diff --git a/include/crm/common/resources.h b/include/crm/common/resources.h index e1a9a23305..2e0fdf3544 100644 --- a/include/crm/common/resources.h +++ b/include/crm/common/resources.h @@ -1,155 +1,152 @@ /* * 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 - // 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 a5ba02eb4f..3f8e0728d2 100644 --- a/include/crm/common/resources_internal.h +++ b/include/crm/common/resources_internal.h @@ -1,402 +1,412 @@ /* * 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; + /* The active nodes are ones where the resource is (or might be, if + * insufficient information is available to be sure) already active at the + * start of the current scheduler transition. + * + * For primitive resources, there should be at most one, but could be more + * if it is (incorrectly) multiply active. For collective resources, this + * combines active nodes of all descendants. + */ + GList *active_nodes; + // The source node, if migrate_to completed but migrate_from has not pcmk_node_t *partial_migration_source; // The destination node, if migrate_to completed but migrate_from has not pcmk_node_t *partial_migration_target; /* 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 9466e6c7aa..55cada39f0 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->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 (container->private->active_nodes != NULL) { + host = container->private->active_nodes->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/libpacemaker_private.h b/lib/pacemaker/libpacemaker_private.h index e4e835fdb5..9e8f5b04bb 100644 --- a/lib/pacemaker/libpacemaker_private.h +++ b/lib/pacemaker/libpacemaker_private.h @@ -1,1181 +1,1191 @@ /* * Copyright 2021-2024 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__LIBPACEMAKER_PRIVATE__H # define PCMK__LIBPACEMAKER_PRIVATE__H /* This header is for the sole use of libpacemaker, so that functions can be * declared with G_GNUC_INTERNAL for efficiency. */ -#include // lrmd_event_data_t -#include // pcmk_action_t, pcmk_node_t, etc. -#include // pcmk__location_t +#include // NULL +#include // uint32_t +#include // bool, false +#include // guint, gpointer, GList, GHashTable +#include // xmlNode + +#include // pcmk_action_t, pcmk_node_t, etc. +#include // pcmk__location_t, etc. +#include // cib_t +#include // lrmd_event_data_t +#include // pe__const_top_resource(), etc. +#include // pcmk_injections_t +#include // pcmk__colocation_t // Colocation flags enum pcmk__coloc_flags { pcmk__coloc_none = 0U, // Primary is affected even if already active pcmk__coloc_influence = (1U << 0), // Colocation was explicitly configured in CIB pcmk__coloc_explicit = (1U << 1), }; // Flags to modify the behavior of add_colocated_node_scores() enum pcmk__coloc_select { // With no other flags, apply all "with this" colocations pcmk__coloc_select_default = 0, // Apply "this with" colocations instead of "with this" colocations pcmk__coloc_select_this_with = (1 << 0), // Apply only colocations with non-negative scores pcmk__coloc_select_nonnegative = (1 << 1), // Apply only colocations with at least one matching node pcmk__coloc_select_active = (1 << 2), }; // Flags the update_ordered_actions() method can return enum pcmk__updated { pcmk__updated_none = 0, // Nothing changed pcmk__updated_first = (1 << 0), // First action was updated pcmk__updated_then = (1 << 1), // Then action was updated }; #define pcmk__set_updated_flags(au_flags, action, flags_to_set) do { \ au_flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Action update", \ (action)->uuid, au_flags, \ (flags_to_set), #flags_to_set); \ } while (0) #define pcmk__clear_updated_flags(au_flags, action, flags_to_clear) do { \ au_flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Action update", \ (action)->uuid, au_flags, \ (flags_to_clear), #flags_to_clear); \ } while (0) // Resource assignment methods struct pcmk__assignment_methods { /*! * \internal * \brief Assign a resource to a node * * \param[in,out] rsc Resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a * node, set next role to stopped and update * existing actions (if \p rsc is not a * primitive, this applies to its primitive * descendants instead) * * \return Node that \p rsc is assigned to, if assigned entirely to one node * * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() * can completely undo the assignment. A successful assignment can be * either undone or left alone as final. A failed assignment has the * same effect as calling pcmk__unassign_resource(); there are no side * effects on roles or actions. */ pcmk_node_t *(*assign)(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail); /*! * \internal * \brief Create all actions needed for a given resource * * \param[in,out] rsc Resource to create actions for */ void (*create_actions)(pcmk_resource_t *rsc); /*! * \internal * \brief Schedule any probes needed for a resource on a node * * \param[in,out] rsc Resource to create probe for * \param[in,out] node Node to create probe on * * \return true if any probe was created, otherwise false */ bool (*create_probe)(pcmk_resource_t *rsc, pcmk_node_t *node); /*! * \internal * \brief Create implicit constraints needed for a resource * * \param[in,out] rsc Resource to create implicit constraints for */ void (*internal_constraints)(pcmk_resource_t *rsc); /*! * \internal * \brief Apply a colocation's score to node scores or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node scores (if we are still placing resources) or priority (if * we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void (*apply_coloc_score)(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); /*! * \internal * \brief Create list of all resources in colocations with a given resource * * Given a resource, create a list of all resources involved in mandatory * colocations with it, whether directly or via chained colocations. * * \param[in] rsc Resource to add to colocated list * \param[in] orig_rsc Resource originally requested * \param[in,out] colocated_rscs Existing list * * \return List of given resource and all resources involved in colocations * * \note This function is recursive; top-level callers should pass NULL as * \p colocated_rscs and \p orig_rsc, and the desired resource as * \p rsc. The recursive calls will use other values. */ GList *(*colocated_resources)(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs); /*! * \internal * \brief Add colocations affecting a resource as primary to a list * * Given a resource being assigned (\p orig_rsc) and a resource somewhere in * its chain of ancestors (\p rsc, which may be \p orig_rsc), get * colocations that affect the ancestor as primary and should affect the * resource, and add them to a given list. * * \param[in] rsc Resource whose colocations should be added * \param[in] orig_rsc Affected resource (\p rsc or a descendant) * \param[in,out] list List of colocations to add to * * \note All arguments should be non-NULL. * \note The pcmk__with_this_colocations() wrapper should usually be used * instead of using this method directly. */ void (*with_this_colocations)(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); /*! * \internal * \brief Add colocations affecting a resource as dependent to a list * * Given a resource being assigned (\p orig_rsc) and a resource somewhere in * its chain of ancestors (\p rsc, which may be \p orig_rsc), get * colocations that affect the ancestor as dependent and should affect the * resource, and add them to a given list. * * * \param[in] rsc Resource whose colocations should be added * \param[in] orig_rsc Affected resource (\p rsc or a descendant) * \param[in,out] list List of colocations to add to * * \note All arguments should be non-NULL. * \note The pcmk__this_with_colocations() wrapper should usually be used * instead of using this method directly. */ void (*this_with_colocations)(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); /*! * \internal * \brief Update nodes with scores of colocated resources' nodes * * Given a table of nodes and a resource, update the nodes' scores with the * scores of the best nodes matching the attribute used for each of the * resource's relevant colocations. * * \param[in,out] source_rsc Resource whose node scores to add * \param[in] target_rsc Resource on whose behalf to update \p *nodes * \param[in] log_id Resource ID for logs (if \c NULL, use * \p source_rsc ID) * \param[in,out] nodes Nodes to update (set initial contents to * \c NULL to copy allowed nodes from * \p source_rsc) * \param[in] colocation Original colocation constraint (used to get * configured primary resource's stickiness, and * to get colocation node attribute; if \c NULL, * source_rsc's own matching node scores * will not be added, and \p *nodes must be * \c NULL as well) * \param[in] factor Incorporate scores multiplied by this factor * \param[in] flags Bitmask of enum pcmk__coloc_select values * * \note \c NULL \p target_rsc, \c NULL \p *nodes, \c NULL \p colocation, * and the \c pcmk__coloc_select_this_with flag are used together (and * only by \c cmp_resources()). * \note The caller remains responsible for freeing \p *nodes. */ void (*add_colocated_node_scores)(pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags); /*! * \internal * \brief Apply a location constraint to a resource's allowed node scores * * \param[in,out] rsc Resource to apply constraint to * \param[in,out] location Location constraint to apply */ void (*apply_location)(pcmk_resource_t *rsc, pcmk__location_t *location); /*! * \internal * \brief Return action flags for a given resource action * * \param[in,out] action Action to get flags for * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node * \note For primitives, this will be the same as action->flags regardless * of node. For collective resources, the flags can differ due to * multiple instances possibly being involved. */ uint32_t (*action_flags)(pcmk_action_t *action, const pcmk_node_t *node); /*! * \internal * \brief Update two actions according to an ordering between them * * Given information about an ordering of two actions, update the actions' * flags (and runnable_before members if appropriate) as appropriate for the * ordering. Effects may cascade to other orderings involving the actions as * well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this * node (only used when interleaving instances) * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates * (may include pcmk_action_optional to affect * only mandatory actions and pcmk_action_runnable * to affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t (*update_ordered_actions)(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler); /*! * \internal * \brief Output a summary of scheduled actions for a resource * * \param[in,out] rsc Resource to output actions for */ void (*output_actions)(pcmk_resource_t *rsc); /*! * \internal * \brief Add a resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void (*add_actions_to_graph)(pcmk_resource_t *rsc); /*! * \internal * \brief Add meta-attributes relevant to transition graph actions to XML * * If a given resource supports variant-specific meta-attributes that are * needed for transition graph actions, add them to a given XML element. * * \param[in] rsc Resource whose meta-attributes should be added * \param[in,out] xml Transition graph action attributes XML to add to */ void (*add_graph_meta)(const pcmk_resource_t *rsc, xmlNode *xml); /*! * \internal * \brief Add a resource's utilization to a table of utilization values * * This function is used when summing the utilization of a resource and all * resources colocated with it, to determine whether a node has sufficient * capacity. Given a resource and a table of utilization values, it will add * the resource's utilization to the existing values, if the resource has * not yet been assigned to a node. * * \param[in] rsc Resource with utilization to add * \param[in] orig_rsc Resource being assigned (for logging only) * \param[in] all_rscs List of all resources that will be summed * \param[in,out] utilization Table of utilization values to add to */ void (*add_utilization)(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); /*! * \internal * \brief Apply a shutdown lock for a resource, if appropriate * * \param[in,out] rsc Resource to check for shutdown lock */ void (*shutdown_lock)(pcmk_resource_t *rsc); }; // Actions (pcmk_sched_actions.c) G_GNUC_INTERNAL void pcmk__update_action_for_orderings(pcmk_action_t *action, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL uint32_t pcmk__update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__log_action(const char *pre_text, const pcmk_action_t *action, bool details); G_GNUC_INTERNAL pcmk_action_t *pcmk__new_cancel_action(pcmk_resource_t *rsc, const char *name, guint interval_ms, const pcmk_node_t *node); G_GNUC_INTERNAL pcmk_action_t *pcmk__new_shutdown_action(pcmk_node_t *node); G_GNUC_INTERNAL bool pcmk__action_locks_rsc_to_node(const pcmk_action_t *action); G_GNUC_INTERNAL void pcmk__deduplicate_action_inputs(pcmk_action_t *action); G_GNUC_INTERNAL void pcmk__output_actions(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL bool pcmk__check_action_config(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *xml_op); G_GNUC_INTERNAL void pcmk__handle_rsc_config_changes(pcmk_scheduler_t *scheduler); // Recurring actions (pcmk_sched_recurring.c) G_GNUC_INTERNAL void pcmk__create_recurring_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__schedule_cancel(pcmk_resource_t *rsc, const char *call_id, const char *task, guint interval_ms, const pcmk_node_t *node, const char *reason); G_GNUC_INTERNAL void pcmk__reschedule_recurring(pcmk_resource_t *rsc, const char *task, guint interval_ms, pcmk_node_t *node); G_GNUC_INTERNAL bool pcmk__action_is_recurring(const pcmk_action_t *action); // Producing transition graphs (pcmk_graph_producer.c) G_GNUC_INTERNAL bool pcmk__graph_has_loop(const pcmk_action_t *init_action, const pcmk_action_t *action, pcmk__related_action_t *input); G_GNUC_INTERNAL void pcmk__add_rsc_actions_to_graph(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_graph(pcmk_scheduler_t *scheduler); // Fencing (pcmk_sched_fencing.c) G_GNUC_INTERNAL void pcmk__order_vs_fence(pcmk_action_t *stonith_op, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__order_vs_unfence(const pcmk_resource_t *rsc, pcmk_node_t *node, pcmk_action_t *action, enum pcmk__action_relation_flags order); G_GNUC_INTERNAL void pcmk__fence_guest(pcmk_node_t *node); G_GNUC_INTERNAL bool pcmk__node_unfenced(const pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__order_restart_vs_unfence(gpointer data, gpointer user_data); // Injected scheduler inputs (pcmk_sched_injections.c) void pcmk__inject_scheduler_input(pcmk_scheduler_t *scheduler, cib_t *cib, const pcmk_injections_t *injections); // Constraints of any type (pcmk_sched_constraints.c) G_GNUC_INTERNAL pcmk_resource_t *pcmk__find_constraint_resource(GList *rsc_list, const char *id); G_GNUC_INTERNAL xmlNode *pcmk__expand_tags_in_sets(xmlNode *xml_obj, const pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL bool pcmk__valid_resource_or_tag(const pcmk_scheduler_t *scheduler, const char *id, pcmk_resource_t **rsc, pcmk_tag_t **tag); G_GNUC_INTERNAL bool pcmk__tag_to_set(xmlNode *xml_obj, xmlNode **rsc_set, const char *attr, bool convert_rsc, const pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__create_internal_constraints(pcmk_scheduler_t *scheduler); // Location constraints G_GNUC_INTERNAL void pcmk__unpack_location(xmlNode *xml_obj, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL pcmk__location_t *pcmk__new_location(const char *id, pcmk_resource_t *rsc, int node_score, const char *discover_mode, pcmk_node_t *foo_node); G_GNUC_INTERNAL void pcmk__apply_locations(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__apply_location(pcmk_resource_t *rsc, pcmk__location_t *constraint); // Colocation constraints (pcmk_sched_colocation.c) enum pcmk__coloc_affects { pcmk__coloc_affects_nothing = 0, pcmk__coloc_affects_location, pcmk__coloc_affects_role, }; /*! * \internal * \brief Get the value of a colocation's node attribute * * \param[in] node Node on which to look up the attribute * \param[in] attr Name of attribute to look up * \param[in] rsc Resource on whose behalf to look up the attribute * * \return Value of \p attr on \p node or on the host of \p node, as appropriate */ static inline const char * pcmk__colocation_node_attr(const pcmk_node_t *node, const char *attr, const pcmk_resource_t *rsc) { const char *target = NULL; /* A resource colocated with a bundle or its primitive can't run on the * bundle node itself (where only the primitive, if any, can run). Instead, * we treat it as a colocation with the bundle's containers, so always look * up colocation node attributes on the container host. */ if (pcmk__is_bundle_node(node) && pcmk__is_bundled(rsc) && (pe__const_top_resource(rsc, false) == pe__bundled_resource(rsc))) { target = PCMK_VALUE_HOST; } else if (rsc != NULL) { target = g_hash_table_lookup(rsc->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); } return pcmk__node_attr(node, attr, target, pcmk__rsc_node_assigned); } G_GNUC_INTERNAL enum pcmk__coloc_affects pcmk__colocation_affects(const pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool preview); G_GNUC_INTERNAL void pcmk__apply_coloc_to_scores(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__apply_coloc_to_priority(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__add_colocated_node_scores(pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags); G_GNUC_INTERNAL void pcmk__add_dependent_scores(gpointer data, gpointer user_data); G_GNUC_INTERNAL void pcmk__colocation_intersect_nodes(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, const GList *primary_nodes, bool merge_scores); G_GNUC_INTERNAL void pcmk__unpack_colocation(xmlNode *xml_obj, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__add_this_with(GList **list, const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__add_this_with_list(GList **list, GList *addition, const pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__add_with_this(GList **list, const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__add_with_this_list(GList **list, GList *addition, const pcmk_resource_t *rsc); G_GNUC_INTERNAL GList *pcmk__with_this_colocations(const pcmk_resource_t *rsc); G_GNUC_INTERNAL GList *pcmk__this_with_colocations(const pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__new_colocation(const char *id, const char *node_attr, int score, pcmk_resource_t *dependent, pcmk_resource_t *primary, const char *dependent_role, const char *primary_role, uint32_t flags); G_GNUC_INTERNAL void pcmk__block_colocation_dependents(pcmk_action_t *action); /*! * \internal * \brief Check whether colocation's dependent preferences should be considered * * \param[in] colocation Colocation constraint * \param[in] rsc Primary instance (normally this will be * colocation->primary, which NULL will be treated as, * but for clones or bundles with multiple instances * this can be a particular instance) * * \return true if colocation influence should be effective, otherwise false */ static inline bool pcmk__colocation_has_influence(const pcmk__colocation_t *colocation, const pcmk_resource_t *rsc) { if (rsc == NULL) { rsc = colocation->primary; } /* A bundle replica colocates its remote connection with its container, * using a finite score so that the container can run on Pacemaker Remote * nodes. * * Moving a connection is lightweight and does not interrupt the service, * while moving a container is heavyweight and does interrupt the service, * so don't move a clean, active container based solely on the preferences * of its connection. * * This also avoids problematic scenarios where two containers want to * perpetually swap places. */ if (pcmk_is_set(colocation->dependent->flags, pcmk__rsc_remote_nesting_allowed) && !pcmk_is_set(rsc->flags, pcmk__rsc_failed) - && pcmk__list_of_1(rsc->running_on)) { + && pcmk__list_of_1(rsc->private->active_nodes)) { return false; } /* The dependent in a colocation influences the primary's location * if the PCMK_XA_INFLUENCE option is true or the primary is not yet active. */ return pcmk_is_set(colocation->flags, pcmk__coloc_influence) - || (rsc->running_on == NULL); + || (rsc->private->active_nodes == NULL); } // Ordering constraints (pcmk_sched_ordering.c) G_GNUC_INTERNAL void pcmk__new_ordering(pcmk_resource_t *first_rsc, char *first_task, pcmk_action_t *first_action, pcmk_resource_t *then_rsc, char *then_task, pcmk_action_t *then_action, uint32_t flags, pcmk_scheduler_t *sched); G_GNUC_INTERNAL void pcmk__unpack_ordering(xmlNode *xml_obj, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__disable_invalid_orderings(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL void pcmk__order_stops_before_shutdown(pcmk_node_t *node, pcmk_action_t *shutdown_op); G_GNUC_INTERNAL void pcmk__apply_orderings(pcmk_scheduler_t *sched); G_GNUC_INTERNAL void pcmk__order_after_each(pcmk_action_t *after, GList *list); /*! * \internal * \brief Create a new ordering between two resource actions * * \param[in,out] first_rsc Resource for 'first' action * \param[in,out] first_task Action key for 'first' action * \param[in] then_rsc Resource for 'then' action * \param[in,out] then_task Action key for 'then' action * \param[in] flags Group of enum pcmk__action_relation_flags */ #define pcmk__order_resource_actions(first_rsc, first_task, \ then_rsc, then_task, flags) \ pcmk__new_ordering((first_rsc), \ pcmk__op_key((first_rsc)->id, (first_task), 0), \ NULL, \ (then_rsc), \ pcmk__op_key((then_rsc)->id, (then_task), 0), \ NULL, (flags), (first_rsc)->private->scheduler) #define pcmk__order_starts(rsc1, rsc2, flags) \ pcmk__order_resource_actions((rsc1), PCMK_ACTION_START, \ (rsc2), PCMK_ACTION_START, (flags)) #define pcmk__order_stops(rsc1, rsc2, flags) \ pcmk__order_resource_actions((rsc1), PCMK_ACTION_STOP, \ (rsc2), PCMK_ACTION_STOP, (flags)) // Ticket constraints (pcmk_sched_tickets.c) G_GNUC_INTERNAL void pcmk__unpack_rsc_ticket(xmlNode *xml_obj, pcmk_scheduler_t *scheduler); // Promotable clone resources (pcmk_sched_promotable.c) G_GNUC_INTERNAL void pcmk__add_promotion_scores(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__require_promotion_tickets(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__set_instance_roles(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_promotable_actions(pcmk_resource_t *clone); G_GNUC_INTERNAL void pcmk__promotable_restart_ordering(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__order_promotable_instances(pcmk_resource_t *clone); G_GNUC_INTERNAL void pcmk__update_dependent_with_promotable(const pcmk_resource_t *primary, pcmk_resource_t *dependent, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__update_promotable_dependent_priority(const pcmk_resource_t *primary, pcmk_resource_t *dependent, const pcmk__colocation_t *colocation); // Pacemaker Remote nodes (pcmk_sched_remote.c) G_GNUC_INTERNAL bool pcmk__is_failed_remote_node(const pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__order_remote_connection_actions(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL bool pcmk__rsc_corresponds_to_guest(const pcmk_resource_t *rsc, const pcmk_node_t *node); G_GNUC_INTERNAL pcmk_node_t *pcmk__connection_host_for_action(const pcmk_action_t *action); G_GNUC_INTERNAL void pcmk__substitute_remote_addr(pcmk_resource_t *rsc, GHashTable *params); G_GNUC_INTERNAL void pcmk__add_guest_meta_to_xml(xmlNode *args_xml, const pcmk_action_t *action); // Primitives (pcmk_sched_primitive.c) G_GNUC_INTERNAL pcmk_node_t *pcmk__primitive_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail); G_GNUC_INTERNAL void pcmk__primitive_create_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__primitive_internal_constraints(pcmk_resource_t *rsc); G_GNUC_INTERNAL uint32_t pcmk__primitive_action_flags(pcmk_action_t *action, const pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__primitive_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_primitive_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__primitive_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__schedule_cleanup(pcmk_resource_t *rsc, const pcmk_node_t *node, bool optional); G_GNUC_INTERNAL void pcmk__primitive_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__primitive_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc); // Groups (pcmk_sched_group.c) G_GNUC_INTERNAL pcmk_node_t *pcmk__group_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail); G_GNUC_INTERNAL void pcmk__group_create_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__group_internal_constraints(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__group_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_group_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__group_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__group_add_colocated_node_scores(pcmk_resource_t *source_rsc, const pcmk_resource_t *target_rsc, const char *log_id, GHashTable **nodes, const pcmk__colocation_t *colocation, float factor, uint32_t flags); G_GNUC_INTERNAL void pcmk__group_apply_location(pcmk_resource_t *rsc, pcmk__location_t *location); G_GNUC_INTERNAL uint32_t pcmk__group_action_flags(pcmk_action_t *action, const pcmk_node_t *node); G_GNUC_INTERNAL uint32_t pcmk__group_update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL GList *pcmk__group_colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs); G_GNUC_INTERNAL void pcmk__group_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__group_shutdown_lock(pcmk_resource_t *rsc); // Clones (pcmk_sched_clone.c) G_GNUC_INTERNAL pcmk_node_t *pcmk__clone_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail); G_GNUC_INTERNAL void pcmk__clone_create_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__clone_create_probe(pcmk_resource_t *rsc, pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__clone_internal_constraints(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__clone_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_clone_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__clone_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__clone_apply_location(pcmk_resource_t *rsc, pcmk__location_t *constraint); G_GNUC_INTERNAL uint32_t pcmk__clone_action_flags(pcmk_action_t *action, const pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__clone_add_actions_to_graph(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__clone_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__clone_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__clone_shutdown_lock(pcmk_resource_t *rsc); // Bundles (pcmk_sched_bundle.c) G_GNUC_INTERNAL pcmk_node_t *pcmk__bundle_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer, bool stop_if_fail); G_GNUC_INTERNAL void pcmk__bundle_create_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__bundle_create_probe(pcmk_resource_t *rsc, pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__bundle_internal_constraints(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__bundle_apply_coloc_score(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_bundle_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__bundle_with_colocations(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__bundle_apply_location(pcmk_resource_t *rsc, pcmk__location_t *constraint); G_GNUC_INTERNAL uint32_t pcmk__bundle_action_flags(pcmk_action_t *action, const pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__output_bundle_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__bundle_add_actions_to_graph(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__bundle_add_utilization(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__bundle_shutdown_lock(pcmk_resource_t *rsc); // Clone instances or bundle replica containers (pcmk_sched_instances.c) G_GNUC_INTERNAL void pcmk__assign_instances(pcmk_resource_t *collective, GList *instances, int max_total, int max_per_node); G_GNUC_INTERNAL void pcmk__create_instance_actions(pcmk_resource_t *rsc, GList *instances); G_GNUC_INTERNAL bool pcmk__instance_matches(const pcmk_resource_t *instance, const pcmk_node_t *node, enum rsc_role_e role, bool current); G_GNUC_INTERNAL pcmk_resource_t *pcmk__find_compatible_instance(const pcmk_resource_t *match_rsc, const pcmk_resource_t *rsc, enum rsc_role_e role, bool current); G_GNUC_INTERNAL uint32_t pcmk__instance_update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL uint32_t pcmk__collective_action_flags(pcmk_action_t *action, const GList *instances, const pcmk_node_t *node); // Injections (pcmk_injections.c) G_GNUC_INTERNAL xmlNode *pcmk__inject_node(cib_t *cib_conn, const char *node, const char *uuid); G_GNUC_INTERNAL xmlNode *pcmk__inject_node_state_change(cib_t *cib_conn, const char *node, bool up); G_GNUC_INTERNAL xmlNode *pcmk__inject_resource_history(pcmk__output_t *out, xmlNode *cib_node, const char *resource, const char *lrm_name, const char *rclass, const char *rtype, const char *rprovider); G_GNUC_INTERNAL void pcmk__inject_failcount(pcmk__output_t *out, cib_t *cib_conn, xmlNode *cib_node, const char *resource, const char *task, guint interval_ms, int rc); G_GNUC_INTERNAL xmlNode *pcmk__inject_action_result(xmlNode *cib_resource, lrmd_event_data_t *op, int target_rc); // Nodes (pcmk_sched_nodes.c) G_GNUC_INTERNAL bool pcmk__node_available(const pcmk_node_t *node, bool consider_score, bool consider_guest); G_GNUC_INTERNAL bool pcmk__any_node_available(GHashTable *nodes); G_GNUC_INTERNAL GHashTable *pcmk__copy_node_table(GHashTable *nodes); G_GNUC_INTERNAL void pcmk__copy_node_tables(const pcmk_resource_t *rsc, GHashTable **copy); G_GNUC_INTERNAL void pcmk__restore_node_tables(pcmk_resource_t *rsc, GHashTable *backup); G_GNUC_INTERNAL GList *pcmk__sort_nodes(GList *nodes, pcmk_node_t *active_node); G_GNUC_INTERNAL void pcmk__apply_node_health(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL pcmk_node_t *pcmk__top_allowed_node(const pcmk_resource_t *rsc, const pcmk_node_t *node); // Functions applying to more than one variant (pcmk_sched_resource.c) G_GNUC_INTERNAL void pcmk__set_assignment_methods(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL bool pcmk__rsc_agent_changed(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *rsc_entry, bool active_on_node); G_GNUC_INTERNAL GList *pcmk__rscs_matching_id(const char *id, const pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL GList *pcmk__colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs); G_GNUC_INTERNAL void pcmk__noop_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__output_resource_actions(pcmk_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__assign_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool force, bool stop_if_fail); G_GNUC_INTERNAL void pcmk__unassign_resource(pcmk_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__threshold_reached(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_resource_t **failed); G_GNUC_INTERNAL void pcmk__sort_resources(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL gint pcmk__cmp_instance(gconstpointer a, gconstpointer b); G_GNUC_INTERNAL gint pcmk__cmp_instance_number(gconstpointer a, gconstpointer b); // Functions related to probes (pcmk_sched_probes.c) G_GNUC_INTERNAL bool pcmk__probe_rsc_on_node(pcmk_resource_t *rsc, pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__order_probes(pcmk_scheduler_t *scheduler); G_GNUC_INTERNAL bool pcmk__probe_resource_list(GList *rscs, pcmk_node_t *node); G_GNUC_INTERNAL void pcmk__schedule_probes(pcmk_scheduler_t *scheduler); // Functions related to live migration (pcmk_sched_migration.c) void pcmk__create_migration_actions(pcmk_resource_t *rsc, const pcmk_node_t *current); void pcmk__abort_dangling_migration(void *data, void *user_data); bool pcmk__rsc_can_migrate(const pcmk_resource_t *rsc, const pcmk_node_t *current); void pcmk__order_migration_equivalents(pcmk__action_relation_t *order); // Functions related to node utilization (pcmk_sched_utilization.c) G_GNUC_INTERNAL int pcmk__compare_node_capacities(const pcmk_node_t *node1, const pcmk_node_t *node2); G_GNUC_INTERNAL void pcmk__consume_node_capacity(GHashTable *current_utilization, const pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__release_node_capacity(GHashTable *current_utilization, const pcmk_resource_t *rsc); G_GNUC_INTERNAL const pcmk_node_t *pcmk__ban_insufficient_capacity(pcmk_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_utilization_constraints(pcmk_resource_t *rsc, const GList *allowed_nodes); G_GNUC_INTERNAL void pcmk__show_node_capacities(const char *desc, pcmk_scheduler_t *scheduler); // Functions related to the scheduler (pcmk_scheduler.c) G_GNUC_INTERNAL int pcmk__init_scheduler(pcmk__output_t *out, xmlNodePtr input, const crm_time_t *date, pcmk_scheduler_t **scheduler); // General setup functions (pcmk_setup.c) G_GNUC_INTERNAL int pcmk__setup_output_cib_sched(pcmk__output_t **out, cib_t **cib, pcmk_scheduler_t **scheduler, xmlNode **xml); #endif // PCMK__LIBPACEMAKER_PRIVATE__H diff --git a/lib/pacemaker/pcmk_output.c b/lib/pacemaker/pcmk_output.c index 8bbf0aa9e6..a5db61bbc3 100644 --- a/lib/pacemaker/pcmk_output.c +++ b/lib/pacemaker/pcmk_output.c @@ -1,2697 +1,2699 @@ /* * Copyright 2019-2024 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 // stonith__* #include #include #include #include #include static char * colocations_header(pcmk_resource_t *rsc, pcmk__colocation_t *cons, bool dependents) { char *retval = NULL; if (cons->primary_role > pcmk_role_started) { retval = crm_strdup_printf("%s (score=%s, %s role=%s, id=%s)", rsc->id, pcmk_readable_score(cons->score), (dependents? "needs" : "with"), pcmk_role_text(cons->primary_role), cons->id); } else { retval = crm_strdup_printf("%s (score=%s, id=%s)", rsc->id, pcmk_readable_score(cons->score), cons->id); } return retval; } static void colocations_xml_node(pcmk__output_t *out, pcmk_resource_t *rsc, pcmk__colocation_t *cons) { xmlNodePtr node = NULL; node = pcmk__output_create_xml_node(out, PCMK_XE_RSC_COLOCATION, PCMK_XA_ID, cons->id, PCMK_XA_RSC, cons->dependent->id, PCMK_XA_WITH_RSC, cons->primary->id, PCMK_XA_SCORE, pcmk_readable_score(cons->score), NULL); if (cons->node_attribute) { xmlSetProp(node, (pcmkXmlStr) PCMK_XA_NODE_ATTRIBUTE, (pcmkXmlStr) cons->node_attribute); } if (cons->dependent_role != pcmk_role_unknown) { xmlSetProp(node, (pcmkXmlStr) PCMK_XA_RSC_ROLE, (pcmkXmlStr) pcmk_role_text(cons->dependent_role)); } if (cons->primary_role != pcmk_role_unknown) { xmlSetProp(node, (pcmkXmlStr) PCMK_XA_WITH_RSC_ROLE, (pcmkXmlStr) pcmk_role_text(cons->primary_role)); } } static int do_locations_list_xml(pcmk__output_t *out, pcmk_resource_t *rsc, bool add_header) { GList *lpc = NULL; int rc = pcmk_rc_no_output; for (lpc = rsc->private->location_constraints; lpc != NULL; lpc = lpc->next) { pcmk__location_t *cons = lpc->data; GList *lpc2 = NULL; for (lpc2 = cons->nodes; lpc2 != NULL; lpc2 = lpc2->next) { pcmk_node_t *node = (pcmk_node_t *) lpc2->data; if (add_header) { PCMK__OUTPUT_LIST_HEADER(out, false, rc, "locations"); } pcmk__output_create_xml_node(out, PCMK_XE_RSC_LOCATION, PCMK_XA_NODE, node->details->uname, PCMK_XA_RSC, rsc->id, PCMK_XA_ID, cons->id, PCMK_XA_SCORE, pcmk_readable_score(node->weight), NULL); } } if (add_header) { PCMK__OUTPUT_LIST_FOOTER(out, rc); } return rc; } PCMK__OUTPUT_ARGS("rsc-action-item", "const char *", "pcmk_resource_t *", "pcmk_node_t *", "pcmk_node_t *", "pcmk_action_t *", "pcmk_action_t *") static int rsc_action_item(pcmk__output_t *out, va_list args) { const char *change = va_arg(args, const char *); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); pcmk_node_t *origin = va_arg(args, pcmk_node_t *); pcmk_node_t *destination = va_arg(args, pcmk_node_t *); pcmk_action_t *action = va_arg(args, pcmk_action_t *); pcmk_action_t *source = va_arg(args, pcmk_action_t *); int len = 0; char *reason = NULL; char *details = NULL; bool same_host = false; bool same_role = false; bool need_role = false; static int rsc_width = 5; static int detail_width = 5; CRM_ASSERT(action); CRM_ASSERT(destination != NULL || origin != NULL); if (source == NULL) { source = action; } len = strlen(rsc->id); if (len > rsc_width) { rsc_width = len + 2; } if ((rsc->role > pcmk_role_started) || (rsc->next_role > pcmk_role_unpromoted)) { need_role = true; } if (pcmk__same_node(origin, destination)) { same_host = true; } if (rsc->role == rsc->next_role) { same_role = true; } if (need_role && (origin == NULL)) { /* Starting and promoting a promotable clone instance */ details = crm_strdup_printf("%s -> %s %s", pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role), pcmk__node_name(destination)); } else if (origin == NULL) { /* Starting a resource */ details = crm_strdup_printf("%s", pcmk__node_name(destination)); } else if (need_role && (destination == NULL)) { /* Stopping a promotable clone instance */ details = crm_strdup_printf("%s %s", pcmk_role_text(rsc->role), pcmk__node_name(origin)); } else if (destination == NULL) { /* Stopping a resource */ details = crm_strdup_printf("%s", pcmk__node_name(origin)); } else if (need_role && same_role && same_host) { /* Recovering, restarting or re-promoting a promotable clone instance */ details = crm_strdup_printf("%s %s", pcmk_role_text(rsc->role), pcmk__node_name(origin)); } else if (same_role && same_host) { /* Recovering or Restarting a normal resource */ details = crm_strdup_printf("%s", pcmk__node_name(origin)); } else if (need_role && same_role) { /* Moving a promotable clone instance */ details = crm_strdup_printf("%s -> %s %s", pcmk__node_name(origin), pcmk__node_name(destination), pcmk_role_text(rsc->role)); } else if (same_role) { /* Moving a normal resource */ details = crm_strdup_printf("%s -> %s", pcmk__node_name(origin), pcmk__node_name(destination)); } else if (same_host) { /* Promoting or demoting a promotable clone instance */ details = crm_strdup_printf("%s -> %s %s", pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role), pcmk__node_name(origin)); } else { /* Moving and promoting/demoting */ details = crm_strdup_printf("%s %s -> %s %s", pcmk_role_text(rsc->role), pcmk__node_name(origin), pcmk_role_text(rsc->next_role), pcmk__node_name(destination)); } len = strlen(details); if (len > detail_width) { detail_width = len; } if ((source->reason != NULL) && !pcmk_is_set(action->flags, pcmk_action_runnable)) { reason = crm_strdup_printf("due to %s (blocked)", source->reason); } else if (source->reason) { reason = crm_strdup_printf("due to %s", source->reason); } else if (!pcmk_is_set(action->flags, pcmk_action_runnable)) { reason = strdup("blocked"); } out->list_item(out, NULL, "%-8s %-*s ( %*s )%s%s", change, rsc_width, rsc->id, detail_width, details, ((reason == NULL)? "" : " "), pcmk__s(reason, "")); free(details); free(reason); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("rsc-action-item", "const char *", "pcmk_resource_t *", "pcmk_node_t *", "pcmk_node_t *", "pcmk_action_t *", "pcmk_action_t *") static int rsc_action_item_xml(pcmk__output_t *out, va_list args) { const char *change = va_arg(args, const char *); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); pcmk_node_t *origin = va_arg(args, pcmk_node_t *); pcmk_node_t *destination = va_arg(args, pcmk_node_t *); pcmk_action_t *action = va_arg(args, pcmk_action_t *); pcmk_action_t *source = va_arg(args, pcmk_action_t *); char *change_str = NULL; bool same_host = false; bool