts, pcmk_show_rsc_only) && pcmk__list_of_multiple(rsc->running_on)) { bool have_nodes = false; for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *n = (pcmk_node_t *) iter->data; have_nodes = add_output_node(outstr, n->details->uname, have_nodes); } if (have_nodes) { g_string_append(outstr, " ]"); } } return g_string_free(outstr, FALSE); } int pe__common_output_html(pcmk__output_t *out, const pcmk_resource_t *rsc, const char *name, const pcmk_node_t *node, uint32_t show_opts) { const char *kind = crm_element_value(rsc->private->xml, PCMK_XA_TYPE); const char *target_role = NULL; const char *cl = NULL; xmlNode *child = NULL; gchar *content = NULL; CRM_ASSERT((kind != NULL) && pcmk__is_primitive(rsc)); if (rsc->meta) { const char *is_internal = g_hash_table_lookup(rsc->meta, PCMK__META_INTERNAL_RSC); if (crm_is_true(is_internal) && !pcmk_is_set(show_opts, pcmk_show_implicit_rscs)) { crm_trace("skipping print of internal resource %s", rsc->id); return pcmk_rc_no_output; } target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); } if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { cl = PCMK__VALUE_RSC_MANAGED; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { cl = PCMK__VALUE_RSC_FAILED; } else if (pcmk__is_primitive(rsc) && (rsc->running_on == NULL)) { cl = PCMK__VALUE_RSC_FAILED; } else if (pcmk__list_of_multiple(rsc->running_on)) { cl = PCMK__VALUE_RSC_MULTIPLE; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_ignore_failure)) { cl = PCMK__VALUE_RSC_FAILURE_IGNORED; } else { cl = PCMK__VALUE_RSC_OK; } child = pcmk__output_create_html_node(out, "li", NULL, NULL, NULL); child = pcmk__html_create(child, PCMK__XE_SPAN, NULL, cl); content = pcmk__native_output_string(rsc, name, node, show_opts, target_role, true); pcmk__xe_set_content(child, "%s", content); g_free(content); return pcmk_rc_ok; } int pe__common_output_text(pcmk__output_t *out, const pcmk_resource_t *rsc, const char *name, const pcmk_node_t *node, uint32_t show_opts) { const char *target_role = NULL; CRM_ASSERT(pcmk__is_primitive(rsc)); if (rsc->meta) { const char *is_internal = g_hash_table_lookup(rsc->meta, PCMK__META_INTERNAL_RSC); if (crm_is_true(is_internal) && !pcmk_is_set(show_opts, pcmk_show_implicit_rscs)) { crm_trace("skipping print of internal resource %s", rsc->id); return pcmk_rc_no_output; } target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); } { gchar *s = pcmk__native_output_string(rsc, name, node, show_opts, target_role, true); out->list_item(out, NULL, "%s", s); g_free(s); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("primitive", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__resource_xml(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); int rc = pcmk_rc_no_output; bool print_pending = pcmk_is_set(show_opts, pcmk_show_pending); const char *class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); const char *prov = crm_element_value(rsc->private->xml, PCMK_XA_PROVIDER); char ra_name[LINE_MAX]; const char *rsc_state = native_displayable_state(rsc, print_pending); const char *target_role = NULL; const char *active = pcmk__btoa(rsc->private->fns->active(rsc, TRUE)); const char *orphaned = pcmk__flag_text(rsc->flags, pcmk__rsc_removed); const char *blocked = pcmk__flag_text(rsc->flags, pcmk__rsc_blocked); const char *maintenance = pcmk__flag_text(rsc->flags, pcmk__rsc_maintenance); const char *managed = pcmk__flag_text(rsc->flags, pcmk__rsc_managed); const char *failed = pcmk__flag_text(rsc->flags, pcmk__rsc_failed); const char *ignored = pcmk__flag_text(rsc->flags, pcmk__rsc_ignore_failure); char *nodes_running_on = NULL; const char *pending = print_pending? native_pending_action(rsc) : NULL; const char *locked_to = NULL; const char *desc = pe__resource_description(rsc, show_opts); CRM_ASSERT(pcmk__is_primitive(rsc)); if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } // Resource information snprintf(ra_name, LINE_MAX, "%s%s%s:%s", class, ((prov == NULL)? "" : ":"), ((prov == NULL)? "" : prov), crm_element_value(rsc->private->xml, PCMK_XA_TYPE)); if (rsc->meta != NULL) { target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); } nodes_running_on = pcmk__itoa(g_list_length(rsc->running_on)); if (rsc->lock_node != NULL) { locked_to = rsc->lock_node->details->uname; } rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_RESOURCE, PCMK_XA_ID, rsc_printable_id(rsc), PCMK_XA_RESOURCE_AGENT, ra_name, PCMK_XA_ROLE, rsc_state, PCMK_XA_TARGET_ROLE, target_role, PCMK_XA_ACTIVE, active, PCMK_XA_ORPHANED, orphaned, PCMK_XA_BLOCKED, blocked, PCMK_XA_MAINTENANCE, maintenance, PCMK_XA_MANAGED, managed, PCMK_XA_FAILED, failed, PCMK_XA_FAILURE_IGNORED, ignored, PCMK_XA_NODES_RUNNING_ON, nodes_running_on, PCMK_XA_PENDING, pending, PCMK_XA_LOCKED_TO, locked_to, PCMK_XA_DESCRIPTION, desc, NULL); free(nodes_running_on); CRM_ASSERT(rc == pcmk_rc_ok); if (rsc->running_on != NULL) { GList *gIter = rsc->running_on; for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = (pcmk_node_t *) gIter->data; const char *cached = pcmk__btoa(node->details->online); rc = pe__name_and_nvpairs_xml(out, false, PCMK_XE_NODE, PCMK_XA_NAME, node->details->uname, PCMK_XA_ID, node->details->id, PCMK_XA_CACHED, cached, NULL); CRM_ASSERT(rc == pcmk_rc_ok); } } pcmk__output_xml_pop_parent(out); return rc; } PCMK__OUTPUT_ARGS("primitive", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__resource_html(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const pcmk_node_t *node = pcmk__current_node(rsc); if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } CRM_ASSERT(pcmk__is_primitive(rsc)); if (node == NULL) { // This is set only if a non-probe action is pending on this node node = rsc->pending_node; } return pe__common_output_html(out, rsc, rsc_printable_id(rsc), node, show_opts); } PCMK__OUTPUT_ARGS("primitive", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__resource_text(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const pcmk_node_t *node = pcmk__current_node(rsc); CRM_ASSERT(pcmk__is_primitive(rsc)); if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } if (node == NULL) { // This is set only if a non-probe action is pending on this node node = rsc->pending_node; } return pe__common_output_text(out, rsc, rsc_printable_id(rsc), node, show_opts); } void native_free(pcmk_resource_t * rsc) { pcmk__rsc_trace(rsc, "Freeing resource action list (not the data)"); common_free(rsc); } enum rsc_role_e native_resource_state(const pcmk_resource_t * rsc, gboolean current) { enum rsc_role_e role = rsc->next_role; if (current) { role = rsc->role; } pcmk__rsc_trace(rsc, "%s state: %s", rsc->id, pcmk_role_text(role)); return role; } /*! * \internal * \brief List nodes where a resource (or any of its children) is * * \param[in] rsc Resource to check * \param[out] list List to add result to * \param[in] current 0 = where allocated, 1 = where running, * 2 = where running or pending * * \return If list contains only one node, that node, or NULL otherwise */ pcmk_node_t * native_location(const pcmk_resource_t *rsc, GList **list, int current) { // @COMPAT: Accept a pcmk__rsc_node argument instead of int current pcmk_node_t *one = NULL; GList *result = NULL; if (rsc->children) { GList *gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = (pcmk_resource_t *) gIter->data; child->private->fns->location(child, &result, current); } } else if (current) { if (rsc->running_on) { result = g_list_copy(rsc->running_on); } if ((current == 2) && rsc->pending_node && !pe_find_node_id(result, rsc->pending_node->details->id)) { result = g_list_append(result, rsc->pending_node); } - } else if (current == FALSE && rsc->allocated_to) { - result = g_list_append(NULL, rsc->allocated_to); + } else if (!current && (rsc->private->assigned_node != NULL)) { + result = g_list_append(NULL, rsc->private->assigned_node); } if (result && (result->next == NULL)) { one = result->data; } if (list) { GList *gIter = result; for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = (pcmk_node_t *) gIter->data; if (*list == NULL || pe_find_node_id(*list, node->details->id) == NULL) { *list = g_list_append(*list, node); } } } g_list_free(result); return one; } static void get_rscs_brief(GList *rsc_list, GHashTable * rsc_table, GHashTable * active_table) { GList *gIter = rsc_list; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; const