same_role = false; bool need_role = false; xmlNode *xml = NULL; CRM_ASSERT(action); CRM_ASSERT(destination != NULL || origin != NULL); if (source == NULL) { source = action; } if ((rsc->role > pcmk_role_started) || (rsc->next_role > pcmk_role_unpromoted)) { need_role = true; } if (pcmk__same_node(origin, destination)) { same_host = true; } if (rsc->role == rsc->next_role) { same_role = true; } change_str = g_ascii_strdown(change, -1); xml = pcmk__output_create_xml_node(out, PCMK_XE_RSC_ACTION, PCMK_XA_ACTION, change_str, PCMK_XA_RESOURCE, rsc->id, NULL); g_free(change_str); if (need_role && (origin == NULL)) { /* Starting and promoting a promotable clone instance */ pcmk__xe_set_props(xml, PCMK_XA_ROLE, pcmk_role_text(rsc->role), PCMK_XA_NEXT_ROLE, pcmk_role_text(rsc->next_role), PCMK_XA_DEST, destination->details->uname, NULL); } else if (origin == NULL) { /* Starting a resource */ crm_xml_add(xml, PCMK_XA_NODE, destination->details->uname); } else if (need_role && (destination == NULL)) { /* Stopping a promotable clone instance */ pcmk__xe_set_props(xml, PCMK_XA_ROLE, pcmk_role_text(rsc->role), PCMK_XA_NODE, origin->details->uname, NULL); } else if (destination == NULL) { /* Stopping a resource */ crm_xml_add(xml, PCMK_XA_NODE, origin->details->uname); } else if (need_role && same_role && same_host) { /* Recovering, restarting or re-promoting a promotable clone instance */ pcmk__xe_set_props(xml, PCMK_XA_ROLE, pcmk_role_text(rsc->role), PCMK_XA_SOURCE, origin->details->uname, NULL); } else if (same_role && same_host) { /* Recovering or Restarting a normal resource */ crm_xml_add(xml, PCMK_XA_SOURCE, origin->details->uname); } else if (need_role && same_role) { /* Moving a promotable clone instance */ pcmk__xe_set_props(xml, PCMK_XA_SOURCE, origin->details->uname, PCMK_XA_DEST, destination->details->uname, PCMK_XA_ROLE, pcmk_role_text(rsc->role), NULL); } else if (same_role) { /* Moving a normal resource */ pcmk__xe_set_props(xml, PCMK_XA_SOURCE, origin->details->uname, PCMK_XA_DEST, destination->details->uname, NULL); } else if (same_host) { /* Promoting or demoting a promotable clone instance */ pcmk__xe_set_props(xml, PCMK_XA_ROLE, pcmk_role_text(rsc->role), PCMK_XA_NEXT_ROLE, pcmk_role_text(rsc->next_role), PCMK_XA_SOURCE, origin->details->uname, NULL); } else { /* Moving and promoting/demoting */ pcmk__xe_set_props(xml, PCMK_XA_ROLE, pcmk_role_text(rsc->role), PCMK_XA_SOURCE, origin->details->uname, PCMK_XA_NEXT_ROLE, pcmk_role_text(rsc->next_role), PCMK_XA_DEST, destination->details->uname, NULL); } if ((source->reason != NULL) && !pcmk_is_set(action->flags, pcmk_action_runnable)) { pcmk__xe_set_props(xml, PCMK_XA_REASON, source->reason, PCMK_XA_BLOCKED, PCMK_VALUE_TRUE, NULL); } else if (source->reason != NULL) { crm_xml_add(xml, PCMK_XA_REASON, source->reason); } else if (!pcmk_is_set(action->flags, pcmk_action_runnable)) { pcmk__xe_set_bool_attr(xml, PCMK_XA_BLOCKED, true); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("rsc-is-colocated-with-list", "pcmk_resource_t *", "bool") static int rsc_is_colocated_with_list(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); bool recursive = va_arg(args, int); int rc = pcmk_rc_no_output; if (pcmk_is_set(rsc->flags, pcmk__rsc_detect_loop)) { return rc; } /* We're listing constraints explicitly involving rsc, so use * rsc->private->this_with_colocations directly rather than call * rsc->private->cmds->this_with_colocations(). */ pcmk__set_rsc_flags(rsc, pcmk__rsc_detect_loop); for (GList *lpc = rsc->private->this_with_colocations; lpc != NULL; lpc = lpc->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data; char *hdr = NULL; PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Resources %s is colocated with", rsc->id); if (pcmk_is_set(cons->primary->flags, pcmk__rsc_detect_loop)) { out->list_item(out, NULL, "%s (id=%s - loop)", cons->primary->id, cons->id); continue; } hdr = colocations_header(cons->primary, cons, false); out->list_item(out, NULL, "%s", hdr); free(hdr); // Empty list header for indentation of information about this resource out->begin_list(out, NULL, NULL, NULL); out->message(out, "locations-list", cons->primary); if (recursive) { out->message(out, "rsc-is-colocated-with-list", cons->primary, recursive); } out->end_list(out); } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("rsc-is-colocated-with-list", "pcmk_resource_t *", "bool") static int rsc_is_colocated_with_list_xml(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); bool recursive = va_arg(args, int); int rc = pcmk_rc_no_output; if (pcmk_is_set(rsc->flags, pcmk__rsc_detect_loop)) { return rc; } /* We're listing constraints explicitly involving rsc, so use * rsc->private->this_with_colocations directly rather than call * rsc->private->cmds->this_with_colocations(). */ pcmk__set_rsc_flags(rsc, pcmk__rsc_detect_loop); for (GList *lpc = rsc->private->this_with_colocations; lpc != NULL; lpc = lpc->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data; if (pcmk_is_set(cons->primary->flags, pcmk__rsc_detect_loop)) { colocations_xml_node(out, cons->primary, cons); continue; } colocations_xml_node(out, cons->primary, cons); do_locations_list_xml(out, cons->primary, false); if (recursive) { out->message(out, "rsc-is-colocated-with-list", cons->primary, recursive); } } return rc; } PCMK__OUTPUT_ARGS("rscs-colocated-with-list", "pcmk_resource_t *", "bool") static int rscs_colocated_with_list(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); bool recursive = va_arg(args, int); int rc = pcmk_rc_no_output; if (pcmk_is_set(rsc->flags, pcmk__rsc_detect_loop)) { return rc; } /* We're listing constraints explicitly involving rsc, so use * rsc->private->with_this_colocations directly rather than * rsc->private->cmds->with_this_colocations(). */ pcmk__set_rsc_flags(rsc, pcmk__rsc_detect_loop); for (GList *lpc = rsc->private->with_this_colocations; lpc != NULL; lpc = lpc->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data; char *hdr = NULL; PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Resources colocated with %s", rsc->id); if (pcmk_is_set(cons->dependent->flags, pcmk__rsc_detect_loop)) { out->list_item(out, NULL, "%s (id=%s - loop)", cons->dependent->id, cons->id); continue; } hdr = colocations_header(cons->dependent, cons, true); out->list_item(out, NULL, "%s", hdr); free(hdr); // Empty list header for indentation of information about this resource out->begin_list(out, NULL, NULL, NULL); out->message(out, "locations-list", cons->dependent); if (recursive) { out->message(out, "rscs-colocated-with-list", cons->dependent, recursive); } out->end_list(out); } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("rscs-colocated-with-list", "pcmk_resource_t *", "bool") static int rscs_colocated_with_list_xml(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); bool recursive = va_arg(args, int); int rc = pcmk_rc_no_output; if (pcmk_is_set(rsc->flags, pcmk__rsc_detect_loop)) { return rc; } /* We're listing constraints explicitly involving rsc, so use * rsc->private->with_this_colocations directly rather than * rsc->private->cmds->with_this_colocations(). */ pcmk__set_rsc_flags(rsc, pcmk__rsc_detect_loop); for (GList *lpc = rsc->private->with_this_colocations; lpc != NULL; lpc = lpc->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data; if (pcmk_is_set(cons->dependent->flags, pcmk__rsc_detect_loop)) { colocations_xml_node(out, cons->dependent, cons); continue; } colocations_xml_node(out, cons->dependent, cons); do_locations_list_xml(out, cons->dependent, false); if (recursive) { out->message(out, "rscs-colocated-with-list", cons->dependent, recursive); } } return rc; } PCMK__OUTPUT_ARGS("locations-list", "pcmk_resource_t *") static int locations_list(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *lpc = NULL; int rc = pcmk_rc_no_output; for (lpc = rsc->private->location_constraints; lpc != NULL; lpc = lpc->next) { pcmk__location_t *cons = lpc->data; GList *lpc2 = NULL; for (lpc2 = cons->nodes; lpc2 != NULL; lpc2 = lpc2->next) { pcmk_node_t *node = (pcmk_node_t *) lpc2->data; PCMK__OUTPUT_LIST_HEADER(out, false, rc, "Locations"); out->list_item(out, NULL, "Node %s (score=%s, id=%s, rsc=%s)", pcmk__node_name(node), pcmk_readable_score(node->weight), cons->id, rsc->id); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } PCMK__OUTPUT_ARGS("locations-list", "pcmk_resource_t *") static int locations_list_xml(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); return do_locations_list_xml(out, rsc, true); } PCMK__OUTPUT_ARGS("locations-and-colocations", "pcmk_resource_t *", "bool", "bool") static int locations_and_colocations(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); bool recursive = va_arg(args, int); bool force = va_arg(args, int); pcmk__unpack_constraints(rsc->private->scheduler); // Constraints apply to group/clone, not member/instance if (!force) { rsc = uber_parent(rsc); } out->message(out, "locations-list", rsc); pe__clear_resource_flags_on_all(rsc->private->scheduler, pcmk__rsc_detect_loop); out->message(out, "rscs-colocated-with-list", rsc, recursive); pe__clear_resource_flags_on_all(rsc->private->scheduler, pcmk__rsc_detect_loop); out->message(out, "rsc-is-colocated-with-list", rsc, recursive); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("locations-and-colocations", "pcmk_resource_t *", "bool", "bool") static int locations_and_colocations_xml(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); bool recursive = va_arg(args, int); bool force = va_arg(args, int); pcmk__unpack_constraints(rsc->private->scheduler); // Constraints apply to group/clone, not member/instance if (!force) { rsc = uber_parent(rsc); } pcmk__output_xml_create_parent(out, PCMK_XE_CONSTRAINTS, NULL); do_locations_list_xml(out, rsc, false); pe__clear_resource_flags_on_all(rsc->private->scheduler, pcmk__rsc_detect_loop); out->message(out, "rscs-colocated-with-list", rsc, recursive); pe__clear_resource_flags_on_all(rsc->private->scheduler, pcmk__rsc_detect_loop); out->message(out, "rsc-is-colocated-with-list", rsc, recursive); pcmk__output_xml_pop_parent(out); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("health", "const char *", "const char *", "const char *", "const char *") static int health(pcmk__output_t *out, va_list args) { const char *sys_from G_GNUC_UNUSED = va_arg(args, const char *); const char *host_from = va_arg(args, const char *); const char *fsa_state = va_arg(args, const char *); const char *result = va_arg(args, const char *); return out->info(out, "Controller on %s in state %s: %s", pcmk__s(host_from, "unknown node"), pcmk__s(fsa_state, "unknown"), pcmk__s(result, "unknown result")); } PCMK__OUTPUT_ARGS("health", "const char *", "const char *", "const char *", "const char *") static int health_text(pcmk__output_t *out, va_list args) { if (!out->is_quiet(out)) { return health(out, args); } else { const char *sys_from G_GNUC_UNUSED = va_arg(args, const char *); const char *host_from G_GNUC_UNUSED = va_arg(args, const char *); const char *fsa_state = va_arg(args, const char *); const char *result G_GNUC_UNUSED = va_arg(args, const char *); if (fsa_state != NULL) { pcmk__formatted_printf(out, "%s\n", fsa_state); return pcmk_rc_ok; } } return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("health", "const char *", "const char *", "const char *", "const char *") static int health_xml(pcmk__output_t *out, va_list args) { const char *sys_from = va_arg(args, const char *); const char *host_from = va_arg(args, const char *); const char *fsa_state = va_arg(args, const char *); const char *result = va_arg(args, const char *); pcmk__output_create_xml_node(out, pcmk__s(sys_from, ""), PCMK_XA_NODE_NAME, pcmk__s(host_from, ""), PCMK_XA_STATE, pcmk__s(fsa_state, ""), PCMK_XA_RESULT, pcmk__s(result, ""), NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("pacemakerd-health", "const char *", "enum pcmk_pacemakerd_state", "const char *", "time_t") static int pacemakerd_health(pcmk__output_t *out, va_list args) { const char *sys_from = va_arg(args, const char *); enum pcmk_pacemakerd_state state = (enum pcmk_pacemakerd_state) va_arg(args, int); const char *state_s = va_arg(args, const char *); time_t last_updated = va_arg(args, time_t); char *last_updated_s = NULL; int rc = pcmk_rc_ok; if (sys_from == NULL) { if (state == pcmk_pacemakerd_state_remote) { sys_from = "pacemaker-remoted"; } else { sys_from = CRM_SYSTEM_MCP; } } if (state_s == NULL) { state_s = pcmk__pcmkd_state_enum2friendly(state); } if (last_updated != 0) { last_updated_s = pcmk__epoch2str(&last_updated, crm_time_log_date |crm_time_log_timeofday |crm_time_log_with_timezone); } rc = out->info(out, "Status of %s: '%s' (last updated %s)", sys_from, state_s, pcmk__s(last_updated_s, "at unknown time")); free(last_updated_s); return rc; } PCMK__OUTPUT_ARGS("pacemakerd-health", "const char *", "enum pcmk_pacemakerd_state", "const char *", "time_t") static int pacemakerd_health_html(pcmk__output_t *out, va_list args) { const char *sys_from = va_arg(args, const char *); enum pcmk_pacemakerd_state state = (enum pcmk_pacemakerd_state) va_arg(args, int); const char *state_s = va_arg(args, const char *); time_t last_updated = va_arg(args, time_t); char *last_updated_s = NULL; char *msg = NULL; if (sys_from == NULL) { if (state == pcmk_pacemakerd_state_remote) { sys_from = "pacemaker-remoted"; } else { sys_from = CRM_SYSTEM_MCP; } } if (state_s == NULL) { state_s = pcmk__pcmkd_state_enum2friendly(state); } if (last_updated != 0) { last_updated_s = pcmk__epoch2str(&last_updated, crm_time_log_date |crm_time_log_timeofday |crm_time_log_with_timezone); } msg = crm_strdup_printf("Status of %s: '%s' (last updated %s)", sys_from, state_s, pcmk__s(last_updated_s, "at unknown time")); pcmk__output_create_html_node(out, "li", NULL, NULL, msg); free(msg); free(last_updated_s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("pacemakerd-health", "const char *", "enum pcmk_pacemakerd_state", "const char *", "time_t") static int pacemakerd_health_text(pcmk__output_t *out, va_list args) { if (!out->is_quiet(out)) { return pacemakerd_health(out, args); } else { const char *sys_from G_GNUC_UNUSED = va_arg(args, const char *); enum pcmk_pacemakerd_state state = (enum pcmk_pacemakerd_state) va_arg(args, int); const char *state_s = va_arg(args, const char *); time_t last_updated G_GNUC_UNUSED = va_arg(args, time_t); if (state_s == NULL) { state_s = pcmk_pacemakerd_api_daemon_state_enum2text(state); } pcmk__formatted_printf(out, "%s\n", state_s); return pcmk_rc_ok; } } PCMK__OUTPUT_ARGS("pacemakerd-health", "const char *", "enum pcmk_pacemakerd_state", "const char *", "time_t") static int pacemakerd_health_xml(pcmk__output_t *out, va_list args) { const char *sys_from = va_arg(args, const char *); enum pcmk_pacemakerd_state state = (enum pcmk_pacemakerd_state) va_arg(args, int); const char *state_s = va_arg(args, const char *); time_t last_updated = va_arg(args, time_t); char *last_updated_s = NULL; if (sys_from == NULL) { if (state == pcmk_pacemakerd_state_remote) { sys_from = "pacemaker-remoted"; } else { sys_from = CRM_SYSTEM_MCP; } } if (state_s == NULL) { state_s = pcmk_pacemakerd_api_daemon_state_enum2text(state); } if (last_updated != 0) { last_updated_s = pcmk__epoch2str(&last_updated, crm_time_log_date |crm_time_log_timeofday |crm_time_log_with_timezone); } pcmk__output_create_xml_node(out, PCMK_XE_PACEMAKERD, PCMK_XA_SYS_FROM, sys_from, PCMK_XA_STATE, state_s, PCMK_XA_LAST_UPDATED, last_updated_s, NULL); free(last_updated_s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("profile", "const char *", "clock_t", "clock_t") static int profile_default(pcmk__output_t *out, va_list args) { const char *xml_file = va_arg(args, const char *); clock_t start = va_arg(args, clock_t); clock_t end = va_arg(args, clock_t); out->list_item(out, NULL, "Testing %s ... %.2f secs", xml_file, (end - start) / (float) CLOCKS_PER_SEC); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("profile", "const char *", "clock_t", "clock_t") static int profile_xml(pcmk__output_t *out, va_list args) { const char *xml_file = va_arg(args, const char *); clock_t start = va_arg(args, clock_t); clock_t end = va_arg(args, clock_t); char *duration = pcmk__ftoa((end - start) / (float) CLOCKS_PER_SEC); pcmk__output_create_xml_node(out, PCMK_XE_TIMING, PCMK_XA_FILE, xml_file, PCMK_XA_DURATION, duration, NULL); free(duration); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("dc", "const char *") static int dc(pcmk__output_t *out, va_list args) { const char *dc = va_arg(args, const char *); return out->info(out, "Designated Controller is: %s", pcmk__s(dc, "not yet elected")); } PCMK__OUTPUT_ARGS("dc", "const char *") static int dc_text(pcmk__output_t *out, va_list args) { if (!out->is_quiet(out)) { return dc(out, args); } else { const char *dc = va_arg(args, const char *); if (dc != NULL) { pcmk__formatted_printf(out, "%s\n", pcmk__s(dc, "")); return pcmk_rc_ok; } } return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("dc", "const char *") static int dc_xml(pcmk__output_t *out, va_list args) { const char *dc = va_arg(args, const char *); pcmk__output_create_xml_node(out, PCMK_XE_DC, PCMK_XA_NODE_NAME, pcmk__s(dc, ""), NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("crmadmin-node", "const char *", "const char *", "const char *", "bool") static int crmadmin_node(pcmk__output_t *out, va_list args) { const char *type = va_arg(args, const char *); const char *name = va_arg(args, const char *); const char *id = va_arg(args, const char *); bool bash_export = va_arg(args, int); if (bash_export) { return out->info(out, "export %s=%s", pcmk__s(name, ""), pcmk__s(id, "")); } else { return out->info(out, "%s node: %s (%s)", type ? type : "cluster", pcmk__s(name, ""), pcmk__s(id, "")); } } PCMK__OUTPUT_ARGS("crmadmin-node", "const char *", "const char *", "const char *", "bool") static int crmadmin_node_text(pcmk__output_t *out, va_list args) { if (!out->is_quiet(out)) { return crmadmin_node(out, args); } else { const char *type G_GNUC_UNUSED = va_arg(args, const char *); const char *name = va_arg(args, const char *); const char *id G_GNUC_UNUSED = va_arg(args, const char *); bool bash_export G_GNUC_UNUSED = va_arg(args, int); pcmk__formatted_printf(out, "%s\n", pcmk__s(name, "")); return pcmk_rc_ok; } } PCMK__OUTPUT_ARGS("crmadmin-node", "const char *", "const char *", "const char *", "bool") static int crmadmin_node_xml(pcmk__output_t *out, va_list args) { const char *type = va_arg(args, const char *); const char *name = va_arg(args, const char *); const char *id = va_arg(args, const char *); bool bash_export G_GNUC_UNUSED = va_arg(args, int); pcmk__output_create_xml_node(out, PCMK_XE_NODE, PCMK_XA_TYPE, pcmk__s(type, "cluster"), PCMK_XA_NAME, pcmk__s(name, ""), PCMK_XA_ID, pcmk__s(id, ""), NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("digests", "const pcmk_resource_t *", "const pcmk_node_t *", "const char *", "guint", "const pcmk__op_digest_t *") static int digests_text(pcmk__output_t *out, va_list args) { const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *); const pcmk_node_t *node = va_arg(args, const pcmk_node_t *); const char *task = va_arg(args, const char *); guint interval_ms = va_arg(args, guint); const pcmk__op_digest_t *digests = va_arg(args, const pcmk__op_digest_t *); char *action_desc = NULL; const char *rsc_desc = "unknown resource"; const char *node_desc = "unknown node"; if (interval_ms != 0) { action_desc = crm_strdup_printf("%ums-interval %s action", interval_ms, ((task == NULL)? "unknown" : task)); } else if (pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_none)) { action_desc = strdup("probe action"); } else { action_desc = crm_strdup_printf("%s action", ((task == NULL)? "unknown" : task)); } if ((rsc != NULL) && (rsc->id != NULL)) { rsc_desc = rsc->id; } if ((node != NULL) && (node->details->uname != NULL)) { node_desc = node->details->uname; } out->begin_list(out, NULL, NULL, "Digests for %s %s on %s", rsc_desc, action_desc, node_desc); free(action_desc); if (digests == NULL) { out->list_item(out, NULL, "none"); out->end_list(out); return pcmk_rc_ok; } if (digests->digest_all_calc != NULL) { out->list_item(out, NULL, "%s (all parameters)", digests->digest_all_calc); } if (digests->digest_secure_calc != NULL) { out->list_item(out, NULL, "%s (non-private parameters)", digests->digest_secure_calc); } if (digests->digest_restart_calc != NULL) { out->list_item(out, NULL, "%s (non-reloadable parameters)", digests->digest_restart_calc); } out->end_list(out); return pcmk_rc_ok; } static void add_digest_xml(xmlNode *parent, const char *type, const char *digest, xmlNode *digest_source) { if (digest != NULL) { xmlNodePtr digest_xml = pcmk__xe_create(parent, PCMK_XE_DIGEST); crm_xml_add(digest_xml, PCMK_XA_TYPE, pcmk__s(type, "unspecified")); crm_xml_add(digest_xml, PCMK_XA_HASH, digest); pcmk__xml_copy(digest_xml, digest_source); } } PCMK__OUTPUT_ARGS("digests", "const pcmk_resource_t *", "const pcmk_node_t *", "const char *", "guint", "const pcmk__op_digest_t *") static int digests_xml(pcmk__output_t *out, va_list args) { const pcmk_resource_t *rsc = va_arg(args, const pcmk_resource_t *); const pcmk_node_t *node = va_arg(args, const pcmk_node_t *); const char *task = va_arg(args, const char *); guint interval_ms = va_arg(args, guint); const pcmk__op_digest_t *digests = va_arg(args, const pcmk__op_digest_t *); char *interval_s = crm_strdup_printf("%ums", interval_ms); xmlNode *xml = NULL; xml = pcmk__output_create_xml_node(out, PCMK_XE_DIGESTS, PCMK_XA_RESOURCE, pcmk__s(rsc->id, ""), PCMK_XA_NODE, pcmk__s(node->details->uname, ""), PCMK_XA_TASK, pcmk__s(task, ""), PCMK_XA_INTERVAL, interval_s, NULL); free(interval_s); if (digests != NULL) { add_digest_xml(xml, "all", digests->digest_all_calc, digests->params_all); add_digest_xml(xml, "nonprivate", digests->digest_secure_calc, digests->params_secure); add_digest_xml(xml, "nonreloadable", digests->digest_restart_calc, digests->params_restart); } return pcmk_rc_ok; } #define STOP_SANITY_ASSERT(lineno) do { \ if ((current != NULL) && current->details->unclean) { \ /* It will be a pseudo op */ \ } else if (stop == NULL) { \ crm_err("%s:%d: No stop action exists for %s", \ __func__, lineno, rsc->id); \ CRM_ASSERT(stop != NULL); \ } else if (pcmk_is_set(stop->flags, pcmk_action_optional)) { \ crm_err("%s:%d: Action %s is still optional", \ __func__, lineno, stop->uuid); \ CRM_ASSERT(!pcmk_is_set(stop->flags, pcmk_action_optional));\ } \ } while (0) PCMK__OUTPUT_ARGS("rsc-action", "pcmk_resource_t *", "pcmk_node_t *", "pcmk_node_t *") static int rsc_action_default(pcmk__output_t *out, va_list args) { pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); pcmk_node_t *current = va_arg(args, pcmk_node_t *); pcmk_node_t *next = va_arg(args, pcmk_node_t *); GList *possible_matches = NULL; char *key = NULL; int rc = pcmk_rc_no_output; bool moving = false; pcmk_node_t *start_node = NULL; pcmk_action_t *start = NULL; pcmk_action_t *stop = NULL; pcmk_action_t *promote = NULL; pcmk_action_t *demote = NULL; pcmk_action_t *reason_op = NULL; if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed) || (current == NULL && next == NULL)) { const bool managed = pcmk_is_set(rsc->flags, pcmk__rsc_managed); pcmk__rsc_info(rsc, "Leave %s\t(%s%s)", rsc->id, pcmk_role_text(rsc->role), (managed? "" : " unmanaged")); return rc; } moving = (current != NULL) && (next != NULL) && !pcmk__same_node(current, next); possible_matches = pe__resource_actions(rsc, next, PCMK_ACTION_START, false); if (possible_matches) { start = possible_matches->data; g_list_free(possible_matches); } if ((start == NULL) || !pcmk_is_set(start->flags, pcmk_action_runnable)) { start_node = NULL; } else { start_node = current; } possible_matches = pe__resource_actions(rsc, start_node, PCMK_ACTION_STOP, false); if (possible_matches) { stop = possible_matches->data; g_list_free(possible_matches); } else if (pcmk_is_set(rsc->flags, pcmk__rsc_stop_unexpected)) { /* The resource is multiply active with PCMK_META_MULTIPLE_ACTIVE set to * PCMK_VALUE_STOP_UNEXPECTED, and not stopping on its current node, but * it should be stopping elsewhere. */ possible_matches = pe__resource_actions(rsc, NULL, PCMK_ACTION_STOP, false); if (possible_matches != NULL) { stop = possible_matches->data; g_list_free(possible_matches); } } possible_matches = pe__resource_actions(rsc, next, PCMK_ACTION_PROMOTE, false); if (possible_matches) { promote = possible_matches->data; g_list_free(possible_matches); } possible_matches = pe__resource_actions(rsc, next, PCMK_ACTION_DEMOTE, false); if (possible_matches) { demote = possible_matches->data; g_list_free(possible_matches); } if (rsc->role == rsc->next_role) { pcmk_action_t *migrate_op = NULL; CRM_CHECK(next != NULL, return rc); possible_matches = pe__resource_actions(rsc, next, PCMK_ACTION_MIGRATE_FROM, false); if (possible_matches) { migrate_op = possible_matches->data; } if ((migrate_op != NULL) && (current != NULL) && pcmk_is_set(migrate_op->flags, pcmk_action_runnable)) { rc = out->message(out, "rsc-action-item", "Migrate", rsc, current, next, start, NULL); } else if (pcmk_is_set(rsc->flags, pcmk__rsc_reload)) { rc = out->message(out, "rsc-action-item", "Reload", rsc, current, next, start, NULL); } else if ((start == NULL) || pcmk_is_set(start->flags, pcmk_action_optional)) { if ((demote != NULL) && (promote != NULL) && !pcmk_is_set(demote->flags, pcmk_action_optional) && !pcmk_is_set(promote->flags, pcmk_action_optional)) { rc = out->message(out, "rsc-action-item", "Re-promote", rsc, current, next, promote, demote); } else { pcmk__rsc_info(rsc, "Leave %s\t(%s %s)", rsc->id, pcmk_role_text(rsc->role), pcmk__node_name(next)); } } else if (!pcmk_is_set(start->flags, pcmk_action_runnable)) { if ((stop == NULL) || (stop->reason == NULL)) { reason_op = start; } else { reason_op = stop; } rc = out->message(out, "rsc-action-item", "Stop", rsc, current, NULL, stop, reason_op); STOP_SANITY_ASSERT(__LINE__); } else if (moving && current) { const bool failed = pcmk_is_set(rsc->flags, pcmk__rsc_failed); rc = out->message(out, "rsc-action-item", (failed? "Recover" : "Move"), rsc, current, next, stop, NULL); } else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { rc = out->message(out, "rsc-action-item", "Recover", rsc, current, NULL, stop, NULL); STOP_SANITY_ASSERT(__LINE__); } else { rc = out->message(out, "rsc-action-item", "Restart", rsc, current, next, start, NULL); #if 0 /* @TODO This can be reached in situations that should really be * "Start" (see for example the migrate-fail-7 regression test) */ STOP_SANITY_ASSERT(__LINE__); #endif } g_list_free(possible_matches); return rc; } if ((stop != NULL) && ((rsc->next_role == pcmk_role_stopped) || ((start != NULL) && !pcmk_is_set(start->flags, pcmk_action_runnable)))) { key = stop_key(rsc); - for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { + for (GList *iter = rsc->private->active_nodes; + iter != NULL; iter = iter->next) { + pcmk_node_t *node = iter->data; pcmk_action_t *stop_op = NULL; reason_op = start; possible_matches = find_actions(rsc->private->actions, key, node); if (possible_matches) { stop_op = possible_matches->data; g_list_free(possible_matches); } if (stop_op != NULL) { if (pcmk_is_set(stop_op->flags, pcmk_action_runnable)) { STOP_SANITY_ASSERT(__LINE__); } if (stop_op->reason != NULL) { reason_op = stop_op; } } if (out->message(out, "rsc-action-item", "Stop", rsc, node, NULL, stop_op, reason_op) == pcmk_rc_ok) { rc = pcmk_rc_ok; } } free(key); } else if ((stop != NULL) && pcmk_all_flags_set(rsc->flags, pcmk__rsc_failed |pcmk__rsc_stop_if_failed)) { /* 'stop' may be NULL if the failure was ignored */ rc = out->message(out, "rsc-action-item", "Recover", rsc, current, next, stop, start); STOP_SANITY_ASSERT(__LINE__); } else if (moving) { rc = out->message(out, "rsc-action-item", "Move", rsc, current, next, stop, NULL); STOP_SANITY_ASSERT(__LINE__); } else if (pcmk_is_set(rsc->flags, pcmk__rsc_reload)) { rc = out->message(out, "rsc-action-item", "Reload", rsc, current, next, start, NULL); } else if ((stop != NULL) && !pcmk_is_set(stop->flags, pcmk_action_optional)) { rc = out->message(out, "rsc-action-item", "Restart", rsc, current, next, start, NULL); STOP_SANITY_ASSERT(__LINE__); } else if (rsc->role == pcmk_role_promoted) { CRM_LOG_ASSERT(current != NULL); rc = out->message(out, "rsc-action-item", "Demote", rsc, current, next, demote, NULL); } else if (rsc->next_role == pcmk_role_promoted) { CRM_LOG_ASSERT(next); rc = out->message(out, "rsc-action-item", "Promote", rsc, current, next, promote, NULL); } else if ((rsc->role == pcmk_role_stopped) && (rsc->next_role > pcmk_role_stopped)) { rc = out->message(out, "rsc-action-item", "Start", rsc, current, next, start, NULL); } return rc; } PCMK__OUTPUT_ARGS("node-action", "const char *", "const char *", "const char *") static int node_action(pcmk__output_t *out, va_list args) { const char *task = va_arg(args, const char *); const char *node_name = va_arg(args, const char *); const char *reason = va_arg(args, const char *); if (task == NULL) { return pcmk_rc_no_output; } else if (reason) { out->list_item(out, NULL, "%s %s '%s'", task, node_name, reason); } else { crm_notice(" * %s %s", task, node_name); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-action", "const char *", "const char *", "const char *") static int node_action_xml(pcmk__output_t *out, va_list args) { const char *task = va_arg(args, const char *); const char *node_name = va_arg(args, const char *); const char *reason = va_arg(args, const char *); if (task == NULL) { return pcmk_rc_no_output; } else if (reason) { pcmk__output_create_xml_node(out, PCMK_XE_NODE_ACTION, PCMK_XA_TASK, task, PCMK_XA_NODE, node_name, PCMK_XA_REASON, reason, NULL); } else { crm_notice(" * %s %s", task, node_name); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("node-info", "uint32_t", "const char *", "const char *", "const char *", "bool", "bool") static int node_info_default(pcmk__output_t *out, va_list args) { uint32_t node_id = va_arg(args, uint32_t); const char *node_name = va_arg(args, const char *); const char *uuid = va_arg(args, const char *); const char *state = va_arg(args, const char *); bool have_quorum = (bool) va_arg(args, int); bool is_remote = (bool) va_arg(args, int); return out->info(out, "Node %" PRIu32 ": %s " "(uuid=%s, state=%s, have_quorum=%s, is_remote=%s)", node_id, pcmk__s(node_name, "unknown"), pcmk__s(uuid, "unknown"), pcmk__s(state, "unknown"), pcmk__btoa(have_quorum), pcmk__btoa(is_remote)); } PCMK__OUTPUT_ARGS("node-info", "uint32_t", "const char *", "const char *", "const char *", "bool", "bool") static int node_info_xml(pcmk__output_t *out, va_list args) { uint32_t node_id = va_arg(args, uint32_t); const char *node_name = va_arg(args, const char *); const char *uuid = va_arg(args, const char *); const char *state = va_arg(args, const char *); bool have_quorum = (bool) va_arg(args, int); bool is_remote = (bool) va_arg(args, int); char *id_s = crm_strdup_printf("%" PRIu32, node_id); pcmk__output_create_xml_node(out, PCMK_XE_NODE_INFO, PCMK_XA_NODEID, id_s, PCMK_XA_UNAME, node_name, PCMK_XA_ID, uuid, PCMK_XA_CRMD, state, PCMK_XA_HAVE_QUORUM, pcmk__btoa(have_quorum), PCMK_XA_REMOTE_NODE, pcmk__btoa(is_remote), NULL); free(id_s); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-cluster-action", "const char *", "const char *", "xmlNode *") static int inject_cluster_action(pcmk__output_t *out, va_list args) { const char *node = va_arg(args, const char *); const char *task = va_arg(args, const char *); xmlNodePtr rsc = va_arg(args, xmlNodePtr); if (out->is_quiet(out)) { return pcmk_rc_no_output; } if (rsc != NULL) { out->list_item(out, NULL, "Cluster action: %s for %s on %s", task, pcmk__xe_id(rsc), node); } else { out->list_item(out, NULL, "Cluster action: %s on %s", task, node); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-cluster-action", "const char *", "const char *", "xmlNode *") static int inject_cluster_action_xml(pcmk__output_t *out, va_list args) { const char *node = va_arg(args, const char *); const char *task = va_arg(args, const char *); xmlNodePtr rsc = va_arg(args, xmlNodePtr); xmlNodePtr xml_node = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } xml_node = pcmk__output_create_xml_node(out, PCMK_XE_CLUSTER_ACTION, PCMK_XA_TASK, task, PCMK_XA_NODE, node, NULL); if (rsc) { crm_xml_add(xml_node, PCMK_XA_ID, pcmk__xe_id(rsc)); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-fencing-action", "const char *", "const char *") static int inject_fencing_action(pcmk__output_t *out, va_list args) { const char *target = va_arg(args, const char *); const char *op = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } out->list_item(out, NULL, "Fencing %s (%s)", target, op); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-fencing-action", "const char *", "const char *") static int inject_fencing_action_xml(pcmk__output_t *out, va_list args) { const char *target = va_arg(args, const char *); const char *op = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } pcmk__output_create_xml_node(out, PCMK_XE_FENCING_ACTION, PCMK_XA_TARGET, target, PCMK_XA_OP, op, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-attr", "const char *", "const char *", "xmlNode *") static int inject_attr(pcmk__output_t *out, va_list args) { const char *name = va_arg(args, const char *); const char *value = va_arg(args, const char *); xmlNodePtr cib_node = va_arg(args, xmlNodePtr); xmlChar *node_path = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } node_path = xmlGetNodePath(cib_node); out->list_item(out, NULL, "Injecting attribute %s=%s into %s '%s'", name, value, node_path, pcmk__xe_id(cib_node)); free(node_path); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-attr", "const char *", "const char *", "xmlNode *") static int inject_attr_xml(pcmk__output_t *out, va_list args) { const char *name = va_arg(args, const char *); const char *value = va_arg(args, const char *); xmlNodePtr cib_node = va_arg(args, xmlNodePtr); xmlChar *node_path = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } node_path = xmlGetNodePath(cib_node); pcmk__output_create_xml_node(out, PCMK_XE_INJECT_ATTR, PCMK_XA_NAME, name, PCMK_XA_VALUE, value, PCMK_XA_NODE_PATH, node_path, PCMK_XA_CIB_NODE, pcmk__xe_id(cib_node), NULL); free(node_path); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-spec", "const char *") static int inject_spec(pcmk__output_t *out, va_list args) { const char *spec = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } out->list_item(out, NULL, "Injecting %s into the configuration", spec); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-spec", "const char *") static int inject_spec_xml(pcmk__output_t *out, va_list args) { const char *spec = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } pcmk__output_create_xml_node(out, PCMK_XE_INJECT_SPEC, PCMK_XA_SPEC, spec, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-modify-config", "const char *", "const char *") static int inject_modify_config(pcmk__output_t *out, va_list args) { const char *quorum = va_arg(args, const char *); const char *watchdog = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } out->begin_list(out, NULL, NULL, "Performing Requested Modifications"); if (quorum) { out->list_item(out, NULL, "Setting quorum: %s", quorum); } if (watchdog) { out->list_item(out, NULL, "Setting watchdog: %s", watchdog); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-modify-config", "const char *", "const char *") static int inject_modify_config_xml(pcmk__output_t *out, va_list args) { const char *quorum = va_arg(args, const char *); const char *watchdog = va_arg(args, const char *); xmlNodePtr node = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } node = pcmk__output_xml_create_parent(out, PCMK_XE_MODIFICATIONS, NULL); if (quorum) { crm_xml_add(node, PCMK_XA_QUORUM, quorum); } if (watchdog) { crm_xml_add(node, PCMK_XA_WATCHDOG, watchdog); } pcmk__output_xml_pop_parent(out); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-modify-node", "const char *", "const char *") static int inject_modify_node(pcmk__output_t *out, va_list args) { const char *action = va_arg(args, const char *); const char *node = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } if (pcmk__str_eq(action, "Online", pcmk__str_none)) { out->list_item(out, NULL, "Bringing node %s online", node); return pcmk_rc_ok; } else if (pcmk__str_eq(action, "Offline", pcmk__str_none)) { out->list_item(out, NULL, "Taking node %s offline", node); return pcmk_rc_ok; } else if (pcmk__str_eq(action, "Failing", pcmk__str_none)) { out->list_item(out, NULL, "Failing node %s", node); return pcmk_rc_ok; } return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("inject-modify-node", "const char *", "const char *") static int inject_modify_node_xml(pcmk__output_t *out, va_list args) { const char *action = va_arg(args, const char *); const char *node = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } pcmk__output_create_xml_node(out, PCMK_XE_MODIFY_NODE, PCMK_XA_ACTION, action, PCMK_XA_NODE, node, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-modify-ticket", "const char *", "const char *") static int inject_modify_ticket(pcmk__output_t *out, va_list args) { const char *action = va_arg(args, const char *); const char *ticket = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } if (pcmk__str_eq(action, "Standby", pcmk__str_none)) { out->list_item(out, NULL, "Making ticket %s standby", ticket); } else { out->list_item(out, NULL, "%s ticket %s", action, ticket); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-modify-ticket", "const char *", "const char *") static int inject_modify_ticket_xml(pcmk__output_t *out, va_list args) { const char *action = va_arg(args, const char *); const char *ticket = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } pcmk__output_create_xml_node(out, PCMK_XE_MODIFY_TICKET, PCMK_XA_ACTION, action, PCMK_XA_TICKET, ticket, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-pseudo-action", "const char *", "const char *") static int inject_pseudo_action(pcmk__output_t *out, va_list args) { const char *node = va_arg(args, const char *); const char *task = va_arg(args, const char *); if (out->is_quiet(out)) { return pcmk_rc_no_output; } out->list_item(out, NULL, "Pseudo action: %s%s%s", task, ((node == NULL)? "" : " on "), pcmk__s(node, "")); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-pseudo-action", "const char *", "const char *") static int inject_pseudo_action_xml(pcmk__output_t *out, va_list args) { const char *node = va_arg(args, const char *); const char *task = va_arg(args, const char *); xmlNodePtr xml_node = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } xml_node = pcmk__output_create_xml_node(out, PCMK_XE_PSEUDO_ACTION, PCMK_XA_TASK, task, NULL); if (node) { crm_xml_add(xml_node, PCMK_XA_NODE, node); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-rsc-action", "const char *", "const char *", "const char *", "guint") static int inject_rsc_action(pcmk__output_t *out, va_list args) { const char *rsc = va_arg(args, const char *); const char *operation = va_arg(args, const char *); const char *node = va_arg(args, const char *); guint interval_ms = va_arg(args, guint); if (out->is_quiet(out)) { return pcmk_rc_no_output; } if (interval_ms) { out->list_item(out, NULL, "Resource action: %-15s %s=%u on %s", rsc, operation, interval_ms, node); } else { out->list_item(out, NULL, "Resource action: %-15s %s on %s", rsc, operation, node); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("inject-rsc-action", "const char *", "const char *", "const char *", "guint") static int inject_rsc_action_xml(pcmk__output_t *out, va_list args) { const char *rsc = va_arg(args, const char *); const char *operation = va_arg(args, const char *); const char *node = va_arg(args, const char *); guint interval_ms = va_arg(args, guint); xmlNodePtr xml_node = NULL; if (out->is_quiet(out)) { return pcmk_rc_no_output; } xml_node = pcmk__output_create_xml_node(out, PCMK_XE_RSC_ACTION, PCMK_XA_RESOURCE, rsc, PCMK_XA_OP, operation, PCMK_XA_NODE, node, NULL); if (interval_ms) { char *interval_s = pcmk__itoa(interval_ms); crm_xml_add(xml_node, PCMK_XA_INTERVAL, interval_s); free(interval_s); } return pcmk_rc_ok; } #define CHECK_RC(retcode, retval) \ if (retval == pcmk_rc_ok) { \ retcode = pcmk_rc_ok; \ } PCMK__OUTPUT_ARGS("cluster-status", "pcmk_scheduler_t *", "enum pcmk_pacemakerd_state", "crm_exit_t", "stonith_history_t *", "enum pcmk__fence_history", "uint32_t", "uint32_t", "const char *", "GList *", "GList *") int pcmk__cluster_status_text(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); enum pcmk_pacemakerd_state pcmkd_state = (enum pcmk_pacemakerd_state) va_arg(args, int); crm_exit_t history_rc = va_arg(args, crm_exit_t); stonith_history_t *stonith_history = va_arg(args, stonith_history_t *); enum pcmk__fence_history fence_history = va_arg(args, int); uint32_t section_opts = va_arg(args, uint32_t); uint32_t show_opts = va_arg(args, uint32_t); const char *prefix = va_arg(args, const char *); GList *unames = va_arg(args, GList *); GList *resources = va_arg(args, GList *); int rc = pcmk_rc_no_output; bool already_printed_failure = false; CHECK_RC(rc, out->message(out, "cluster-summary", scheduler, pcmkd_state, section_opts, show_opts)); if (pcmk_is_set(section_opts, pcmk_section_nodes) && unames) { CHECK_RC(rc, out->message(out, "node-list", scheduler->nodes, unames, resources, show_opts, rc == pcmk_rc_ok)); } /* Print resources section, if needed */ if (pcmk_is_set(section_opts, pcmk_section_resources)) { CHECK_RC(rc, out->message(out, "resource-list", scheduler, show_opts, true, unames, resources, rc == pcmk_rc_ok)); } /* print Node Attributes section if requested */ if (pcmk_is_set(section_opts, pcmk_section_attributes)) { CHECK_RC(rc, out->message(out, "node-attribute-list", scheduler, show_opts, (rc == pcmk_rc_ok), unames, resources)); } /* If requested, print resource operations (which includes failcounts) * or just failcounts */ if (pcmk_any_flags_set(section_opts, pcmk_section_operations|pcmk_section_failcounts)) { CHECK_RC(rc, out->message(out, "node-summary", scheduler, unames, resources, section_opts, show_opts, (rc == pcmk_rc_ok))); } /* If there were any failed actions, print them */ if (pcmk_is_set(section_opts, pcmk_section_failures) && (scheduler->failed != NULL) && (scheduler->failed->children != NULL)) { CHECK_RC(rc, out->message(out, "failed-action-list", scheduler, unames, resources, show_opts, rc == pcmk_rc_ok)); } /* Print failed stonith actions */ if (pcmk_is_set(section_opts, pcmk_section_fence_failed) && fence_history != pcmk__fence_history_none) { if (history_rc == 0) { stonith_history_t *hp = NULL; hp = stonith__first_matching_event(stonith_history, stonith__event_state_eq, GINT_TO_POINTER(st_failed)); if (hp) { CHECK_RC(rc, out->message(out, "failed-fencing-list", stonith_history, unames, section_opts, show_opts, rc == pcmk_rc_ok)); } } else { PCMK__OUTPUT_SPACER_IF(out, rc == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Failed Fencing Actions"); out->list_item(out, NULL, "Failed to get fencing history: %s", crm_exit_str(history_rc)); out->end_list(out); already_printed_failure = true; } } /* Print tickets if requested */ if (pcmk_is_set(section_opts, pcmk_section_tickets)) { CHECK_RC(rc, out->message(out, "ticket-list", scheduler->tickets, (rc == pcmk_rc_ok), false, false)); } /* Print negative location constraints if requested */ if (pcmk_is_set(section_opts, pcmk_section_bans)) { CHECK_RC(rc, out->message(out, "ban-list", scheduler, prefix, resources, show_opts, rc == pcmk_rc_ok)); } /* Print stonith history */ if (pcmk_any_flags_set(section_opts, pcmk_section_fencing_all) && fence_history != pcmk__fence_history_none) { if (history_rc != 0) { if (!already_printed_failure) { PCMK__OUTPUT_SPACER_IF(out, rc == pcmk_rc_ok); out->begin_list(out, NULL, NULL, "Failed Fencing Actions"); out->list_item(out, NULL, "Failed to get fencing history: %s", crm_exit_str(history_rc)); out->end_list(out); } } else if (pcmk_is_set(section_opts, pcmk_section_fence_worked)) { stonith_history_t *hp = NULL; hp = stonith__first_matching_event(stonith_history, stonith__event_state_neq, GINT_TO_POINTER(st_failed)); if (hp) { CHECK_RC(rc, out->message(out, "fencing-list", hp, unames, section_opts, show_opts, rc == pcmk_rc_ok)); } } else if (pcmk_is_set(section_opts, pcmk_section_fence_pending)) { stonith_history_t *hp = NULL; hp = stonith__first_matching_event(stonith_history, stonith__event_state_pending, NULL); if (hp) { CHECK_RC(rc, out->message(out, "pending-fencing-list", hp, unames, section_opts, show_opts, rc == pcmk_rc_ok)); } } } return rc; } PCMK__OUTPUT_ARGS("cluster-status", "pcmk_scheduler_t *", "enum pcmk_pacemakerd_state", "crm_exit_t", "stonith_history_t *", "enum pcmk__fence_history", "uint32_t", "uint32_t", "const char *", "GList *", "GList *") static int cluster_status_xml(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); enum pcmk_pacemakerd_state pcmkd_state = (enum pcmk_pacemakerd_state) va_arg(args, int); crm_exit_t history_rc = va_arg(args, crm_exit_t); stonith_history_t *stonith_history = va_arg(args, stonith_history_t *); enum pcmk__fence_history fence_history = va_arg(args, int); uint32_t section_opts = va_arg(args, uint32_t); uint32_t show_opts = va_arg(args, uint32_t); const char *prefix = va_arg(args, const char *); GList *unames = va_arg(args, GList *); GList *resources = va_arg(args, GList *); out->message(out, "cluster-summary", scheduler, pcmkd_state, section_opts, show_opts); /*** NODES ***/ if (pcmk_is_set(section_opts, pcmk_section_nodes)) { out->message(out, "node-list", scheduler->nodes, unames, resources, show_opts, false); } /* Print resources section, if needed */ if (pcmk_is_set(section_opts, pcmk_section_resources)) { /* XML output always displays full details. */ uint32_t full_show_opts = show_opts & ~pcmk_show_brief; out->message(out, "resource-list", scheduler, full_show_opts, false, unames, resources, false); } /* print Node Attributes section if requested */ if (pcmk_is_set(section_opts, pcmk_section_attributes)) { out->message(out, "node-attribute-list", scheduler, show_opts, false, unames, resources); } /* If requested, print resource operations (which includes failcounts) * or just failcounts */ if (pcmk_any_flags_set(section_opts, pcmk_section_operations|pcmk_section_failcounts)) { out->message(out, "node-summary", scheduler, unames, resources, section_opts, show_opts, false); } /* If there were any failed actions, print them */ if (pcmk_is_set(section_opts, pcmk_section_failures) && (scheduler->failed != NULL) && (scheduler->failed->children != NULL)) { out->message(out, "failed-action-list", scheduler, unames, resources, show_opts, false); } /* Print stonith history */ if (pcmk_is_set(section_opts, pcmk_section_fencing_all) && fence_history != pcmk__fence_history_none) { out->message(out, "full-fencing-list", history_rc, stonith_history, unames, section_opts, show_opts, false); } /* Print tickets if requested */ if (pcmk_is_set(section_opts, pcmk_section_tickets)) { out->message(out, "ticket-list", scheduler->tickets, false, false, false); } /* Print negative location constraints if requested */ if (pcmk_is_set(section_opts, pcmk_section_bans)) { out->message(out, "ban-list", scheduler, prefix, resources, show_opts, false); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cluster-status", "pcmk_scheduler_t *", "enum pcmk_pacemakerd_state", "crm_exit_t", "stonith_history_t *", "enum pcmk__fence_history", "uint32_t", "uint32_t", "const char *", "GList *", "GList *") static int cluster_status_html(pcmk__output_t *out, va_list args) { pcmk_scheduler_t *scheduler = va_arg(args, pcmk_scheduler_t *); enum pcmk_pacemakerd_state pcmkd_state = (enum pcmk_pacemakerd_state) va_arg(args, int); crm_exit_t history_rc = va_arg(args, crm_exit_t); stonith_history_t *stonith_history = va_arg(args, stonith_history_t *); enum pcmk__fence_history fence_history = va_arg(args, int); uint32_t section_opts = va_arg(args, uint32_t); uint32_t show_opts = va_arg(args, uint32_t); const char *prefix = va_arg(args, const char *); GList *unames = va_arg(args, GList *); GList *resources = va_arg(args, GList *); bool already_printed_failure = false; out->message(out, "cluster-summary", scheduler, pcmkd_state, section_opts, show_opts); /*** NODE LIST ***/ if (pcmk_is_set(section_opts, pcmk_section_nodes) && unames) { out->message(out, "node-list", scheduler->nodes, unames, resources, show_opts, false); } /* Print resources section, if needed */ if (pcmk_is_set(section_opts, pcmk_section_resources)) { out->message(out, "resource-list", scheduler, show_opts, true, unames, resources, false); } /* print Node Attributes section if requested */ if (pcmk_is_set(section_opts, pcmk_section_attributes)) { out->message(out, "node-attribute-list", scheduler, show_opts, false, unames, resources); } /* If requested, print resource operations (which includes failcounts) * or just failcounts */ if (pcmk_any_flags_set(section_opts, pcmk_section_operations|pcmk_section_failcounts)) { out->message(out, "node-summary", scheduler, unames, resources, section_opts, show_opts, false); } /* If there were any failed actions, print them */ if (pcmk_is_set(section_opts, pcmk_section_failures) && (scheduler->failed != NULL) && (scheduler->failed->children != NULL)) { out->message(out, "failed-action-list", scheduler, unames, resources, show_opts, false); } /* Print failed stonith actions */ if (pcmk_is_set(section_opts, pcmk_section_fence_failed) && fence_history != pcmk__fence_history_none) { if (history_rc == 0) { stonith_history_t *hp = NULL; hp = stonith__first_matching_event(stonith_history, stonith__event_state_eq, GINT_TO_POINTER(st_failed)); if (hp) { out->message(out, "failed-fencing-list", stonith_history, unames, section_opts, show_opts, false); } } else { out->begin_list(out, NULL, NULL, "Failed Fencing Actions"); out->list_item(out, NULL, "Failed to get fencing history: %s", crm_exit_str(history_rc)); out->end_list(out); } } /* Print stonith history */ if (pcmk_any_flags_set(section_opts, pcmk_section_fencing_all) && fence_history != pcmk__fence_history_none) { if (history_rc != 0) { if (!already_printed_failure) { out->begin_list(out, NULL, NULL, "Failed Fencing Actions"); out->list_item(out, NULL, "Failed to get fencing history: %s", crm_exit_str(history_rc)); out->end_list(out); } } else if (pcmk_is_set(section_opts, pcmk_section_fence_worked)) { stonith_history_t *hp = NULL; hp = stonith__first_matching_event(stonith_history, stonith__event_state_neq, GINT_TO_POINTER(st_failed)); if (hp) { out->message(out, "fencing-list", hp, unames, section_opts, show_opts, false); } } else if (pcmk_is_set(section_opts, pcmk_section_fence_pending)) { stonith_history_t *hp = NULL; hp = stonith__first_matching_event(stonith_history, stonith__event_state_pending, NULL); if (hp) { out->message(out, "pending-fencing-list", hp, unames, section_opts, show_opts, false); } } } /* Print tickets if requested */ if (pcmk_is_set(section_opts, pcmk_section_tickets)) { out->message(out, "ticket-list", scheduler->tickets, false, false, false); } /* Print negative location constraints if requested */ if (pcmk_is_set(section_opts, pcmk_section_bans)) { out->message(out, "ban-list", scheduler, prefix, resources, show_opts, false); } return pcmk_rc_ok; } #define KV_PAIR(k, v) do { \ if (legacy) { \ pcmk__g_strcat(s, k "=", pcmk__s(v, ""), " ", NULL); \ } else { \ pcmk__g_strcat(s, k "=\"", pcmk__s(v, ""), "\" ", NULL); \ } \ } while (0) PCMK__OUTPUT_ARGS("attribute", "const char *", "const char *", "const char *", "const char *", "const char *", "bool", "bool") static int attribute_default(pcmk__output_t *out, va_list args) { const char *scope = va_arg(args, const char *); const char *instance = va_arg(args, const char *); const char *name = va_arg(args, const char *); const char *value = va_arg(args, const char *); const char *host = va_arg(args, const char *); bool quiet = va_arg(args, int); bool legacy = va_arg(args, int); gchar *value_esc = NULL; GString *s = NULL; if (quiet) { if (value != NULL) { /* Quiet needs to be turned off for ->info() to do anything */ bool was_quiet = out->is_quiet(out); if (was_quiet) { out->quiet = false; } out->info(out, "%s", value); out->quiet = was_quiet; } return pcmk_rc_ok; } s = g_string_sized_new(50); if (pcmk__xml_needs_escape(value, pcmk__xml_escape_attr_pretty)) { value_esc = pcmk__xml_escape(value, pcmk__xml_escape_attr_pretty); value = value_esc; } if (!pcmk__str_empty(scope)) { KV_PAIR(PCMK_XA_SCOPE, scope); } if (!pcmk__str_empty(instance)) { KV_PAIR(PCMK_XA_ID, instance); } KV_PAIR(PCMK_XA_NAME, name); if (!pcmk__str_empty(host)) { KV_PAIR(PCMK_XA_HOST, host); } if (legacy) { pcmk__g_strcat(s, PCMK_XA_VALUE "=", pcmk__s(value, "(null)"), NULL); } else { pcmk__g_strcat(s, PCMK_XA_VALUE "=\"", pcmk__s(value, ""), "\"", NULL); } out->info(out, "%s", s->str); g_free(value_esc); g_string_free(s, TRUE); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("attribute", "const char *", "const char *", "const char *", "const char *", "const char *", "bool", "bool") static int attribute_xml(pcmk__output_t *out, va_list args) { const char *scope = va_arg(args, const char *); const char *instance = va_arg(args, const char *); const char *name = va_arg(args, const char *); const char *value = va_arg(args, const char *); const char *host = va_arg(args, const char *); bool quiet G_GNUC_UNUSED = va_arg(args, int); bool legacy G_GNUC_UNUSED = va_arg(args, int); xmlNodePtr node = NULL; node = pcmk__output_create_xml_node(out, PCMK_XE_ATTRIBUTE, PCMK_XA_NAME, name, PCMK_XA_VALUE, pcmk__s(value, ""), NULL); if (!pcmk__str_empty(scope)) { crm_xml_add(node, PCMK_XA_SCOPE, scope); } if (!pcmk__str_empty(instance)) { crm_xml_add(node, PCMK_XA_ID, instance); } if (!pcmk__str_empty(host)) { crm_xml_add(node, PCMK_XA_HOST, host); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("rule-check", "const char *", "int", "const char *") static int rule_check_default(pcmk__output_t *out, va_list args) { const char *rule_id = va_arg(args, const char *); int result = va_arg(args, int); const char *error = va_arg(args, const char *); switch (result) { case pcmk_rc_within_range: return out->info(out, "Rule %s is still in effect", rule_id); case pcmk_rc_ok: return out->info(out, "Rule %s satisfies conditions", rule_id); case pcmk_rc_after_range: return out->info(out, "Rule %s is expired", rule_id); case pcmk_rc_before_range: return out->info(out, "Rule %s has not yet taken effect", rule_id); case pcmk_rc_op_unsatisfied: return out->info(out, "Rule %s does not satisfy conditions", rule_id); default: out->err(out, "Could not determine whether rule %s is in effect: %s", rule_id, ((error != NULL)? error : "unexpected error")); return pcmk_rc_ok; } } PCMK__OUTPUT_ARGS("rule-check", "const char *", "int", "const char *") static int rule_check_xml(pcmk__output_t *out, va_list args) { const char *rule_id = va_arg(args, const char *); int result = va_arg(args, int); const char *error = va_arg(args, const char *); char *rc_str = pcmk__itoa(pcmk_rc2exitc(result)); pcmk__output_create_xml_node(out, PCMK_XE_RULE_CHECK, PCMK_XA_RULE_ID, rule_id, PCMK_XA_RC, rc_str, NULL); free(rc_str); switch (result) { case pcmk_rc_within_range: case pcmk_rc_ok: case pcmk_rc_after_range: case pcmk_rc_before_range: case pcmk_rc_op_unsatisfied: return pcmk_rc_ok; default: out->err(out, "Could not determine whether rule %s is in effect: %s", rule_id, ((error != NULL)? error : "unexpected error")); return pcmk_rc_ok; } } PCMK__OUTPUT_ARGS("result-code", "int", "const char *", "const char *") static int result_code_none(pcmk__output_t *out, va_list args) { return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("result-code", "int", "const char *", "const char *") static int result_code_text(pcmk__output_t *out, va_list args) { int code = va_arg(args, int); const char *name = va_arg(args, const char *); const char *desc = va_arg(args, const char *); static int code_width = 0; if (out->is_quiet(out)) { /* If out->is_quiet(), don't print the code. Print name and/or desc in a * compact format for text output, or print nothing at all for none-type * output. */ if ((name != NULL) && (desc != NULL)) { pcmk__formatted_printf(out, "%s - %s\n", name, desc); } else if ((name != NULL) || (desc != NULL)) { pcmk__formatted_printf(out, "%s\n", ((name != NULL)? name : desc)); } return pcmk_rc_ok; } /* Get length of longest (most negative) standard Pacemaker return code * This should be longer than all the values of any other type of return * code. */ if (code_width == 0) { long long most_negative = pcmk_rc_error - (long long) pcmk__n_rc + 1; code_width = (int) snprintf(NULL, 0, "%lld", most_negative); } if ((name != NULL) && (desc != NULL)) { static int name_width = 0; if (name_width == 0) { // Get length of longest standard Pacemaker return code name for (int lpc = 0; lpc < pcmk__n_rc; lpc++) { int len = (int) strlen(pcmk_rc_name(pcmk_rc_error - lpc)); name_width = QB_MAX(name_width, len); } } return out->info(out, "% *d: %-*s %s", code_width, code, name_width, name, desc); } if ((name != NULL) || (desc != NULL)) { return out->info(out, "% *d: %s", code_width, code, ((name != NULL)? name : desc)); } return out->info(out, "% *d", code_width, code); } PCMK__OUTPUT_ARGS("result-code", "int", "const char *", "const char *") static int result_code_xml(pcmk__output_t *out, va_list args) { int code = va_arg(args, int); const char *name = va_arg(args, const char *); const char *desc = va_arg(args, const char *); char *code_str = pcmk__itoa(code); pcmk__output_create_xml_node(out, PCMK_XE_RESULT_CODE, PCMK_XA_CODE, code_str, PCMK_XA_NAME, name, PCMK_XA_DESCRIPTION, desc, NULL); free(code_str); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ticket-attribute", "const char *", "const char *", "const char *") static int ticket_attribute_default(pcmk__output_t *out, va_list args) { const char *ticket_id G_GNUC_UNUSED = va_arg(args, const char *); const char *name G_GNUC_UNUSED = va_arg(args, const char *); const char *value = va_arg(args, const char *); out->info(out, "%s", value); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ticket-attribute", "const char *", "const char *", "const char *") static int ticket_attribute_xml(pcmk__output_t *out, va_list args) { const char *ticket_id = va_arg(args, const char *); const char *name = va_arg(args, const char *); const char *value = va_arg(args, const char *); /* Create: * * * * * */ pcmk__output_xml_create_parent(out, PCMK_XE_TICKETS, NULL); pcmk__output_xml_create_parent(out, PCMK_XE_TICKET, PCMK_XA_ID, ticket_id, NULL); pcmk__output_create_xml_node(out, PCMK_XA_ATTRIBUTE, PCMK_XA_NAME, name, PCMK_XA_VALUE, value, NULL); pcmk__output_xml_pop_parent(out); pcmk__output_xml_pop_parent(out); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ticket-constraints", "xmlNode *") static int ticket_constraints_default(pcmk__output_t *out, va_list args) { xmlNode *constraint_xml = va_arg(args, xmlNode *); /* constraint_xml can take two forms: * * * * for when there's only one ticket in the CIB, or when the user asked * for a specific ticket (crm_ticket -c -t for instance) * * * * * * * for when there's multiple tickets in the and the user did not ask for * a specific one. * * In both cases, we simply output a element for each ticket * in the results. */ out->info(out, "Constraints XML:\n"); if (pcmk__xe_is(constraint_xml, PCMK__XE_XPATH_QUERY)) { xmlNode *child = pcmk__xe_first_child(constraint_xml, NULL, NULL, NULL); do { GString *buf = g_string_sized_new(1024); pcmk__xml_string(child, pcmk__xml_fmt_pretty, buf, 0); out->output_xml(out, PCMK_XE_CONSTRAINT, buf->str); g_string_free(buf, TRUE); child = pcmk__xe_next(child); } while (child != NULL); } else { GString *buf = g_string_sized_new(1024); pcmk__xml_string(constraint_xml, pcmk__xml_fmt_pretty, buf, 0); out->output_xml(out, PCMK_XE_CONSTRAINT, buf->str); g_string_free(buf, TRUE); } return pcmk_rc_ok; } static int add_ticket_element_with_constraints(xmlNode *node, void *userdata) { pcmk__output_t *out = (pcmk__output_t *) userdata; const char *ticket_id = crm_element_value(node, PCMK_XA_TICKET); pcmk__output_xml_create_parent(out, PCMK_XE_TICKET, PCMK_XA_ID, ticket_id, NULL); pcmk__output_xml_create_parent(out, PCMK_XE_CONSTRAINTS, NULL); pcmk__output_xml_add_node_copy(out, node); /* Pop two parents so now we are back under the element */ pcmk__output_xml_pop_parent(out); pcmk__output_xml_pop_parent(out); return pcmk_rc_ok; } static int add_resource_element(xmlNode *node, void *userdata) { pcmk__output_t *out = (pcmk__output_t *) userdata; const char *rsc = crm_element_value(node, PCMK_XA_RSC); pcmk__output_create_xml_node(out, PCMK_XE_RESOURCE, PCMK_XA_ID, rsc, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ticket-constraints", "xmlNode *") static int ticket_constraints_xml(pcmk__output_t *out, va_list args) { xmlNode *constraint_xml = va_arg(args, xmlNode *); /* Create: * * * * * * * ... * */ pcmk__output_xml_create_parent(out, PCMK_XE_TICKETS, NULL); if (pcmk__xe_is(constraint_xml, PCMK__XE_XPATH_QUERY)) { /* Iterate through the list of children once to create all the * ticket/constraint elements. */ pcmk__xe_foreach_child(constraint_xml, NULL, add_ticket_element_with_constraints, out); /* Put us back at the same level as where was created. */ pcmk__output_xml_pop_parent(out); /* Constraints can reference a resource ID that is defined in the XML * schema as an IDREF. This requires some other element to be present * with an id= attribute that matches. * * Iterate through the list of children a second time to create the * following: * * * * ... * */ pcmk__output_xml_create_parent(out, PCMK_XE_RESOURCES, NULL); pcmk__xe_foreach_child(constraint_xml, NULL, add_resource_element, out); pcmk__output_xml_pop_parent(out); } else { /* Creating the output for a single constraint is much easier. All the * comments in the above block apply here. */ add_ticket_element_with_constraints(constraint_xml, out); pcmk__output_xml_pop_parent(out); pcmk__output_xml_create_parent(out, PCMK_XE_RESOURCES, NULL); add_resource_element(constraint_xml, out); pcmk__output_xml_pop_parent(out); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ticket-state", "xmlNode *") static int ticket_state_default(pcmk__output_t *out, va_list args) { xmlNode *state_xml = va_arg(args, xmlNode *); GString *buf = g_string_sized_new(1024); out->info(out, "State XML:\n"); pcmk__xml_string(state_xml, pcmk__xml_fmt_pretty, buf, 0); out->output_xml(out, PCMK__XE_TICKET_STATE, buf->str); g_string_free(buf, TRUE); return pcmk_rc_ok; } static int add_ticket_element(xmlNode *node, void *userdata) { pcmk__output_t *out = (pcmk__output_t *) userdata; xmlNode *ticket_node = NULL; ticket_node = pcmk__output_create_xml_node(out, PCMK_XE_TICKET, NULL); pcmk__xe_copy_attrs(ticket_node, node, pcmk__xaf_none); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("ticket-state", "xmlNode *") static int ticket_state_xml(pcmk__output_t *out, va_list args) { xmlNode *state_xml = va_arg(args, xmlNode *); /* Create: * * * ... * */ pcmk__output_xml_create_parent(out, PCMK_XE_TICKETS, NULL); if (state_xml->children != NULL) { /* Iterate through the list of children once to create all the * ticket elements. */ pcmk__xe_foreach_child(state_xml, PCMK__XE_TICKET_STATE, add_ticket_element, out); } else { add_ticket_element(state_xml, out); } pcmk__output_xml_pop_parent(out); return pcmk_rc_ok; } static pcmk__message_entry_t fmt_functions[] = { { "attribute", "default", attribute_default }, { "attribute", "xml", attribute_xml }, { "cluster-status", "default", pcmk__cluster_status_text }, { "cluster-status", "html", cluster_status_html }, { "cluster-status", "xml", cluster_status_xml }, { "crmadmin-node", "default", crmadmin_node }, { "crmadmin-node", "text", crmadmin_node_text }, { "crmadmin-node", "xml", crmadmin_node_xml }, { "dc", "default", dc }, { "dc", "text", dc_text }, { "dc", "xml", dc_xml }, { "digests", "default", digests_text }, { "digests", "xml", digests_xml }, { "health", "default", health }, { "health", "text", health_text }, { "health", "xml", health_xml }, { "inject-attr", "default", inject_attr }, { "inject-attr", "xml", inject_attr_xml }, { "inject-cluster-action", "default", inject_cluster_action }, { "inject-cluster-action", "xml", inject_cluster_action_xml }, { "inject-fencing-action", "default", inject_fencing_action }, { "inject-fencing-action", "xml", inject_fencing_action_xml }, { "inject-modify-config", "default", inject_modify_config }, { "inject-modify-config", "xml", inject_modify_config_xml }, { "inject-modify-node", "default", inject_modify_node }, { "inject-modify-node", "xml", inject_modify_node_xml }, { "inject-modify-ticket", "default", inject_modify_ticket }, { "inject-modify-ticket", "xml", inject_modify_ticket_xml }, { "inject-pseudo-action", "default", inject_pseudo_action }, { "inject-pseudo-action", "xml", inject_pseudo_action_xml }, { "inject-rsc-action", "default", inject_rsc_action }, { "inject-rsc-action", "xml", inject_rsc_action_xml }, { "inject-spec", "default", inject_spec }, { "inject-spec", "xml", inject_spec_xml }, { "locations-and-colocations", "default", locations_and_colocations }, { "locations-and-colocations", "xml", locations_and_colocations_xml }, { "locations-list", "default", locations_list }, { "locations-list", "xml", locations_list_xml }, { "node-action", "default", node_action }, { "node-action", "xml", node_action_xml }, { "node-info", "default", node_info_default }, { "node-info", "xml", node_info_xml }, { "pacemakerd-health", "default", pacemakerd_health }, { "pacemakerd-health", "html", pacemakerd_health_html }, { "pacemakerd-health", "text", pacemakerd_health_text }, { "pacemakerd-health", "xml", pacemakerd_health_xml }, { "profile", "default", profile_default, }, { "profile", "xml", profile_xml }, { "result-code", PCMK_VALUE_NONE, result_code_none }, { "result-code", "text", result_code_text }, { "result-code", "xml", result_code_xml }, { "rsc-action", "default", rsc_action_default }, { "rsc-action-item", "default", rsc_action_item }, { "rsc-action-item", "xml", rsc_action_item_xml }, { "rsc-is-colocated-with-list", "default", rsc_is_colocated_with_list }, { "rsc-is-colocated-with-list", "xml", rsc_is_colocated_with_list_xml }, { "rscs-colocated-with-list", "default", rscs_colocated_with_list }, { "rscs-colocated-with-list", "xml", rscs_colocated_with_list_xml }, { "rule-check", "default", rule_check_default }, { "rule-check", "xml", rule_check_xml }, { "ticket-attribute", "default", ticket_attribute_default }, { "ticket-attribute", "xml", ticket_attribute_xml }, { "ticket-constraints", "default", ticket_constraints_default }, { "ticket-constraints", "xml", ticket_constraints_xml }, { "ticket-state", "default", ticket_state_default }, { "ticket-state", "xml", ticket_state_xml }, { NULL, NULL, NULL } }; void pcmk__register_lib_messages(pcmk__output_t *out) { pcmk__register_messages(out, fmt_functions); } diff --git a/lib/pacemaker/pcmk_sched_actions.c b/lib/pacemaker/pcmk_sched_actions.c index 0376b54f93..58246e4922 100644 --- a/lib/pacemaker/pcmk_sched_actions.c +++ b/lib/pacemaker/pcmk_sched_actions.c @@ -1,1941 +1,1943 @@ /* * 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" /*! * \internal * \brief Get the action flags relevant to ordering constraints * * \param[in,out] action Action to check * \param[in] node Node that *other* action in the ordering is on * (used only for clone resource actions) * * \return Action flags that should be used for orderings */ static uint32_t action_flags_for_ordering(pcmk_action_t *action, const pcmk_node_t *node) { bool runnable = false; uint32_t flags; // For non-resource actions, return the action flags if (action->rsc == NULL) { return action->flags; } /* For non-clone resources, or a clone action not assigned to a node, * return the flags as determined by the resource method without a node * specified. */ flags = action->rsc->private->cmds->action_flags(action, NULL); if ((node == NULL) || !pcmk__is_clone(action->rsc)) { return flags; } /* Otherwise (i.e., for clone resource actions on a specific node), first * remember whether the non-node-specific action is runnable. */ runnable = pcmk_is_set(flags, pcmk_action_runnable); // Then recheck the resource method with the node flags = action->rsc->private->cmds->action_flags(action, node); /* For clones in ordering constraints, the node-specific "runnable" doesn't * matter, just the non-node-specific setting (i.e., is the action runnable * anywhere). * * This applies only to runnable, and only for ordering constraints. This * function shouldn't be used for other types of constraints without * changes. Not very satisfying, but it's logical and appears to work well. */ if (runnable && !pcmk_is_set(flags, pcmk_action_runnable)) { pcmk__set_raw_action_flags(flags, action->rsc->id, pcmk_action_runnable); } return flags; } /*! * \internal * \brief Get action UUID that should be used with a resource ordering * * When an action is ordered relative to an action for a collective resource * (clone, group, or bundle), it actually needs to be ordered after all * instances of the collective have completed the relevant action (for example, * given "start CLONE then start RSC", RSC must wait until all instances of * CLONE have started). Given the UUID and resource of the first action in an * ordering, this returns the UUID of the action that should actually be used * for ordering (for example, "CLONE_started_0" instead of "CLONE_start_0"). * * \param[in] first_uuid UUID of first action in ordering * \param[in] first_rsc Resource of first action in ordering * * \return Newly allocated copy of UUID to use with ordering * \note It is the caller's responsibility to free the return value. */ static char * action_uuid_for_ordering(const char *first_uuid, const pcmk_resource_t *first_rsc) { guint interval_ms = 0; char *uuid = NULL; char *rid = NULL; char *first_task_str = NULL; enum action_tasks first_task = pcmk_action_unspecified; enum action_tasks remapped_task = pcmk_action_unspecified; // Only non-notify actions for collective resources need remapping if ((strstr(first_uuid, PCMK_ACTION_NOTIFY) != NULL) || (first_rsc->private->variant < pcmk__rsc_variant_group)) { goto done; } // Only non-recurring actions need remapping CRM_ASSERT(parse_op_key(first_uuid, &rid, &first_task_str, &interval_ms)); if (interval_ms > 0) { goto done; } first_task = pcmk_parse_action(first_task_str); switch (first_task) { case pcmk_action_stop: case pcmk_action_start: case pcmk_action_notify: case pcmk_action_promote: case pcmk_action_demote: remapped_task = first_task + 1; break; case pcmk_action_stopped: case pcmk_action_started: case pcmk_action_notified: case pcmk_action_promoted: case pcmk_action_demoted: remapped_task = first_task; break; case pcmk_action_monitor: case pcmk_action_shutdown: case pcmk_action_fence: break; default: crm_err("Unknown action '%s' in ordering", first_task_str); break; } if (remapped_task != pcmk_action_unspecified) { /* If a clone or bundle has notifications enabled, the ordering will be * relative to when notifications have been sent for the remapped task. */ if (pcmk_is_set(first_rsc->flags, pcmk__rsc_notify) && (pcmk__is_clone(first_rsc) || pcmk__is_bundled(first_rsc))) { uuid = pcmk__notify_key(rid, "confirmed-post", pcmk_action_text(remapped_task)); } else { uuid = pcmk__op_key(rid, pcmk_action_text(remapped_task), 0); } pcmk__rsc_trace(first_rsc, "Remapped action UUID %s to %s for ordering purposes", first_uuid, uuid); } done: free(first_task_str); free(rid); return (uuid != NULL)? uuid : pcmk__str_copy(first_uuid); } /*! * \internal * \brief Get actual action that should be used with an ordering * * When an action is ordered relative to an action for a collective resource * (clone, group, or bundle), it actually needs to be ordered after all * instances of the collective have completed the relevant action (for example, * given "start CLONE then start RSC", RSC must wait until all instances of * CLONE have started). Given the first action in an ordering, this returns the * the action that should actually be used for ordering (for example, the * started action instead of the start action). * * \param[in] action First action in an ordering * * \return Actual action that should be used for the ordering */ static pcmk_action_t * action_for_ordering(pcmk_action_t *action) { pcmk_action_t *result = action; pcmk_resource_t *rsc = action->rsc; if (rsc == NULL) { return result; } if ((rsc->private->variant >= pcmk__rsc_variant_group) && (action->uuid != NULL)) { char *uuid = action_uuid_for_ordering(action->uuid, rsc); result = find_first_action(rsc->private->actions, uuid, NULL, NULL); if (result == NULL) { crm_warn("Not remapping %s to %s because %s does not have " "remapped action", action->uuid, uuid, rsc->id); result = action; } free(uuid); } return result; } /*! * \internal * \brief Wrapper for update_ordered_actions() method for readability * * \param[in,out] rsc Resource to call method for * \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 pe_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 */ static inline uint32_t update(pcmk_resource_t *rsc, 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) { return rsc->private->cmds->update_ordered_actions(first, then, node, flags, filter, type, scheduler); } /*! * \internal * \brief Update flags for ordering's actions appropriately for ordering's flags * * \param[in,out] first First action in an ordering * \param[in,out] then Then action in an ordering * \param[in] first_flags Action flags for \p first for ordering purposes * \param[in] then_flags Action flags for \p then for ordering purposes * \param[in,out] order Action wrapper for \p first in ordering * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags */ static uint32_t update_action_for_ordering_flags(pcmk_action_t *first, pcmk_action_t *then, uint32_t first_flags, uint32_t then_flags, pcmk__related_action_t *order, pcmk_scheduler_t *scheduler) { uint32_t changed = pcmk__updated_none; /* The node will only be used for clones. If interleaved, node will be NULL, * otherwise the ordering scope will be limited to the node. Normally, the * whole 'then' clone should restart if 'first' is restarted, so then->node * is needed. */ pcmk_node_t *node = then->node; if (pcmk_is_set(order->type, pcmk__ar_first_implies_same_node_then)) { /* For unfencing, only instances of 'then' on the same node as 'first' * (the unfencing operation) should restart, so reset node to * first->node, at which point this case is handled like a normal * pcmk__ar_first_implies_then. */ pcmk__clear_relation_flags(order->type, pcmk__ar_first_implies_same_node_then); pcmk__set_relation_flags(order->type, pcmk__ar_first_implies_then); node = first->node; pcmk__rsc_trace(then->rsc, "%s then %s: mapped " "pcmk__ar_first_implies_same_node_then to " "pcmk__ar_first_implies_then on %s", first->uuid, then->uuid, pcmk__node_name(node)); } if (pcmk_is_set(order->type, pcmk__ar_first_implies_then)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags & pcmk_action_optional, pcmk_action_optional, pcmk__ar_first_implies_then, scheduler); } else if (!pcmk_is_set(first_flags, pcmk_action_optional) && pcmk_is_set(then->flags, pcmk_action_optional)) { pcmk__clear_action_flags(then, pcmk_action_optional); pcmk__set_updated_flags(changed, first, pcmk__updated_then); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_first_implies_then", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->type, pcmk__ar_intermediate_stop) && (then->rsc != NULL)) { enum pe_action_flags restart = pcmk_action_optional |pcmk_action_runnable; changed |= update(then->rsc, first, then, node, first_flags, restart, pcmk__ar_intermediate_stop, scheduler); pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_intermediate_stop", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->type, pcmk__ar_then_implies_first)) { if (first->rsc != NULL) { changed |= update(first->rsc, first, then, node, first_flags, pcmk_action_optional, pcmk__ar_then_implies_first, scheduler); } else if (!pcmk_is_set(first_flags, pcmk_action_optional) && pcmk_is_set(first->flags, pcmk_action_runnable)) { pcmk__clear_action_flags(first, pcmk_action_runnable); pcmk__set_updated_flags(changed, first, pcmk__updated_first); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_then_implies_first", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->type, pcmk__ar_promoted_then_implies_first)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags & pcmk_action_optional, pcmk_action_optional, pcmk__ar_promoted_then_implies_first, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after " "pcmk__ar_promoted_then_implies_first", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->type, pcmk__ar_min_runnable)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk_action_runnable, pcmk__ar_min_runnable, scheduler); } else if (pcmk_is_set(first_flags, pcmk_action_runnable)) { // We have another runnable instance of "first" then->runnable_before++; /* Mark "then" as runnable if it requires a certain number of * "before" instances to be runnable, and they now are. */ if ((then->runnable_before >= then->required_runnable_before) && !pcmk_is_set(then->flags, pcmk_action_runnable)) { pcmk__set_action_flags(then, pcmk_action_runnable); pcmk__set_updated_flags(changed, first, pcmk__updated_then); } } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_min_runnable", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->type, pcmk__ar_nested_remote_probe) && (then->rsc != NULL)) { if (!pcmk_is_set(first_flags, pcmk_action_runnable) - && (first->rsc != NULL) && (first->rsc->running_on != NULL)) { + && (first->rsc != NULL) + && (first->rsc->private->active_nodes != NULL)) { pcmk__rsc_trace(then->rsc, "%s then %s: ignoring because first is stopping", first->uuid, then->uuid); order->type = (enum pe_ordering) pcmk__ar_none; } else { changed |= update(then->rsc, first, then, node, first_flags, pcmk_action_runnable, pcmk__ar_unrunnable_first_blocks, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_nested_remote_probe", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->type, pcmk__ar_unrunnable_first_blocks)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk_action_runnable, pcmk__ar_unrunnable_first_blocks, scheduler); } else if (!pcmk_is_set(first_flags, pcmk_action_runnable) && pcmk_is_set(then->flags, pcmk_action_runnable)) { pcmk__clear_action_flags(then, pcmk_action_runnable); pcmk__set_updated_flags(changed, first, pcmk__updated_then); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_unrunnable_first_blocks", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->type, pcmk__ar_unmigratable_then_blocks)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk_action_optional, pcmk__ar_unmigratable_then_blocks, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after " "pcmk__ar_unmigratable_then_blocks", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->type, pcmk__ar_first_else_then)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk_action_optional, pcmk__ar_first_else_then, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_first_else_then", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->type, pcmk__ar_ordered)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk_action_runnable, pcmk__ar_ordered, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_ordered", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(order->type, pcmk__ar_asymmetric)) { if (then->rsc != NULL) { changed |= update(then->rsc, first, then, node, first_flags, pcmk_action_runnable, pcmk__ar_asymmetric, scheduler); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after pcmk__ar_asymmetric", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } if (pcmk_is_set(first->flags, pcmk_action_runnable) && pcmk_is_set(order->type, pcmk__ar_first_implies_then_graphed) && !pcmk_is_set(first_flags, pcmk_action_optional)) { pcmk__rsc_trace(then->rsc, "%s will be in graph because %s is required", then->uuid, first->uuid); pcmk__set_action_flags(then, pcmk_action_always_in_graph); // Don't bother marking 'then' as changed just for this } if (pcmk_is_set(order->type, pcmk__ar_then_implies_first_graphed) && !pcmk_is_set(then_flags, pcmk_action_optional)) { pcmk__rsc_trace(then->rsc, "%s will be in graph because %s is required", first->uuid, then->uuid); pcmk__set_action_flags(first, pcmk_action_always_in_graph); // Don't bother marking 'first' as changed just for this } if (pcmk_any_flags_set(order->type, pcmk__ar_first_implies_then |pcmk__ar_then_implies_first |pcmk__ar_intermediate_stop) && (first->rsc != NULL) && !pcmk_is_set(first->rsc->flags, pcmk__rsc_managed) && pcmk_is_set(first->rsc->flags, pcmk__rsc_blocked) && !pcmk_is_set(first->flags, pcmk_action_runnable) && pcmk__str_eq(first->task, PCMK_ACTION_STOP, pcmk__str_none)) { if (pcmk_is_set(then->flags, pcmk_action_runnable)) { pcmk__clear_action_flags(then, pcmk_action_runnable); pcmk__set_updated_flags(changed, first, pcmk__updated_then); } pcmk__rsc_trace(then->rsc, "%s then %s: %s after checking whether first " "is blocked, unmanaged, unrunnable stop", first->uuid, then->uuid, (changed? "changed" : "unchanged")); } return changed; } // 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 Update an action's flags for all orderings where it is "then" * * \param[in,out] then Action to update * \param[in,out] scheduler Scheduler data */ void pcmk__update_action_for_orderings(pcmk_action_t *then, pcmk_scheduler_t *scheduler) { GList *lpc = NULL; uint32_t changed = pcmk__updated_none; int last_flags = then->flags; pcmk__rsc_trace(then->rsc, "Updating %s %s (%s %s) on %s", action_type_str(then->flags), then->uuid, action_optional_str(then->flags), action_runnable_str(then->flags), action_node_str(then)); if (pcmk_is_set(then->flags, pcmk_action_min_runnable)) { /* Initialize current known "runnable before" actions. As * update_action_for_ordering_flags() is called for each of then's * before actions, this number will increment as runnable 'first' * actions are encountered. */ then->runnable_before = 0; if (then->required_runnable_before == 0) { /* @COMPAT This ordering constraint uses the deprecated * PCMK_XA_REQUIRE_ALL=PCMK_VALUE_FALSE attribute. Treat it like * PCMK_META_CLONE_MIN=1. */ then->required_runnable_before = 1; } /* The pcmk__ar_min_runnable clause of * update_action_for_ordering_flags() (called below) * will reset runnable if appropriate. */ pcmk__clear_action_flags(then, pcmk_action_runnable); } for (lpc = then->actions_before; lpc != NULL; lpc = lpc->next) { pcmk__related_action_t *other = lpc->data; pcmk_action_t *first = other->action; pcmk_node_t *then_node = then->node; pcmk_node_t *first_node = first->node; if ((first->rsc != NULL) && pcmk__is_group(first->rsc) && pcmk__str_eq(first->task, PCMK_ACTION_START, pcmk__str_none)) { first_node = first->rsc->private->fns->location(first->rsc, NULL, FALSE); if (first_node != NULL) { pcmk__rsc_trace(first->rsc, "Found %s for 'first' %s", pcmk__node_name(first_node), first->uuid); } } if (pcmk__is_group(then->rsc) && pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none)) { then_node = then->rsc->private->fns->location(then->rsc, NULL, FALSE); if (then_node != NULL) { pcmk__rsc_trace(then->rsc, "Found %s for 'then' %s", pcmk__node_name(then_node), then->uuid); } } // Disable constraint if it only applies when on same node, but isn't if (pcmk_is_set(other->type, pcmk__ar_if_on_same_node) && (first_node != NULL) && (then_node != NULL) && !pcmk__same_node(first_node, then_node)) { pcmk__rsc_trace(then->rsc, "Disabled ordering %s on %s then %s on %s: " "not same node", other->action->uuid, pcmk__node_name(first_node), then->uuid, pcmk__node_name(then_node)); other->type = (enum pe_ordering) pcmk__ar_none; continue; } pcmk__clear_updated_flags(changed, then, pcmk__updated_first); if ((first->rsc != NULL) && pcmk_is_set(other->type, pcmk__ar_then_cancels_first) && !pcmk_is_set(then->flags, pcmk_action_optional)) { /* 'then' is required, so we must abandon 'first' * (e.g. a required stop cancels any agent reload). */ pcmk__set_action_flags(other->action, pcmk_action_optional); if (!strcmp(first->task, PCMK_ACTION_RELOAD_AGENT)) { pcmk__clear_rsc_flags(first->rsc, pcmk__rsc_reload); } } if ((first->rsc != NULL) && (then->rsc != NULL) && (first->rsc != then->rsc) && !is_parent(then->rsc, first->rsc)) { first = action_for_ordering(first); } if (first != other->action) { pcmk__rsc_trace(then->rsc, "Ordering %s after %s instead of %s", then->uuid, first->uuid, other->action->uuid); } pcmk__rsc_trace(then->rsc, "%s (%#.6x) then %s (%#.6x): type=%#.6x node=%s", first->uuid, first->flags, then->uuid, then->flags, other->type, action_node_str(first)); if (first == other->action) { /* 'first' was not remapped (e.g. from 'start' to 'running'), which * could mean it is a non-resource action, a primitive resource * action, or already expanded. */ uint32_t first_flags, then_flags; first_flags = action_flags_for_ordering(first, then_node); then_flags = action_flags_for_ordering(then, first_node); changed |= update_action_for_ordering_flags(first, then, first_flags, then_flags, other, scheduler); /* 'first' was for a complex resource (clone, group, etc), * create a new dependency if necessary */ } else if (order_actions(first, then, other->type)) { /* This was the first time 'first' and 'then' were associated, * start again to get the new actions_before list */ pcmk__set_updated_flags(changed, then, pcmk__updated_then); pcmk__rsc_trace(then->rsc, "Disabled ordering %s then %s in favor of %s " "then %s", other->action->uuid, then->uuid, first->uuid, then->uuid); other->type = (enum pe_ordering) pcmk__ar_none; } if (pcmk_is_set(changed, pcmk__updated_first)) { crm_trace("Re-processing %s and its 'after' actions " "because it changed", first->uuid); for (GList *lpc2 = first->actions_after; lpc2 != NULL; lpc2 = lpc2->next) { pcmk__related_action_t *other = lpc2->data; pcmk__update_action_for_orderings(other->action, scheduler); } pcmk__update_action_for_orderings(first, scheduler); } } if (pcmk_is_set(then->flags, pcmk_action_min_runnable)) { if (last_flags == then->flags) { pcmk__clear_updated_flags(changed, then, pcmk__updated_then); } else { pcmk__set_updated_flags(changed, then, pcmk__updated_then); } } if (pcmk_is_set(changed, pcmk__updated_then)) { crm_trace("Re-processing %s and its 'after' actions because it changed", then->uuid); if (pcmk_is_set(last_flags, pcmk_action_runnable) && !pcmk_is_set(then->flags, pcmk_action_runnable)) { pcmk__block_colocation_dependents(then); } pcmk__update_action_for_orderings(then, scheduler); for (lpc = then->actions_after; lpc != NULL; lpc = lpc->next) { pcmk__related_action_t *other = lpc->data; pcmk__update_action_for_orderings(other->action, scheduler); } } } static inline bool is_primitive_action(const pcmk_action_t *action) { return (action != NULL) && pcmk__is_primitive(action->rsc); } /*! * \internal * \brief Clear a single action flag and set reason text * * \param[in,out] action Action whose flag should be cleared * \param[in] flag Action flag that should be cleared * \param[in] reason Action that is the reason why flag is being cleared */ #define clear_action_flag_because(action, flag, reason) do { \ if (pcmk_is_set((action)->flags, (flag))) { \ pcmk__clear_action_flags(action, flag); \ if ((action)->rsc != (reason)->rsc) { \ char *reason_text = pe__action2reason((reason), (flag)); \ pe_action_set_reason((action), reason_text, false); \ free(reason_text); \ } \ } \ } while (0) /*! * \internal * \brief Update actions in an asymmetric ordering * * If the "first" action in an asymmetric ordering is unrunnable, make the * "second" action unrunnable as well, if appropriate. * * \param[in] first 'First' action in an asymmetric ordering * \param[in,out] then 'Then' action in an asymmetric ordering */ static void handle_asymmetric_ordering(const pcmk_action_t *first, pcmk_action_t *then) { /* Only resource actions after an unrunnable 'first' action need updates for * asymmetric ordering. */ if ((then->rsc == NULL) || pcmk_is_set(first->flags, pcmk_action_runnable)) { return; } // Certain optional 'then' actions are unaffected by unrunnable 'first' if (pcmk_is_set(then->flags, pcmk_action_optional)) { enum rsc_role_e then_rsc_role; then_rsc_role = then->rsc->private->fns->state(then->rsc, TRUE); if ((then_rsc_role == pcmk_role_stopped) && pcmk__str_eq(then->task, PCMK_ACTION_STOP, pcmk__str_none)) { /* If 'then' should stop after 'first' but is already stopped, the * ordering is irrelevant. */ return; } else if ((then_rsc_role >= pcmk_role_started) && pcmk__str_eq(then->task, PCMK_ACTION_START, pcmk__str_none) && pe__rsc_running_on_only(then->rsc, then->node)) { /* Similarly if 'then' should start after 'first' but is already * started on a single node. */ return; } } // 'First' can't run, so 'then' can't either clear_action_flag_because(then, pcmk_action_optional, first); clear_action_flag_because(then, pcmk_action_runnable, first); } /*! * \internal * \brief Set action bits appropriately when pcmk__ar_intermediate_stop is used * * \param[in,out] first 'First' action in ordering * \param[in,out] then 'Then' action in ordering * \param[in] filter What action flags to care about * * \note pcmk__ar_intermediate_stop is set for "stop resource before starting * it" and "stop later group member before stopping earlier group member" */ static void handle_restart_ordering(pcmk_action_t *first, pcmk_action_t *then, uint32_t filter) { const char *reason = NULL; CRM_ASSERT(is_primitive_action(first)); CRM_ASSERT(is_primitive_action(then)); // We need to update the action in two cases: // ... if 'then' is required if (pcmk_is_set(filter, pcmk_action_optional) && !pcmk_is_set(then->flags, pcmk_action_optional)) { reason = "restart"; } /* ... if 'then' is unrunnable action on same resource (if a resource * should restart but can't start, we still want to stop) */ if (pcmk_is_set(filter, pcmk_action_runnable) && !pcmk_is_set(then->flags, pcmk_action_runnable) && pcmk_is_set(then->rsc->flags, pcmk__rsc_managed) && (first->rsc == then->rsc)) { reason = "stop"; } if (reason == NULL) { return; } pcmk__rsc_trace(first->rsc, "Handling %s -> %s for %s", first->uuid, then->uuid, reason); // Make 'first' required if it is runnable if (pcmk_is_set(first->flags, pcmk_action_runnable)) { clear_action_flag_because(first, pcmk_action_optional, then); } // Make 'first' required if 'then' is required if (!pcmk_is_set(then->flags, pcmk_action_optional)) { clear_action_flag_because(first, pcmk_action_optional, then); } // Make 'first' unmigratable if 'then' is unmigratable if (!pcmk_is_set(then->flags, pcmk_action_migratable)) { clear_action_flag_because(first, pcmk_action_migratable, then); } // Make 'then' unrunnable if 'first' is required but unrunnable if (!pcmk_is_set(first->flags, pcmk_action_optional) && !pcmk_is_set(first->flags, pcmk_action_runnable)) { clear_action_flag_because(then, pcmk_action_runnable, first); } } /*! * \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 * (ignored) * \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__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; uint32_t then_flags = 0U; uint32_t first_flags = 0U; CRM_ASSERT((first != NULL) && (then != NULL) && (scheduler != NULL)); then_flags = then->flags; first_flags = first->flags; if (pcmk_is_set(type, pcmk__ar_asymmetric)) { handle_asymmetric_ordering(first, then); } if (pcmk_is_set(type, pcmk__ar_then_implies_first) && !pcmk_is_set(then_flags, pcmk_action_optional)) { // Then is required, and implies first should be, too if (pcmk_is_set(filter, pcmk_action_optional) && !pcmk_is_set(flags, pcmk_action_optional) && pcmk_is_set(first_flags, pcmk_action_optional)) { clear_action_flag_because(first, pcmk_action_optional, then); } if (pcmk_is_set(flags, pcmk_action_migratable) && !pcmk_is_set(then->flags, pcmk_action_migratable)) { clear_action_flag_because(first, pcmk_action_migratable, then); } } if (pcmk_is_set(type, pcmk__ar_promoted_then_implies_first) && (then->rsc != NULL) && (then->rsc->role == pcmk_role_promoted) && pcmk_is_set(filter, pcmk_action_optional) && !pcmk_is_set(then->flags, pcmk_action_optional)) { clear_action_flag_because(first, pcmk_action_optional, then); if (pcmk_is_set(first->flags, pcmk_action_migratable) && !pcmk_is_set(then->flags, pcmk_action_migratable)) { clear_action_flag_because(first, pcmk_action_migratable, then); } } if (pcmk_is_set(type, pcmk__ar_unmigratable_then_blocks) && pcmk_is_set(filter, pcmk_action_optional)) { if (!pcmk_all_flags_set(then->flags, pcmk_action_migratable |pcmk_action_runnable)) { clear_action_flag_because(first, pcmk_action_runnable, then); } if (!pcmk_is_set(then->flags, pcmk_action_optional)) { clear_action_flag_because(first, pcmk_action_optional, then); } } if (pcmk_is_set(type, pcmk__ar_first_else_then) && pcmk_is_set(filter, pcmk_action_optional) && !pcmk_is_set(first->flags, pcmk_action_runnable)) { clear_action_flag_because(then, pcmk_action_migratable, first); pcmk__clear_action_flags(then, pcmk_action_pseudo); } if (pcmk_is_set(type, pcmk__ar_unrunnable_first_blocks) && pcmk_is_set(filter, pcmk_action_runnable) && pcmk_is_set(then->flags, pcmk_action_runnable) && !pcmk_is_set(flags, pcmk_action_runnable)) { clear_action_flag_because(then, pcmk_action_runnable, first); clear_action_flag_because(then, pcmk_action_migratable, first); } if (pcmk_is_set(type, pcmk__ar_first_implies_then) && pcmk_is_set(filter, pcmk_action_optional) && pcmk_is_set(then->flags, pcmk_action_optional) && !pcmk_is_set(flags, pcmk_action_optional) && !pcmk_is_set(first->flags, pcmk_action_migratable)) { clear_action_flag_because(then, pcmk_action_optional, first); } if (pcmk_is_set(type, pcmk__ar_intermediate_stop)) { handle_restart_ordering(first, then, filter); } if (then_flags != then->flags) { pcmk__set_updated_flags(changed, first, pcmk__updated_then); pcmk__rsc_trace(then->rsc, "%s on %s: flags are now %#.6x (was %#.6x) " "because of 'first' %s (%#.6x)", then->uuid, pcmk__node_name(then->node), then->flags, then_flags, first->uuid, first->flags); if ((then->rsc != NULL) && (then->rsc->private->parent != NULL)) { // Required to handle "X_stop then X_start" for cloned groups pcmk__update_action_for_orderings(then, scheduler); } } if (first_flags != first->flags) { pcmk__set_updated_flags(changed, first, pcmk__updated_first); pcmk__rsc_trace(first->rsc, "%s on %s: flags are now %#.6x (was %#.6x) " "because of 'then' %s (%#.6x)", first->uuid, pcmk__node_name(first->node), first->flags, first_flags, then->uuid, then->flags); } return changed; } /*! * \internal * \brief Trace-log an action (optionally with its dependent actions) * * \param[in] pre_text If not NULL, prefix the log with this plus ": " * \param[in] action Action to log * \param[in] details If true, recursively log dependent actions */ void pcmk__log_action(const char *pre_text, const pcmk_action_t *action, bool details) { const char *node_uname = NULL; const char *node_uuid = NULL; const char *desc = NULL; CRM_CHECK(action != NULL, return); if (!pcmk_is_set(action->flags, pcmk_action_pseudo)) { if (action->node != NULL) { node_uname = action->node->details->uname; node_uuid = action->node->details->id; } else { node_uname = ""; } } switch (pcmk_parse_action(action->task)) { case pcmk_action_fence: case pcmk_action_shutdown: if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { desc = "Pseudo "; } else if (pcmk_is_set(action->flags, pcmk_action_optional)) { desc = "Optional "; } else if (!pcmk_is_set(action->flags, pcmk_action_runnable)) { desc = "!!Non-Startable!! "; } else { desc = "(Provisional) "; } crm_trace("%s%s%sAction %d: %s%s%s%s%s%s", ((pre_text == NULL)? "" : pre_text), ((pre_text == NULL)? "" : ": "), desc, action->id, action->uuid, (node_uname? "\ton " : ""), (node_uname? node_uname : ""), (node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""), (node_uuid? ")" : "")); break; default: if (pcmk_is_set(action->flags, pcmk_action_optional)) { desc = "Optional "; } else if (pcmk_is_set(action->flags, pcmk_action_pseudo)) { desc = "Pseudo "; } else if (!pcmk_is_set(action->flags, pcmk_action_runnable)) { desc = "!!Non-Startable!! "; } else { desc = "(Provisional) "; } crm_trace("%s%s%sAction %d: %s %s%s%s%s%s%s", ((pre_text == NULL)? "" : pre_text), ((pre_text == NULL)? "" : ": "), desc, action->id, action->uuid, (action->rsc? action->rsc->id : ""), (node_uname? "\ton " : ""), (node_uname? node_uname : ""), (node_uuid? "\t\t(" : ""), (node_uuid? node_uuid : ""), (node_uuid? ")" : "")); break; } if (details) { const GList *iter = NULL; const pcmk__related_action_t *other = NULL; crm_trace("\t\t====== Preceding Actions"); for (iter = action->actions_before; iter != NULL; iter = iter->next) { other = (const pcmk__related_action_t *) iter->data; pcmk__log_action("\t\t", other->action, false); } crm_trace("\t\t====== Subsequent Actions"); for (iter = action->actions_after; iter != NULL; iter = iter->next) { other = (const pcmk__related_action_t *) iter->data; pcmk__log_action("\t\t", other->action, false); } crm_trace("\t\t====== End"); } else { crm_trace("\t\t(before=%d, after=%d)", g_list_length(action->actions_before), g_list_length(action->actions_after)); } } /*! * \internal * \brief Create a new shutdown action for a node * * \param[in,out] node Node being shut down * * \return Newly created shutdown action for \p node */ pcmk_action_t * pcmk__new_shutdown_action(pcmk_node_t *node) { char *shutdown_id = NULL; pcmk_action_t *shutdown_op = NULL; CRM_ASSERT(node != NULL); shutdown_id = crm_strdup_printf("%s-%s", PCMK_ACTION_DO_SHUTDOWN, node->details->uname); shutdown_op = custom_action(NULL, shutdown_id, PCMK_ACTION_DO_SHUTDOWN, node, FALSE, node->details->data_set); pcmk__order_stops_before_shutdown(node, shutdown_op); pcmk__insert_meta(shutdown_op, PCMK__META_OP_NO_WAIT, PCMK_VALUE_TRUE); return shutdown_op; } /*! * \internal * \brief Calculate and add an operation digest to XML * * Calculate an operation digest, which enables us to later determine when a * restart is needed due to the resource's parameters being changed, and add it * to given XML. * * \param[in] op Operation result from executor * \param[in,out] update XML to add digest to */ static void add_op_digest_to_xml(const lrmd_event_data_t *op, xmlNode *update) { char *digest = NULL; xmlNode *args_xml = NULL; if (op->params == NULL) { return; } args_xml = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS); g_hash_table_foreach(op->params, hash2field, args_xml); pcmk__filter_op_for_digest(args_xml); digest = pcmk__digest_operation(args_xml); crm_xml_add(update, PCMK__XA_OP_DIGEST, digest); pcmk__xml_free(args_xml); free(digest); } #define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" /*! * \internal * \brief Create XML for resource operation history update * * \param[in,out] parent Parent XML node to add to * \param[in,out] op Operation event data * \param[in] caller_version DC feature set * \param[in] target_rc Expected result of operation * \param[in] node Name of node on which operation was performed * \param[in] origin Arbitrary description of update source * * \return Newly created XML node for history update */ xmlNode * pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *op, const char *caller_version, int target_rc, const char *node, const char *origin) { char *key = NULL; char *magic = NULL; char *op_id = NULL; char *op_id_additional = NULL; char *local_user_data = NULL; const char *exit_reason = NULL; xmlNode *xml_op = NULL; const char *task = NULL; CRM_CHECK(op != NULL, return NULL); crm_trace("Creating history XML for %s-interval %s action for %s on %s " "(DC version: %s, origin: %s)", pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id, ((node == NULL)? "no node" : node), caller_version, origin); task = op->op_type; /* Record a successful agent reload as a start, and a failed one as a * monitor, to make life easier for the scheduler when determining the * current state. * * @COMPAT We should check "reload" here only if the operation was for a * pre-OCF-1.1 resource agent, but we don't know that here, and we should * only ever get results for actions scheduled by us, so we can reasonably * assume any "reload" is actually a pre-1.1 agent reload. */ if (pcmk__str_any_of(task, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, NULL)) { if (op->op_status == PCMK_EXEC_DONE) { task = PCMK_ACTION_START; } else { task = PCMK_ACTION_MONITOR; } } key = pcmk__op_key(op->rsc_id, task, op->interval_ms); if (pcmk__str_eq(task, PCMK_ACTION_NOTIFY, pcmk__str_none)) { const char *n_type = crm_meta_value(op->params, "notify_type"); const char *n_task = crm_meta_value(op->params, "notify_operation"); CRM_LOG_ASSERT(n_type != NULL); CRM_LOG_ASSERT(n_task != NULL); op_id = pcmk__notify_key(op->rsc_id, n_type, n_task); if (op->op_status != PCMK_EXEC_PENDING) { /* Ignore notify errors. * * @TODO It might be better to keep the correct result here, and * ignore it in process_graph_event(). */ lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); } /* Migration history is preserved separately, which usually matters for * multiple nodes and is important for future cluster transitions. */ } else if (pcmk__str_any_of(op->op_type, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { op_id = strdup(key); } else if (did_rsc_op_fail(op, target_rc)) { op_id = pcmk__op_key(op->rsc_id, "last_failure", 0); if (op->interval_ms == 0) { /* Ensure 'last' gets updated, in case PCMK_META_RECORD_PENDING is * true */ op_id_additional = pcmk__op_key(op->rsc_id, "last", 0); } exit_reason = op->exit_reason; } else if (op->interval_ms > 0) { op_id = strdup(key); } else { op_id = pcmk__op_key(op->rsc_id, "last", 0); } again: xml_op = pcmk__xe_first_child(parent, PCMK__XE_LRM_RSC_OP, PCMK_XA_ID, op_id); if (xml_op == NULL) { xml_op = pcmk__xe_create(parent, PCMK__XE_LRM_RSC_OP); } if (op->user_data == NULL) { crm_debug("Generating fake transition key for: " PCMK__OP_FMT " %d from %s", op->rsc_id, op->op_type, op->interval_ms, op->call_id, origin); local_user_data = pcmk__transition_key(-1, op->call_id, target_rc, FAKE_TE_ID); op->user_data = local_user_data; } if (magic == NULL) { magic = crm_strdup_printf("%d:%d;%s", op->op_status, op->rc, (const char *) op->user_data); } crm_xml_add(xml_op, PCMK_XA_ID, op_id); crm_xml_add(xml_op, PCMK__XA_OPERATION_KEY, key); crm_xml_add(xml_op, PCMK_XA_OPERATION, task); crm_xml_add(xml_op, PCMK_XA_CRM_DEBUG_ORIGIN, origin); crm_xml_add(xml_op, PCMK_XA_CRM_FEATURE_SET, caller_version); crm_xml_add(xml_op, PCMK__XA_TRANSITION_KEY, op->user_data); crm_xml_add(xml_op, PCMK__XA_TRANSITION_MAGIC, magic); crm_xml_add(xml_op, PCMK_XA_EXIT_REASON, pcmk__s(exit_reason, "")); crm_xml_add(xml_op, PCMK__META_ON_NODE, node); // For context during triage crm_xml_add_int(xml_op, PCMK__XA_CALL_ID, op->call_id); crm_xml_add_int(xml_op, PCMK__XA_RC_CODE, op->rc); crm_xml_add_int(xml_op, PCMK__XA_OP_STATUS, op->op_status); crm_xml_add_ms(xml_op, PCMK_META_INTERVAL, op->interval_ms); if (compare_version("2.1", caller_version) <= 0) { if (op->t_run || op->t_rcchange || op->exec_time || op->queue_time) { crm_trace("Timing data (" PCMK__OP_FMT "): last=%u change=%u exec=%u queue=%u", op->rsc_id, op->op_type, op->interval_ms, op->t_run, op->t_rcchange, op->exec_time, op->queue_time); if ((op->interval_ms != 0) && (op->t_rcchange != 0)) { // Recurring ops may have changed rc after initial run crm_xml_add_ll(xml_op, PCMK_XA_LAST_RC_CHANGE, (long long) op->t_rcchange); } else { crm_xml_add_ll(xml_op, PCMK_XA_LAST_RC_CHANGE, (long long) op->t_run); } crm_xml_add_int(xml_op, PCMK_XA_EXEC_TIME, op->exec_time); crm_xml_add_int(xml_op, PCMK_XA_QUEUE_TIME, op->queue_time); } } if (pcmk__str_any_of(op->op_type, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { /* Record PCMK__META_MIGRATE_SOURCE and PCMK__META_MIGRATE_TARGET always * for migrate ops. */ const char *name = PCMK__META_MIGRATE_SOURCE; crm_xml_add(xml_op, name, crm_meta_value(op->params, name)); name = PCMK__META_MIGRATE_TARGET; crm_xml_add(xml_op, name, crm_meta_value(op->params, name)); } add_op_digest_to_xml(op, xml_op); if (op_id_additional) { free(op_id); op_id = op_id_additional; op_id_additional = NULL; goto again; } if (local_user_data) { free(local_user_data); op->user_data = NULL; } free(magic); free(op_id); free(key); return xml_op; } /*! * \internal * \brief Check whether an action shutdown-locks a resource to a node * * If the PCMK_OPT_SHUTDOWN_LOCK cluster property is set, resources will not be * recovered on a different node if cleanly stopped, and may start only on that * same node. This function checks whether that applies to a given action, so * that the transition graph can be marked appropriately. * * \param[in] action Action to check * * \return true if \p action locks its resource to the action's node, * otherwise false */ bool pcmk__action_locks_rsc_to_node(const pcmk_action_t *action) { // Only resource actions taking place on resource's lock node are locked if ((action == NULL) || (action->rsc == NULL) || !pcmk__same_node(action->node, action->rsc->lock_node)) { return false; } /* During shutdown, only stops are locked (otherwise, another action such as * a demote would cause the controller to clear the lock) */ if (action->node->details->shutdown && (action->task != NULL) && (strcmp(action->task, PCMK_ACTION_STOP) != 0)) { return false; } return true; } /* lowest to highest */ static gint sort_action_id(gconstpointer a, gconstpointer b) { const pcmk__related_action_t *action_wrapper2 = a; const pcmk__related_action_t *action_wrapper1 = b; if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (action_wrapper1->action->id < action_wrapper2->action->id) { return 1; } if (action_wrapper1->action->id > action_wrapper2->action->id) { return -1; } return 0; } /*! * \internal * \brief Remove any duplicate action inputs, merging action flags * * \param[in,out] action Action whose inputs should be checked */ void pcmk__deduplicate_action_inputs(pcmk_action_t *action) { GList *item = NULL; GList *next = NULL; pcmk__related_action_t *last_input = NULL; action->actions_before = g_list_sort(action->actions_before, sort_action_id); for (item = action->actions_before; item != NULL; item = next) { pcmk__related_action_t *input = item->data; next = item->next; if ((last_input != NULL) && (input->action->id == last_input->action->id)) { crm_trace("Input %s (%d) duplicate skipped for action %s (%d)", input->action->uuid, input->action->id, action->uuid, action->id); /* For the purposes of scheduling, the ordering flags no longer * matter, but crm_simulate looks at certain ones when creating a * dot graph. Combining the flags is sufficient for that purpose. */ last_input->type |= input->type; if (input->state == pe_link_dumped) { last_input->state = pe_link_dumped; } free(item->data); action->actions_before = g_list_delete_link(action->actions_before, item); } else { last_input = input; input->state = pe_link_not_dumped; } } } /*! * \internal * \brief Output all scheduled actions * * \param[in,out] scheduler Scheduler data */ void pcmk__output_actions(pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv; // Output node (non-resource) actions for (GList *iter = scheduler->actions; iter != NULL; iter = iter->next) { char *node_name = NULL; char *task = NULL; pcmk_action_t *action = (pcmk_action_t *) iter->data; if (action->rsc != NULL) { continue; // Resource actions will be output later } else if (pcmk_is_set(action->flags, pcmk_action_optional)) { continue; // This action was not scheduled } if (pcmk__str_eq(action->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { task = strdup("Shutdown"); } else if (pcmk__str_eq(action->task, PCMK_ACTION_STONITH, pcmk__str_none)) { const char *op = g_hash_table_lookup(action->meta, PCMK__META_STONITH_ACTION); task = crm_strdup_printf("Fence (%s)", op); } else { continue; // Don't display other node action types } if (pcmk__is_guest_or_bundle_node(action->node)) { const pcmk_resource_t *remote = action->node->details->remote_rsc; node_name = crm_strdup_printf("%s (resource: %s)", pcmk__node_name(action->node), remote->container->id); } else if (action->node != NULL) { node_name = crm_strdup_printf("%s", pcmk__node_name(action->node)); } out->message(out, "node-action", task, node_name, action->reason); free(node_name); free(task); } // Output resource actions for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; rsc->private->cmds->output_actions(rsc); } } /*! * \internal * \brief Get action name needed to compare digest for configuration changes * * \param[in] task Action name from history * \param[in] interval_ms Action interval (in milliseconds) * * \return Action name whose digest should be compared */ static const char * task_for_digest(const char *task, guint interval_ms) { /* Certain actions need to be compared against the parameters used to start * the resource. */ if ((interval_ms == 0) && pcmk__str_any_of(task, PCMK_ACTION_MONITOR, PCMK_ACTION_MIGRATE_FROM, PCMK_ACTION_PROMOTE, NULL)) { task = PCMK_ACTION_START; } return task; } /*! * \internal * \brief Check whether only sanitized parameters to an action changed * * When collecting CIB files for troubleshooting, crm_report will mask * sensitive resource parameters. If simulations were run using that, affected * resources would appear to need a restart, which would complicate * troubleshooting. To avoid that, we save a "secure digest" of non-sensitive * parameters. This function used that digest to check whether only masked * parameters are different. * * \param[in] xml_op Resource history entry with secure digest * \param[in] digest_data Operation digest information being compared * \param[in] scheduler Scheduler data * * \return true if only sanitized parameters changed, otherwise false */ static bool only_sanitized_changed(const xmlNode *xml_op, const pcmk__op_digest_t *digest_data, const pcmk_scheduler_t *scheduler) { const char *digest_secure = NULL; if (!pcmk_is_set(scheduler->flags, pcmk_sched_sanitized)) { // The scheduler is not being run as a simulation return false; } digest_secure = crm_element_value(xml_op, PCMK__XA_OP_SECURE_DIGEST); return (digest_data->rc != pcmk__digest_match) && (digest_secure != NULL) && (digest_data->digest_secure_calc != NULL) && (strcmp(digest_data->digest_secure_calc, digest_secure) == 0); } /*! * \internal * \brief Force a restart due to a configuration change * * \param[in,out] rsc Resource that action is for * \param[in] task Name of action whose configuration changed * \param[in] interval_ms Action interval (in milliseconds) * \param[in,out] node Node where resource should be restarted */ static void force_restart(pcmk_resource_t *rsc, const char *task, guint interval_ms, pcmk_node_t *node) { char *key = pcmk__op_key(rsc->id, task, interval_ms); pcmk_action_t *required = custom_action(rsc, key, task, NULL, FALSE, rsc->private->scheduler); pe_action_set_reason(required, "resource definition change", true); trigger_unfencing(rsc, node, "Device parameters changed", NULL, rsc->private->scheduler); } /*! * \internal * \brief Schedule a reload of a resource on a node * * \param[in,out] data Resource to reload * \param[in] user_data Where resource should be reloaded */ static void schedule_reload(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; const pcmk_node_t *node = user_data; pcmk_action_t *reload = NULL; // For collective resources, just call recursively for children if (rsc->private->variant > pcmk__rsc_variant_primitive) { g_list_foreach(rsc->children, schedule_reload, user_data); return; } // Skip the reload in certain situations if ((node == NULL) || !pcmk_is_set(rsc->flags, pcmk__rsc_managed) || pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { pcmk__rsc_trace(rsc, "Skip reload of %s:%s%s %s", rsc->id, pcmk_is_set(rsc->flags, pcmk__rsc_managed)? "" : " unmanaged", pcmk_is_set(rsc->flags, pcmk__rsc_failed)? " failed" : "", (node == NULL)? "inactive" : node->details->uname); return; } /* If a resource's configuration changed while a start was pending, * force a full restart instead of a reload. */ if (pcmk_is_set(rsc->flags, pcmk__rsc_start_pending)) { pcmk__rsc_trace(rsc, "%s: preventing agent reload because start pending", rsc->id); custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, node, FALSE, rsc->private->scheduler); return; } // Schedule the reload pcmk__set_rsc_flags(rsc, pcmk__rsc_reload); reload = custom_action(rsc, reload_key(rsc), PCMK_ACTION_RELOAD_AGENT, node, FALSE, rsc->private->scheduler); pe_action_set_reason(reload, "resource definition change", FALSE); // Set orderings so that a required stop or demote cancels the reload pcmk__new_ordering(NULL, NULL, reload, rsc, stop_key(rsc), NULL, pcmk__ar_ordered|pcmk__ar_then_cancels_first, rsc->private->scheduler); pcmk__new_ordering(NULL, NULL, reload, rsc, demote_key(rsc), NULL, pcmk__ar_ordered|pcmk__ar_then_cancels_first, rsc->private->scheduler); } /*! * \internal * \brief Handle any configuration change for an action * * Given an action from resource history, if the resource's configuration * changed since the action was done, schedule any actions needed (restart, * reload, unfencing, rescheduling recurring actions, etc.). * * \param[in,out] rsc Resource that action is for * \param[in,out] node Node that action was on * \param[in] xml_op Action XML from resource history * * \return true if action configuration changed, otherwise false */ bool pcmk__check_action_config(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *xml_op) { guint interval_ms = 0; const char *task = NULL; const pcmk__op_digest_t *digest_data = NULL; CRM_CHECK((rsc != NULL) && (node != NULL) && (xml_op != NULL), return false); task = crm_element_value(xml_op, PCMK_XA_OPERATION); CRM_CHECK(task != NULL, return false); crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &interval_ms); // If this is a recurring action, check whether it has been orphaned if (interval_ms > 0) { if (pcmk__find_action_config(rsc, task, interval_ms, false) != NULL) { pcmk__rsc_trace(rsc, "%s-interval %s for %s on %s is in configuration", pcmk__readable_interval(interval_ms), task, rsc->id, pcmk__node_name(node)); } else if (pcmk_is_set(rsc->private->scheduler->flags, pcmk_sched_cancel_removed_actions)) { pcmk__schedule_cancel(rsc, crm_element_value(xml_op, PCMK__XA_CALL_ID), task, interval_ms, node, "orphan"); return true; } else { pcmk__rsc_debug(rsc, "%s-interval %s for %s on %s is orphaned", pcmk__readable_interval(interval_ms), task, rsc->id, pcmk__node_name(node)); return true; } } crm_trace("Checking %s-interval %s for %s on %s for configuration changes", pcmk__readable_interval(interval_ms), task, rsc->id, pcmk__node_name(node)); task = task_for_digest(task, interval_ms); digest_data = rsc_action_digest_cmp(rsc, xml_op, node, rsc->private->scheduler); if (only_sanitized_changed(xml_op, digest_data, rsc->private->scheduler)) { if (!pcmk__is_daemon && (rsc->private->scheduler->priv != NULL)) { pcmk__output_t *out = rsc->private->scheduler->priv; out->info(out, "Only 'private' parameters to %s-interval %s for %s " "on %s changed: %s", pcmk__readable_interval(interval_ms), task, rsc->id, pcmk__node_name(node), crm_element_value(xml_op, PCMK__XA_TRANSITION_MAGIC)); } return false; } switch (digest_data->rc) { case pcmk__digest_restart: crm_log_xml_debug(digest_data->params_restart, "params:restart"); force_restart(rsc, task, interval_ms, node); return true; case pcmk__digest_unknown: case pcmk__digest_mismatch: // Changes that can potentially be handled by an agent reload if (interval_ms > 0) { /* Recurring actions aren't reloaded per se, they are just * re-scheduled so the next run uses the new parameters. * The old instance will be cancelled automatically. */ crm_log_xml_debug(digest_data->params_all, "params:reschedule"); pcmk__reschedule_recurring(rsc, task, interval_ms, node); } else if (crm_element_value(xml_op, PCMK__XA_OP_RESTART_DIGEST) != NULL) { // Agent supports reload, so use it trigger_unfencing(rsc, node, "Device parameters changed (reload)", NULL, rsc->private->scheduler); crm_log_xml_debug(digest_data->params_all, "params:reload"); schedule_reload((gpointer) rsc, (gpointer) node); } else { pcmk__rsc_trace(rsc, "Restarting %s " "because agent doesn't support reload", rsc->id); crm_log_xml_debug(digest_data->params_restart, "params:restart"); force_restart(rsc, task, interval_ms, node); } return true; default: break; } return false; } /*! * \internal * \brief Create a list of resource's action history entries, sorted by call ID * * \param[in] rsc_entry Resource's \c PCMK__XE_LRM_RSC_OP status XML * \param[out] start_index Where to store index of start-like action, if any * \param[out] stop_index Where to store index of stop action, if any */ static GList * rsc_history_as_list(const xmlNode *rsc_entry, int *start_index, int *stop_index) { GList *ops = NULL; for (xmlNode *rsc_op = pcmk__xe_first_child(rsc_entry, PCMK__XE_LRM_RSC_OP, NULL, NULL); rsc_op != NULL; rsc_op = pcmk__xe_next_same(rsc_op)) { ops = g_list_prepend(ops, rsc_op); } ops = g_list_sort(ops, sort_op_by_callid); calculate_active_ops(ops, start_index, stop_index); return ops; } /*! * \internal * \brief Process a resource's action history from the CIB status * * Given a resource's action history, if the resource's configuration * changed since the actions were done, schedule any actions needed (restart, * reload, unfencing, rescheduling recurring actions, clean-up, etc.). * (This also cancels recurring actions for maintenance mode, which is not * entirely related but convenient to do here.) * * \param[in] rsc_entry Resource's \c PCMK__XE_LRM_RSC_OP status XML * \param[in,out] rsc Resource whose history is being processed * \param[in,out] node Node whose history is being processed */ static void process_rsc_history(const xmlNode *rsc_entry, pcmk_resource_t *rsc, pcmk_node_t *node) { int offset = -1; int stop_index = 0; int start_index = 0; GList *sorted_op_list = NULL; if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { if (pcmk__is_anonymous_clone(pe__const_top_resource(rsc, false))) { pcmk__rsc_trace(rsc, "Skipping configuration check " "for orphaned clone instance %s", rsc->id); } else { pcmk__rsc_trace(rsc, "Skipping configuration check and scheduling " "clean-up for orphaned resource %s", rsc->id); pcmk__schedule_cleanup(rsc, node, false); } return; } - if (pe_find_node_id(rsc->running_on, node->details->id) == NULL) { + if (pe_find_node_id(rsc->private->active_nodes, + node->details->id) == NULL) { if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, false)) { pcmk__schedule_cleanup(rsc, node, false); } pcmk__rsc_trace(rsc, "Skipping configuration check for %s " "because no longer active on %s", rsc->id, pcmk__node_name(node)); return; } pcmk__rsc_trace(rsc, "Checking for configuration changes for %s on %s", rsc->id, pcmk__node_name(node)); if (pcmk__rsc_agent_changed(rsc, node, rsc_entry, true)) { pcmk__schedule_cleanup(rsc, node, false); } sorted_op_list = rsc_history_as_list(rsc_entry, &start_index, &stop_index); if (start_index < stop_index) { return; // Resource is stopped } for (GList *iter = sorted_op_list; iter != NULL; iter = iter->next) { xmlNode *rsc_op = (xmlNode *) iter->data; const char *task = NULL; guint interval_ms = 0; if (++offset < start_index) { // Skip actions that happened before a start continue; } task = crm_element_value(rsc_op, PCMK_XA_OPERATION); crm_element_value_ms(rsc_op, PCMK_META_INTERVAL, &interval_ms); if ((interval_ms > 0) && (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance) || node->details->maintenance)) { // Maintenance mode cancels recurring operations pcmk__schedule_cancel(rsc, crm_element_value(rsc_op, PCMK__XA_CALL_ID), task, interval_ms, node, "maintenance mode"); } else if ((interval_ms > 0) || pcmk__strcase_any_of(task, PCMK_ACTION_MONITOR, PCMK_ACTION_START, PCMK_ACTION_PROMOTE, PCMK_ACTION_MIGRATE_FROM, NULL)) { /* If a resource operation failed, and the operation's definition * has changed, clear any fail count so they can be retried fresh. */ if (pe__bundle_needs_remote_name(rsc)) { /* We haven't assigned resources to nodes yet, so if the * REMOTE_CONTAINER_HACK is used, we may calculate the digest * based on the literal "#uname" value rather than the properly * substituted value. That would mistakenly make the action * definition appear to have been changed. Defer the check until * later in this case. */ pe__add_param_check(rsc_op, rsc, node, pcmk__check_active, rsc->private->scheduler); } else if (pcmk__check_action_config(rsc, node, rsc_op) && (pe_get_failcount(node, rsc, NULL, pcmk__fc_effective, NULL) != 0)) { pe__clear_failcount(rsc, node, "action definition changed", rsc->private->scheduler); } } } g_list_free(sorted_op_list); } /*! * \internal * \brief Process a node's action history from the CIB status * * Given a node's resource history, if the resource's configuration changed * since the actions were done, schedule any actions needed (restart, * reload, unfencing, rescheduling recurring actions, clean-up, etc.). * (This also cancels recurring actions for maintenance mode, which is not * entirely related but convenient to do here.) * * \param[in,out] node Node whose history is being processed * \param[in] lrm_rscs Node's \c PCMK__XE_LRM_RESOURCES from CIB status XML */ static void process_node_history(pcmk_node_t *node, const xmlNode *lrm_rscs) { crm_trace("Processing node history for %s", pcmk__node_name(node)); for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rscs, PCMK__XE_LRM_RESOURCE, NULL, NULL); rsc_entry != NULL; rsc_entry = pcmk__xe_next_same(rsc_entry)) { if (rsc_entry->children != NULL) { GList *result = pcmk__rscs_matching_id(pcmk__xe_id(rsc_entry), node->details->data_set); for (GList *iter = result; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; if (pcmk__is_primitive(rsc)) { process_rsc_history(rsc_entry, rsc, node); } } g_list_free(result); } } } // XPath to find a node's resource history #define XPATH_NODE_HISTORY "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE \ "[@" PCMK_XA_UNAME "='%s']" \ "/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES /*! * \internal * \brief Process any resource configuration changes in the CIB status * * Go through all nodes' resource history, and if a resource's configuration * changed since its actions were done, schedule any actions needed (restart, * reload, unfencing, rescheduling recurring actions, clean-up, etc.). * (This also cancels recurring actions for maintenance mode, which is not * entirely related but convenient to do here.) * * \param[in,out] scheduler Scheduler data */ void pcmk__handle_rsc_config_changes(pcmk_scheduler_t *scheduler) { crm_trace("Check resource and action configuration for changes"); /* Rather than iterate through the status section, iterate through the nodes * and search for the appropriate status subsection for each. This skips * orphaned nodes and lets us eliminate some cases before searching the XML. */ for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; /* Don't bother checking actions for a node that can't run actions ... * unless it's in maintenance mode, in which case we still need to * cancel any existing recurring monitors. */ if (node->details->maintenance || pcmk__node_available(node, false, false)) { char *xpath = NULL; xmlNode *history = NULL; xpath = crm_strdup_printf(XPATH_NODE_HISTORY, node->details->uname); history = get_xpath_object(xpath, scheduler->input, LOG_NEVER); free(xpath); process_node_history(node, history); } } } diff --git a/lib/pacemaker/pcmk_sched_group.c b/lib/pacemaker/pcmk_sched_group.c index 7ad47753a4..0116edf011 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->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)) { + if ((member->private->active_nodes != NULL) + && (member_data->previous_member->private->active_nodes == 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->private->active_nodes != NULL) { if (member_data->ordered && (member_data->previous_member != NULL) - && (member_data->previous_member->running_on == NULL) + && (member_data->previous_member->private->active_nodes == NULL) && (member_data->last_active != NULL) - && (member_data->last_active->running_on != NULL)) { + && (member_data->last_active->private->active_nodes != 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_instances.c b/lib/pacemaker/pcmk_sched_instances.c index 638fa44f44..356bbb042c 100644 --- a/lib/pacemaker/pcmk_sched_instances.c +++ b/lib/pacemaker/pcmk_sched_instances.c @@ -1,1700 +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. */ /* This file is intended for code usable with both clone instances and bundle * replica containers. */ #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Check whether a node is allowed to run an instance * * \param[in] instance Clone instance or bundle container to check * \param[in] node Node to check * \param[in] max_per_node Maximum number of instances allowed to run on a node * * \return true if \p node is allowed to run \p instance, otherwise false */ static bool can_run_instance(const pcmk_resource_t *instance, const pcmk_node_t *node, int max_per_node) { pcmk_node_t *allowed_node = NULL; if (pcmk_is_set(instance->flags, pcmk__rsc_removed)) { pcmk__rsc_trace(instance, "%s cannot run on %s: orphaned", instance->id, pcmk__node_name(node)); return false; } if (!pcmk__node_available(node, false, false)) { pcmk__rsc_trace(instance, "%s cannot run on %s: node cannot run resources", instance->id, pcmk__node_name(node)); return false; } allowed_node = pcmk__top_allowed_node(instance, node); if (allowed_node == NULL) { crm_warn("%s cannot run on %s: node not allowed", instance->id, pcmk__node_name(node)); return false; } if (allowed_node->weight < 0) { pcmk__rsc_trace(instance, "%s cannot run on %s: parent score is %s there", instance->id, pcmk__node_name(node), pcmk_readable_score(allowed_node->weight)); return false; } if (allowed_node->count >= max_per_node) { pcmk__rsc_trace(instance, "%s cannot run on %s: node already has %d instance%s", instance->id, pcmk__node_name(node), max_per_node, pcmk__plural_s(max_per_node)); return false; } pcmk__rsc_trace(instance, "%s can run on %s (%d already running)", instance->id, pcmk__node_name(node), allowed_node->count); return true; } /*! * \internal * \brief Ban a clone instance or bundle replica from unavailable allowed nodes * * \param[in,out] instance Clone instance or bundle replica to ban * \param[in] max_per_node Maximum instances allowed to run on a node */ static void ban_unavailable_allowed_nodes(pcmk_resource_t *instance, int max_per_node) { if (instance->allowed_nodes != NULL) { GHashTableIter iter; pcmk_node_t *node = NULL; g_hash_table_iter_init(&iter, instance->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (!can_run_instance(instance, node, max_per_node)) { pcmk__rsc_trace(instance, "Banning %s from unavailable node %s", instance->id, pcmk__node_name(node)); node->weight = -PCMK_SCORE_INFINITY; for (GList *child_iter = instance->children; child_iter != NULL; child_iter = child_iter->next) { pcmk_resource_t *child = child_iter->data; pcmk_node_t *child_node = NULL; child_node = g_hash_table_lookup(child->allowed_nodes, node->details->id); if (child_node != NULL) { pcmk__rsc_trace(instance, "Banning %s child %s " "from unavailable node %s", instance->id, child->id, pcmk__node_name(node)); child_node->weight = -PCMK_SCORE_INFINITY; } } } } } } /*! * \internal * \brief Create a hash table with a single node in it * * \param[in] node Node to copy into new table * * \return Newly created hash table containing a copy of \p node * \note The caller is responsible for freeing the result with * g_hash_table_destroy(). */ static GHashTable * new_node_table(pcmk_node_t *node) { GHashTable *table = pcmk__strkey_table(NULL, free); node = pe__copy_node(node); g_hash_table_insert(table, (gpointer) node->details->id, node); return table; } /*! * \internal * \brief Apply a resource's parent's colocation scores to a node table * * \param[in] rsc Resource whose colocations should be applied * \param[in,out] nodes Node table to apply colocations to */ static void apply_parent_colocations(const pcmk_resource_t *rsc, GHashTable **nodes) { GList *colocations = pcmk__this_with_colocations(rsc); for (const GList *iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *colocation = iter->data; pcmk_resource_t *other = colocation->primary; float factor = colocation->score / (float) PCMK_SCORE_INFINITY; other->private->cmds->add_colocated_node_scores(other, rsc, rsc->id, nodes, colocation, factor, pcmk__coloc_select_default); } g_list_free(colocations); colocations = pcmk__with_this_colocations(rsc); for (const GList *iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *colocation = iter->data; pcmk_resource_t *other = colocation->dependent; float factor = colocation->score / (float) PCMK_SCORE_INFINITY; if (!pcmk__colocation_has_influence(colocation, rsc)) { continue; } other->private->cmds->add_colocated_node_scores(other, rsc, rsc->id, nodes, colocation, factor, pcmk__coloc_select_nonnegative); } g_list_free(colocations); } /*! * \internal * \brief Compare clone or bundle instances based on colocation scores * * Determine the relative order in which two clone or bundle instances should be * assigned to nodes, considering the scores of colocation constraints directly * or indirectly involving them. * * \param[in] instance1 First instance to compare * \param[in] instance2 Second instance to compare * * \return A negative number if \p instance1 should be assigned first, * a positive number if \p instance2 should be assigned first, * or 0 if assignment order doesn't matter */ static int cmp_instance_by_colocation(const pcmk_resource_t *instance1, const pcmk_resource_t *instance2) { int rc = 0; pcmk_node_t *node1 = NULL; pcmk_node_t *node2 = NULL; pcmk_node_t *current_node1 = pcmk__current_node(instance1); pcmk_node_t *current_node2 = pcmk__current_node(instance2); GHashTable *colocated_scores1 = NULL; GHashTable *colocated_scores2 = NULL; CRM_ASSERT((instance1 != NULL) && (instance1->private->parent != NULL) && (instance2 != NULL) && (instance2->private->parent != NULL) && (current_node1 != NULL) && (current_node2 != NULL)); // Create node tables initialized with each node colocated_scores1 = new_node_table(current_node1); colocated_scores2 = new_node_table(current_node2); // Apply parental colocations apply_parent_colocations(instance1, &colocated_scores1); apply_parent_colocations(instance2, &colocated_scores2); // Find original nodes again, with scores updated for colocations node1 = g_hash_table_lookup(colocated_scores1, current_node1->details->id); node2 = g_hash_table_lookup(colocated_scores2, current_node2->details->id); // Compare nodes by updated scores if (node1->weight < node2->weight) { crm_trace("Assign %s (%d on %s) after %s (%d on %s)", instance1->id, node1->weight, pcmk__node_name(node1), instance2->id, node2->weight, pcmk__node_name(node2)); rc = 1; } else if (node1->weight > node2->weight) { crm_trace("Assign %s (%d on %s) before %s (%d on %s)", instance1->id, node1->weight, pcmk__node_name(node1), instance2->id, node2->weight, pcmk__node_name(node2)); rc = -1; } g_hash_table_destroy(colocated_scores1); g_hash_table_destroy(colocated_scores2); return rc; } /*! * \internal * \brief Check whether a resource or any of its children are failed * * \param[in] rsc Resource to check * * \return true if \p rsc or any of its children are failed, otherwise false */ static bool did_fail(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { return true; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { if (did_fail((const pcmk_resource_t *) iter->data)) { return true; } } return false; } /*! * \internal * \brief Check whether a node is allowed to run a resource * * \param[in] rsc Resource to check * \param[in,out] node Node to check (will be set NULL if not allowed) * * \return true if *node is either NULL or allowed for \p rsc, otherwise false */ static bool node_is_allowed(const pcmk_resource_t *rsc, pcmk_node_t **node) { if (*node != NULL) { pcmk_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes, (*node)->details->id); if ((allowed == NULL) || (allowed->weight < 0)) { pcmk__rsc_trace(rsc, "%s: current location (%s) is unavailable", rsc->id, pcmk__node_name(*node)); *node = NULL; return false; } } return true; } /*! * \internal * \brief Compare two clone or bundle instances' instance numbers * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a's instance number is lower, * a positive number if \p b's instance number is lower, * or 0 if their instance numbers are the same */ gint pcmk__cmp_instance_number(gconstpointer a, gconstpointer b) { const pcmk_resource_t *instance1 = (const pcmk_resource_t *) a; const pcmk_resource_t *instance2 = (const pcmk_resource_t *) b; char *div1 = NULL; char *div2 = NULL; CRM_ASSERT((instance1 != NULL) && (instance2 != NULL)); // Clone numbers are after a colon, bundle numbers after a dash div1 = strrchr(instance1->id, ':'); if (div1 == NULL) { div1 = strrchr(instance1->id, '-'); } div2 = strrchr(instance2->id, ':'); if (div2 == NULL) { div2 = strrchr(instance2->id, '-'); } CRM_ASSERT((div1 != NULL) && (div2 != NULL)); return (gint) (strtol(div1 + 1, NULL, 10) - strtol(div2 + 1, NULL, 10)); } /*! * \internal * \brief Compare clone or bundle instances according to assignment order * * Compare two clone or bundle instances according to the order they should be * assigned to nodes, preferring (in order): * * - Active instance that is less multiply active * - Instance that is not active on a disallowed node * - Instance with higher configured priority * - Active instance whose current node can run resources * - Active instance whose parent is allowed on current node * - Active instance whose current node has fewer other instances * - Active instance * - Instance that isn't failed * - Instance whose colocations result in higher score on current node * - Instance with lower ID in lexicographic order * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a should be assigned first, * a positive number if \p b should be assigned first, * or 0 if assignment order doesn't matter */ gint pcmk__cmp_instance(gconstpointer a, gconstpointer b) { int rc = 0; pcmk_node_t *node1 = NULL; pcmk_node_t *node2 = NULL; unsigned int nnodes1 = 0; unsigned int nnodes2 = 0; bool can1 = true; bool can2 = true; const pcmk_resource_t *instance1 = (const pcmk_resource_t *) a; const pcmk_resource_t *instance2 = (const pcmk_resource_t *) b; CRM_ASSERT((instance1 != NULL) && (instance2 != NULL)); node1 = instance1->private->fns->active_node(instance1, &nnodes1, NULL); node2 = instance2->private->fns->active_node(instance2, &nnodes2, NULL); /* If both instances are running and at least one is multiply * active, prefer instance that's running on fewer nodes. */ if ((nnodes1 > 0) && (nnodes2 > 0)) { if (nnodes1 < nnodes2) { crm_trace("Assign %s (active on %d) before %s (active on %d): " "less multiply active", instance1->id, nnodes1, instance2->id, nnodes2); return -1; } else if (nnodes1 > nnodes2) { crm_trace("Assign %s (active on %d) after %s (active on %d): " "more multiply active", instance1->id, nnodes1, instance2->id, nnodes2); return 1; } } /* An instance that is either inactive or active on an allowed node is * preferred over an instance that is active on a no-longer-allowed node. */ can1 = node_is_allowed(instance1, &node1); can2 = node_is_allowed(instance2, &node2); if (can1 && !can2) { crm_trace("Assign %s before %s: not active on a disallowed node", instance1->id, instance2->id); return -1; } else if (!can1 && can2) { crm_trace("Assign %s after %s: active on a disallowed node", instance1->id, instance2->id); return 1; } // Prefer instance with higher configured priority if (instance1->private->priority > instance2->private->priority) { crm_trace("Assign %s before %s: priority (%d > %d)", instance1->id, instance2->id, instance1->private->priority, instance2->private->priority); return -1; } else if (instance1->private->priority < instance2->private->priority) { crm_trace("Assign %s after %s: priority (%d < %d)", instance1->id, instance2->id, instance1->private->priority, instance2->private->priority); return 1; } // Prefer active instance if ((node1 == NULL) && (node2 == NULL)) { crm_trace("No assignment preference for %s vs. %s: inactive", instance1->id, instance2->id); return 0; } else if (node1 == NULL) { crm_trace("Assign %s after %s: active", instance1->id, instance2->id); return 1; } else if (node2 == NULL) { crm_trace("Assign %s before %s: active", instance1->id, instance2->id); return -1; } // Prefer instance whose current node can run resources can1 = pcmk__node_available(node1, false, false); can2 = pcmk__node_available(node2, false, false); if (can1 && !can2) { crm_trace("Assign %s before %s: current node can run resources", instance1->id, instance2->id); return -1; } else if (!can1 && can2) { crm_trace("Assign %s after %s: current node can't run resources", instance1->id, instance2->id); return 1; } // Prefer instance whose parent is allowed to run on instance's current node node1 = pcmk__top_allowed_node(instance1, node1); node2 = pcmk__top_allowed_node(instance2, node2); if ((node1 == NULL) && (node2 == NULL)) { crm_trace("No assignment preference for %s vs. %s: " "parent not allowed on either instance's current node", instance1->id, instance2->id); return 0; } else if (node1 == NULL) { crm_trace("Assign %s after %s: parent not allowed on current node", instance1->id, instance2->id); return 1; } else if (node2 == NULL) { crm_trace("Assign %s before %s: parent allowed on current node", instance1->id, instance2->id); return -1; } // Prefer instance whose current node is running fewer other instances if (node1->count < node2->count) { crm_trace("Assign %s before %s: fewer active instances on current node", instance1->id, instance2->id); return -1; } else if (node1->count > node2->count) { crm_trace("Assign %s after %s: more active instances on current node", instance1->id, instance2->id); return 1; } // Prefer instance that isn't failed can1 = did_fail(instance1); can2 = did_fail(instance2); if (!can1 && can2) { crm_trace("Assign %s before %s: not failed", instance1->id, instance2->id); return -1; } else if (can1 && !can2) { crm_trace("Assign %s after %s: failed", instance1->id, instance2->id); return 1; } // Prefer instance with higher cumulative colocation score on current node rc = cmp_instance_by_colocation(instance1, instance2); if (rc != 0) { return rc; } // Prefer instance with lower instance number rc = pcmk__cmp_instance_number(instance1, instance2); if (rc < 0) { crm_trace("Assign %s before %s: instance number", instance1->id, instance2->id); } else if (rc > 0) { crm_trace("Assign %s after %s: instance number", instance1->id, instance2->id); } else { crm_trace("No assignment preference for %s vs. %s", instance1->id, instance2->id); } return rc; } /*! * \internal * \brief Increment the parent's instance count after assigning an instance * * An instance's parent tracks how many instances have been assigned to each * node via its pcmk_node_t:count member. After assigning an instance to a node, * find the corresponding node in the parent's allowed table and increment it. * * \param[in,out] instance Instance whose parent to update * \param[in] assigned_to Node to which the instance was assigned */ static void increment_parent_count(pcmk_resource_t *instance, const pcmk_node_t *assigned_to) { pcmk_node_t *allowed = NULL; if (assigned_to == NULL) { return; } allowed = pcmk__top_allowed_node(instance, assigned_to); if (allowed == NULL) { /* The instance is allowed on the node, but its parent isn't. This * shouldn't be possible if the resource is managed, and we won't be * able to limit the number of instances assigned to the node. */ CRM_LOG_ASSERT(!pcmk_is_set(instance->flags, pcmk__rsc_managed)); } else { allowed->count++; } } /*! * \internal * \brief Assign an instance to a node * * \param[in,out] instance Clone instance or bundle replica container * \param[in] prefer If not NULL, attempt early assignment to this * node, if still the best choice; otherwise, * perform final assignment * \param[in] max_per_node Assign at most this many instances to one node * * \return Node to which \p instance is assigned */ static const pcmk_node_t * assign_instance(pcmk_resource_t *instance, const pcmk_node_t *prefer, int max_per_node) { pcmk_node_t *chosen = NULL; pcmk__rsc_trace(instance, "Assigning %s (preferring %s)", instance->id, ((prefer == NULL)? "no node" : prefer->details->uname)); if (pcmk_is_set(instance->flags, pcmk__rsc_assigning)) { pcmk__rsc_debug(instance, "Assignment loop detected involving %s colocations", instance->id); return NULL; } ban_unavailable_allowed_nodes(instance, max_per_node); // Failed early assignments are reversible (stop_if_fail=false) chosen = instance->private->cmds->assign(instance, prefer, (prefer == NULL)); increment_parent_count(instance, chosen); return chosen; } /*! * \internal * \brief Try to assign an instance to its current node early * * \param[in] rsc Clone or bundle being assigned (for logs only) * \param[in] instance Clone instance or bundle replica container * \param[in] current Instance's current node * \param[in] max_per_node Maximum number of instances per node * \param[in] available Number of instances still available for assignment * * \return \c true if \p instance was successfully assigned to its current node, * or \c false otherwise */ static bool assign_instance_early(const pcmk_resource_t *rsc, pcmk_resource_t *instance, const pcmk_node_t *current, int max_per_node, int available) { const pcmk_node_t *chosen = NULL; int reserved = 0; pcmk_resource_t *parent = instance->private->parent; GHashTable *allowed_orig = NULL; GHashTable *allowed_orig_parent = parent->allowed_nodes; const pcmk_node_t *allowed_node = NULL; pcmk__rsc_trace(instance, "Trying to assign %s to its current node %s", instance->id, pcmk__node_name(current)); allowed_node = g_hash_table_lookup(instance->allowed_nodes, current->details->id); if (!pcmk__node_available(allowed_node, true, false)) { pcmk__rsc_info(instance, "Not assigning %s to current node %s: unavailable", instance->id, pcmk__node_name(current)); return false; } /* On each iteration, if instance gets assigned to a node other than its * current one, we reserve one instance for the chosen node, unassign * instance, restore instance's original node tables, and try again. This * way, instances are proportionally assigned to nodes based on preferences, * but shuffling of specific instances is minimized. If a node will be * assigned instances at all, it preferentially receives instances that are * currently active there. * * parent->allowed_nodes tracks the number of instances assigned to each * node. If a node already has max_per_node instances assigned, * ban_unavailable_allowed_nodes() marks it as unavailable. * * In the end, we restore the original parent->allowed_nodes to undo the * changes to counts during tentative assignments. If we successfully * assigned instance to its current node, we increment that node's counter. */ // Back up the allowed node tables of instance and its children recursively pcmk__copy_node_tables(instance, &allowed_orig); // Update instances-per-node counts in a scratch table parent->allowed_nodes = pcmk__copy_node_table(parent->allowed_nodes); while (reserved < available) { chosen = assign_instance(instance, current, max_per_node); if (pcmk__same_node(chosen, current)) { // Successfully assigned to current node break; } // Assignment updates scores, so restore to original state pcmk__rsc_debug(instance, "Rolling back node scores for %s", instance->id); pcmk__restore_node_tables(instance, allowed_orig); if (chosen == NULL) { // Assignment failed, so give up pcmk__rsc_info(instance, "Not assigning %s to current node %s: unavailable", instance->id, pcmk__node_name(current)); pcmk__set_rsc_flags(instance, pcmk__rsc_unassigned); break; } // We prefer more strongly to assign an instance to the chosen node pcmk__rsc_debug(instance, "Not assigning %s to current node %s: %s is better", instance->id, pcmk__node_name(current), pcmk__node_name(chosen)); // Reserve one instance for the chosen node and try again if (++reserved >= available) { pcmk__rsc_info(instance, "Not assigning %s to current node %s: " "other assignments are more important", instance->id, pcmk__node_name(current)); } else { pcmk__rsc_debug(instance, "Reserved an instance of %s for %s. Retrying " "assignment of %s to %s", rsc->id, pcmk__node_name(chosen), instance->id, pcmk__node_name(current)); } // Clear this assignment (frees chosen); leave instance counts in parent pcmk__unassign_resource(instance); chosen = NULL; } g_hash_table_destroy(allowed_orig); // Restore original instances-per-node counts g_hash_table_destroy(parent->allowed_nodes); parent->allowed_nodes = allowed_orig_parent; if (chosen == NULL) { // Couldn't assign instance to current node return false; } pcmk__rsc_trace(instance, "Assigned %s to current node %s", instance->id, pcmk__node_name(current)); increment_parent_count(instance, chosen); return true; } /*! * \internal * \brief Reset the node counts of a resource's allowed nodes to zero * * \param[in,out] rsc Resource to reset * * \return Number of nodes that are available to run resources */ static unsigned int reset_allowed_node_counts(pcmk_resource_t *rsc) { unsigned int available_nodes = 0; pcmk_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { node->count = 0; if (pcmk__node_available(node, false, false)) { available_nodes++; } } return available_nodes; } /*! * \internal * \brief Check whether an instance has a preferred node * * \param[in] instance Clone instance or bundle replica container * \param[in] optimal_per_node Optimal number of instances per node * * \return Instance's current node if still available, otherwise NULL */ static const pcmk_node_t * preferred_node(const pcmk_resource_t *instance, int optimal_per_node) { const pcmk_node_t *node = NULL; const pcmk_node_t *parent_node = NULL; // Check whether instance is active, healthy, and not yet assigned - if ((instance->running_on == NULL) + if ((instance->private->active_nodes == NULL) || !pcmk_is_set(instance->flags, pcmk__rsc_unassigned) || pcmk_is_set(instance->flags, pcmk__rsc_failed)) { return NULL; } // Check whether instance's current node can run resources node = pcmk__current_node(instance); if (!pcmk__node_available(node, true, false)) { pcmk__rsc_trace(instance, "Not assigning %s to %s early (unavailable)", instance->id, pcmk__node_name(node)); return NULL; } // Check whether node already has optimal number of instances assigned parent_node = pcmk__top_allowed_node(instance, node); if ((parent_node != NULL) && (parent_node->count >= optimal_per_node)) { pcmk__rsc_trace(instance, "Not assigning %s to %s early " "(optimal instances already assigned)", instance->id, pcmk__node_name(node)); return NULL; } return node; } /*! * \internal * \brief Assign collective instances to nodes * * \param[in,out] collective Clone or bundle resource being assigned * \param[in,out] instances List of clone instances or bundle containers * \param[in] max_total Maximum instances to assign in total * \param[in] max_per_node Maximum instances to assign to any one node */ void pcmk__assign_instances(pcmk_resource_t *collective, GList *instances, int max_total, int max_per_node) { // Reuse node count to track number of assigned instances unsigned int available_nodes = reset_allowed_node_counts(collective); int optimal_per_node = 0; int assigned = 0; GList *iter = NULL; pcmk_resource_t *instance = NULL; const pcmk_node_t *current = NULL; if (available_nodes > 0) { optimal_per_node = max_total / available_nodes; } if (optimal_per_node < 1) { optimal_per_node = 1; } pcmk__rsc_debug(collective, "Assigning up to %d %s instance%s to up to %u node%s " "(at most %d per host, %d optimal)", max_total, collective->id, pcmk__plural_s(max_total), available_nodes, pcmk__plural_s(available_nodes), max_per_node, optimal_per_node); // Assign as many instances as possible to their current location for (iter = instances; (iter != NULL) && (assigned < max_total); iter = iter->next) { int available = max_total - assigned; instance = iter->data; if (!pcmk_is_set(instance->flags, pcmk__rsc_unassigned)) { continue; // Already assigned } current = preferred_node(instance, optimal_per_node); if ((current != NULL) && assign_instance_early(collective, instance, current, max_per_node, available)) { assigned++; } } pcmk__rsc_trace(collective, "Assigned %d of %d instance%s to current node", assigned, max_total, pcmk__plural_s(max_total)); for (iter = instances; iter != NULL; iter = iter->next) { instance = (pcmk_resource_t *) iter->data; if (!pcmk_is_set(instance->flags, pcmk__rsc_unassigned)) { continue; // Already assigned } - if (instance->running_on != NULL) { + if (instance->private->active_nodes != NULL) { current = pcmk__current_node(instance); if (pcmk__top_allowed_node(instance, current) == NULL) { const char *unmanaged = ""; if (!pcmk_is_set(instance->flags, pcmk__rsc_managed)) { unmanaged = "Unmanaged resource "; } crm_notice("%s%s is running on %s which is no longer allowed", unmanaged, instance->id, pcmk__node_name(current)); } } if (assigned >= max_total) { pcmk__rsc_debug(collective, "Not assigning %s because maximum %d instances " "already assigned", instance->id, max_total); resource_location(instance, NULL, -PCMK_SCORE_INFINITY, "collective_limit_reached", collective->private->scheduler); } else if (assign_instance(instance, NULL, max_per_node) != NULL) { assigned++; } } pcmk__rsc_debug(collective, "Assigned %d of %d possible instance%s of %s", assigned, max_total, pcmk__plural_s(max_total), collective->id); } enum instance_state { instance_starting = (1 << 0), instance_stopping = (1 << 1), /* This indicates that some instance is restarting. It's not the same as * instance_starting|instance_stopping, which would indicate that some * instance is starting, and some instance (not necessarily the same one) is * stopping. */ instance_restarting = (1 << 2), instance_active = (1 << 3), instance_all = instance_starting|instance_stopping |instance_restarting|instance_active, }; /*! * \internal * \brief Check whether an instance is active, starting, and/or stopping * * \param[in] instance Clone instance or bundle replica container * \param[in,out] state Whether any instance is starting, stopping, etc. */ static void check_instance_state(const pcmk_resource_t *instance, uint32_t *state) { const GList *iter = NULL; uint32_t instance_state = 0; // State of just this instance // No need to check further if all conditions have already been detected if (pcmk_all_flags_set(*state, instance_all)) { return; } // If instance is a collective (a cloned group), check its children instead if (instance->private->variant > pcmk__rsc_variant_primitive) { for (iter = instance->children; (iter != NULL) && !pcmk_all_flags_set(*state, instance_all); iter = iter->next) { check_instance_state((const pcmk_resource_t *) iter->data, state); } return; } // If we get here, instance is a primitive - if (instance->running_on != NULL) { + if (instance->private->active_nodes != NULL) { instance_state |= instance_active; } // Check each of the instance's actions for runnable start or stop for (iter = instance->private->actions; (iter != NULL) && !pcmk_all_flags_set(instance_state, instance_starting |instance_stopping); iter = iter->next) { const pcmk_action_t *action = (const pcmk_action_t *) iter->data; const bool optional = pcmk_is_set(action->flags, pcmk_action_optional); if (pcmk__str_eq(PCMK_ACTION_START, action->task, pcmk__str_none)) { if (!optional && pcmk_is_set(action->flags, pcmk_action_runnable)) { pcmk__rsc_trace(instance, "Instance is starting due to %s", action->uuid); instance_state |= instance_starting; } else { pcmk__rsc_trace(instance, "%s doesn't affect %s state (%s)", action->uuid, instance->id, (optional? "optional" : "unrunnable")); } } else if (pcmk__str_eq(PCMK_ACTION_STOP, action->task, pcmk__str_none)) { /* Only stop actions can be pseudo-actions for primitives. That * indicates that the node they are on is being fenced, so the stop * is implied rather than actually executed. */ if (!optional && pcmk_any_flags_set(action->flags, pcmk_action_pseudo |pcmk_action_runnable)) { pcmk__rsc_trace(instance, "Instance is stopping due to %s", action->uuid); instance_state |= instance_stopping; } else { pcmk__rsc_trace(instance, "%s doesn't affect %s state (%s)", action->uuid, instance->id, (optional? "optional" : "unrunnable")); } } } if (pcmk_all_flags_set(instance_state, instance_starting|instance_stopping)) { instance_state |= instance_restarting; } *state |= instance_state; } /*! * \internal * \brief Create actions for collective resource instances * * \param[in,out] collective Clone or bundle resource to create actions for * \param[in,out] instances List of clone instances or bundle containers */ void pcmk__create_instance_actions(pcmk_resource_t *collective, GList *instances) { uint32_t state = 0; pcmk_action_t *stop = NULL; pcmk_action_t *stopped = NULL; pcmk_action_t *start = NULL; pcmk_action_t *started = NULL; pcmk__rsc_trace(collective, "Creating collective instance actions for %s", collective->id); // Create actions for each instance appropriate to its variant for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; instance->private->cmds->create_actions(instance); check_instance_state(instance, &state); } // Create pseudo-actions for rsc start and started start = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_START, !pcmk_is_set(state, instance_starting), true); started = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_RUNNING, !pcmk_is_set(state, instance_starting), false); started->priority = PCMK_SCORE_INFINITY; if (pcmk_any_flags_set(state, instance_active|instance_starting)) { pcmk__set_action_flags(started, pcmk_action_runnable); } // Create pseudo-actions for rsc stop and stopped stop = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_STOP, !pcmk_is_set(state, instance_stopping), true); stopped = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_STOPPED, !pcmk_is_set(state, instance_stopping), true); stopped->priority = PCMK_SCORE_INFINITY; if (!pcmk_is_set(state, instance_restarting)) { pcmk__set_action_flags(stop, pcmk_action_migratable); } if (pcmk__is_clone(collective)) { pe__create_clone_notif_pseudo_ops(collective, start, started, stop, stopped); } } /*! * \internal * \brief Get a list of clone instances or bundle replica containers * * \param[in] rsc Clone or bundle resource * * \return Clone instances if \p rsc is a clone, or a newly created list of * \p rsc's replica containers if \p rsc is a bundle * \note The caller must call free_instance_list() on the result when the list * is no longer needed. */ static inline GList * get_instance_list(const pcmk_resource_t *rsc) { if (pcmk__is_bundle(rsc)) { return pe__bundle_containers(rsc); } else { return rsc->children; } } /*! * \internal * \brief Free any memory created by get_instance_list() * * \param[in] rsc Clone or bundle resource passed to get_instance_list() * \param[in,out] list Return value of get_instance_list() for \p rsc */ static inline void free_instance_list(const pcmk_resource_t *rsc, GList *list) { if (list != rsc->children) { g_list_free(list); } } /*! * \internal * \brief Check whether an instance is compatible with a role and node * * \param[in] instance Clone instance or bundle replica container * \param[in] node Instance must match this node * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return true if \p instance is compatible with \p node and \p role, * otherwise false */ bool pcmk__instance_matches(const pcmk_resource_t *instance, const pcmk_node_t *node, enum rsc_role_e role, bool current) { pcmk_node_t *instance_node = NULL; CRM_CHECK((instance != NULL) && (node != NULL), return false); if ((role != pcmk_role_unknown) && (role != instance->private->fns->state(instance, current))) { pcmk__rsc_trace(instance, "%s is not a compatible instance (role is not %s)", instance->id, pcmk_role_text(role)); return false; } if (!is_set_recursive(instance, pcmk__rsc_blocked, true)) { // We only want instances that haven't failed instance_node = instance->private->fns->location(instance, NULL, current); } if (instance_node == NULL) { pcmk__rsc_trace(instance, "%s is not a compatible instance " "(not assigned to a node)", instance->id); return false; } if (!pcmk__same_node(instance_node, node)) { pcmk__rsc_trace(instance, "%s is not a compatible instance " "(assigned to %s not %s)", instance->id, pcmk__node_name(instance_node), pcmk__node_name(node)); return false; } return true; } #define display_role(r) \ (((r) == pcmk_role_unknown)? "matching" : pcmk_role_text(r)) /*! * \internal * \brief Find an instance that matches a given resource by node and role * * \param[in] match_rsc Resource that instance must match (for logging only) * \param[in] rsc Clone or bundle resource to check for matching instance * \param[in] node Instance must match this node * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return \p rsc instance matching \p node and \p role if any, otherwise NULL */ static pcmk_resource_t * find_compatible_instance_on_node(const pcmk_resource_t *match_rsc, const pcmk_resource_t *rsc, const pcmk_node_t *node, enum rsc_role_e role, bool current) { GList *instances = NULL; instances = get_instance_list(rsc); for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; if (pcmk__instance_matches(instance, node, role, current)) { pcmk__rsc_trace(match_rsc, "Found %s %s instance %s compatible with %s on %s", display_role(role), rsc->id, instance->id, match_rsc->id, pcmk__node_name(node)); free_instance_list(rsc, instances); // Only frees list, not contents return instance; } } free_instance_list(rsc, instances); pcmk__rsc_trace(match_rsc, "No %s %s instance found compatible with %s on %s", display_role(role), rsc->id, match_rsc->id, pcmk__node_name(node)); return NULL; } /*! * \internal * \brief Find a clone instance or bundle container compatible with a resource * * \param[in] match_rsc Resource that instance must match * \param[in] rsc Clone or bundle resource to check for matching instance * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return Compatible (by \p role and \p match_rsc location) instance of \p rsc * if any, otherwise NULL */ pcmk_resource_t * pcmk__find_compatible_instance(const pcmk_resource_t *match_rsc, const pcmk_resource_t *rsc, enum rsc_role_e role, bool current) { pcmk_resource_t *instance = NULL; GList *nodes = NULL; const pcmk_node_t *node = NULL; // If match_rsc has a node, check only that node node = match_rsc->private->fns->location(match_rsc, NULL, current); if (node != NULL) { return find_compatible_instance_on_node(match_rsc, rsc, node, role, current); } // Otherwise check for an instance matching any of match_rsc's allowed nodes nodes = pcmk__sort_nodes(g_hash_table_get_values(match_rsc->allowed_nodes), NULL); for (GList *iter = nodes; (iter != NULL) && (instance == NULL); iter = iter->next) { instance = find_compatible_instance_on_node(match_rsc, rsc, (pcmk_node_t *) iter->data, role, current); } if (instance == NULL) { pcmk__rsc_debug(rsc, "No %s instance found compatible with %s", rsc->id, match_rsc->id); } g_list_free(nodes); return instance; } /*! * \internal * \brief Unassign an instance if mandatory ordering has no interleave match * * \param[in] first 'First' action in an ordering * \param[in] then 'Then' action in an ordering * \param[in,out] then_instance 'Then' instance that has no interleave match * \param[in] type Group of enum pcmk__action_relation_flags * \param[in] current If true, "then" action is stopped or demoted * * \return true if \p then_instance was unassigned, otherwise false */ static bool unassign_if_mandatory(const pcmk_action_t *first, const pcmk_action_t *then, pcmk_resource_t *then_instance, uint32_t type, bool current) { // Allow "then" instance to go down even without an interleave match if (current) { pcmk__rsc_trace(then->rsc, "%s has no instance to order before stopping " "or demoting %s", first->rsc->id, then_instance->id); /* If the "first" action must be runnable, but there is no "first" * instance, the "then" instance must not be allowed to come up. */ } else if (pcmk_any_flags_set(type, pcmk__ar_unrunnable_first_blocks |pcmk__ar_first_implies_then)) { pcmk__rsc_info(then->rsc, "Inhibiting %s from being active " "because there is no %s instance to interleave", then_instance->id, first->rsc->id); return pcmk__assign_resource(then_instance, NULL, true, true); } return false; } /*! * \internal * \brief Find first matching action for a clone instance or bundle container * * \param[in] action Action in an interleaved ordering * \param[in] instance Clone instance or bundle container being interleaved * \param[in] action_name Action to look for * \param[in] node If not NULL, require action to be on this node * \param[in] for_first If true, \p instance is the 'first' resource in the * ordering, otherwise it is the 'then' resource * * \return First action for \p instance (or in some cases if \p instance is a * bundle container, its containerized resource) that matches * \p action_name and \p node if any, otherwise NULL */ static pcmk_action_t * find_instance_action(const pcmk_action_t *action, const pcmk_resource_t *instance, const char *action_name, const pcmk_node_t *node, bool for_first) { const pcmk_resource_t *rsc = NULL; pcmk_action_t *matching_action = NULL; /* If instance is a bundle container, sometimes we should interleave the * action for the container itself, and sometimes for the containerized * resource. * * For example, given "start bundle A then bundle B", B likely requires the * service inside A's container to be active, rather than just the * container, so we should interleave the action for A's containerized * resource. On the other hand, it's possible B's container itself requires * something from A, so we should interleave the action for B's container. * * Essentially, for 'first', we should use the containerized resource for * everything except stop, and for 'then', we should use the container for * everything except promote and demote (which can only be performed on the * containerized resource). */ if ((for_first && !pcmk__str_any_of(action->task, PCMK_ACTION_STOP, PCMK_ACTION_STOPPED, NULL)) || (!for_first && pcmk__str_any_of(action->task, PCMK_ACTION_PROMOTE, PCMK_ACTION_PROMOTED, PCMK_ACTION_DEMOTE, PCMK_ACTION_DEMOTED, NULL))) { rsc = pe__get_rsc_in_container(instance); } if (rsc == NULL) { rsc = instance; // No containerized resource, use instance itself } else { node = NULL; // Containerized actions are on bundle-created guest } matching_action = find_first_action(rsc->private->actions, NULL, action_name, node); if (matching_action != NULL) { return matching_action; } if (pcmk_is_set(instance->flags, pcmk__rsc_removed) || pcmk__str_any_of(action_name, PCMK_ACTION_STOP, PCMK_ACTION_DEMOTE, NULL)) { crm_trace("No %s action found for %s%s", action_name, pcmk_is_set(instance->flags, pcmk__rsc_removed)? "orphan " : "", instance->id); } else { crm_err("No %s action found for %s to interleave (bug?)", action_name, instance->id); } return NULL; } /*! * \internal * \brief Get the original action name of a bundle or clone action * * Given an action for a bundle or clone, get the original action name, * mapping notify to the action being notified, and if the instances are * primitives, mapping completion actions to the action that was completed * (for example, stopped to stop). * * \param[in] action Clone or bundle action to check * * \return Original action name for \p action */ static const char * orig_action_name(const pcmk_action_t *action) { // Any instance will do const pcmk_resource_t *instance = action->rsc->children->data; char *action_type = NULL; const char *action_name = action->task; enum action_tasks orig_task = pcmk_action_unspecified; if (pcmk__strcase_any_of(action->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_NOTIFIED, NULL)) { // action->uuid is RSC_(confirmed-){pre,post}_notify_ACTION_INTERVAL CRM_CHECK(parse_op_key(action->uuid, NULL, &action_type, NULL), return pcmk_action_text(pcmk_action_unspecified)); action_name = strstr(action_type, "_notify_"); CRM_CHECK(action_name != NULL, return pcmk_action_text(pcmk_action_unspecified)); action_name += strlen("_notify_"); } orig_task = get_complex_task(instance, action_name); free(action_type); return pcmk_action_text(orig_task); } /*! * \internal * \brief Update two interleaved actions according to an ordering between them * * Given information about an ordering of two interleaved 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 * \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 * * \return Group of enum pcmk__updated flags indicating what was updated */ static uint32_t update_interleaved_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t filter, uint32_t type) { GList *instances = NULL; uint32_t changed = pcmk__updated_none; const char *orig_first_task = orig_action_name(first); // Stops and demotes must be interleaved with instance on current node bool current = pcmk__ends_with(first->uuid, "_" PCMK_ACTION_STOPPED "_0") || pcmk__ends_with(first->uuid, "_" PCMK_ACTION_DEMOTED "_0"); // Update the specified actions for each "then" instance individually instances = get_instance_list(then->rsc); for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *first_instance = NULL; pcmk_resource_t *then_instance = iter->data; pcmk_action_t *first_action = NULL; pcmk_action_t *then_action = NULL; // Find a "first" instance to interleave with this "then" instance first_instance = pcmk__find_compatible_instance(then_instance, first->rsc, pcmk_role_unknown, current); if (first_instance == NULL) { // No instance can be interleaved if (unassign_if_mandatory(first, then, then_instance, type, current)) { pcmk__set_updated_flags(changed, first, pcmk__updated_then); } continue; } first_action = find_instance_action(first, first_instance, orig_first_task, node, true); if (first_action == NULL) { continue; } then_action = find_instance_action(then, then_instance, then->task, node, false); if (then_action == NULL) { continue; } if (order_actions(first_action, then_action, type)) { pcmk__set_updated_flags(changed, first, pcmk__updated_first|pcmk__updated_then); } changed |= then_instance->private->cmds->update_ordered_actions( first_action, then_action, node, first_instance->private->cmds->action_flags(first_action, node), filter, type, then->rsc->private->scheduler); } free_instance_list(then->rsc, instances); return changed; } /*! * \internal * \brief Check whether two actions in an ordering can be interleaved * * \param[in] first 'First' action in the ordering * \param[in] then 'Then' action in the ordering * * \return true if \p first and \p then can be interleaved, otherwise false */ static bool can_interleave_actions(const pcmk_action_t *first, const pcmk_action_t *then) { bool interleave = false; pcmk_resource_t *rsc = NULL; if ((first->rsc == NULL) || (then->rsc == NULL)) { crm_trace("Not interleaving %s with %s: not resource actions", first->uuid, then->uuid); return false; } if (first->rsc == then->rsc) { crm_trace("Not interleaving %s with %s: same resource", first->uuid, then->uuid); return false; } if ((first->rsc->private->variant < pcmk__rsc_variant_clone) || (then->rsc->private->variant < pcmk__rsc_variant_clone)) { crm_trace("Not interleaving %s with %s: not clones or bundles", first->uuid, then->uuid); return false; } if (pcmk__ends_with(then->uuid, "_stop_0") || pcmk__ends_with(then->uuid, "_demote_0")) { rsc = first->rsc; } else { rsc = then->rsc; } interleave = crm_is_true(g_hash_table_lookup(rsc->meta, PCMK_META_INTERLEAVE)); pcmk__rsc_trace(rsc, "'%s then %s' will %sbe interleaved (based on %s)", first->uuid, then->uuid, (interleave? "" : "not "), rsc->id); return interleave; } /*! * \internal * \brief Update non-interleaved instance actions according to an ordering * * Given information about an ordering of two non-interleaved 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] instance Clone instance or bundle container * \param[in,out] first "First" action in ordering * \param[in] then "Then" action in ordering (for \p instance's parent) * \param[in] node If not NULL, limit scope of ordering to this node * \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 * * \return Group of enum pcmk__updated flags indicating what was updated */ static uint32_t update_noninterleaved_actions(pcmk_resource_t *instance, pcmk_action_t *first, const pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type) { pcmk_action_t *instance_action = NULL; pcmk_scheduler_t *scheduler = instance->private->scheduler; uint32_t instance_flags = 0; uint32_t changed = pcmk__updated_none; // Check whether instance has an equivalent of "then" action instance_action = find_first_action(instance->private->actions, NULL, then->task, node); if (instance_action == NULL) { return changed; } // Check whether action is runnable instance_flags = instance->private->cmds->action_flags(instance_action, node); if (!pcmk_is_set(instance_flags, pcmk_action_runnable)) { return changed; } // If so, update actions for the instance changed = instance->private->cmds->update_ordered_actions(first, instance_action, node, flags, filter, type, scheduler); // Propagate any changes to later actions if (pcmk_is_set(changed, pcmk__updated_then)) { for (GList *after_iter = instance_action->actions_after; after_iter != NULL; after_iter = after_iter->next) { pcmk__related_action_t *after = after_iter->data; pcmk__update_action_for_orderings(after->action, scheduler); } } return changed; } /*! * \internal * \brief Update two actions according to an ordering between them * * Given information about an ordering of two clone or bundle 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__instance_update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler) { CRM_ASSERT((first != NULL) && (then != NULL) && (scheduler != NULL)); if (then->rsc == NULL) { return pcmk__updated_none; } else if (can_interleave_actions(first, then)) { return update_interleaved_actions(first, then, node, filter, type); } else { uint32_t changed = pcmk__updated_none; GList *instances = get_instance_list(then->rsc); // Update actions for the clone or bundle resource itself changed |= pcmk__update_ordered_actions(first, then, node, flags, filter, type, scheduler); // Update the 'then' clone instances or bundle containers individually for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = iter->data; changed |= update_noninterleaved_actions(instance, first, then, node, flags, filter, type); } free_instance_list(then->rsc, instances); return changed; } } #define pe__clear_action_summary_flags(flags, action, flag) do { \ flags = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Action summary", action->rsc->id, \ flags, flag, #flag); \ } while (0) /*! * \internal * \brief Return action flags for a given clone or bundle action * * \param[in,out] action Action for a clone or bundle * \param[in] instances Clone instances or bundle containers * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__collective_action_flags(pcmk_action_t *action, const GList *instances, const pcmk_node_t *node) { bool any_runnable = false; const char *action_name = orig_action_name(action); // Set original assumptions (optional and runnable may be cleared below) uint32_t flags = pcmk_action_optional |pcmk_action_runnable |pcmk_action_pseudo; for (const GList *iter = instances; iter != NULL; iter = iter->next) { const pcmk_resource_t *instance = iter->data; const pcmk_node_t *instance_node = NULL; pcmk_action_t *instance_action = NULL; uint32_t instance_flags; // Node is relevant only to primitive instances if (pcmk__is_primitive(instance)) { instance_node = node; } instance_action = find_first_action(instance->private->actions, NULL, action_name, instance_node); if (instance_action == NULL) { pcmk__rsc_trace(action->rsc, "%s has no %s action on %s", instance->id, action_name, pcmk__node_name(node)); continue; } pcmk__rsc_trace(action->rsc, "%s has %s for %s on %s", instance->id, instance_action->uuid, action_name, pcmk__node_name(node)); instance_flags = instance->private->cmds->action_flags(instance_action, node); // If any instance action is mandatory, so is the collective action if (pcmk_is_set(flags, pcmk_action_optional) && !pcmk_is_set(instance_flags, pcmk_action_optional)) { pcmk__rsc_trace(instance, "%s is mandatory because %s is", action->uuid, instance_action->uuid); pe__clear_action_summary_flags(flags, action, pcmk_action_optional); pcmk__clear_action_flags(action, pcmk_action_optional); } // If any instance action is runnable, so is the collective action if (pcmk_is_set(instance_flags, pcmk_action_runnable)) { any_runnable = true; } } if (!any_runnable) { pcmk__rsc_trace(action->rsc, "%s is not runnable because no instance can run %s", action->uuid, action_name); pe__clear_action_summary_flags(flags, action, pcmk_action_runnable); if (node == NULL) { pcmk__clear_action_flags(action, pcmk_action_runnable); } } return flags; } diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c index 3d92082688..dcf1f09739 100644 --- a/lib/pacemaker/pcmk_sched_primitive.c +++ b/lib/pacemaker/pcmk_sched_primitive.c @@ -1,1704 +1,1708 @@ /* * 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->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->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->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->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->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->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) { + } else if ((rsc->private->active_nodes != NULL) && stop_if_fail) { pcmk__rsc_info(rsc, "Stopping removed resource %s", rsc->id); } } pcmk__clear_rsc_flags(rsc, pcmk__rsc_assigning); if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { remote_connection_assigned(rsc); } return rsc->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->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->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->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->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; pcmk_node_t *migration_target = 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->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->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->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 migration_target = rsc->private->partial_migration_target; if ((rsc->private->partial_migration_source != NULL) && (migration_target != NULL) && allow_migrate && (num_all_active == 2) && pcmk__same_node(current, rsc->private->partial_migration_source) && pcmk__same_node(rsc->private->assigned_node, migration_target)) { /* A partial migration is in progress, and the migration target remains * the same as when the migration began. */ pcmk__rsc_trace(rsc, "Partial migration of %s from %s to %s will continue", rsc->id, pcmk__node_name(rsc->private->partial_migration_source), pcmk__node_name(migration_target)); } else if ((rsc->private->partial_migration_source != NULL) || (migration_target != NULL)) { // A partial migration is in progress but can't be continued if (num_all_active > 2) { // The resource is migrating *and* multiply active! crm_notice("Forcing recovery of %s because it is migrating " "from %s to %s and possibly active elsewhere", rsc->id, pcmk__node_name(rsc->private->partial_migration_source), pcmk__node_name(migration_target)); } else { // The migration source or target isn't available crm_notice("Forcing recovery of %s because it can no longer " "migrate from %s to %s", rsc->id, pcmk__node_name(rsc->private->partial_migration_source), pcmk__node_name(migration_target)); } need_stop = true; rsc->private->partial_migration_source = NULL; rsc->private->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->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->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->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) { + for (GList *iter = rsc->private->active_nodes; + iter != NULL; iter = iter->next) { + pcmk_node_t *current = (pcmk_node_t *) iter->data; pcmk_action_t *stop = NULL; if (is_expected_node(rsc, current)) { /* We are scheduling restart actions for a multiply active resource * with PCMK_META_MULTIPLE_ACTIVE=PCMK_VALUE_STOP_UNEXPECTED, and * this is where it should not be stopped. */ pcmk__rsc_trace(rsc, "Skipping stop of multiply active resource %s " "on expected node %s", rsc->id, pcmk__node_name(current)); continue; } if (rsc->private->partial_migration_target != NULL) { // Continue migration if node originally was and remains target if (pcmk__same_node(current, rsc->private->partial_migration_target) && 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->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) { + for (GList *iter = rsc->private->active_nodes; + iter != NULL; iter = iter->next) { + pcmk_node_t *current = (pcmk_node_t *) iter->data; if (is_expected_node(rsc, current)) { pcmk__rsc_trace(rsc, "Skipping demote of multiply active resource %s " "on expected node %s", rsc->id, pcmk__node_name(current)); } else { pcmk__rsc_trace(rsc, "Scheduling %s demotion of %s on %s", (optional? "optional" : "required"), rsc->id, pcmk__node_name(current)); demote_action(rsc, current, optional); } } } static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional) { 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) { + if (rsc->private->active_nodes != NULL) { /* The resource was started elsewhere even though it is now * considered locked. This shouldn't be possible, but as a * failsafe, we don't want to disturb the resource now. */ pcmk__rsc_info(rsc, "Cancelling shutdown lock " "because %s is already active", rsc->id); pe__clear_resource_history(rsc, rsc->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; + } else if (pcmk__list_of_1(rsc->private->active_nodes)) { + pcmk_node_t *node = rsc->private->active_nodes->data; if (node->details->shutdown) { if (node->details->unclean) { pcmk__rsc_debug(rsc, "Not locking %s to unclean %s for shutdown", rsc->id, pcmk__node_name(node)); } else { rsc->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 d9d7f71622..7fdf2a9a21 100644 --- a/lib/pacemaker/pcmk_sched_probes.c +++ b/lib/pacemaker/pcmk_sched_probes.c @@ -1,914 +1,915 @@ /* * 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); + pcmk_node_t *running = pe_find_node_id(rsc->private->active_nodes, + 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) { const pcmk_node_t *rsc1_node = rsc1->private->assigned_node; if ((rsc1_node != NULL) && (g_hash_table_lookup(rsc1->known_on, 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_node != NULL) - && pcmk__find_node_in_list(guest_rsc->running_on, + && pcmk__find_node_in_list(guest_rsc->private->active_nodes, 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)) { + && (top->private->active_nodes == 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 + if ((then->action->rsc->private->active_nodes != NULL) || (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_promotable.c b/lib/pacemaker/pcmk_sched_promotable.c index 52b96ec038..159f2a8468 100644 --- a/lib/pacemaker/pcmk_sched_promotable.c +++ b/lib/pacemaker/pcmk_sched_promotable.c @@ -1,1341 +1,1343 @@ /* * 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 "libpacemaker_private.h" /*! * \internal * \brief Add implicit promotion ordering for a promotable instance * * \param[in,out] clone Clone resource * \param[in,out] child Instance of \p clone being ordered * \param[in,out] last Previous instance ordered (NULL if \p child is first) */ static void order_instance_promotion(pcmk_resource_t *clone, pcmk_resource_t *child, pcmk_resource_t *last) { // "Promote clone" -> promote instance -> "clone promoted" pcmk__order_resource_actions(clone, PCMK_ACTION_PROMOTE, child, PCMK_ACTION_PROMOTE, pcmk__ar_ordered); pcmk__order_resource_actions(child, PCMK_ACTION_PROMOTE, clone, PCMK_ACTION_PROMOTED, pcmk__ar_ordered); // If clone is ordered, order this instance relative to last if ((last != NULL) && pe__clone_is_ordered(clone)) { pcmk__order_resource_actions(last, PCMK_ACTION_PROMOTE, child, PCMK_ACTION_PROMOTE, pcmk__ar_ordered); } } /*! * \internal * \brief Add implicit demotion ordering for a promotable instance * * \param[in,out] clone Clone resource * \param[in,out] child Instance of \p clone being ordered * \param[in] last Previous instance ordered (NULL if \p child is first) */ static void order_instance_demotion(pcmk_resource_t *clone, pcmk_resource_t *child, pcmk_resource_t *last) { // "Demote clone" -> demote instance -> "clone demoted" pcmk__order_resource_actions(clone, PCMK_ACTION_DEMOTE, child, PCMK_ACTION_DEMOTE, pcmk__ar_then_implies_first_graphed); pcmk__order_resource_actions(child, PCMK_ACTION_DEMOTE, clone, PCMK_ACTION_DEMOTED, pcmk__ar_first_implies_then_graphed); // If clone is ordered, order this instance relative to last if ((last != NULL) && pe__clone_is_ordered(clone)) { pcmk__order_resource_actions(child, PCMK_ACTION_DEMOTE, last, PCMK_ACTION_DEMOTE, pcmk__ar_ordered); } } /*! * \internal * \brief Check whether an instance will be promoted or demoted * * \param[in] rsc Instance to check * \param[out] demoting If \p rsc will be demoted, this will be set to true * \param[out] promoting If \p rsc will be promoted, this will be set to true */ static void check_for_role_change(const pcmk_resource_t *rsc, bool *demoting, bool *promoting) { const GList *iter = NULL; // If this is a cloned group, check group members recursively if (rsc->children != NULL) { for (iter = rsc->children; iter != NULL; iter = iter->next) { check_for_role_change((const pcmk_resource_t *) iter->data, demoting, promoting); } return; } for (iter = rsc->private->actions; iter != NULL; iter = iter->next) { const pcmk_action_t *action = (const pcmk_action_t *) iter->data; if (*promoting && *demoting) { return; } else if (pcmk_is_set(action->flags, pcmk_action_optional)) { continue; } else if (pcmk__str_eq(PCMK_ACTION_DEMOTE, action->task, pcmk__str_none)) { *demoting = true; } else if (pcmk__str_eq(PCMK_ACTION_PROMOTE, action->task, pcmk__str_none)) { *promoting = true; } } } /*! * \internal * \brief Add promoted-role location constraint scores to an instance's priority * * Adjust a promotable clone instance's promotion priority by the scores of any * location constraints in a list that are both limited to the promoted role and * for the node where the instance will be placed. * * \param[in,out] child Promotable clone instance * \param[in] location_constraints List of location constraints to apply * \param[in] chosen Node where \p child will be placed */ static void apply_promoted_locations(pcmk_resource_t *child, const GList *location_constraints, const pcmk_node_t *chosen) { for (const GList *iter = location_constraints; iter; iter = iter->next) { const pcmk__location_t *location = iter->data; const pcmk_node_t *constraint_node = NULL; if (location->role_filter == pcmk_role_promoted) { constraint_node = pe_find_node_id(location->nodes, chosen->details->id); } if (constraint_node != NULL) { int new_priority = pcmk__add_scores(child->private->priority, constraint_node->weight); pcmk__rsc_trace(child, "Applying location %s to %s promotion priority on " "%s: %s + %s = %s", location->id, child->id, pcmk__node_name(constraint_node), pcmk_readable_score(child->private->priority), pcmk_readable_score(constraint_node->weight), pcmk_readable_score(new_priority)); child->private->priority = new_priority; } } } /*! * \internal * \brief Get the node that an instance will be promoted on * * \param[in] rsc Promotable clone instance to check * * \return Node that \p rsc will be promoted on, or NULL if none */ static pcmk_node_t * node_to_be_promoted_on(const pcmk_resource_t *rsc) { pcmk_node_t *node = NULL; pcmk_node_t *local_node = NULL; const pcmk_resource_t *parent = NULL; // If this is a cloned group, bail if any group member can't be promoted for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) iter->data; if (node_to_be_promoted_on(child) == NULL) { pcmk__rsc_trace(rsc, "%s can't be promoted because member %s can't", rsc->id, child->id); return NULL; } } node = rsc->private->fns->location(rsc, NULL, FALSE); if (node == NULL) { pcmk__rsc_trace(rsc, "%s can't be promoted because it won't be active", rsc->id); return NULL; } else if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { if (rsc->private->fns->state(rsc, TRUE) == pcmk_role_promoted) { crm_notice("Unmanaged instance %s will be left promoted on %s", rsc->id, pcmk__node_name(node)); } else { pcmk__rsc_trace(rsc, "%s can't be promoted because it is unmanaged", rsc->id); return NULL; } } else if (rsc->private->priority < 0) { pcmk__rsc_trace(rsc, "%s can't be promoted because its promotion priority " "%d is negative", rsc->id, rsc->private->priority); return NULL; } else if (!pcmk__node_available(node, false, true)) { pcmk__rsc_trace(rsc, "%s can't be promoted because %s can't run resources", rsc->id, pcmk__node_name(node)); return NULL; } parent = pe__const_top_resource(rsc, false); local_node = g_hash_table_lookup(parent->allowed_nodes, node->details->id); if (local_node == NULL) { /* It should not be possible for the scheduler to have assigned the * instance to a node where its parent is not allowed, but it's good to * have a fail-safe. */ if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__sched_err("%s can't be promoted because %s is not allowed " "on %s (scheduler bug?)", rsc->id, parent->id, pcmk__node_name(node)); } // else the instance is unmanaged and already promoted return NULL; } else if ((local_node->count >= pe__clone_promoted_node_max(parent)) && pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__rsc_trace(rsc, "%s can't be promoted because %s has " "maximum promoted instances already", rsc->id, pcmk__node_name(node)); return NULL; } return local_node; } /*! * \internal * \brief Compare two promotable clone instances by promotion priority * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a has higher promotion priority, * a positive number if \p b has higher promotion priority, * or 0 if promotion priorities are equal */ static gint cmp_promotable_instance(gconstpointer a, gconstpointer b) { const pcmk_resource_t *rsc1 = (const pcmk_resource_t *) a; const pcmk_resource_t *rsc2 = (const pcmk_resource_t *) b; enum rsc_role_e role1 = pcmk_role_unknown; enum rsc_role_e role2 = pcmk_role_unknown; CRM_ASSERT((rsc1 != NULL) && (rsc2 != NULL)); // Check promotion priority set by pcmk__set_instance_roles() if (rsc1->private->promotion_priority > rsc2->private->promotion_priority) { pcmk__rsc_trace(rsc1, "%s has higher promotion priority (%s) than %s (%d)", rsc1->id, pcmk_readable_score(rsc1->private->promotion_priority), rsc2->id, rsc2->private->promotion_priority); return -1; } if (rsc1->private->promotion_priority < rsc2->private->promotion_priority) { pcmk__rsc_trace(rsc1, "%s has lower promotion priority (%s) than %s (%d)", rsc1->id, pcmk_readable_score(rsc1->private->promotion_priority), rsc2->id, rsc2->private->promotion_priority); return 1; } // If those are the same, prefer instance whose current role is higher role1 = rsc1->private->fns->state(rsc1, TRUE); role2 = rsc2->private->fns->state(rsc2, TRUE); if (role1 > role2) { pcmk__rsc_trace(rsc1, "%s has higher promotion priority than %s " "(higher current role)", rsc1->id, rsc2->id); return -1; } else if (role1 < role2) { pcmk__rsc_trace(rsc1, "%s has lower promotion priority than %s " "(lower current role)", rsc1->id, rsc2->id); return 1; } // Finally, do normal clone instance sorting return pcmk__cmp_instance(a, b); } /*! * \internal * \brief Add promotable clone instance's promotion priority to its node's score * * Add a promotable clone instance's promotion priority (which sums its * promotion preferences and scores of relevant location constraints for the * promoted role) to the node score of the instance's assigned node. * * \param[in] data Promotable clone instance * \param[in,out] user_data Clone parent of \p data */ static void add_promotion_priority_to_node_score(gpointer data, gpointer user_data) { const pcmk_resource_t *child = (const pcmk_resource_t *) data; pcmk_resource_t *clone = (pcmk_resource_t *) user_data; pcmk_node_t *node = NULL; const pcmk_node_t *chosen = NULL; const int promotion_priority = child->private->promotion_priority; if (promotion_priority < 0) { pcmk__rsc_trace(clone, "Not adding promotion priority of %s: negative (%s)", child->id, pcmk_readable_score(promotion_priority)); return; } chosen = child->private->fns->location(child, NULL, FALSE); if (chosen == NULL) { pcmk__rsc_trace(clone, "Not adding promotion priority of %s: inactive", child->id); return; } node = g_hash_table_lookup(clone->allowed_nodes, chosen->details->id); CRM_ASSERT(node != NULL); node->weight = pcmk__add_scores(promotion_priority, node->weight); pcmk__rsc_trace(clone, "Added cumulative priority of %s (%s) to score on %s " "(now %d)", child->id, pcmk_readable_score(promotion_priority), pcmk__node_name(node), node->weight); } /*! * \internal * \brief Apply colocation to dependent's node scores if for promoted role * * \param[in,out] data Colocation constraint to apply * \param[in,out] user_data Promotable clone that is constraint's dependent */ static void apply_coloc_to_dependent(gpointer data, gpointer user_data) { pcmk__colocation_t *colocation = data; pcmk_resource_t *clone = user_data; pcmk_resource_t *primary = colocation->primary; uint32_t flags = pcmk__coloc_select_default; float factor = colocation->score / (float) PCMK_SCORE_INFINITY; if (colocation->dependent_role != pcmk_role_promoted) { return; } if (colocation->score < PCMK_SCORE_INFINITY) { flags = pcmk__coloc_select_active; } pcmk__rsc_trace(clone, "Applying colocation %s (promoted %s with %s) @%s", colocation->id, colocation->dependent->id, colocation->primary->id, pcmk_readable_score(colocation->score)); primary->private->cmds->add_colocated_node_scores(primary, clone, clone->id, &clone->allowed_nodes, colocation, factor, flags); } /*! * \internal * \brief Apply colocation to primary's node scores if for promoted role * * \param[in,out] data Colocation constraint to apply * \param[in,out] user_data Promotable clone that is constraint's primary */ static void apply_coloc_to_primary(gpointer data, gpointer user_data) { pcmk__colocation_t *colocation = data; pcmk_resource_t *clone = user_data; pcmk_resource_t *dependent = colocation->dependent; const float factor = colocation->score / (float) PCMK_SCORE_INFINITY; const uint32_t flags = pcmk__coloc_select_active |pcmk__coloc_select_nonnegative; if ((colocation->primary_role != pcmk_role_promoted) || !pcmk__colocation_has_influence(colocation, NULL)) { return; } pcmk__rsc_trace(clone, "Applying colocation %s (%s with promoted %s) @%s", colocation->id, colocation->dependent->id, colocation->primary->id, pcmk_readable_score(colocation->score)); dependent->private->cmds->add_colocated_node_scores(dependent, clone, clone->id, &clone->allowed_nodes, colocation, factor, flags); } /*! * \internal * \brief Set clone instance's promotion priority to its node's score * * \param[in,out] data Promotable clone instance * \param[in] user_data Parent clone of \p data */ static void set_promotion_priority_to_node_score(gpointer data, gpointer user_data) { pcmk_resource_t *child = (pcmk_resource_t *) data; const pcmk_resource_t *clone = (const pcmk_resource_t *) user_data; pcmk_node_t *chosen = child->private->fns->location(child, NULL, FALSE); if (!pcmk_is_set(child->flags, pcmk__rsc_managed) && (child->next_role == pcmk_role_promoted)) { child->private->promotion_priority = PCMK_SCORE_INFINITY; pcmk__rsc_trace(clone, "Final promotion priority for %s is %s " "(unmanaged promoted)", child->id, pcmk_readable_score(PCMK_SCORE_INFINITY)); } else if ((chosen == NULL) || (child->private->promotion_priority < 0)) { pcmk__rsc_trace(clone, "Final promotion priority for %s is %s " "(ignoring node score)", child->id, pcmk_readable_score(child->private->promotion_priority)); } else { const pcmk_node_t *node = g_hash_table_lookup(clone->allowed_nodes, chosen->details->id); CRM_ASSERT(node != NULL); child->private->promotion_priority = node->weight; pcmk__rsc_trace(clone, "Adding scores for %s: " "final promotion priority for %s is %s", clone->id, child->id, pcmk_readable_score(child->private->promotion_priority)); } } /*! * \internal * \brief Sort a promotable clone's instances by descending promotion priority * * \param[in,out] clone Promotable clone to sort */ static void sort_promotable_instances(pcmk_resource_t *clone) { GList *colocations = NULL; if (pe__set_clone_flag(clone, pcmk__clone_promotion_constrained) == pcmk_rc_already) { return; } pcmk__set_rsc_flags(clone, pcmk__rsc_updating_nodes); for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) iter->data; pcmk__rsc_trace(clone, "Adding scores for %s: " "initial promotion priority for %s is %s", clone->id, child->id, pcmk_readable_score(child->private->promotion_priority)); } pe__show_node_scores(true, clone, "Before", clone->allowed_nodes, clone->private->scheduler); g_list_foreach(clone->children, add_promotion_priority_to_node_score, clone); colocations = pcmk__this_with_colocations(clone); g_list_foreach(colocations, apply_coloc_to_dependent, clone); g_list_free(colocations); colocations = pcmk__with_this_colocations(clone); g_list_foreach(colocations, apply_coloc_to_primary, clone); g_list_free(colocations); // Ban resource from all nodes if it needs a ticket but doesn't have it pcmk__require_promotion_tickets(clone); pe__show_node_scores(true, clone, "After", clone->allowed_nodes, clone->private->scheduler); // Reset promotion priorities to final node scores g_list_foreach(clone->children, set_promotion_priority_to_node_score, clone); // Finally, sort instances in descending order of promotion priority clone->children = g_list_sort(clone->children, cmp_promotable_instance); pcmk__clear_rsc_flags(clone, pcmk__rsc_updating_nodes); } /*! * \internal * \brief Find the active instance (if any) of an anonymous clone on a node * * \param[in] clone Anonymous clone to check * \param[in] id Instance ID (without instance number) to check * \param[in] node Node to check * * \return */ static pcmk_resource_t * find_active_anon_instance(const pcmk_resource_t *clone, const char *id, const pcmk_node_t *node) { for (GList *iter = clone->children; iter; iter = iter->next) { pcmk_resource_t *child = iter->data; pcmk_resource_t *active = NULL; // Use ->find_rsc() in case this is a cloned group active = clone->private->fns->find_rsc(child, id, node, pcmk_rsc_match_clone_only |pcmk_rsc_match_current_node); if (active != NULL) { return active; } } return NULL; } /* * \brief Check whether an anonymous clone instance is known on a node * * \param[in] clone Anonymous clone to check * \param[in] id Instance ID (without instance number) to check * \param[in] node Node to check * * \return true if \p id instance of \p clone is known on \p node, * otherwise false */ static bool anonymous_known_on(const pcmk_resource_t *clone, const char *id, const pcmk_node_t *node) { for (GList *iter = clone->children; iter; iter = iter->next) { pcmk_resource_t *child = iter->data; /* Use ->find_rsc() because this might be a cloned group, and knowing * that other members of the group are known here implies nothing. */ child = clone->private->fns->find_rsc(child, id, NULL, pcmk_rsc_match_clone_only); CRM_LOG_ASSERT(child != NULL); if (child != NULL) { if (g_hash_table_lookup(child->known_on, node->details->id)) { return true; } } } return false; } /*! * \internal * \brief Check whether a node is allowed to run a resource * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p node is allowed to run \p rsc, otherwise false */ static bool is_allowed(const pcmk_resource_t *rsc, const pcmk_node_t *node) { pcmk_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); return (allowed != NULL) && (allowed->weight >= 0); } /*! * \brief Check whether a clone instance's promotion score should be considered * * \param[in] rsc Promotable clone instance to check * \param[in] node Node where score would be applied * * \return true if \p rsc's promotion score should be considered on \p node, * otherwise false */ static bool promotion_score_applies(const pcmk_resource_t *rsc, const pcmk_node_t *node) { char *id = clone_strip(rsc->id); const pcmk_resource_t *parent = pe__const_top_resource(rsc, false); pcmk_resource_t *active = NULL; const char *reason = "allowed"; // Some checks apply only to anonymous clone instances if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { // If instance is active on the node, its score definitely applies active = find_active_anon_instance(parent, id, node); if (active == rsc) { reason = "active"; goto check_allowed; } /* If *no* instance is active on this node, this instance's score will * count if it has been probed on this node. */ if ((active == NULL) && anonymous_known_on(parent, id, node)) { reason = "probed"; goto check_allowed; } } /* If this clone's status is unknown on *all* nodes (e.g. cluster