char *class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); const char *kind = crm_element_value(rsc->private->xml, PCMK_XA_TYPE); int offset = 0; char buffer[LINE_MAX]; int *rsc_counter = NULL; int *active_counter = NULL; if (!pcmk__is_primitive(rsc)) { continue; } offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", class); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) { const char *prov = crm_element_value(rsc->private->xml, PCMK_XA_PROVIDER); if (prov != NULL) { offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s", prov); } } offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s", kind); CRM_LOG_ASSERT(offset > 0); if (rsc_table) { rsc_counter = g_hash_table_lookup(rsc_table, buffer); if (rsc_counter == NULL) { rsc_counter = pcmk__assert_alloc(1, sizeof(int)); *rsc_counter = 0; g_hash_table_insert(rsc_table, strdup(buffer), rsc_counter); } (*rsc_counter)++; } if (active_table) { GList *gIter2 = rsc->running_on; for (; gIter2 != NULL; gIter2 = gIter2->next) { pcmk_node_t *node = (pcmk_node_t *) gIter2->data; GHashTable *node_table = NULL; if (node->details->unclean == FALSE && node->details->online == FALSE && pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { continue; } node_table = g_hash_table_lookup(active_table, node->details->uname); if (node_table == NULL) { node_table = pcmk__strkey_table(free, free); g_hash_table_insert(active_table, strdup(node->details->uname), node_table); } active_counter = g_hash_table_lookup(node_table, buffer); if (active_counter == NULL) { active_counter = pcmk__assert_alloc(1, sizeof(int)); *active_counter = 0; g_hash_table_insert(node_table, strdup(buffer), active_counter); } (*active_counter)++; } } } } static void destroy_node_table(gpointer data) { GHashTable *node_table = data; if (node_table) { g_hash_table_destroy(node_table); } } int pe__rscs_brief_output(pcmk__output_t *out, GList *rsc_list, uint32_t show_opts) { GHashTable *rsc_table = pcmk__strkey_table(free, free); GHashTable *active_table = pcmk__strkey_table(free, destroy_node_table); GList *sorted_rscs; int rc = pcmk_rc_no_output; get_rscs_brief(rsc_list, rsc_table, active_table); /* Make a list of the rsc_table keys so that it can be sorted. This is to make sure * output order stays consistent between systems. */ sorted_rscs = g_hash_table_get_keys(rsc_table); sorted_rscs = g_list_sort(sorted_rscs, (GCompareFunc) strcmp); for (GList *gIter = sorted_rscs; gIter; gIter = gIter->next) { char *type = (char *) gIter->data; int *rsc_counter = g_hash_table_lookup(rsc_table, type); GList *sorted_nodes = NULL; int active_counter_all = 0; /* Also make a list of the active_table keys so it can be sorted. If there's * more than one instance of a type of resource running, we need the nodes to * be sorted to make sure output order stays consistent between systems. */ sorted_nodes = g_hash_table_get_keys(active_table); sorted_nodes = g_list_sort(sorted_nodes, (GCompareFunc) pcmk__numeric_strcasecmp); for (GList *gIter2 = sorted_nodes; gIter2; gIter2 = gIter2->next) { char *node_name = (char *) gIter2->data; GHashTable *node_table = g_hash_table_lookup(active_table, node_name); int *active_counter = NULL; if (node_table == NULL) { continue; } active_counter = g_hash_table_lookup(node_table, type); if (active_counter == NULL || *active_counter == 0) { continue; } else { active_counter_all += *active_counter; } if (pcmk_is_set(show_opts, pcmk_show_rsc_only)) { node_name = NULL; } if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { out->list_item(out, NULL, "%d/%d\t(%s):\tActive %s", *active_counter, rsc_counter ? *rsc_counter : 0, type, (*active_counter > 0) && node_name ? node_name : ""); } else { out->list_item(out, NULL, "%d\t(%s):\tActive %s", *active_counter, type, (*active_counter > 0) && node_name ? node_name : ""); } rc = pcmk_rc_ok; } if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs) && active_counter_all == 0) { out->list_item(out, NULL, "%d/%d\t(%s):\tActive", active_counter_all, rsc_counter ? *rsc_counter : 0, type); rc = pcmk_rc_ok; } if (sorted_nodes) { g_list_free(sorted_nodes); } } if (rsc_table) { g_hash_table_destroy(rsc_table); rsc_table = NULL; } if (active_table) { g_hash_table_destroy(active_table); active_table = NULL; } if (sorted_rscs) { g_list_free(sorted_rscs); } return rc; } gboolean pe__native_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent) { if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) || pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches)) { return FALSE; } else if (check_parent && (rsc->private->parent != NULL)) { const pcmk_resource_t *up = pe__const_top_resource(rsc, true); return up->private->fns->is_filtered(up, only_rsc, FALSE); } return TRUE; } /*! * \internal * \brief Get maximum primitive resource instances per node * * \param[in] rsc Primitive resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int pe__primitive_max_per_node(const pcmk_resource_t *rsc) { CRM_ASSERT(pcmk__is_primitive(rsc)); return 1U; } diff --git a/lib/pengine/pe_notif.c b/lib/pengine/pe_notif.c index f4e68a1fec..e7b3ded4e1 100644 --- a/lib/pengine/pe_notif.c +++ b/lib/pengine/pe_notif.c @@ -1,1006 +1,1006 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "pe_status_private.h" typedef struct notify_entry_s { const pcmk_resource_t *rsc; const pcmk_node_t *node; } notify_entry_t; /*! * \internal * \brief Compare two notification entries * * Compare two notification entries, where the one with the alphabetically first * resource name (or if equal, node name) sorts as first, with NULL sorting as * less than non-NULL. * * \param[in] a First notification entry to compare * \param[in] b Second notification entry to compare * * \return -1 if \p a sorts before \p b, 0 if they are equal, otherwise 1 */ static gint compare_notify_entries(gconstpointer a, gconstpointer b) { int tmp; const notify_entry_t *entry_a = a; const notify_entry_t *entry_b = b; // NULL a or b is not actually possible if ((entry_a == NULL) && (entry_b == NULL)) { return 0; } if (entry_a == NULL) { return 1; } if (entry_b == NULL) { return -1; } // NULL resources sort first if ((entry_a->rsc == NULL) && (entry_b->rsc == NULL)) { return 0; } if (entry_a->rsc == NULL) { return 1; } if (entry_b->rsc == NULL) { return -1; } // Compare resource names tmp = strcmp(entry_a->rsc->id, entry_b->rsc->id); if (tmp != 0) { return tmp; } // Otherwise NULL nodes sort first if ((entry_a->node == NULL) && (entry_b->node == NULL)) { return 0; } if (entry_a->node == NULL) { return 1; } if (entry_b->node == NULL) { return -1; } // Finally, compare node names return strcmp(entry_a->node->details->id, entry_b->node->details->id); } /*! * \internal * \brief Duplicate a notification entry * * \param[in] entry Entry to duplicate * * \return Newly allocated duplicate of \p entry * \note It is the caller's responsibility to free the return value. */ static notify_entry_t * dup_notify_entry(const notify_entry_t *entry) { notify_entry_t *dup = pcmk__assert_alloc(1, sizeof(notify_entry_t)); dup->rsc = entry->rsc; dup->node = entry->node; return dup; } /*! * \internal * \brief Given a list of nodes, create strings with node names * * \param[in] list List of nodes (as pcmk_node_t *) * \param[out] all_node_names If not NULL, will be set to space-separated list * of the names of all nodes in \p list * \param[out] host_node_names Same as \p all_node_names, except active * guest nodes will list the name of their host * * \note The caller is responsible for freeing the output argument values using * \p g_string_free(). */ static void get_node_names(const GList *list, GString **all_node_names, GString **host_node_names) { if (all_node_names != NULL) { *all_node_names = NULL; } if (host_node_names != NULL) { *host_node_names = NULL; } for (const GList *iter = list; iter != NULL; iter = iter->next) { const pcmk_node_t *node = (const pcmk_node_t *) iter->data; if (node->details->uname == NULL) { continue; } // Always add to list of all node names if (all_node_names != NULL) { pcmk__add_word(all_node_names, 1024, node->details->uname); } // Add to host node name list if appropriate if (host_node_names != NULL) { if (pcmk__is_guest_or_bundle_node(node) && (node->details->remote_rsc->container->running_on != NULL)) { node = pcmk__current_node(node->details->remote_rsc->container); if (node->details->uname == NULL) { continue; } } pcmk__add_word(host_node_names, 1024, node->details->uname); } } if ((all_node_names != NULL) && (*all_node_names == NULL)) { *all_node_names = g_string_new(" "); } if ((host_node_names != NULL) && (*host_node_names == NULL)) { *host_node_names = g_string_new(" "); } } /*! * \internal * \brief Create strings of instance and node names from notification entries * * \param[in,out] list List of notification entries (will be sorted here) * \param[out] rsc_names If not NULL, will be set to space-separated list * of clone instances from \p list * \param[out] node_names If not NULL, will be set to space-separated list * of node names from \p list * * \return (Possibly new) head of sorted \p list * \note The caller is responsible for freeing the output argument values using * \p g_list_free_full() and \p g_string_free(). */ static GList * notify_entries_to_strings(GList *list, GString **rsc_names, GString **node_names) { const char *last_rsc_id = NULL; // Initialize output lists to NULL if (rsc_names != NULL) { *rsc_names = NULL; } if (node_names != NULL) { *node_names = NULL; } // Sort input list for user-friendliness (and ease of filtering duplicates) list = g_list_sort(list, compare_notify_entries); for (GList *gIter = list; gIter != NULL; gIter = gIter->next) { notify_entry_t *entry = (notify_entry_t *) gIter->data; // Entry must have a resource (with ID) CRM_LOG_ASSERT((entry != NULL) && (entry->rsc != NULL) && (entry->rsc->id != NULL)); if ((entry == NULL) || (entry->rsc == NULL) || (entry->rsc->id == NULL)) { continue; } // Entry must have a node unless listing inactive resources CRM_LOG_ASSERT((node_names == NULL) || (entry->node != NULL)); if ((node_names != NULL) && (entry->node == NULL)) { continue; } // Don't add duplicates of a particular clone instance if (pcmk__str_eq(entry->rsc->id, last_rsc_id, pcmk__str_none)) { continue; } last_rsc_id = entry->rsc->id; if (rsc_names != NULL) { pcmk__add_word(rsc_names, 1024, entry->rsc->id); } if ((node_names != NULL) && (entry->node->details->uname != NULL)) { pcmk__add_word(node_names, 1024, entry->node->details->uname); } } // If there are no entries, return "empty" lists if ((rsc_names != NULL) && (*rsc_names == NULL)) { *rsc_names = g_string_new(" "); } if ((node_names != NULL) && (*node_names == NULL)) { *node_names = g_string_new(" "); } return list; } /*! * \internal * \brief Copy a meta-attribute into a notify action * * \param[in] key Name of meta-attribute to copy * \param[in] value Value of meta-attribute to copy * \param[in,out] user_data Notify action to copy into */ static void copy_meta_to_notify(gpointer key, gpointer value, gpointer user_data) { pcmk_action_t *notify = (pcmk_action_t *) user_data; /* Any existing meta-attributes (for example, the action timeout) are for * the notify action itself, so don't override those. */ if (g_hash_table_lookup(notify->meta, (const char *) key) != NULL) { return; } pcmk__insert_dup(notify->meta, (const char *) key, (const char *) value); } static void add_notify_data_to_action_meta(const notify_data_t *n_data, pcmk_action_t *action) { for (const GSList *item = n_data->keys; item; item = item->next) { const pcmk_nvpair_t *nvpair = (const pcmk_nvpair_t *) item->data; pcmk__insert_meta(action, nvpair->name, nvpair->value); } } /*! * \internal * \brief Create a new notify pseudo-action for a clone resource * * \param[in,out] rsc Clone resource that notification is for * \param[in] action Action to use in notify action key * \param[in] notif_action PCMK_ACTION_NOTIFY or PCMK_ACTION_NOTIFIED * \param[in] notif_type "pre", "post", "confirmed-pre", "confirmed-post" * * \return Newly created notify pseudo-action */ static pcmk_action_t * new_notify_pseudo_action(pcmk_resource_t *rsc, const pcmk_action_t *action, const char *notif_action, const char *notif_type) { pcmk_action_t *notify = NULL; notify = custom_action(rsc, pcmk__notify_key(rsc->id, notif_type, action->task), notif_action, NULL, pcmk_is_set(action->flags, pcmk_action_optional), rsc->private->scheduler); pcmk__set_action_flags(notify, pcmk_action_pseudo); pcmk__insert_meta(notify, "notify_key_type", notif_type); pcmk__insert_meta(notify, "notify_key_operation", action->task); return notify; } /*! * \internal * \brief Create a new notify action for a clone instance * * \param[in,out] rsc Clone instance that notification is for * \param[in] node Node that notification is for * \param[in,out] op Action that notification is for * \param[in,out] notify_done Parent pseudo-action for notifications complete * \param[in] n_data Notification values to add to action meta-data * * \return Newly created notify action */ static pcmk_action_t * new_notify_action(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_action_t *op, pcmk_action_t *notify_done, const notify_data_t *n_data) { char *key = NULL; pcmk_action_t *notify_action = NULL; const char *value = NULL; const char *task = NULL; const char *skip_reason = NULL; CRM_CHECK((rsc != NULL) && (node != NULL), return NULL); // Ensure we have all the info we need if (op == NULL) { skip_reason = "no action"; } else if (notify_done == NULL) { skip_reason = "no parent notification"; } else if (!node->details->online) { skip_reason = "node offline"; } else if (!pcmk_is_set(op->flags, pcmk_action_runnable)) { skip_reason = "original action not runnable"; } if (skip_reason != NULL) { pcmk__rsc_trace(rsc, "Skipping notify action for %s on %s: %s", rsc->id, pcmk__node_name(node), skip_reason); return NULL; } value = g_hash_table_lookup(op->meta, "notify_type"); // "pre" or "post" task = g_hash_table_lookup(op->meta, "notify_operation"); // original action pcmk__rsc_trace(rsc, "Creating notify action for %s on %s (%s-%s)", rsc->id, pcmk__node_name(node), value, task); // Create the notify action key = pcmk__notify_key(rsc->id, value, task); notify_action = custom_action(rsc, key, op->task, node, pcmk_is_set(op->flags, pcmk_action_optional), rsc->private->scheduler); // Add meta-data to notify action g_hash_table_foreach(op->meta, copy_meta_to_notify, notify_action); add_notify_data_to_action_meta(n_data, notify_action); // Order notify after original action and before parent notification order_actions(op, notify_action, pcmk__ar_ordered); order_actions(notify_action, notify_done, pcmk__ar_ordered); return notify_action; } /*! * \internal * \brief Create a new "post-" notify action for a clone instance * * \param[in,out] rsc Clone instance that notification is for * \param[in] node Node that notification is for * \param[in,out] n_data Notification values to add to action meta-data */ static void new_post_notify_action(pcmk_resource_t *rsc, const pcmk_node_t *node, notify_data_t *n_data) { pcmk_action_t *notify = NULL; CRM_ASSERT(n_data != NULL); // Create the "post-" notify action for specified instance notify = new_notify_action(rsc, node, n_data->post, n_data->post_done, n_data); if (notify != NULL) { notify->priority = PCMK_SCORE_INFINITY; } // Order recurring monitors after all "post-" notifications complete if (n_data->post_done == NULL) { return; } for (GList *iter = rsc->private->actions; iter != NULL; iter = iter->next) { pcmk_action_t *mon = (pcmk_action_t *) iter->data; const char *interval_ms_s = NULL; interval_ms_s = g_hash_table_lookup(mon->meta, PCMK_META_INTERVAL); if (pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches) || pcmk__str_eq(mon->task, PCMK_ACTION_CANCEL, pcmk__str_none)) { continue; // Not a recurring monitor } order_actions(n_data->post_done, mon, pcmk__ar_ordered); } } /*! * \internal * \brief Create and order notification pseudo-actions for a clone action * * In addition to the actual notify actions needed for each clone instance, * clone notifications also require pseudo-actions to provide ordering points * in the notification process. This creates the notification data, along with * appropriate pseudo-actions and their orderings. * * For example, the ordering sequence for starting a clone is: * * "pre-" notify pseudo-action for clone * -> "pre-" notify actions for each clone instance * -> "pre-" notifications complete pseudo-action for clone * -> start actions for each clone instance * -> "started" pseudo-action for clone * -> "post-" notify pseudo-action for clone * -> "post-" notify actions for each clone instance * -> "post-" notifications complete pseudo-action for clone * * \param[in,out] rsc Clone that notifications are for * \param[in] task Name of action that notifications are for * \param[in,out] action If not NULL, create a "pre-" pseudo-action ordered * before a "pre-" complete pseudo-action, ordered * before this action * \param[in,out] complete If not NULL, create a "post-" pseudo-action ordered * after this action, and a "post-" complete * pseudo-action ordered after that * * \return Newly created notification data */ notify_data_t * pe__action_notif_pseudo_ops(pcmk_resource_t *rsc, const char *task, pcmk_action_t *action, pcmk_action_t *complete) { notify_data_t *n_data = NULL; if (!pcmk_is_set(rsc->flags, pcmk__rsc_notify)) { return NULL; } n_data = pcmk__assert_alloc(1, sizeof(notify_data_t)); n_data->action = task; if (action != NULL) { // Need "pre-" pseudo-actions // Create "pre-" notify pseudo-action for clone n_data->pre = new_notify_pseudo_action(rsc, action, PCMK_ACTION_NOTIFY, "pre"); pcmk__set_action_flags(n_data->pre, pcmk_action_runnable); pcmk__insert_meta(n_data->pre, "notify_type", "pre"); pcmk__insert_meta(n_data->pre, "notify_operation", n_data->action); // Create "pre-" notifications complete pseudo-action for clone n_data->pre_done = new_notify_pseudo_action(rsc, action, PCMK_ACTION_NOTIFIED, "confirmed-pre"); pcmk__set_action_flags(n_data->pre_done, pcmk_action_runnable); pcmk__insert_meta(n_data->pre_done, "notify_type", "pre"); pcmk__insert_meta(n_data->pre_done, "notify_operation", n_data->action); // Order "pre-" -> "pre-" complete -> original action order_actions(n_data->pre, n_data->pre_done, pcmk__ar_ordered); order_actions(n_data->pre_done, action, pcmk__ar_ordered); } if (complete != NULL) { // Need "post-" pseudo-actions // Create "post-" notify pseudo-action for clone n_data->post = new_notify_pseudo_action(rsc, complete, PCMK_ACTION_NOTIFY, "post"); n_data->post->priority = PCMK_SCORE_INFINITY; if (pcmk_is_set(complete->flags, pcmk_action_runnable)) { pcmk__set_action_flags(n_data->post, pcmk_action_runnable); } else { pcmk__clear_action_flags(n_data->post, pcmk_action_runnable); } pcmk__insert_meta(n_data->post, "notify_type", "post"); pcmk__insert_meta(n_data->post, "notify_operation", n_data->action); // Create "post-" notifications complete pseudo-action for clone n_data->post_done = new_notify_pseudo_action(rsc, complete, PCMK_ACTION_NOTIFIED, "confirmed-post"); n_data->post_done->priority = PCMK_SCORE_INFINITY; if (pcmk_is_set(complete->flags, pcmk_action_runnable)) { pcmk__set_action_flags(n_data->post_done, pcmk_action_runnable); } else { pcmk__clear_action_flags(n_data->post_done, pcmk_action_runnable); } pcmk__insert_meta(n_data->post_done, "notify_type", "post"); pcmk__insert_meta(n_data->post_done, "notify_operation", n_data->action); // Order original action complete -> "post-" -> "post-" complete order_actions(complete, n_data->post, pcmk__ar_first_implies_then); order_actions(n_data->post, n_data->post_done, pcmk__ar_first_implies_then); } // If we created both, order "pre-" complete -> "post-" if ((action != NULL) && (complete != NULL)) { order_actions(n_data->pre_done, n_data->post, pcmk__ar_ordered); } return n_data; } /*! * \internal * \brief Create a new notification entry * * \param[in] rsc Resource for notification * \param[in] node Node for notification * * \return Newly allocated notification entry * \note The caller is responsible for freeing the return value. */ static notify_entry_t * new_notify_entry(const pcmk_resource_t *rsc, const pcmk_node_t *node) { notify_entry_t *entry = pcmk__assert_alloc(1, sizeof(notify_entry_t)); entry->rsc = rsc; entry->node = node; return entry; } /*! * \internal * \brief Add notification data for resource state and optionally actions * * \param[in] rsc Clone or clone instance being notified * \param[in] activity Whether to add notification entries for actions * \param[in,out] n_data Notification data for clone */ static void collect_resource_data(const pcmk_resource_t *rsc, bool activity, notify_data_t *n_data) { const GList *iter = NULL; notify_entry_t *entry = NULL; const pcmk_node_t *node = NULL; if (n_data == NULL) { return; } if (n_data->allowed_nodes == NULL) { n_data->allowed_nodes = rsc->allowed_nodes; } // If this is a clone, call recursively for each instance if (rsc->children != NULL) { for (iter = rsc->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *child = (const pcmk_resource_t *) iter->data; collect_resource_data(child, activity, n_data); } return; } // This is a notification for a single clone instance if (rsc->running_on != NULL) { node = rsc->running_on->data; // First is sufficient } entry = new_notify_entry(rsc, node); // Add notification indicating the resource state switch (rsc->role) { case pcmk_role_stopped: n_data->inactive = g_list_prepend(n_data->inactive, entry); break; case pcmk_role_started: n_data->active = g_list_prepend(n_data->active, entry); break; case pcmk_role_unpromoted: n_data->unpromoted = g_list_prepend(n_data->unpromoted, entry); n_data->active = g_list_prepend(n_data->active, dup_notify_entry(entry)); break; case pcmk_role_promoted: n_data->promoted = g_list_prepend(n_data->promoted, entry); n_data->active = g_list_prepend(n_data->active, dup_notify_entry(entry)); break; default: pcmk__sched_err("Resource %s role on %s (%s) is not supported for " "notifications (bug?)", rsc->id, pcmk__node_name(node), pcmk_role_text(rsc->role)); free(entry); break; } if (!activity) { return; } // Add notification entries for each of the resource's actions for (iter = rsc->private->actions; iter != NULL; iter = iter->next) { const pcmk_action_t *op = (const pcmk_action_t *) iter->data; if (!pcmk_is_set(op->flags, pcmk_action_optional) && (op->node != NULL)) { enum action_tasks task = pcmk_parse_action(op->task); if ((task == pcmk_action_stop) && op->node->details->unclean) { // Create anyway (additional noise if node can't be fenced) } else if (!pcmk_is_set(op->flags, pcmk_action_runnable)) { continue; } entry = new_notify_entry(rsc, op->node); switch (task) { case pcmk_action_start: n_data->start = g_list_prepend(n_data->start, entry); break; case pcmk_action_stop: n_data->stop = g_list_prepend(n_data->stop, entry); break; case pcmk_action_promote: n_data->promote = g_list_prepend(n_data->promote, entry); break; case pcmk_action_demote: n_data->demote = g_list_prepend(n_data->demote, entry); break; default: free(entry); break; } } } } // For (char *) value #define add_notify_env(n_data, key, value) do { \ n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, value); \ } while (0) // For (GString *) value #define add_notify_env_gs(n_data, key, value) do { \ n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, \ (const char *) value->str); \ } while (0) // For (GString *) value #define add_notify_env_free_gs(n_data, key, value) do { \ n_data->keys = pcmk_prepend_nvpair(n_data->keys, key, \ (const char *) value->str); \ g_string_free(value, TRUE); value = NULL; \ } while (0) /*! * \internal * \brief Create notification name/value pairs from structured data * * \param[in] rsc Resource that notification is for * \param[in,out] n_data Notification data */ static void add_notif_keys(const pcmk_resource_t *rsc, notify_data_t *n_data) { bool required = false; // Whether to make notify actions required GString *rsc_list = NULL; GString *node_list = NULL; GString *metal_list = NULL; const char *source = NULL; GList *nodes = NULL; n_data->stop = notify_entries_to_strings(n_data->stop, &rsc_list, &node_list); if ((strcmp(" ", (const char *) rsc_list->str) != 0) && pcmk__str_eq(n_data->action, PCMK_ACTION_STOP, pcmk__str_none)) { required = true; } add_notify_env_free_gs(n_data, "notify_stop_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_stop_uname", node_list); if ((n_data->start != NULL) && pcmk__str_eq(n_data->action, PCMK_ACTION_START, pcmk__str_none)) { required = true; } n_data->start = notify_entries_to_strings(n_data->start, &rsc_list, &node_list); add_notify_env_free_gs(n_data, "notify_start_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_start_uname", node_list); if ((n_data->demote != NULL) && pcmk__str_eq(n_data->action, PCMK_ACTION_DEMOTE, pcmk__str_none)) { required = true; } n_data->demote = notify_entries_to_strings(n_data->demote, &rsc_list, &node_list); add_notify_env_free_gs(n_data, "notify_demote_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_demote_uname", node_list); if ((n_data->promote != NULL) && pcmk__str_eq(n_data->action, PCMK_ACTION_PROMOTE, pcmk__str_none)) { required = true; } n_data->promote = notify_entries_to_strings(n_data->promote, &rsc_list, &node_list); add_notify_env_free_gs(n_data, "notify_promote_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_promote_uname", node_list); n_data->active = notify_entries_to_strings(n_data->active, &rsc_list, &node_list); add_notify_env_free_gs(n_data, "notify_active_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_active_uname", node_list); n_data->unpromoted = notify_entries_to_strings(n_data->unpromoted, &rsc_list, &node_list); add_notify_env_gs(n_data, "notify_unpromoted_resource", rsc_list); add_notify_env_gs(n_data, "notify_unpromoted_uname", node_list); // Deprecated: kept for backward compatibility with older resource agents add_notify_env_free_gs(n_data, "notify_slave_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_slave_uname", node_list); n_data->promoted = notify_entries_to_strings(n_data->promoted, &rsc_list, &node_list); add_notify_env_gs(n_data, "notify_promoted_resource", rsc_list); add_notify_env_gs(n_data, "notify_promoted_uname", node_list); // Deprecated: kept for backward compatibility with older resource agents add_notify_env_free_gs(n_data, "notify_master_resource", rsc_list); add_notify_env_free_gs(n_data, "notify_master_uname", node_list); n_data->inactive = notify_entries_to_strings(n_data->inactive, &rsc_list, NULL); add_notify_env_free_gs(n_data, "notify_inactive_resource", rsc_list); nodes = g_hash_table_get_values(n_data->allowed_nodes); if (!pcmk__is_daemon) { /* For display purposes, sort the node list, for consistent * regression test output (while avoiding the performance hit * for the live cluster). */ nodes = g_list_sort(nodes, pe__cmp_node_name); } get_node_names(nodes, &node_list, NULL); add_notify_env_free_gs(n_data, "notify_available_uname", node_list); g_list_free(nodes); source = g_hash_table_lookup(rsc->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); if (pcmk__str_eq(PCMK_VALUE_HOST, source, pcmk__str_none)) { get_node_names(rsc->private->scheduler->nodes, &node_list, &metal_list); add_notify_env_free_gs(n_data, "notify_all_hosts", metal_list); } else { get_node_names(rsc->private->scheduler->nodes, &node_list, NULL); } add_notify_env_free_gs(n_data, "notify_all_uname", node_list); if (required && (n_data->pre != NULL)) { pcmk__clear_action_flags(n_data->pre, pcmk_action_optional); pcmk__clear_action_flags(n_data->pre_done, pcmk_action_optional); } if (required && (n_data->post != NULL)) { pcmk__clear_action_flags(n_data->post, pcmk_action_optional); pcmk__clear_action_flags(n_data->post_done, pcmk_action_optional); } } /* * \internal * \brief Find any remote connection start relevant to an action * * \param[in] action Action to check * * \return If action is behind a remote connection, connection's start */ static pcmk_action_t * find_remote_start(pcmk_action_t *action) { if ((action != NULL) && (action->node != NULL)) { pcmk_resource_t *remote_rsc = action->node->details->remote_rsc; if (remote_rsc != NULL) { return find_first_action(remote_rsc->private->actions, NULL, PCMK_ACTION_START, NULL); } } return NULL; } /*! * \internal * \brief Create notify actions, and add notify data to original actions * * \param[in,out] rsc Clone or clone instance that notification is for * \param[in,out] n_data Clone notification data for some action */ static void create_notify_actions(pcmk_resource_t *rsc, notify_data_t *n_data) { GList *iter = NULL; pcmk_action_t *stop = NULL; pcmk_action_t *start = NULL; enum action_tasks task = pcmk_parse_action(n_data->action); // If this is a clone, call recursively for each instance if (rsc->children != NULL) { g_list_foreach(rsc->children, (GFunc) create_notify_actions, n_data); return; } // Add notification meta-attributes to original actions for (iter = rsc->private->actions; iter != NULL; iter = iter->next) { pcmk_action_t *op = (pcmk_action_t *) iter->data; if (!pcmk_is_set(op->flags, pcmk_action_optional) && (op->node != NULL)) { switch (pcmk_parse_action(op->task)) { case pcmk_action_start: case pcmk_action_stop: case pcmk_action_promote: case pcmk_action_demote: add_notify_data_to_action_meta(n_data, op); break; default: break; } } } // Skip notify action itself if original action was not needed switch (task) { case pcmk_action_start: if (n_data->start == NULL) { pcmk__rsc_trace(rsc, "No notify action needed for %s %s", rsc->id, n_data->action); return; } break; case pcmk_action_promote: if (n_data->promote == NULL) { pcmk__rsc_trace(rsc, "No notify action needed for %s %s", rsc->id, n_data->action); return; } break; case pcmk_action_demote: if (n_data->demote == NULL) { pcmk__rsc_trace(rsc, "No notify action needed for %s %s", rsc->id, n_data->action); return; } break; default: // We cannot do same for stop because it might be implied by fencing break; } pcmk__rsc_trace(rsc, "Creating notify actions for %s %s", rsc->id, n_data->action); // Create notify actions for stop or demote if ((rsc->role != pcmk_role_stopped) && ((task == pcmk_action_stop) || (task == pcmk_action_demote))) { stop = find_first_action(rsc->private->actions, NULL, PCMK_ACTION_STOP, NULL); for (iter = rsc->running_on; iter != NULL; iter = iter->next) { pcmk_node_t *current_node = (pcmk_node_t *) iter->data; /* If a stop is a pseudo-action implied by fencing, don't try to * notify the node getting fenced. */ if ((stop != NULL) && pcmk_is_set(stop->flags, pcmk_action_pseudo) && (current_node->details->unclean || current_node->details->remote_requires_reset)) { continue; } new_notify_action(rsc, current_node, n_data->pre, n_data->pre_done, n_data); if ((task == pcmk_action_demote) || (stop == NULL) || pcmk_is_set(stop->flags, pcmk_action_optional)) { new_post_notify_action(rsc, current_node, n_data); } } } // Create notify actions for start or promote if ((rsc->next_role != pcmk_role_stopped) && ((task == pcmk_action_start) || (task == pcmk_action_promote))) { start = find_first_action(rsc->private->actions, NULL, PCMK_ACTION_START, NULL); if (start != NULL) { pcmk_action_t *remote_start = find_remote_start(start); if ((remote_start != NULL) && !pcmk_is_set(remote_start->flags, pcmk_action_runnable)) { /* Start and promote actions for a clone instance behind * a Pacemaker Remote connection happen after the * connection starts. If the connection start is blocked, do * not schedule notifications for these actions. */ return; } } - if (rsc->allocated_to == NULL) { + if (rsc->private->assigned_node == NULL) { pcmk__sched_err("Next role '%s' but %s is not allocated", pcmk_role_text(rsc->next_role), rsc->id); return; } if ((task != pcmk_action_start) || (start == NULL) || pcmk_is_set(start->flags, pcmk_action_optional)) { - new_notify_action(rsc, rsc->allocated_to, n_data->pre, + new_notify_action(rsc, rsc->private->assigned_node, n_data->pre, n_data->pre_done, n_data); } - new_post_notify_action(rsc, rsc->allocated_to, n_data); + new_post_notify_action(rsc, rsc->private->assigned_node, n_data); } } /*! * \internal * \brief Create notification data and actions for one clone action * * \param[in,out] rsc Clone resource that notification is for * \param[in,out] n_data Clone notification data for some action */ void pe__create_action_notifications(pcmk_resource_t *rsc, notify_data_t *n_data) { if ((rsc == NULL) || (n_data == NULL)) { return; } collect_resource_data(rsc, true, n_data); add_notif_keys(rsc, n_data); create_notify_actions(rsc, n_data); } /*! * \internal * \brief Free notification data for one action * * \param[in,out] n_data Notification data to free */ void pe__free_action_notification_data(notify_data_t *n_data) { if (n_data == NULL) { return; } g_list_free_full(n_data->stop, free); g_list_free_full(n_data->start, free); g_list_free_full(n_data->demote, free); g_list_free_full(n_data->promote, free); g_list_free_full(n_data->promoted, free); g_list_free_full(n_data->unpromoted, free); g_list_free_full(n_data->active, free); g_list_free_full(n_data->inactive, free); pcmk_free_nvpairs(n_data->keys); free(n_data); } /*! * \internal * \brief Order clone "notifications complete" pseudo-action after fencing * * If a stop action is implied by fencing, the usual notification pseudo-actions * will not be sufficient to order things properly, or even create all needed * notifications if the clone is also stopping on another node, and another * clone is ordered after it. This function creates new notification * pseudo-actions relative to the fencing to ensure everything works properly. * * \param[in] stop Stop action implied by fencing * \param[in,out] rsc Clone resource that notification is for * \param[in,out] stonith_op Fencing action that implies \p stop */ void pe__order_notifs_after_fencing(const pcmk_action_t *stop, pcmk_resource_t *rsc, pcmk_action_t *stonith_op) { notify_data_t *n_data; crm_info("Ordering notifications for implied %s after fencing", stop->uuid); n_data = pe__action_notif_pseudo_ops(rsc, PCMK_ACTION_STOP, NULL, stonith_op); if (n_data != NULL) { collect_resource_data(rsc, false, n_data); add_notify_env(n_data, "notify_stop_resource", rsc->id); add_notify_env(n_data, "notify_stop_uname", stop->node->details->uname); create_notify_actions(uber_parent(rsc), n_data); pe__free_action_notification_data(n_data); } } diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c index 44f2dc21ea..1acc2b00ad 100644 --- a/lib/pengine/utils.c +++ b/lib/pengine/utils.c @@ -1,927 +1,928 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include "pe_status_private.h" extern bool pcmk__is_daemon; gboolean ghash_free_str_str(gpointer key, gpointer value, gpointer user_data); /*! * \internal * \brief Check whether we can fence a particular node * * \param[in] scheduler Scheduler data * \param[in] node Name of node to check * * \return true if node can be fenced, false otherwise */ bool pe_can_fence(const pcmk_scheduler_t *scheduler, const pcmk_node_t *node) { if (pcmk__is_guest_or_bundle_node(node)) { /* Guest nodes are fenced by stopping their container resource. We can * do that if the container's host is either online or fenceable. */ pcmk_resource_t *rsc = node->details->remote_rsc->container; for (GList *n = rsc->running_on; n != NULL; n = n->next) { pcmk_node_t *container_node = n->data; if (!container_node->details->online && !pe_can_fence(scheduler, container_node)) { return false; } } return true; } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { return false; /* Turned off */ } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) { return false; /* No devices */ } else if (pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) { return true; } else if (scheduler->no_quorum_policy == pcmk_no_quorum_ignore) { return true; } else if(node == NULL) { return false; } else if(node->details->online) { crm_notice("We can fence %s without quorum because they're in our membership", pcmk__node_name(node)); return true; } crm_trace("Cannot fence %s", pcmk__node_name(node)); return false; } /*! * \internal * \brief Copy a node object * * \param[in] this_node Node object to copy * * \return Newly allocated shallow copy of this_node * \note This function asserts on errors and is guaranteed to return non-NULL. */ pcmk_node_t * pe__copy_node(const pcmk_node_t *this_node) { pcmk_node_t *new_node = NULL; CRM_ASSERT(this_node != NULL); new_node = pcmk__assert_alloc(1, sizeof(pcmk_node_t)); new_node->rsc_discover_mode = this_node->rsc_discover_mode; new_node->weight = this_node->weight; new_node->fixed = this_node->fixed; // @COMPAT deprecated and unused new_node->count = this_node->count; new_node->details = this_node->details; return new_node; } /*! * \internal * \brief Create a node hash table from a node list * * \param[in] list Node list * * \return Hash table equivalent of node list */ GHashTable * pe__node_list2table(const GList *list) { GHashTable *result = NULL; result = pcmk__strkey_table(NULL, free); for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) { pcmk_node_t *new_node = NULL; new_node = pe__copy_node((const pcmk_node_t *) gIter->data); g_hash_table_insert(result, (gpointer) new_node->details->id, new_node); } return result; } /*! * \internal * \brief Compare two nodes by name, with numeric portions sorted numerically * * Sort two node names case-insensitively like strcasecmp(), but with any * numeric portions of the name sorted numerically. For example, "node10" will * sort higher than "node9" but lower than "remotenode9". * * \param[in] a First node to compare (can be \c NULL) * \param[in] b Second node to compare (can be \c NULL) * * \retval -1 \c a comes before \c b (or \c a is \c NULL and \c b is not) * \retval 0 \c a and \c b are equal (or both are \c NULL) * \retval 1 \c a comes after \c b (or \c b is \c NULL and \c a is not) */ gint pe__cmp_node_name(gconstpointer a, gconstpointer b) { const pcmk_node_t *node1 = (const pcmk_node_t *) a; const pcmk_node_t *node2 = (const pcmk_node_t *) b; if ((node1 == NULL) && (node2 == NULL)) { return 0; } if (node1 == NULL) { return -1; } if (node2 == NULL) { return 1; } return pcmk__numeric_strcasecmp(node1->details->uname, node2->details->uname); } /*! * \internal * \brief Output node weights to stdout * * \param[in] rsc Use allowed nodes for this resource * \param[in] comment Text description to prefix lines with * \param[in] nodes If rsc is not specified, use these nodes * \param[in,out] scheduler Scheduler data */ static void pe__output_node_weights(const pcmk_resource_t *rsc, const char *comment, GHashTable *nodes, pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv; // Sort the nodes so the output is consistent for regression tests GList *list = g_list_sort(g_hash_table_get_values(nodes), pe__cmp_node_name); for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) { const pcmk_node_t *node = (const pcmk_node_t *) gIter->data; out->message(out, "node-weight", rsc, comment, node->details->uname, pcmk_readable_score(node->weight)); } g_list_free(list); } /*! * \internal * \brief Log node weights at trace level * * \param[in] file Caller's filename * \param[in] function Caller's function name * \param[in] line Caller's line number * \param[in] rsc If not NULL, include this resource's ID in logs * \param[in] comment Text description to prefix lines with * \param[in] nodes Nodes whose scores should be logged */ static void pe__log_node_weights(const char *file, const char *function, int line, const pcmk_resource_t *rsc, const char *comment, GHashTable *nodes) { GHashTableIter iter; pcmk_node_t *node = NULL; // Don't waste time if we're not tracing at this point pcmk__if_tracing({}, return); g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (rsc) { qb_log_from_external_source(function, file, "%s: %s allocation score on %s: %s", LOG_TRACE, line, 0, comment, rsc->id, pcmk__node_name(node), pcmk_readable_score(node->weight)); } else { qb_log_from_external_source(function, file, "%s: %s = %s", LOG_TRACE, line, 0, comment, pcmk__node_name(node), pcmk_readable_score(node->weight)); } } } /*! * \internal * \brief Log or output node weights * * \param[in] file Caller's filename * \param[in] function Caller's function name * \param[in] line Caller's line number * \param[in] to_log Log if true, otherwise output * \param[in] rsc If not NULL, use this resource's ID in logs, * and show scores recursively for any children * \param[in] comment Text description to prefix lines with * \param[in] nodes Nodes whose scores should be shown * \param[in,out] scheduler Scheduler data */ void pe__show_node_scores_as(const char *file, const char *function, int line, bool to_log, const pcmk_resource_t *rsc, const char *comment, GHashTable *nodes, pcmk_scheduler_t *scheduler) { if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { // Don't show allocation scores for orphans return; } if (nodes == NULL) { // Nothing to show return; } if (to_log) { pe__log_node_weights(file, function, line, rsc, comment, nodes); } else { pe__output_node_weights(rsc, comment, nodes, scheduler); } // If this resource has children, repeat recursively for each if (rsc && rsc->children) { for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = (pcmk_resource_t *) gIter->data; pe__show_node_scores_as(file, function, line, to_log, child, comment, child->allowed_nodes, scheduler); } } } /*! * \internal * \brief Compare two resources by priority * * \param[in] a First resource to compare (can be \c NULL) * \param[in] b Second resource to compare (can be \c NULL) * * \retval -1 a's priority > b's priority (or \c b is \c NULL and \c a is not) * \retval 0 a's priority == b's priority (or both \c a and \c b are \c NULL) * \retval 1 a's priority < b's priority (or \c a is \c NULL and \c b is not) */ gint pe__cmp_rsc_priority(gconstpointer a, gconstpointer b) { const pcmk_resource_t *resource1 = (const pcmk_resource_t *)a; const pcmk_resource_t *resource2 = (const pcmk_resource_t *)b; if (a == NULL && b == NULL) { return 0; } if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (resource1->private->priority > resource2->private->priority) { return -1; } if (resource1->private->priority < resource2->private->priority) { return 1; } return 0; } static void resource_node_score(pcmk_resource_t *rsc, const pcmk_node_t *node, int score, const char *tag) { pcmk_node_t *match = NULL; if ((pcmk_is_set(rsc->flags, pcmk__rsc_exclusive_probes) || (node->rsc_discover_mode == pcmk_probe_never)) && pcmk__str_eq(tag, "symmetric_default", pcmk__str_casei)) { /* This string comparision may be fragile, but exclusive resources and * exclusive nodes should not have the symmetric_default constraint * applied to them. */ return; } else if (rsc->children) { GList *gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; resource_node_score(child_rsc, node, score, tag); } } match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (match == NULL) { match = pe__copy_node(node); g_hash_table_insert(rsc->allowed_nodes, (gpointer) match->details->id, match); } match->weight = pcmk__add_scores(match->weight, score); pcmk__rsc_trace(rsc, "Enabling %s preference (%s) for %s on %s (now %s)", tag, pcmk_readable_score(score), rsc->id, pcmk__node_name(node), pcmk_readable_score(match->weight)); } void resource_location(pcmk_resource_t *rsc, const pcmk_node_t *node, int score, const char *tag, pcmk_scheduler_t *scheduler) { if (node != NULL) { resource_node_score(rsc, node, score, tag); } else if (scheduler != NULL) { GList *gIter = scheduler->nodes; for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node_iter = (pcmk_node_t *) gIter->data; resource_node_score(rsc, node_iter, score, tag); } } else { GHashTableIter iter; pcmk_node_t *node_iter = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node_iter)) { resource_node_score(rsc, node_iter, score, tag); } } - if ((node == NULL) && (score == -PCMK_SCORE_INFINITY)) { - if (rsc->allocated_to) { - crm_info("Deallocating %s from %s", - rsc->id, pcmk__node_name(rsc->allocated_to)); - free(rsc->allocated_to); - rsc->allocated_to = NULL; - } + if ((node == NULL) && (score == -PCMK_SCORE_INFINITY) + && (rsc->private->assigned_node != NULL)) { + + // @TODO Should this be more like pcmk__unassign_resource()? + crm_info("Unassigning %s from %s", + rsc->id, pcmk__node_name(rsc->private->assigned_node)); + free(rsc->private->assigned_node); + rsc->private->assigned_node = NULL; } } time_t get_effective_time(pcmk_scheduler_t *scheduler) { if(scheduler) { if (scheduler->now == NULL) { crm_trace("Recording a new 'now'"); scheduler->now = crm_time_new(NULL); } return crm_time_get_seconds_since_epoch(scheduler->now); } crm_trace("Defaulting to 'now'"); return time(NULL); } gboolean get_target_role(const pcmk_resource_t *rsc, enum rsc_role_e *role) { enum rsc_role_e local_role = pcmk_role_unknown; const char *value = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); CRM_CHECK(role != NULL, return FALSE); if (pcmk__str_eq(value, PCMK_ROLE_STARTED, pcmk__str_null_matches|pcmk__str_casei)) { return FALSE; } if (pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) { // @COMPAT Deprecated since 2.1.8 pcmk__config_warn("Support for setting " PCMK_META_TARGET_ROLE " to the explicit value '" PCMK_VALUE_DEFAULT "' is deprecated and will be removed in a " "future release (just leave it unset)"); return FALSE; } local_role = pcmk_parse_role(value); if (local_role == pcmk_role_unknown) { pcmk__config_err("Ignoring '" PCMK_META_TARGET_ROLE "' for %s " "because '%s' is not valid", rsc->id, value); return FALSE; } else if (local_role > pcmk_role_started) { if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_promotable)) { if (local_role > pcmk_role_unpromoted) { /* This is what we'd do anyway, just leave the default to avoid messing up the placement algorithm */ return FALSE; } } else { pcmk__config_err("Ignoring '" PCMK_META_TARGET_ROLE "' for %s " "because '%s' only makes sense for promotable " "clones", rsc->id, value); return FALSE; } } *role = local_role; return TRUE; } gboolean order_actions(pcmk_action_t *lh_action, pcmk_action_t *rh_action, uint32_t flags) { GList *gIter = NULL; pcmk__related_action_t *wrapper = NULL; GList *list = NULL; if (flags == pcmk__ar_none) { return FALSE; } if (lh_action == NULL || rh_action == NULL) { return FALSE; } crm_trace("Creating action wrappers for ordering: %s then %s", lh_action->uuid, rh_action->uuid); /* Ensure we never create a dependency on ourselves... it's happened */ CRM_ASSERT(lh_action != rh_action); /* Filter dups, otherwise update_action_states() has too much work to do */ gIter = lh_action->actions_after; for (; gIter != NULL; gIter = gIter->next) { pcmk__related_action_t *after = gIter->data; if (after->action == rh_action && (after->type & flags)) { return FALSE; } } wrapper = pcmk__assert_alloc(1, sizeof(pcmk__related_action_t)); wrapper->action = rh_action; wrapper->type = flags; list = lh_action->actions_after; list = g_list_prepend(list, wrapper); lh_action->actions_after = list; wrapper = pcmk__assert_alloc(1, sizeof(pcmk__related_action_t)); wrapper->action = lh_action; wrapper->type = flags; list = rh_action->actions_before; list = g_list_prepend(list, wrapper); rh_action->actions_before = list; return TRUE; } void destroy_ticket(gpointer data) { pcmk_ticket_t *ticket = data; if (ticket->state) { g_hash_table_destroy(ticket->state); } free(ticket->id); free(ticket); } pcmk_ticket_t * ticket_new(const char *ticket_id, pcmk_scheduler_t *scheduler) { pcmk_ticket_t *ticket = NULL; if (pcmk__str_empty(ticket_id)) { return NULL; } if (scheduler->tickets == NULL) { scheduler->tickets = pcmk__strkey_table(free, destroy_ticket); } ticket = g_hash_table_lookup(scheduler->tickets, ticket_id); if (ticket == NULL) { ticket = calloc(1, sizeof(pcmk_ticket_t)); if (ticket == NULL) { pcmk__sched_err("Cannot allocate ticket '%s'", ticket_id); return NULL; } crm_trace("Creating ticket entry for %s", ticket_id); ticket->id = strdup(ticket_id); ticket->granted = FALSE; ticket->last_granted = -1; ticket->standby = FALSE; ticket->state = pcmk__strkey_table(free, free); g_hash_table_insert(scheduler->tickets, strdup(ticket->id), ticket); } return ticket; } const char * rsc_printable_id(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { return rsc->id; } return pcmk__xe_id(rsc->private->xml); } void pe__clear_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags) { pcmk__clear_rsc_flags(rsc, flags); for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe__clear_resource_flags_recursive((pcmk_resource_t *) gIter->data, flags); } } void pe__clear_resource_flags_on_all(pcmk_scheduler_t *scheduler, uint64_t flag) { for (GList *lpc = scheduler->resources; lpc != NULL; lpc = lpc->next) { pcmk_resource_t *r = (pcmk_resource_t *) lpc->data; pe__clear_resource_flags_recursive(r, flag); } } void pe__set_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags) { pcmk__set_rsc_flags(rsc, flags); for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe__set_resource_flags_recursive((pcmk_resource_t *) gIter->data, flags); } } void trigger_unfencing(pcmk_resource_t *rsc, pcmk_node_t *node, const char *reason, pcmk_action_t *dependency, pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_enable_unfencing)) { /* No resources require it */ return; } else if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) { /* Wasn't a stonith device */ return; } else if(node && node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { pcmk_action_t *unfence = pe_fence_op(node, PCMK_ACTION_ON, FALSE, reason, FALSE, scheduler); if(dependency) { order_actions(unfence, dependency, pcmk__ar_ordered); } } else if(rsc) { GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if(node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { trigger_unfencing(rsc, node, reason, dependency, scheduler); } } } } gboolean add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref) { pcmk_tag_t *tag = NULL; GList *gIter = NULL; gboolean is_existing = FALSE; CRM_CHECK(tags && tag_name && obj_ref, return FALSE); tag = g_hash_table_lookup(tags, tag_name); if (tag == NULL) { tag = calloc(1, sizeof(pcmk_tag_t)); if (tag == NULL) { pcmk__sched_err("Could not allocate memory for tag %s", tag_name); return FALSE; } tag->id = strdup(tag_name); tag->refs = NULL; g_hash_table_insert(tags, strdup(tag_name), tag); } for (gIter = tag->refs; gIter != NULL; gIter = gIter->next) { const char *existing_ref = (const char *) gIter->data; if (pcmk__str_eq(existing_ref, obj_ref, pcmk__str_none)){ is_existing = TRUE; break; } } if (is_existing == FALSE) { tag->refs = g_list_append(tag->refs, strdup(obj_ref)); crm_trace("Added: tag=%s ref=%s", tag->id, obj_ref); } return TRUE; } /*! * \internal * \brief Check whether shutdown has been requested for a node * * \param[in] node Node to check * * \return TRUE if node has shutdown attribute set and nonzero, FALSE otherwise * \note This differs from simply using node->details->shutdown in that it can * be used before that has been determined (and in fact to determine it), * and it can also be used to distinguish requested shutdown from implicit * shutdown of remote nodes by virtue of their connection stopping. */ bool pe__shutdown_requested(const pcmk_node_t *node) { const char *shutdown = pcmk__node_attr(node, PCMK__NODE_ATTR_SHUTDOWN, NULL, pcmk__rsc_node_current); return !pcmk__str_eq(shutdown, "0", pcmk__str_null_matches); } /*! * \internal * \brief Update "recheck by" time in scheduler data * * \param[in] recheck Epoch time when recheck should happen * \param[in,out] scheduler Scheduler data * \param[in] reason What time is being updated for (for logs) */ void pe__update_recheck_time(time_t recheck, pcmk_scheduler_t *scheduler, const char *reason) { if ((recheck > get_effective_time(scheduler)) && ((scheduler->recheck_by == 0) || (scheduler->recheck_by > recheck))) { scheduler->recheck_by = recheck; crm_debug("Updated next scheduler recheck to %s for %s", pcmk__trim(ctime(&recheck)), reason); } } /*! * \internal * \brief Extract nvpair blocks contained by a CIB XML element into a hash table * * \param[in] xml_obj XML element containing blocks of nvpair elements * \param[in] set_name If not NULL, only use blocks of this element * \param[in] rule_data Matching parameters to use when unpacking * \param[out] hash Where to store extracted name/value pairs * \param[in] always_first If not NULL, process block with this ID first * \param[in] overwrite Whether to replace existing values with same name * \param[in,out] scheduler Scheduler data containing \p xml_obj */ void pe__unpack_dataset_nvpairs(const xmlNode *xml_obj, const char *set_name, const pe_rule_eval_data_t *rule_data, GHashTable *hash, const char *always_first, gboolean overwrite, pcmk_scheduler_t *scheduler) { crm_time_t *next_change = crm_time_new_undefined(); pe_eval_nvpairs(scheduler->input, xml_obj, set_name, rule_data, hash, always_first, overwrite, next_change); if (crm_time_is_defined(next_change)) { time_t recheck = (time_t) crm_time_get_seconds_since_epoch(next_change); pe__update_recheck_time(recheck, scheduler, "rule evaluation"); } crm_time_free(next_change); } bool pe__resource_is_disabled(const pcmk_resource_t *rsc) { const char *target_role = NULL; CRM_CHECK(rsc != NULL, return false); target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); if (target_role) { // If invalid, we've already logged an error when unpacking enum rsc_role_e target_role_e = pcmk_parse_role(target_role); if ((target_role_e == pcmk_role_stopped) || ((target_role_e == pcmk_role_unpromoted) && pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_promotable))) { return true; } } return false; } /*! * \internal * \brief Check whether a resource is running only on given node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p rsc is running only on \p node, otherwise false */ bool pe__rsc_running_on_only(const pcmk_resource_t *rsc, const pcmk_node_t *node) { return (rsc != NULL) && pcmk__list_of_1(rsc->running_on) && pcmk__same_node((const pcmk_node_t *) rsc->running_on->data, node); } bool pe__rsc_running_on_any(pcmk_resource_t *rsc, GList *node_list) { for (GList *ele = rsc->running_on; ele; ele = ele->next) { pcmk_node_t *node = (pcmk_node_t *) ele->data; if (pcmk__str_in_list(node->details->uname, node_list, pcmk__str_star_matches|pcmk__str_casei)) { return true; } } return false; } bool pcmk__rsc_filtered_by_node(pcmk_resource_t *rsc, GList *only_node) { return rsc->private->fns->active(rsc, FALSE) && !pe__rsc_running_on_any(rsc, only_node); } GList * pe__filter_rsc_list(GList *rscs, GList *filter) { GList *retval = NULL; for (GList *gIter = rscs; gIter; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; /* I think the second condition is safe here for all callers of this * function. If not, it needs to move into pe__node_text. */ if (pcmk__str_in_list(rsc_printable_id(rsc), filter, pcmk__str_star_matches) || ((rsc->private->parent != NULL) && pcmk__str_in_list(rsc_printable_id(rsc->private->parent), filter, pcmk__str_star_matches))) { retval = g_list_prepend(retval, rsc); } } return retval; } GList * pe__build_node_name_list(pcmk_scheduler_t *scheduler, const char *s) { GList *nodes = NULL; if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) { /* Nothing was given so return a list of all node names. Or, '*' was * given. This would normally fall into the pe__unames_with_tag branch * where it will return an empty list. Catch it here instead. */ nodes = g_list_prepend(nodes, strdup("*")); } else { pcmk_node_t *node = pcmk_find_node(scheduler, s); if (node) { /* The given string was a valid uname for a node. Return a * singleton list containing just that uname. */ nodes = g_list_prepend(nodes, strdup(s)); } else { /* The given string was not a valid uname. It's either a tag or * it's a typo or something. In the first case, we'll return a * list of all the unames of the nodes with the given tag. In the * second case, we'll return a NULL pointer and nothing will * get displayed. */ nodes = pe__unames_with_tag(scheduler, s); } } return nodes; } GList * pe__build_rsc_list(pcmk_scheduler_t *scheduler, const char *s) { GList *resources = NULL; if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) { resources = g_list_prepend(resources, strdup("*")); } else { const uint32_t flags = pcmk_rsc_match_history|pcmk_rsc_match_basename; pcmk_resource_t *rsc = pe_find_resource_with_flags(scheduler->resources, s, flags); if (rsc) { /* A colon in the name we were given means we're being asked to filter * on a specific instance of a cloned resource. Put that exact string * into the filter list. Otherwise, use the printable ID of whatever * resource was found that matches what was asked for. */ if (strstr(s, ":") != NULL) { resources = g_list_prepend(resources, strdup(rsc->id)); } else { resources = g_list_prepend(resources, strdup(rsc_printable_id(rsc))); } } else { /* The given string was not a valid resource name. It's a tag or a * typo or something. See pe__build_node_name_list() for more * detail. */ resources = pe__rscs_with_tag(scheduler, s); } } return resources; } xmlNode * pe__failed_probe_for_rsc(const pcmk_resource_t *rsc, const char *name) { const pcmk_resource_t *parent = pe__const_top_resource(rsc, false); const char *rsc_id = rsc->id; if (pcmk__is_clone(parent)) { rsc_id = pe__clone_child_id(parent); } for (xmlNode *xml_op = pcmk__xe_first_child(rsc->private->scheduler->failed, NULL, NULL, NULL); xml_op != NULL; xml_op = pcmk__xe_next(xml_op)) { const char *value = NULL; char *op_id = NULL; /* This resource operation is not a failed probe. */ if (!pcmk_xe_mask_probe_failure(xml_op)) { continue; } /* This resource operation was not run on the given node. Note that if name is * NULL, this will always succeed. */ value = crm_element_value(xml_op, PCMK__META_ON_NODE); if (value == NULL || !pcmk__str_eq(value, name, pcmk__str_casei|pcmk__str_null_matches)) { continue; } if (!parse_op_key(pcmk__xe_history_key(xml_op), &op_id, NULL, NULL)) { continue; // This history entry is missing an operation key } /* This resource operation's ID does not match the rsc_id we are looking for. */ if (!pcmk__str_eq(op_id, rsc_id, pcmk__str_none)) { free(op_id); continue; } free(op_id); return xml_op; } return NULL; }