startup), * take all instances' scores into account, to make sure we use any * permanent promotion scores. */ - if ((rsc->running_on == NULL) && (g_hash_table_size(rsc->known_on) == 0)) { + if ((rsc->private->active_nodes == NULL) + && (g_hash_table_size(rsc->known_on) == 0)) { reason = "none probed"; goto check_allowed; } /* Otherwise, we've probed and/or started the resource *somewhere*, so * consider promotion scores on nodes where we know the status. */ if ((g_hash_table_lookup(rsc->known_on, node->details->id) != NULL) - || (pe_find_node_id(rsc->running_on, node->details->id) != NULL)) { + || (pe_find_node_id(rsc->private->active_nodes, + node->details->id) != NULL)) { reason = "known"; } else { pcmk__rsc_trace(rsc, "Ignoring %s promotion score (for %s) on %s: " "not probed", rsc->id, id, pcmk__node_name(node)); free(id); return false; } check_allowed: if (is_allowed(rsc, node)) { pcmk__rsc_trace(rsc, "Counting %s promotion score (for %s) on %s: %s", rsc->id, id, pcmk__node_name(node), reason); free(id); return true; } pcmk__rsc_trace(rsc, "Ignoring %s promotion score (for %s) on %s: not allowed", rsc->id, id, pcmk__node_name(node)); free(id); return false; } /*! * \internal * \brief Get the value of a promotion score node attribute * * \param[in] rsc Promotable clone instance to get promotion score for * \param[in] node Node to get promotion score for * \param[in] name Resource name to use in promotion score attribute name * * \return Value of promotion score node attribute for \p rsc on \p node */ static const char * promotion_attr_value(const pcmk_resource_t *rsc, const pcmk_node_t *node, const char *name) { char *attr_name = NULL; const char *attr_value = NULL; const char *target = NULL; enum pcmk__rsc_node node_type = pcmk__rsc_node_assigned; if (pcmk_is_set(rsc->flags, pcmk__rsc_unassigned)) { // Not assigned yet node_type = pcmk__rsc_node_current; } target = g_hash_table_lookup(rsc->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); attr_name = pcmk_promotion_score_name(name); attr_value = pcmk__node_attr(node, attr_name, target, node_type); free(attr_name); return attr_value; } /*! * \internal * \brief Get the promotion score for a clone instance on a node * * \param[in] rsc Promotable clone instance to get score for * \param[in] node Node to get score for * \param[out] is_default If non-NULL, will be set true if no score available * * \return Promotion score for \p rsc on \p node (or 0 if none) */ static int promotion_score(const pcmk_resource_t *rsc, const pcmk_node_t *node, bool *is_default) { const char *name = NULL; const char *attr_value = NULL; if (is_default != NULL) { *is_default = true; } CRM_CHECK((rsc != NULL) && (node != NULL), return 0); /* If this is an instance of a cloned group, the promotion score is the sum * of all members' promotion scores. */ if (rsc->children != NULL) { int score = 0; for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *child = (const pcmk_resource_t *) iter->data; bool child_default = false; int child_score = promotion_score(child, node, &child_default); if (!child_default && (is_default != NULL)) { *is_default = false; } score += child_score; } return score; } if (!promotion_score_applies(rsc, node)) { return 0; } /* For the promotion score attribute name, use the name the resource is * known as in resource history, since that's what crm_attribute --promotion * would have used. */ name = pcmk__s(rsc->private->history_id, rsc->id); attr_value = promotion_attr_value(rsc, node, name); if (attr_value != NULL) { pcmk__rsc_trace(rsc, "Promotion score for %s on %s = %s", name, pcmk__node_name(node), pcmk__s(attr_value, "(unset)")); } else if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { /* If we don't have any resource history yet, we won't have history_id. * In that case, for anonymous clones, try the resource name without * any instance number. */ char *rsc_name = clone_strip(rsc->id); if (strcmp(rsc->id, rsc_name) != 0) { attr_value = promotion_attr_value(rsc, node, rsc_name); pcmk__rsc_trace(rsc, "Promotion score for %s on %s (for %s) = %s", rsc_name, pcmk__node_name(node), rsc->id, pcmk__s(attr_value, "(unset)")); } free(rsc_name); } if (attr_value == NULL) { return 0; } if (is_default != NULL) { *is_default = false; } return char2score(attr_value); } /*! * \internal * \brief Include promotion scores in instances' node scores and priorities * * \param[in,out] rsc Promotable clone resource to update */ void pcmk__add_promotion_scores(pcmk_resource_t *rsc) { if (pe__set_clone_flag(rsc, pcmk__clone_promotion_added) == pcmk_rc_already) { return; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) iter->data; GHashTableIter iter; pcmk_node_t *node = NULL; int score, new_score; g_hash_table_iter_init(&iter, child_rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (!pcmk__node_available(node, false, false)) { /* This node will never be promoted, so don't apply the * promotion score, as that may lead to clone shuffling. */ continue; } score = promotion_score(child_rsc, node, NULL); if (score > 0) { new_score = pcmk__add_scores(node->weight, score); if (new_score != node->weight) { // Could remain INFINITY node->weight = new_score; pcmk__rsc_trace(rsc, "Added %s promotion priority (%s) to score " "on %s (now %s)", child_rsc->id, pcmk_readable_score(score), pcmk__node_name(node), pcmk_readable_score(new_score)); } } if (score > child_rsc->private->priority) { pcmk__rsc_trace(rsc, "Updating %s priority to promotion score " "(%d->%d)", child_rsc->id, child_rsc->private->priority, score); child_rsc->private->priority = score; } } } } /*! * \internal * \brief If a resource's current role is started, change it to unpromoted * * \param[in,out] data Resource to update * \param[in] user_data Ignored */ static void set_current_role_unpromoted(void *data, void *user_data) { pcmk_resource_t *rsc = (pcmk_resource_t *) data; if (rsc->role == pcmk_role_started) { // Promotable clones should use unpromoted role instead of started rsc->role = pcmk_role_unpromoted; } g_list_foreach(rsc->children, set_current_role_unpromoted, NULL); } /*! * \internal * \brief Set a resource's next role to unpromoted (or stopped if unassigned) * * \param[in,out] data Resource to update * \param[in] user_data Ignored */ static void set_next_role_unpromoted(void *data, void *user_data) { pcmk_resource_t *rsc = (pcmk_resource_t *) data; GList *assigned = NULL; rsc->private->fns->location(rsc, &assigned, FALSE); if (assigned == NULL) { pe__set_next_role(rsc, pcmk_role_stopped, "stopped instance"); } else { pe__set_next_role(rsc, pcmk_role_unpromoted, "unpromoted instance"); g_list_free(assigned); } g_list_foreach(rsc->children, set_next_role_unpromoted, NULL); } /*! * \internal * \brief Set a resource's next role to promoted if not already set * * \param[in,out] data Resource to update * \param[in] user_data Ignored */ static void set_next_role_promoted(void *data, gpointer user_data) { pcmk_resource_t *rsc = (pcmk_resource_t *) data; if (rsc->next_role == pcmk_role_unknown) { pe__set_next_role(rsc, pcmk_role_promoted, "promoted instance"); } g_list_foreach(rsc->children, set_next_role_promoted, NULL); } /*! * \internal * \brief Show instance's promotion score on node where it will be active * * \param[in,out] instance Promotable clone instance to show */ static void show_promotion_score(pcmk_resource_t *instance) { pcmk_node_t *chosen = instance->private->fns->location(instance, NULL, FALSE); const char *score_s = NULL; score_s = pcmk_readable_score(instance->private->promotion_priority); if (pcmk_is_set(instance->private->scheduler->flags, pcmk_sched_output_scores) && !pcmk__is_daemon && (instance->private->scheduler->priv != NULL)) { pcmk__output_t *out = instance->private->scheduler->priv; out->message(out, "promotion-score", instance, chosen, score_s); } else { pcmk__rsc_debug(pe__const_top_resource(instance, false), "%s promotion score on %s: sort=%s priority=%d", instance->id, ((chosen == NULL)? "none" : pcmk__node_name(chosen)), score_s, instance->private->priority); } } /*! * \internal * \brief Set a clone instance's promotion priority * * \param[in,out] data Promotable clone instance to update * \param[in] user_data Instance's parent clone */ static void set_instance_priority(gpointer data, gpointer user_data) { pcmk_resource_t *instance = (pcmk_resource_t *) data; const pcmk_resource_t *clone = (const pcmk_resource_t *) user_data; const pcmk_node_t *chosen = NULL; enum rsc_role_e next_role = pcmk_role_unknown; GList *list = NULL; pcmk__rsc_trace(clone, "Assigning priority for %s: %s", instance->id, pcmk_role_text(instance->next_role)); if (instance->private->fns->state(instance, TRUE) == pcmk_role_started) { set_current_role_unpromoted(instance, NULL); } // Only an instance that will be active can be promoted chosen = instance->private->fns->location(instance, &list, FALSE); if (pcmk__list_of_multiple(list)) { pcmk__config_err("Cannot promote non-colocated child %s", instance->id); } g_list_free(list); if (chosen == NULL) { return; } next_role = instance->private->fns->state(instance, FALSE); switch (next_role) { case pcmk_role_started: case pcmk_role_unknown: // Set instance priority to its promotion score (or -1 if none) { bool is_default = false; instance->private->priority = promotion_score(instance, chosen, &is_default); if (is_default) { /* Default to -1 if no value is set. This allows instances * eligible for promotion to be specified based solely on * PCMK_XE_RSC_LOCATION constraints, but prevents any * instance from being promoted if neither a constraint nor * a promotion score is present. */ instance->private->priority = -1; } } break; case pcmk_role_unpromoted: case pcmk_role_stopped: // Instance can't be promoted instance->private->priority = -PCMK_SCORE_INFINITY; break; case pcmk_role_promoted: // Nothing needed (re-creating actions after scheduling fencing) break; default: CRM_CHECK(FALSE, crm_err("Unknown resource role %d for %s", next_role, instance->id)); } // Add relevant location constraint scores for promoted role apply_promoted_locations(instance, instance->private->location_constraints, chosen); apply_promoted_locations(instance, clone->private->location_constraints, chosen); // Consider instance's role-based colocations with other resources list = pcmk__this_with_colocations(instance); for (GList *iter = list; iter != NULL; iter = iter->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) iter->data; instance->private->cmds->apply_coloc_score(instance, cons->primary, cons, true); } g_list_free(list); instance->private->promotion_priority = instance->private->priority; if (next_role == pcmk_role_promoted) { instance->private->promotion_priority = PCMK_SCORE_INFINITY; } pcmk__rsc_trace(clone, "Assigning %s priority = %d", instance->id, instance->private->priority); } /*! * \internal * \brief Set a promotable clone instance's role * * \param[in,out] data Promotable clone instance to update * \param[in,out] user_data Pointer to count of instances chosen for promotion */ static void set_instance_role(gpointer data, gpointer user_data) { pcmk_resource_t *instance = (pcmk_resource_t *) data; int *count = (int *) user_data; const pcmk_resource_t *clone = pe__const_top_resource(instance, false); const pcmk_scheduler_t *scheduler = instance->private->scheduler; pcmk_node_t *chosen = NULL; show_promotion_score(instance); if (instance->private->promotion_priority < 0) { pcmk__rsc_trace(clone, "Not supposed to promote instance %s", instance->id); } else if ((*count < pe__clone_promoted_max(instance)) || !pcmk_is_set(clone->flags, pcmk__rsc_managed)) { chosen = node_to_be_promoted_on(instance); } if (chosen == NULL) { set_next_role_unpromoted(instance, NULL); return; } if ((instance->role < pcmk_role_promoted) && !pcmk_is_set(scheduler->flags, pcmk_sched_quorate) && (scheduler->no_quorum_policy == pcmk_no_quorum_freeze)) { crm_notice("Clone instance %s cannot be promoted without quorum", instance->id); set_next_role_unpromoted(instance, NULL); return; } chosen->count++; pcmk__rsc_info(clone, "Choosing %s (%s) on %s for promotion", instance->id, pcmk_role_text(instance->role), pcmk__node_name(chosen)); set_next_role_promoted(instance, NULL); (*count)++; } /*! * \internal * \brief Set roles for all instances of a promotable clone * * \param[in,out] rsc Promotable clone resource to update */ void pcmk__set_instance_roles(pcmk_resource_t *rsc) { int promoted = 0; GHashTableIter iter; pcmk_node_t *node = NULL; // Repurpose count to track the number of promoted instances assigned g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { node->count = 0; } // Set instances' promotion priorities and sort by highest priority first g_list_foreach(rsc->children, set_instance_priority, rsc); sort_promotable_instances(rsc); // Choose the first N eligible instances to be promoted g_list_foreach(rsc->children, set_instance_role, &promoted); pcmk__rsc_info(rsc, "%s: Promoted %d instances of a possible %d", rsc->id, promoted, pe__clone_promoted_max(rsc)); } /*! * * \internal * \brief Create actions for promotable clone instances * * \param[in,out] clone Promotable clone to create actions for * \param[out] any_promoting Will be set true if any instance is promoting * \param[out] any_demoting Will be set true if any instance is demoting */ static void create_promotable_instance_actions(pcmk_resource_t *clone, bool *any_promoting, bool *any_demoting) { for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; instance->private->cmds->create_actions(instance); check_for_role_change(instance, any_demoting, any_promoting); } } /*! * \internal * \brief Reset each promotable instance's resource priority * * Reset the priority of each instance of a promotable clone to the clone's * priority (after promotion actions are scheduled, when instance priorities * were repurposed as promotion scores). * * \param[in,out] clone Promotable clone to reset */ static void reset_instance_priorities(pcmk_resource_t *clone) { for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; instance->private->priority = clone->private->priority; } } /*! * \internal * \brief Create actions specific to promotable clones * * \param[in,out] clone Promotable clone to create actions for */ void pcmk__create_promotable_actions(pcmk_resource_t *clone) { bool any_promoting = false; bool any_demoting = false; // Create actions for each clone instance individually create_promotable_instance_actions(clone, &any_promoting, &any_demoting); // Create pseudo-actions for clone as a whole pe__create_promotable_pseudo_ops(clone, any_promoting, any_demoting); // Undo our temporary repurposing of resource priority for instances reset_instance_priorities(clone); } /*! * \internal * \brief Create internal orderings for a promotable clone's instances * * \param[in,out] clone Promotable clone instance to order */ void pcmk__order_promotable_instances(pcmk_resource_t *clone) { pcmk_resource_t *previous = NULL; // Needed for ordered clones pcmk__promotable_restart_ordering(clone); for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; // Demote before promote pcmk__order_resource_actions(instance, PCMK_ACTION_DEMOTE, instance, PCMK_ACTION_PROMOTE, pcmk__ar_ordered); order_instance_promotion(clone, instance, previous); order_instance_demotion(clone, instance, previous); previous = instance; } } /*! * \internal * \brief Update dependent's allowed nodes for colocation with promotable * * \param[in,out] dependent Dependent resource to update * \param[in] primary Primary resource * \param[in] primary_node Node where an instance of the primary will be * \param[in] colocation Colocation constraint to apply */ static void update_dependent_allowed_nodes(pcmk_resource_t *dependent, const pcmk_resource_t *primary, const pcmk_node_t *primary_node, const pcmk__colocation_t *colocation) { GHashTableIter iter; pcmk_node_t *node = NULL; const char *primary_value = NULL; const char *attr = colocation->node_attribute; if (colocation->score >= PCMK_SCORE_INFINITY) { return; // Colocation is mandatory, so allowed node scores don't matter } primary_value = pcmk__colocation_node_attr(primary_node, attr, primary); pcmk__rsc_trace(colocation->primary, "Applying %s (%s with %s on %s by %s @%d) to %s", colocation->id, colocation->dependent->id, colocation->primary->id, pcmk__node_name(primary_node), attr, colocation->score, dependent->id); g_hash_table_iter_init(&iter, dependent->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { const char *dependent_value = pcmk__colocation_node_attr(node, attr, dependent); if (pcmk__str_eq(primary_value, dependent_value, pcmk__str_casei)) { node->weight = pcmk__add_scores(node->weight, colocation->score); pcmk__rsc_trace(colocation->primary, "Added %s score (%s) to %s (now %s)", colocation->id, pcmk_readable_score(colocation->score), pcmk__node_name(node), pcmk_readable_score(node->weight)); } } } /*! * \brief Update dependent for a colocation with a promotable clone * * \param[in] primary Primary resource in the colocation * \param[in,out] dependent Dependent resource in the colocation * \param[in] colocation Colocation constraint to apply */ void pcmk__update_dependent_with_promotable(const pcmk_resource_t *primary, pcmk_resource_t *dependent, const pcmk__colocation_t *colocation) { GList *affected_nodes = NULL; /* Build a list of all nodes where an instance of the primary will be, and * (for optional colocations) update the dependent's allowed node scores for * each one. */ for (GList *iter = primary->children; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; pcmk_node_t *node = instance->private->fns->location(instance, NULL, FALSE); if (node == NULL) { continue; } if (instance->private->fns->state(instance, FALSE) == colocation->primary_role) { update_dependent_allowed_nodes(dependent, primary, node, colocation); affected_nodes = g_list_prepend(affected_nodes, node); } } /* For mandatory colocations, add the primary's node score to the * dependent's node score for each affected node, and ban the dependent * from all other nodes. * * However, skip this for promoted-with-promoted colocations, otherwise * inactive dependent instances can't start (in the unpromoted role). */ if ((colocation->score >= PCMK_SCORE_INFINITY) && ((colocation->dependent_role != pcmk_role_promoted) || (colocation->primary_role != pcmk_role_promoted))) { pcmk__rsc_trace(colocation->primary, "Applying %s (mandatory %s with %s) to %s", colocation->id, colocation->dependent->id, colocation->primary->id, dependent->id); pcmk__colocation_intersect_nodes(dependent, primary, colocation, affected_nodes, true); } g_list_free(affected_nodes); } /*! * \internal * \brief Update dependent priority for colocation with promotable * * \param[in] primary Primary resource in the colocation * \param[in,out] dependent Dependent resource in the colocation * \param[in] colocation Colocation constraint to apply */ void pcmk__update_promotable_dependent_priority(const pcmk_resource_t *primary, pcmk_resource_t *dependent, const pcmk__colocation_t *colocation) { pcmk_resource_t *primary_instance = NULL; // Look for a primary instance where dependent will be primary_instance = pcmk__find_compatible_instance(dependent, primary, colocation->primary_role, false); if (primary_instance != NULL) { // Add primary instance's priority to dependent's int new_priority = pcmk__add_scores(dependent->private->priority, colocation->score); pcmk__rsc_trace(colocation->primary, "Applying %s (%s with %s) to %s priority " "(%s + %s = %s)", colocation->id, colocation->dependent->id, colocation->primary->id, dependent->id, pcmk_readable_score(dependent->private->priority), pcmk_readable_score(colocation->score), pcmk_readable_score(new_priority)); dependent->private->priority = new_priority; } else if (colocation->score >= PCMK_SCORE_INFINITY) { // Mandatory colocation, but primary won't be here pcmk__rsc_trace(colocation->primary, "Applying %s (%s with %s) to %s: can't be promoted", colocation->id, colocation->dependent->id, colocation->primary->id, dependent->id); dependent->private->priority = -PCMK_SCORE_INFINITY; } } diff --git a/lib/pacemaker/pcmk_sched_remote.c b/lib/pacemaker/pcmk_sched_remote.c index 0ecb2fc642..070280ab90 100644 --- a/lib/pacemaker/pcmk_sched_remote.c +++ b/lib/pacemaker/pcmk_sched_remote.c @@ -1,739 +1,739 @@ /* * 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->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) + } else if (pcmk__list_of_multiple(remote_rsc->private->active_nodes) && (remote_rsc->private->partial_migration_source != NULL) && (remote_rsc->private->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; pcmk_resource_t *remote = NULL; if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_none) || !pcmk__is_pacemaker_remote_node(action->node)) { return NULL; } remote = action->node->details->remote_rsc; CRM_ASSERT(remote != NULL); began_on = pcmk__current_node(remote); ended_on = remote->private->assigned_node; if ((remote->container == NULL) && (remote->private->partial_migration_target != NULL)) { 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(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 = 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 61c1e5bd51..91a9c90c35 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->private->assigned_node; - if (rsc->running_on) { + if (rsc->private->active_nodes != NULL) { 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->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->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->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->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) { + if (resource1->private->active_nodes != NULL) { r1_node = pcmk__current_node(resource1); } - if (resource2->running_on != NULL) { + if (resource2->private->active_nodes != 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/pacemaker/pcmk_sched_tickets.c b/lib/pacemaker/pcmk_sched_tickets.c index 81327f4c95..536305da4b 100644 --- a/lib/pacemaker/pcmk_sched_tickets.c +++ b/lib/pacemaker/pcmk_sched_tickets.c @@ -1,543 +1,545 @@ /* * 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" enum loss_ticket_policy { loss_ticket_stop, loss_ticket_demote, loss_ticket_fence, loss_ticket_freeze }; typedef struct { const char *id; pcmk_resource_t *rsc; pcmk_ticket_t *ticket; enum loss_ticket_policy loss_policy; int role; } rsc_ticket_t; /*! * \brief Check whether a ticket constraint matches a resource by role * * \param[in] rsc_ticket Ticket constraint * \param[in] rsc Resource to compare with ticket * * \param[in] true if constraint has no role or resource's role matches * constraint's, otherwise false */ static bool ticket_role_matches(const pcmk_resource_t *rsc, const rsc_ticket_t *rsc_ticket) { if ((rsc_ticket->role == pcmk_role_unknown) || (rsc_ticket->role == rsc->role)) { return true; } pcmk__rsc_trace(rsc, "Skipping constraint: \"%s\" state filter", pcmk_role_text(rsc_ticket->role)); return false; } /*! * \brief Create location constraints and fencing as needed for a ticket * * \param[in,out] rsc Resource affected by ticket * \param[in] rsc_ticket Ticket */ static void constraints_for_ticket(pcmk_resource_t *rsc, const rsc_ticket_t *rsc_ticket) { GList *iter = NULL; CRM_CHECK((rsc != NULL) && (rsc_ticket != NULL), return); if (rsc_ticket->ticket->granted && !rsc_ticket->ticket->standby) { return; } if (rsc->children) { pcmk__rsc_trace(rsc, "Processing ticket dependencies from %s", rsc->id); for (iter = rsc->children; iter != NULL; iter = iter->next) { constraints_for_ticket((pcmk_resource_t *) iter->data, rsc_ticket); } return; } pcmk__rsc_trace(rsc, "%s: Processing ticket dependency on %s (%s, %s)", rsc->id, rsc_ticket->ticket->id, rsc_ticket->id, pcmk_role_text(rsc_ticket->role)); - if (!rsc_ticket->ticket->granted && (rsc->running_on != NULL)) { + if (!rsc_ticket->ticket->granted && (rsc->private->active_nodes != NULL)) { switch (rsc_ticket->loss_policy) { case loss_ticket_stop: resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__loss_of_ticket__", rsc->private->scheduler); break; case loss_ticket_demote: // Promotion score will be set to -INFINITY in promotion_order() if (rsc_ticket->role != pcmk_role_promoted) { resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__loss_of_ticket__", rsc->private->scheduler); } break; case loss_ticket_fence: if (!ticket_role_matches(rsc, rsc_ticket)) { return; } resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__loss_of_ticket__", rsc->private->scheduler); - for (iter = rsc->running_on; iter != NULL; iter = iter->next) { + for (iter = rsc->private->active_nodes; + iter != NULL; iter = iter->next) { + pe_fence_node(rsc->private->scheduler, (pcmk_node_t *) iter->data, "deadman ticket was lost", FALSE); } break; case loss_ticket_freeze: if (!ticket_role_matches(rsc, rsc_ticket)) { return; } - if (rsc->running_on != NULL) { + if (rsc->private->active_nodes != NULL) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(rsc, pcmk__rsc_blocked); } break; } } else if (!rsc_ticket->ticket->granted) { if ((rsc_ticket->role != pcmk_role_promoted) || (rsc_ticket->loss_policy == loss_ticket_stop)) { resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__no_ticket__", rsc->private->scheduler); } } else if (rsc_ticket->ticket->standby) { if ((rsc_ticket->role != pcmk_role_promoted) || (rsc_ticket->loss_policy == loss_ticket_stop)) { resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__ticket_standby__", rsc->private->scheduler); } } } static void rsc_ticket_new(const char *id, pcmk_resource_t *rsc, pcmk_ticket_t *ticket, const char *state, const char *loss_policy) { rsc_ticket_t *new_rsc_ticket = NULL; if (rsc == NULL) { pcmk__config_err("Ignoring ticket '%s' because resource " "does not exist", id); return; } new_rsc_ticket = calloc(1, sizeof(rsc_ticket_t)); if (new_rsc_ticket == NULL) { return; } if (pcmk__str_eq(state, PCMK_ROLE_STARTED, pcmk__str_null_matches|pcmk__str_casei)) { state = PCMK__ROLE_UNKNOWN; } new_rsc_ticket->id = id; new_rsc_ticket->ticket = ticket; new_rsc_ticket->rsc = rsc; new_rsc_ticket->role = pcmk_parse_role(state); if (pcmk__str_eq(loss_policy, PCMK_VALUE_FENCE, pcmk__str_casei)) { if (pcmk_is_set(rsc->private->scheduler->flags, pcmk_sched_fencing_enabled)) { new_rsc_ticket->loss_policy = loss_ticket_fence; } else { pcmk__config_err("Resetting '" PCMK_XA_LOSS_POLICY "' " "for ticket '%s' to '" PCMK_VALUE_STOP "' " "because fencing is not configured", ticket->id); loss_policy = PCMK_VALUE_STOP; } } if (new_rsc_ticket->loss_policy == loss_ticket_fence) { crm_debug("On loss of ticket '%s': Fence the nodes running %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); } else if (pcmk__str_eq(loss_policy, PCMK_VALUE_FREEZE, pcmk__str_casei)) { crm_debug("On loss of ticket '%s': Freeze %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); new_rsc_ticket->loss_policy = loss_ticket_freeze; } else if (pcmk__str_eq(loss_policy, PCMK_VALUE_DEMOTE, pcmk__str_casei)) { crm_debug("On loss of ticket '%s': Demote %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); new_rsc_ticket->loss_policy = loss_ticket_demote; } else if (pcmk__str_eq(loss_policy, PCMK_VALUE_STOP, pcmk__str_casei)) { crm_debug("On loss of ticket '%s': Stop %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); new_rsc_ticket->loss_policy = loss_ticket_stop; } else { if (new_rsc_ticket->role == pcmk_role_promoted) { crm_debug("On loss of ticket '%s': Default to demote %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); new_rsc_ticket->loss_policy = loss_ticket_demote; } else { crm_debug("On loss of ticket '%s': Default to stop %s (%s)", new_rsc_ticket->ticket->id, new_rsc_ticket->rsc->id, pcmk_role_text(new_rsc_ticket->role)); new_rsc_ticket->loss_policy = loss_ticket_stop; } } pcmk__rsc_trace(rsc, "%s (%s) ==> %s", rsc->id, pcmk_role_text(new_rsc_ticket->role), ticket->id); rsc->private->ticket_constraints = g_list_append(rsc->private->ticket_constraints, new_rsc_ticket); rsc->private->scheduler->ticket_constraints = g_list_append(rsc->private->scheduler->ticket_constraints, new_rsc_ticket); if (!(new_rsc_ticket->ticket->granted) || new_rsc_ticket->ticket->standby) { constraints_for_ticket(rsc, new_rsc_ticket); } } // \return Standard Pacemaker return code static int unpack_rsc_ticket_set(xmlNode *set, pcmk_ticket_t *ticket, const char *loss_policy, pcmk_scheduler_t *scheduler) { const char *set_id = NULL; const char *role = NULL; CRM_CHECK(set != NULL, return EINVAL); CRM_CHECK(ticket != NULL, return EINVAL); set_id = pcmk__xe_id(set); if (set_id == NULL) { pcmk__config_err("Ignoring <" PCMK_XE_RESOURCE_SET "> without " PCMK_XA_ID); return pcmk_rc_unpack_error; } role = crm_element_value(set, PCMK_XA_ROLE); for (xmlNode *xml_rsc = pcmk__xe_first_child(set, PCMK_XE_RESOURCE_REF, NULL, NULL); xml_rsc != NULL; xml_rsc = pcmk__xe_next_same(xml_rsc)) { pcmk_resource_t *resource = NULL; resource = pcmk__find_constraint_resource(scheduler->resources, pcmk__xe_id(xml_rsc)); if (resource == NULL) { pcmk__config_err("%s: No resource found for %s", set_id, pcmk__xe_id(xml_rsc)); return pcmk_rc_unpack_error; } pcmk__rsc_trace(resource, "Resource '%s' depends on ticket '%s'", resource->id, ticket->id); rsc_ticket_new(set_id, resource, ticket, role, loss_policy); } return pcmk_rc_ok; } static void unpack_simple_rsc_ticket(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *ticket_str = crm_element_value(xml_obj, PCMK_XA_TICKET); const char *loss_policy = crm_element_value(xml_obj, PCMK_XA_LOSS_POLICY); pcmk_ticket_t *ticket = NULL; const char *rsc_id = crm_element_value(xml_obj, PCMK_XA_RSC); const char *state = crm_element_value(xml_obj, PCMK_XA_RSC_ROLE); // @COMPAT: Deprecated since 2.1.5 const char *instance = crm_element_value(xml_obj, PCMK__XA_RSC_INSTANCE); pcmk_resource_t *rsc = NULL; if (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"); } CRM_CHECK(xml_obj != NULL, return); id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("Ignoring <%s> constraint without " PCMK_XA_ID, xml_obj->name); return; } if (ticket_str == NULL) { pcmk__config_err("Ignoring constraint '%s' without ticket specified", id); return; } else { ticket = g_hash_table_lookup(scheduler->tickets, ticket_str); } if (ticket == NULL) { pcmk__config_err("Ignoring constraint '%s' because ticket '%s' " "does not exist", id, ticket_str); return; } if (rsc_id == NULL) { pcmk__config_err("Ignoring constraint '%s' without resource", id); return; } else { rsc = pcmk__find_constraint_resource(scheduler->resources, rsc_id); } if (rsc == NULL) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "does not exist", id, rsc_id); return; } else if ((instance != NULL) && !pcmk__is_clone(rsc)) { pcmk__config_err("Ignoring constraint '%s' because resource '%s' " "is not a clone but instance '%s' was requested", id, rsc_id, instance); return; } if (instance != NULL) { rsc = find_clone_instance(rsc, instance); if (rsc == NULL) { pcmk__config_warn("Ignoring constraint '%s' because resource '%s' " "does not have an instance '%s'", "'%s'", id, rsc_id, instance); return; } } rsc_ticket_new(id, rsc, ticket, state, loss_policy); } // \return Standard Pacemaker return code static int unpack_rsc_ticket_tags(xmlNode *xml_obj, xmlNode **expanded_xml, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *rsc_id = NULL; const char *state = NULL; pcmk_resource_t *rsc = NULL; pcmk_tag_t *tag = NULL; xmlNode *rsc_set = NULL; *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 rsc_ticket"); return pcmk_rc_ok; } rsc_id = crm_element_value(xml_obj, PCMK_XA_RSC); if (rsc_id == NULL) { return pcmk_rc_ok; } if (!pcmk__valid_resource_or_tag(scheduler, rsc_id, &rsc, &tag)) { pcmk__config_err("Ignoring constraint '%s' because '%s' is not a " "valid resource or tag", id, rsc_id); return pcmk_rc_unpack_error; } else if (rsc != NULL) { // No template or tag is referenced return pcmk_rc_ok; } state = crm_element_value(xml_obj, PCMK_XA_RSC_ROLE); *expanded_xml = pcmk__xml_copy(NULL, xml_obj); /* Convert any template or tag reference in "rsc" into ticket * PCMK_XE_RESOURCE_SET */ if (!pcmk__tag_to_set(*expanded_xml, &rsc_set, PCMK_XA_RSC, false, scheduler)) { pcmk__xml_free(*expanded_xml); *expanded_xml = NULL; return pcmk_rc_unpack_error; } if (rsc_set != NULL) { if (state != NULL) { /* Move PCMK_XA_RSC_ROLE into converted PCMK_XE_RESOURCE_SET as a * PCMK_XA_ROLE attribute */ crm_xml_add(rsc_set, PCMK_XA_ROLE, state); pcmk__xe_remove_attr(*expanded_xml, PCMK_XA_RSC_ROLE); } } else { pcmk__xml_free(*expanded_xml); *expanded_xml = NULL; } return pcmk_rc_ok; } void pcmk__unpack_rsc_ticket(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { xmlNode *set = NULL; bool any_sets = false; const char *id = NULL; const char *ticket_str = NULL; pcmk_ticket_t *ticket = NULL; xmlNode *orig_xml = NULL; xmlNode *expanded_xml = NULL; CRM_CHECK(xml_obj != NULL, return); id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("Ignoring <%s> constraint without " PCMK_XA_ID, xml_obj->name); return; } if (scheduler->tickets == NULL) { scheduler->tickets = pcmk__strkey_table(free, destroy_ticket); } ticket_str = crm_element_value(xml_obj, PCMK_XA_TICKET); if (ticket_str == NULL) { pcmk__config_err("Ignoring constraint '%s' without ticket", id); return; } else { ticket = g_hash_table_lookup(scheduler->tickets, ticket_str); } if (ticket == NULL) { ticket = ticket_new(ticket_str, scheduler); if (ticket == NULL) { return; } } if (unpack_rsc_ticket_tags(xml_obj, &expanded_xml, scheduler) != pcmk_rc_ok) { return; } if (expanded_xml != NULL) { orig_xml = xml_obj; xml_obj = expanded_xml; } for (set = pcmk__xe_first_child(xml_obj, PCMK_XE_RESOURCE_SET, NULL, NULL); set != NULL; set = pcmk__xe_next_same(set)) { const char *loss_policy = NULL; any_sets = true; set = pcmk__xe_resolve_idref(set, scheduler->input); loss_policy = crm_element_value(xml_obj, PCMK_XA_LOSS_POLICY); if ((set == NULL) // Configuration error, message already logged || (unpack_rsc_ticket_set(set, ticket, loss_policy, scheduler) != pcmk_rc_ok)) { if (expanded_xml != NULL) { pcmk__xml_free(expanded_xml); } return; } } if (expanded_xml) { pcmk__xml_free(expanded_xml); xml_obj = orig_xml; } if (!any_sets) { unpack_simple_rsc_ticket(xml_obj, scheduler); } } /*! * \internal * \brief Ban resource from a node if it doesn't have a promotion ticket * * If a resource has tickets for the promoted role, and the ticket is either not * granted or set to standby, then ban the resource from all nodes. * * \param[in,out] rsc Resource to check */ void pcmk__require_promotion_tickets(pcmk_resource_t *rsc) { for (GList *item = rsc->private->ticket_constraints; item != NULL; item = item->next) { rsc_ticket_t *rsc_ticket = (rsc_ticket_t *) item->data; if ((rsc_ticket->role == pcmk_role_promoted) && (!rsc_ticket->ticket->granted || rsc_ticket->ticket->standby)) { resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__stateful_without_ticket__", rsc->private->scheduler); } } } diff --git a/lib/pacemaker/pcmk_sched_utilization.c b/lib/pacemaker/pcmk_sched_utilization.c index 19436821ff..368bc8f9b5 100644 --- a/lib/pacemaker/pcmk_sched_utilization.c +++ b/lib/pacemaker/pcmk_sched_utilization.c @@ -1,474 +1,474 @@ /* * 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 "libpacemaker_private.h" /*! * \internal * \brief Get integer utilization from a string * * \param[in] s String representation of a node utilization value * * \return Integer equivalent of \p s * \todo It would make sense to restrict utilization values to nonnegative * integers, but the documentation just says "integers" and we didn't * restrict them initially, so for backward compatibility, allow any * integer. */ static int utilization_value(const char *s) { int value = 0; if ((s != NULL) && (pcmk__scan_min_int(s, &value, INT_MIN) == EINVAL)) { pcmk__config_warn("Using 0 for utilization instead of " "invalid value '%s'", value); value = 0; } return value; } /* * Functions for comparing node capacities */ struct compare_data { const pcmk_node_t *node1; const pcmk_node_t *node2; bool node2_only; int result; }; /*! * \internal * \brief Compare a single utilization attribute for two nodes * * Compare one utilization attribute for two nodes, decrementing the result if * the first node has greater capacity, and incrementing it if the second node * has greater capacity. * * \param[in] key Utilization attribute name to compare * \param[in] value Utilization attribute value to compare * \param[in,out] user_data Comparison data (as struct compare_data*) */ static void compare_utilization_value(gpointer key, gpointer value, gpointer user_data) { int node1_capacity = 0; int node2_capacity = 0; struct compare_data *data = user_data; const char *node2_value = NULL; if (data->node2_only) { if (g_hash_table_lookup(data->node1->details->utilization, key)) { return; // We've already compared this attribute } } else { node1_capacity = utilization_value((const char *) value); } node2_value = g_hash_table_lookup(data->node2->details->utilization, key); node2_capacity = utilization_value(node2_value); if (node1_capacity > node2_capacity) { data->result--; } else if (node1_capacity < node2_capacity) { data->result++; } } /*! * \internal * \brief Compare utilization capacities of two nodes * * \param[in] node1 First node to compare * \param[in] node2 Second node to compare * * \return Negative integer if node1 has more free capacity, * 0 if the capacities are equal, or a positive integer * if node2 has more free capacity */ int pcmk__compare_node_capacities(const pcmk_node_t *node1, const pcmk_node_t *node2) { struct compare_data data = { .node1 = node1, .node2 = node2, .node2_only = false, .result = 0, }; // Compare utilization values that node1 and maybe node2 have g_hash_table_foreach(node1->details->utilization, compare_utilization_value, &data); // Compare utilization values that only node2 has data.node2_only = true; g_hash_table_foreach(node2->details->utilization, compare_utilization_value, &data); return data.result; } /* * Functions for updating node capacities */ struct calculate_data { GHashTable *current_utilization; bool plus; }; /*! * \internal * \brief Update a single utilization attribute with a new value * * \param[in] key Name of utilization attribute to update * \param[in] value Value to add or substract * \param[in,out] user_data Calculation data (as struct calculate_data *) */ static void update_utilization_value(gpointer key, gpointer value, gpointer user_data) { int result = 0; const char *current = NULL; struct calculate_data *data = user_data; current = g_hash_table_lookup(data->current_utilization, key); if (data->plus) { result = utilization_value(current) + utilization_value(value); } else if (current) { result = utilization_value(current) - utilization_value(value); } g_hash_table_replace(data->current_utilization, strdup(key), pcmk__itoa(result)); } /*! * \internal * \brief Subtract a resource's utilization from node capacity * * \param[in,out] current_utilization Current node utilization attributes * \param[in] rsc Resource with utilization to subtract */ void pcmk__consume_node_capacity(GHashTable *current_utilization, const pcmk_resource_t *rsc) { struct calculate_data data = { .current_utilization = current_utilization, .plus = false, }; g_hash_table_foreach(rsc->utilization, update_utilization_value, &data); } /*! * \internal * \brief Add a resource's utilization to node capacity * * \param[in,out] current_utilization Current node utilization attributes * \param[in] rsc Resource with utilization to add */ void pcmk__release_node_capacity(GHashTable *current_utilization, const pcmk_resource_t *rsc) { struct calculate_data data = { .current_utilization = current_utilization, .plus = true, }; g_hash_table_foreach(rsc->utilization, update_utilization_value, &data); } /* * Functions for checking for sufficient node capacity */ struct capacity_data { const pcmk_node_t *node; const char *rsc_id; bool is_enough; }; /*! * \internal * \brief Check whether a single utilization attribute has sufficient capacity * * \param[in] key Name of utilization attribute to check * \param[in] value Amount of utilization required * \param[in,out] user_data Capacity data (as struct capacity_data *) */ static void check_capacity(gpointer key, gpointer value, gpointer user_data) { int required = 0; int remaining = 0; const char *node_value_s = NULL; struct capacity_data *data = user_data; node_value_s = g_hash_table_lookup(data->node->details->utilization, key); required = utilization_value(value); remaining = utilization_value(node_value_s); if (required > remaining) { crm_debug("Remaining capacity for %s on %s (%d) is insufficient " "for resource %s usage (%d)", (const char *) key, pcmk__node_name(data->node), remaining, data->rsc_id, required); data->is_enough = false; } } /*! * \internal * \brief Check whether a node has sufficient capacity for a resource * * \param[in] node Node to check * \param[in] rsc_id ID of resource to check (for debug logs only) * \param[in] utilization Required utilization amounts * * \return true if node has sufficient capacity for resource, otherwise false */ static bool have_enough_capacity(const pcmk_node_t *node, const char *rsc_id, GHashTable *utilization) { struct capacity_data data = { .node = node, .rsc_id = rsc_id, .is_enough = true, }; g_hash_table_foreach(utilization, check_capacity, &data); return data.is_enough; } /*! * \internal * \brief Sum the utilization requirements of a list of resources * * \param[in] orig_rsc Resource being assigned (for logging purposes) * \param[in] rscs Resources whose utilization should be summed * * \return Newly allocated hash table with sum of all utilization values * \note It is the caller's responsibility to free the return value using * g_hash_table_destroy(). */ static GHashTable * sum_resource_utilization(const pcmk_resource_t *orig_rsc, GList *rscs) { GHashTable *utilization = pcmk__strkey_table(free, free); for (GList *iter = rscs; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; rsc->private->cmds->add_utilization(rsc, orig_rsc, rscs, utilization); } return utilization; } /*! * \internal * \brief Ban resource from nodes with insufficient utilization capacity * * \param[in,out] rsc Resource to check * * \return Allowed node for \p rsc with most spare capacity, if there are no * nodes with enough capacity for \p rsc and all its colocated resources */ const pcmk_node_t * pcmk__ban_insufficient_capacity(pcmk_resource_t *rsc) { bool any_capable = false; char *rscs_id = NULL; pcmk_node_t *node = NULL; const pcmk_node_t *most_capable_node = NULL; GList *colocated_rscs = NULL; GHashTable *unassigned_utilization = NULL; GHashTableIter iter; CRM_CHECK(rsc != NULL, return NULL); // The default placement strategy ignores utilization if (pcmk__str_eq(rsc->private->scheduler->placement_strategy, PCMK_VALUE_DEFAULT, pcmk__str_casei)) { return NULL; } // Check whether any resources are colocated with this one colocated_rscs = rsc->private->cmds->colocated_resources(rsc, NULL, NULL); if (colocated_rscs == NULL) { return NULL; } rscs_id = crm_strdup_printf("%s and its colocated resources", rsc->id); // If rsc isn't in the list, add it so we include its utilization if (g_list_find(colocated_rscs, rsc) == NULL) { colocated_rscs = g_list_append(colocated_rscs, rsc); } // Sum utilization of colocated resources that haven't been assigned yet unassigned_utilization = sum_resource_utilization(rsc, colocated_rscs); // Check whether any node has enough capacity for all the resources g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (!pcmk__node_available(node, true, false)) { continue; } if (have_enough_capacity(node, rscs_id, unassigned_utilization)) { any_capable = true; } // Keep track of node with most free capacity if ((most_capable_node == NULL) || (pcmk__compare_node_capacities(node, most_capable_node) < 0)) { most_capable_node = node; } } if (any_capable) { // If so, ban resource from any node with insufficient capacity g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (pcmk__node_available(node, true, false) && !have_enough_capacity(node, rscs_id, unassigned_utilization)) { pcmk__rsc_debug(rsc, "%s does not have enough capacity for %s", pcmk__node_name(node), rscs_id); resource_location(rsc, node, -PCMK_SCORE_INFINITY, "__limit_utilization__", rsc->private->scheduler); } } most_capable_node = NULL; } else { // Otherwise, ban from nodes with insufficient capacity for rsc alone g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (pcmk__node_available(node, true, false) && !have_enough_capacity(node, rsc->id, rsc->utilization)) { pcmk__rsc_debug(rsc, "%s does not have enough capacity for %s", pcmk__node_name(node), rsc->id); resource_location(rsc, node, -PCMK_SCORE_INFINITY, "__limit_utilization__", rsc->private->scheduler); } } } g_hash_table_destroy(unassigned_utilization); g_list_free(colocated_rscs); free(rscs_id); pe__show_node_scores(true, rsc, "Post-utilization", rsc->allowed_nodes, rsc->private->scheduler); return most_capable_node; } /*! * \internal * \brief Create a new load_stopped pseudo-op for a node * * \param[in,out] node Node to create op for * * \return Newly created load_stopped op */ static pcmk_action_t * new_load_stopped_op(pcmk_node_t *node) { char *load_stopped_task = crm_strdup_printf(PCMK_ACTION_LOAD_STOPPED "_%s", node->details->uname); pcmk_action_t *load_stopped = get_pseudo_op(load_stopped_task, node->details->data_set); if (load_stopped->node == NULL) { load_stopped->node = pe__copy_node(node); pcmk__clear_action_flags(load_stopped, pcmk_action_optional); } free(load_stopped_task); return load_stopped; } /*! * \internal * \brief Create utilization-related internal constraints for a resource * * \param[in,out] rsc Resource to create constraints for * \param[in] allowed_nodes List of allowed next nodes for \p rsc */ void pcmk__create_utilization_constraints(pcmk_resource_t *rsc, const GList *allowed_nodes) { const GList *iter = NULL; pcmk_action_t *load_stopped = NULL; pcmk__rsc_trace(rsc, "Creating utilization constraints for %s - strategy: %s", rsc->id, rsc->private->scheduler->placement_strategy); // "stop rsc then load_stopped" constraints for current nodes - for (iter = rsc->running_on; iter != NULL; iter = iter->next) { + for (iter = rsc->private->active_nodes; iter != NULL; iter = iter->next) { load_stopped = new_load_stopped_op(iter->data); pcmk__new_ordering(rsc, stop_key(rsc), NULL, NULL, NULL, load_stopped, pcmk__ar_if_on_same_node_or_target, rsc->private->scheduler); } // "load_stopped then start/migrate_to rsc" constraints for allowed nodes for (iter = allowed_nodes; iter; iter = iter->next) { load_stopped = new_load_stopped_op(iter->data); pcmk__new_ordering(NULL, NULL, load_stopped, rsc, start_key(rsc), NULL, pcmk__ar_if_on_same_node_or_target, rsc->private->scheduler); pcmk__new_ordering(NULL, NULL, load_stopped, rsc, pcmk__op_key(rsc->id, PCMK_ACTION_MIGRATE_TO, 0), NULL, pcmk__ar_if_on_same_node_or_target, rsc->private->scheduler); } } /*! * \internal * \brief Output node capacities if enabled * * \param[in] desc Prefix for output * \param[in,out] scheduler Scheduler data */ void pcmk__show_node_capacities(const char *desc, pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_show_utilization)) { return; } for (const GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { const pcmk_node_t *node = (const pcmk_node_t *) iter->data; pcmk__output_t *out = scheduler->priv; out->message(out, "node-capacity", node, desc); } } diff --git a/lib/pacemaker/pcmk_scheduler.c b/lib/pacemaker/pcmk_scheduler.c index ab4d86bc0a..bc7cae7ed7 100644 --- a/lib/pacemaker/pcmk_scheduler.c +++ b/lib/pacemaker/pcmk_scheduler.c @@ -1,900 +1,900 @@ /* * 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 #include "libpacemaker_private.h" CRM_TRACE_INIT_DATA(pacemaker); /*! * \internal * \brief Do deferred action checks after assignment * * When unpacking the resource history, the scheduler checks for resource * configurations that have changed since an action was run. However, at that * time, bundles using the REMOTE_CONTAINER_HACK don't have their final * parameter information, so instead they add a deferred check to a list. This * function processes one entry in that list. * * \param[in,out] rsc Resource that action history is for * \param[in,out] node Node that action history is for * \param[in] rsc_op Action history entry * \param[in] check Type of deferred check to do */ static void check_params(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *rsc_op, enum pcmk__check_parameters check) { const char *reason = NULL; pcmk__op_digest_t *digest_data = NULL; switch (check) { case pcmk__check_active: if (pcmk__check_action_config(rsc, node, rsc_op) && pe_get_failcount(node, rsc, NULL, pcmk__fc_effective, NULL)) { reason = "action definition changed"; } break; case pcmk__check_last_failure: digest_data = rsc_action_digest_cmp(rsc, rsc_op, node, rsc->private->scheduler); switch (digest_data->rc) { case pcmk__digest_unknown: crm_trace("Resource %s history entry %s on %s has " "no digest to compare", rsc->id, pcmk__xe_id(rsc_op), node->details->id); break; case pcmk__digest_match: break; default: reason = "resource parameters have changed"; break; } break; } if (reason != NULL) { pe__clear_failcount(rsc, node, reason, rsc->private->scheduler); } } /*! * \internal * \brief Check whether a resource has failcount clearing scheduled on a node * * \param[in] node Node to check * \param[in] rsc Resource to check * * \return true if \p rsc has failcount clearing scheduled on \p node, * otherwise false */ static bool failcount_clear_action_exists(const pcmk_node_t *node, const pcmk_resource_t *rsc) { GList *list = pe__resource_actions(rsc, node, PCMK_ACTION_CLEAR_FAILCOUNT, TRUE); if (list != NULL) { g_list_free(list); return true; } return false; } /*! * \internal * \brief Ban a resource from a node if it reached its failure threshold there * * \param[in,out] data Resource to check failure threshold for * \param[in] user_data Node to check resource on */ static void check_failure_threshold(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; const pcmk_node_t *node = user_data; // If this is a collective resource, apply recursively to children instead if (rsc->children != NULL) { g_list_foreach(rsc->children, check_failure_threshold, user_data); return; } if (!failcount_clear_action_exists(node, rsc)) { /* Don't force the resource away from this node due to a failcount * that's going to be cleared. * * @TODO Failcount clearing can be scheduled in * pcmk__handle_rsc_config_changes() via process_rsc_history(), or in * schedule_resource_actions() via check_params(). This runs well before * then, so it cannot detect those, meaning we might check the migration * threshold when we shouldn't. Worst case, we stop or move the * resource, then move it back in the next transition. */ pcmk_resource_t *failed = NULL; if (pcmk__threshold_reached(rsc, node, &failed)) { resource_location(failed, node, -PCMK_SCORE_INFINITY, "__fail_limit__", rsc->private->scheduler); } } } /*! * \internal * \brief If resource has exclusive discovery, ban node if not allowed * * Location constraints have a PCMK_XA_RESOURCE_DISCOVERY option that allows * users to specify where probes are done for the affected resource. If this is * set to \c exclusive, probes will only be done on nodes listed in exclusive * constraints. This function bans the resource from the node if the node is not * listed. * * \param[in,out] data Resource to check * \param[in] user_data Node to check resource on */ static void apply_exclusive_discovery(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; const pcmk_node_t *node = user_data; if (pcmk_is_set(rsc->flags, pcmk__rsc_exclusive_probes) || pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_exclusive_probes)) { pcmk_node_t *match = NULL; // If this is a collective resource, apply recursively to children g_list_foreach(rsc->children, apply_exclusive_discovery, user_data); match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if ((match != NULL) && (match->rsc_discover_mode != pcmk_probe_exclusive)) { match->weight = -PCMK_SCORE_INFINITY; } } } /*! * \internal * \brief Apply stickiness to a resource if appropriate * * \param[in,out] data Resource to check for stickiness * \param[in] user_data Ignored */ static void apply_stickiness(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; pcmk_node_t *node = NULL; // If this is a collective resource, apply recursively to children instead if (rsc->children != NULL) { g_list_foreach(rsc->children, apply_stickiness, NULL); return; } /* A resource is sticky if it is managed, has stickiness configured, and is * active on a single node. */ if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed) || (rsc->private->stickiness < 1) - || !pcmk__list_of_1(rsc->running_on)) { + || !pcmk__list_of_1(rsc->private->active_nodes)) { return; } - node = rsc->running_on->data; + node = rsc->private->active_nodes->data; /* In a symmetric cluster, stickiness can always be used. In an * asymmetric cluster, we have to check whether the resource is still * allowed on the node, so we don't keep the resource somewhere it is no * longer explicitly enabled. */ if (!pcmk_is_set(rsc->private->scheduler->flags, pcmk_sched_symmetric_cluster) && (g_hash_table_lookup(rsc->allowed_nodes, node->details->id) == NULL)) { pcmk__rsc_debug(rsc, "Ignoring %s stickiness because the cluster is " "asymmetric and %s is not explicitly allowed", rsc->id, pcmk__node_name(node)); return; } pcmk__rsc_debug(rsc, "Resource %s has %d stickiness on %s", rsc->id, rsc->private->stickiness, pcmk__node_name(node)); resource_location(rsc, node, rsc->private->stickiness, "stickiness", rsc->private->scheduler); } /*! * \internal * \brief Apply shutdown locks for all resources as appropriate * * \param[in,out] scheduler Scheduler data */ static void apply_shutdown_locks(pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { return; } for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; rsc->private->cmds->shutdown_lock(rsc); } } /*! * \internal * \brief Calculate the number of available nodes in the cluster * * \param[in,out] scheduler Scheduler data */ static void count_available_nodes(pcmk_scheduler_t *scheduler) { if (pcmk_is_set(scheduler->flags, pcmk_sched_no_compat)) { return; } // @COMPAT for API backward compatibility only (cluster does not use value) for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; if ((node != NULL) && (node->weight >= 0) && node->details->online && (node->details->type != node_ping)) { scheduler->max_valid_nodes++; } } crm_trace("Online node count: %d", scheduler->max_valid_nodes); } /* * \internal * \brief Apply node-specific scheduling criteria * * After the CIB has been unpacked, process node-specific scheduling criteria * including shutdown locks, location constraints, resource stickiness, * migration thresholds, and exclusive resource discovery. */ static void apply_node_criteria(pcmk_scheduler_t *scheduler) { crm_trace("Applying node-specific scheduling criteria"); apply_shutdown_locks(scheduler); count_available_nodes(scheduler); pcmk__apply_locations(scheduler); g_list_foreach(scheduler->resources, apply_stickiness, NULL); for (GList *node_iter = scheduler->nodes; node_iter != NULL; node_iter = node_iter->next) { for (GList *rsc_iter = scheduler->resources; rsc_iter != NULL; rsc_iter = rsc_iter->next) { check_failure_threshold(rsc_iter->data, node_iter->data); apply_exclusive_discovery(rsc_iter->data, node_iter->data); } } } /*! * \internal * \brief Assign resources to nodes * * \param[in,out] scheduler Scheduler data */ static void assign_resources(pcmk_scheduler_t *scheduler) { GList *iter = NULL; crm_trace("Assigning resources to nodes"); if (!pcmk__str_eq(scheduler->placement_strategy, PCMK_VALUE_DEFAULT, pcmk__str_casei)) { pcmk__sort_resources(scheduler); } pcmk__show_node_capacities("Original", scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_have_remote_nodes)) { /* Assign remote connection resources first (which will also assign any * colocation dependencies). If the connection is migrating, always * prefer the partial migration target. */ for (iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; const pcmk_node_t *target = rsc->private->partial_migration_target; if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { pcmk__rsc_trace(rsc, "Assigning remote connection resource '%s'", rsc->id); rsc->private->cmds->assign(rsc, target, true); } } } /* now do the rest of the resources */ for (iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; if (!pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { pcmk__rsc_trace(rsc, "Assigning %s resource '%s'", rsc->private->xml->name, rsc->id); rsc->private->cmds->assign(rsc, NULL, true); } } pcmk__show_node_capacities("Remaining", scheduler); } /*! * \internal * \brief Schedule fail count clearing on online nodes if resource is orphaned * * \param[in,out] data Resource to check * \param[in] user_data Ignored */ static void clear_failcounts_if_orphaned(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { return; } crm_trace("Clear fail counts for orphaned resource %s", rsc->id); /* There's no need to recurse into rsc->children because those * should just be unassigned clone instances. */ for (GList *iter = rsc->private->scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; pcmk_action_t *clear_op = NULL; if (!node->details->online) { continue; } if (pe_get_failcount(node, rsc, NULL, pcmk__fc_effective, NULL) == 0) { continue; } clear_op = pe__clear_failcount(rsc, node, "it is orphaned", rsc->private->scheduler); /* We can't use order_action_then_stop() here because its * pcmk__ar_guest_allowed breaks things */ pcmk__new_ordering(clear_op->rsc, NULL, clear_op, rsc, stop_key(rsc), NULL, pcmk__ar_ordered, rsc->private->scheduler); } } /*! * \internal * \brief Schedule any resource actions needed * * \param[in,out] scheduler Scheduler data */ static void schedule_resource_actions(pcmk_scheduler_t *scheduler) { // Process deferred action checks pe__foreach_param_check(scheduler, check_params); pe__free_param_checks(scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_probe_resources)) { crm_trace("Scheduling probes"); pcmk__schedule_probes(scheduler); } if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { g_list_foreach(scheduler->resources, clear_failcounts_if_orphaned, NULL); } crm_trace("Scheduling resource actions"); for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data; rsc->private->cmds->create_actions(rsc); } } /*! * \internal * \brief Check whether a resource or any of its descendants are managed * * \param[in] rsc Resource to check * * \return true if resource or any descendant is managed, otherwise false */ static bool is_managed(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { return true; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { if (is_managed((pcmk_resource_t *) iter->data)) { return true; } } return false; } /*! * \internal * \brief Check whether any resources in the cluster are managed * * \param[in] scheduler Scheduler data * * \return true if any resource is managed, otherwise false */ static bool any_managed_resources(const pcmk_scheduler_t *scheduler) { for (const GList *iter = scheduler->resources; iter != NULL; iter = iter->next) { if (is_managed((const pcmk_resource_t *) iter->data)) { return true; } } return false; } /*! * \internal * \brief Check whether a node requires fencing * * \param[in] node Node to check * \param[in] have_managed Whether any resource in cluster is managed * * \return true if \p node should be fenced, otherwise false */ static bool needs_fencing(const pcmk_node_t *node, bool have_managed) { return have_managed && node->details->unclean && pe_can_fence(node->details->data_set, node); } /*! * \internal * \brief Check whether a node requires shutdown * * \param[in] node Node to check * * \return true if \p node should be shut down, otherwise false */ static bool needs_shutdown(const pcmk_node_t *node) { if (pcmk__is_pacemaker_remote_node(node)) { /* Do not send shutdown actions for Pacemaker Remote nodes. * @TODO We might come up with a good use for this in the future. */ return false; } return node->details->online && node->details->shutdown; } /*! * \internal * \brief Track and order non-DC fencing * * \param[in,out] list List of existing non-DC fencing actions * \param[in,out] action Fencing action to prepend to \p list * \param[in] scheduler Scheduler data * * \return (Possibly new) head of \p list */ static GList * add_nondc_fencing(GList *list, pcmk_action_t *action, const pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing) && (list != NULL)) { /* Concurrent fencing is disabled, so order each non-DC * fencing in a chain. If there is any DC fencing or * shutdown, it will be ordered after the last action in the * chain later. */ order_actions((pcmk_action_t *) list->data, action, pcmk__ar_ordered); } return g_list_prepend(list, action); } /*! * \internal * \brief Schedule a node for fencing * * \param[in,out] node Node that requires fencing */ static pcmk_action_t * schedule_fencing(pcmk_node_t *node) { pcmk_action_t *fencing = pe_fence_op(node, NULL, FALSE, "node is unclean", FALSE, node->details->data_set); pcmk__sched_warn("Scheduling node %s for fencing", pcmk__node_name(node)); pcmk__order_vs_fence(fencing, node->details->data_set); return fencing; } /*! * \internal * \brief Create and order node fencing and shutdown actions * * \param[in,out] scheduler Scheduler data */ static void schedule_fencing_and_shutdowns(pcmk_scheduler_t *scheduler) { pcmk_action_t *dc_down = NULL; bool integrity_lost = false; bool have_managed = any_managed_resources(scheduler); GList *fencing_ops = NULL; GList *shutdown_ops = NULL; crm_trace("Scheduling fencing and shutdowns as needed"); if (!have_managed) { crm_notice("No fencing will be done until there are resources " "to manage"); } // Check each node for whether it needs fencing or shutdown for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; pcmk_action_t *fencing = NULL; /* Guest nodes are "fenced" by recovering their container resource, * so handle them separately. */ if (pcmk__is_guest_or_bundle_node(node)) { if (node->details->remote_requires_reset && have_managed && pe_can_fence(scheduler, node)) { pcmk__fence_guest(node); } continue; } if (needs_fencing(node, have_managed)) { fencing = schedule_fencing(node); // Track DC and non-DC fence actions separately if (node->details->is_dc) { dc_down = fencing; } else { fencing_ops = add_nondc_fencing(fencing_ops, fencing, scheduler); } } else if (needs_shutdown(node)) { pcmk_action_t *down_op = pcmk__new_shutdown_action(node); // Track DC and non-DC shutdown actions separately if (node->details->is_dc) { dc_down = down_op; } else { shutdown_ops = g_list_prepend(shutdown_ops, down_op); } } if ((fencing == NULL) && node->details->unclean) { integrity_lost = true; pcmk__config_warn("Node %s is unclean but cannot be fenced", pcmk__node_name(node)); } } if (integrity_lost) { if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("Resource functionality and data integrity " "cannot be guaranteed (configure, enable, " "and test fencing to correct this)"); } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) { crm_notice("Unclean nodes will not be fenced until quorum is " "attained or " PCMK_OPT_NO_QUORUM_POLICY " is set to " PCMK_VALUE_IGNORE); } } if (dc_down != NULL) { /* Order any non-DC shutdowns before any DC shutdown, to avoid repeated * DC elections. However, we don't want to order non-DC shutdowns before * a DC *fencing*, because even though we don't want a node that's * shutting down to become DC, the DC fencing could be ordered before a * clone stop that's also ordered before the shutdowns, thus leading to * a graph loop. */ if (pcmk__str_eq(dc_down->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { pcmk__order_after_each(dc_down, shutdown_ops); } // Order any non-DC fencing before any DC fencing or shutdown if (pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing)) { /* With concurrent fencing, order each non-DC fencing action * separately before any DC fencing or shutdown. */ pcmk__order_after_each(dc_down, fencing_ops); } else if (fencing_ops != NULL) { /* Without concurrent fencing, the non-DC fencing actions are * already ordered relative to each other, so we just need to order * the DC fencing after the last action in the chain (which is the * first item in the list). */ order_actions((pcmk_action_t *) fencing_ops->data, dc_down, pcmk__ar_ordered); } } g_list_free(fencing_ops); g_list_free(shutdown_ops); } static void log_resource_details(pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv; GList *all = NULL; /* Due to the `crm_mon --node=` feature, out->message() for all the * resource-related messages expects a list of nodes that we are allowed to * output information for. Here, we create a wildcard to match all nodes. */ all = g_list_prepend(all, (gpointer) "*"); for (GList *item = scheduler->resources; item != NULL; item = item->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) item->data; // Log all resources except inactive orphans if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed) || (rsc->role != pcmk_role_stopped)) { out->message(out, pcmk__map_element_name(rsc->private->xml), 0UL, rsc, all, all); } } g_list_free(all); } static void log_all_actions(pcmk_scheduler_t *scheduler) { /* This only ever outputs to the log, so ignore whatever output object was * previously set and just log instead. */ pcmk__output_t *prev_out = scheduler->priv; pcmk__output_t *out = NULL; if (pcmk__log_output_new(&out) != pcmk_rc_ok) { return; } pe__register_messages(out); pcmk__register_lib_messages(out); pcmk__output_set_log_level(out, LOG_NOTICE); scheduler->priv = out; out->begin_list(out, NULL, NULL, "Actions"); pcmk__output_actions(scheduler); out->end_list(out); out->finish(out, CRM_EX_OK, true, NULL); pcmk__output_free(out); scheduler->priv = prev_out; } /*! * \internal * \brief Log all required but unrunnable actions at trace level * * \param[in] scheduler Scheduler data */ static void log_unrunnable_actions(const pcmk_scheduler_t *scheduler) { const uint64_t flags = pcmk_action_optional |pcmk_action_runnable |pcmk_action_pseudo; crm_trace("Required but unrunnable actions:"); for (const GList *iter = scheduler->actions; iter != NULL; iter = iter->next) { const pcmk_action_t *action = (const pcmk_action_t *) iter->data; if (!pcmk_any_flags_set(action->flags, flags)) { pcmk__log_action("\t", action, true); } } } /*! * \internal * \brief Unpack the CIB for scheduling * * \param[in,out] cib CIB XML to unpack (may be NULL if already unpacked) * \param[in] flags Scheduler flags to set in addition to defaults * \param[in,out] scheduler Scheduler data */ static void unpack_cib(xmlNode *cib, unsigned long long flags, pcmk_scheduler_t *scheduler) { const char* localhost_save = NULL; if (pcmk_is_set(scheduler->flags, pcmk_sched_have_status)) { crm_trace("Reusing previously calculated cluster status"); pcmk__set_scheduler_flags(scheduler, flags); return; } if (scheduler->localhost) { localhost_save = scheduler->localhost; } CRM_ASSERT(cib != NULL); crm_trace("Calculating cluster status"); /* This will zero the entire struct without freeing anything first, so * callers should never call pcmk__schedule_actions() with a populated data * set unless pcmk_sched_have_status is set (i.e. cluster_status() was * previously called, whether directly or via pcmk__schedule_actions()). */ set_working_set_defaults(scheduler); if (localhost_save) { scheduler->localhost = localhost_save; } pcmk__set_scheduler_flags(scheduler, flags); scheduler->input = cib; cluster_status(scheduler); // Sets pcmk_sched_have_status } /*! * \internal * \brief Run the scheduler for a given CIB * * \param[in,out] cib CIB XML to use as scheduler input * \param[in] flags Scheduler flags to set in addition to defaults * \param[in,out] scheduler Scheduler data */ void pcmk__schedule_actions(xmlNode *cib, unsigned long long flags, pcmk_scheduler_t *scheduler) { unpack_cib(cib, flags, scheduler); pcmk__set_assignment_methods(scheduler); pcmk__apply_node_health(scheduler); pcmk__unpack_constraints(scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_validate_only)) { return; } if (!pcmk_is_set(scheduler->flags, pcmk_sched_location_only) && pcmk__is_daemon) { log_resource_details(scheduler); } apply_node_criteria(scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { return; } pcmk__create_internal_constraints(scheduler); pcmk__handle_rsc_config_changes(scheduler); assign_resources(scheduler); schedule_resource_actions(scheduler); /* Remote ordering constraints need to happen prior to calculating fencing * because it is one more place we can mark nodes as needing fencing. */ pcmk__order_remote_connection_actions(scheduler); schedule_fencing_and_shutdowns(scheduler); pcmk__apply_orderings(scheduler); log_all_actions(scheduler); pcmk__create_graph(scheduler); if (get_crm_log_level() == LOG_TRACE) { log_unrunnable_actions(scheduler); } } /*! * \internal * \brief Initialize scheduler data * * Make our own copies of the CIB XML and date/time object, if they're not * \c NULL. This way we don't have to take ownership of the objects passed via * the API. * * This function is most useful for public API functions that want the caller * to retain ownership of the CIB object * * \param[in,out] out Output object * \param[in] input The CIB XML to check (if \c NULL, use current CIB) * \param[in] date Date and time to use in the scheduler (if \c NULL, * use current date and time). This can be used for * checking whether a rule is in effect at a certa * date and time. * \param[out] scheduler Where to store initialized scheduler data * * \return Standard Pacemaker return code */ int pcmk__init_scheduler(pcmk__output_t *out, xmlNodePtr input, const crm_time_t *date, pcmk_scheduler_t **scheduler) { // Allows for cleaner syntax than dereferencing the scheduler argument pcmk_scheduler_t *new_scheduler = NULL; new_scheduler = pe_new_working_set(); if (new_scheduler == NULL) { return ENOMEM; } pcmk__set_scheduler_flags(new_scheduler, pcmk_sched_no_counts|pcmk_sched_no_compat); // Populate the scheduler data // Make our own copy of the given input or fetch the CIB and use that if (input != NULL) { new_scheduler->input = pcmk__xml_copy(NULL, input); if (new_scheduler->input == NULL) { out->err(out, "Failed to copy input XML"); pe_free_working_set(new_scheduler); return ENOMEM; } } else { int rc = cib__signon_query(out, NULL, &(new_scheduler->input)); if (rc != pcmk_rc_ok) { pe_free_working_set(new_scheduler); return rc; } } // Make our own copy of the given crm_time_t object; otherwise // cluster_status() populates with the current time if (date != NULL) { // pcmk_copy_time() guarantees non-NULL new_scheduler->now = pcmk_copy_time(date); } // Unpack everything cluster_status(new_scheduler); *scheduler = new_scheduler; return pcmk_rc_ok; } diff --git a/lib/pengine/bundle.c b/lib/pengine/bundle.c index dd25cfbf69..0c6268e116 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->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) { + if (replica->container->private->active_nodes != 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->private->active_nodes; + node_iter != NULL; node_iter = node_iter->next) { - 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/clone.c b/lib/pengine/clone.c index 2368d83af3..559bdfe9d8 100644 --- a/lib/pengine/clone.c +++ b/lib/pengine/clone.c @@ -1,1261 +1,1261 @@ /* * 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 typedef struct clone_variant_data_s { int clone_max; int clone_node_max; int promoted_max; int promoted_node_max; int total_clones; uint32_t flags; // Group of enum pcmk__clone_flags notify_data_t *stop_notify; notify_data_t *start_notify; notify_data_t *demote_notify; notify_data_t *promote_notify; xmlNode *xml_obj_child; } clone_variant_data_t; #define get_clone_variant_data(data, rsc) do { \ CRM_ASSERT(pcmk__is_clone(rsc)); \ data = rsc->private->variant_opaque; \ } while (0) /*! * \internal * \brief Return the maximum number of clone instances allowed to be run * * \param[in] clone Clone or clone instance to check * * \return Maximum instances for \p clone */ int pe__clone_max(const pcmk_resource_t *clone) { const clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, pe__const_top_resource(clone, false)); return clone_data->clone_max; } /*! * \internal * \brief Return the maximum number of clone instances allowed per node * * \param[in] clone Promotable clone or clone instance to check * * \return Maximum allowed instances per node for \p clone */ int pe__clone_node_max(const pcmk_resource_t *clone) { const clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, pe__const_top_resource(clone, false)); return clone_data->clone_node_max; } /*! * \internal * \brief Return the maximum number of clone instances allowed to be promoted * * \param[in] clone Promotable clone or clone instance to check * * \return Maximum promoted instances for \p clone */ int pe__clone_promoted_max(const pcmk_resource_t *clone) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, pe__const_top_resource(clone, false)); return clone_data->promoted_max; } /*! * \internal * \brief Return the maximum number of clone instances allowed to be promoted * * \param[in] clone Promotable clone or clone instance to check * * \return Maximum promoted instances for \p clone */ int pe__clone_promoted_node_max(const pcmk_resource_t *clone) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, pe__const_top_resource(clone, false)); return clone_data->promoted_node_max; } static GList * sorted_hash_table_values(GHashTable *table) { GList *retval = NULL; GHashTableIter iter; gpointer key, value; g_hash_table_iter_init(&iter, table); while (g_hash_table_iter_next(&iter, &key, &value)) { if (!g_list_find_custom(retval, value, (GCompareFunc) strcmp)) { retval = g_list_prepend(retval, (char *) value); } } retval = g_list_sort(retval, (GCompareFunc) strcmp); return retval; } static GList * nodes_with_status(GHashTable *table, const char *status) { GList *retval = NULL; GHashTableIter iter; gpointer key, value; g_hash_table_iter_init(&iter, table); while (g_hash_table_iter_next(&iter, &key, &value)) { if (!strcmp((char *) value, status)) { retval = g_list_prepend(retval, key); } } retval = g_list_sort(retval, (GCompareFunc) pcmk__numeric_strcasecmp); return retval; } static GString * node_list_to_str(const GList *list) { GString *retval = NULL; for (const GList *iter = list; iter != NULL; iter = iter->next) { pcmk__add_word(&retval, 1024, (const char *) iter->data); } return retval; } static void clone_header(pcmk__output_t *out, int *rc, const pcmk_resource_t *rsc, clone_variant_data_t *clone_data, const char *desc) { GString *attrs = NULL; if (pcmk_is_set(rsc->flags, pcmk__rsc_promotable)) { pcmk__add_separated_word(&attrs, 64, "promotable", ", "); } if (pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { pcmk__add_separated_word(&attrs, 64, "unique", ", "); } if (pe__resource_is_disabled(rsc)) { pcmk__add_separated_word(&attrs, 64, "disabled", ", "); } if (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance)) { pcmk__add_separated_word(&attrs, 64, "maintenance", ", "); } else if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__add_separated_word(&attrs, 64, "unmanaged", ", "); } if (attrs != NULL) { PCMK__OUTPUT_LIST_HEADER(out, FALSE, *rc, "Clone Set: %s [%s] (%s)%s%s%s", rsc->id, pcmk__xe_id(clone_data->xml_obj_child), (const char *) attrs->str, desc ? " (" : "", desc ? desc : "", desc ? ")" : ""); g_string_free(attrs, TRUE); } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, *rc, "Clone Set: %s [%s]%s%s%s", rsc->id, pcmk__xe_id(clone_data->xml_obj_child), desc ? " (" : "", desc ? desc : "", desc ? ")" : ""); } } void pe__force_anon(const char *standard, pcmk_resource_t *rsc, const char *rid, pcmk_scheduler_t *scheduler) { if (pcmk__is_clone(rsc)) { clone_variant_data_t *clone_data = rsc->private->variant_opaque; pcmk__config_warn("Ignoring " PCMK_META_GLOBALLY_UNIQUE " for %s " "because %s resources such as %s can be used only as " "anonymous clones", rsc->id, standard, rid); clone_data->clone_node_max = 1; clone_data->clone_max = QB_MIN(clone_data->clone_max, g_list_length(scheduler->nodes)); } } pcmk_resource_t * find_clone_instance(const pcmk_resource_t *rsc, const char *sub_id) { char *child_id = NULL; pcmk_resource_t *child = NULL; const char *child_base = NULL; clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); child_base = pcmk__xe_id(clone_data->xml_obj_child); child_id = crm_strdup_printf("%s:%s", child_base, sub_id); child = pe_find_resource(rsc->children, child_id); free(child_id); return child; } pcmk_resource_t * pe__create_clone_child(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { gboolean as_orphan = FALSE; char *inc_num = NULL; char *inc_max = NULL; pcmk_resource_t *child_rsc = NULL; xmlNode *child_copy = NULL; clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); CRM_CHECK(clone_data->xml_obj_child != NULL, return FALSE); if (clone_data->total_clones >= clone_data->clone_max) { // If we've already used all available instances, this is an orphan as_orphan = TRUE; } // Allocate instance numbers in numerical order (starting at 0) inc_num = pcmk__itoa(clone_data->total_clones); inc_max = pcmk__itoa(clone_data->clone_max); child_copy = pcmk__xml_copy(NULL, clone_data->xml_obj_child); crm_xml_add(child_copy, PCMK__META_CLONE, inc_num); if (pe__unpack_resource(child_copy, &child_rsc, rsc, scheduler) != pcmk_rc_ok) { goto bail; } /* child_rsc->globally_unique = rsc->globally_unique; */ CRM_ASSERT(child_rsc); clone_data->total_clones += 1; pcmk__rsc_trace(child_rsc, "Setting clone attributes for: %s", child_rsc->id); rsc->children = g_list_append(rsc->children, child_rsc); if (as_orphan) { pe__set_resource_flags_recursive(child_rsc, pcmk__rsc_removed); } pcmk__insert_meta(child_rsc, PCMK_META_CLONE_MAX, inc_max); pcmk__rsc_trace(rsc, "Added %s instance %s", rsc->id, child_rsc->id); bail: free(inc_num); free(inc_max); return child_rsc; } /*! * \internal * \brief Unpack a nonnegative integer value from a resource meta-attribute * * \param[in] rsc Resource with meta-attribute * \param[in] meta_name Name of meta-attribute to unpack * \param[in] deprecated_name If not NULL, try unpacking this * if \p meta_name is unset * \param[in] default_value Value to use if unset * * \return Integer parsed from resource's specified meta-attribute if a valid * nonnegative integer, \p default_value if unset, or 0 if invalid */ static int unpack_meta_int(const pcmk_resource_t *rsc, const char *meta_name, const char *deprecated_name, int default_value) { int integer = default_value; const char *value = g_hash_table_lookup(rsc->meta, meta_name); if ((value == NULL) && (deprecated_name != NULL)) { value = g_hash_table_lookup(rsc->meta, deprecated_name); if (value != NULL) { if (pcmk__str_eq(deprecated_name, PCMK__META_PROMOTED_MAX_LEGACY, pcmk__str_none)) { pcmk__warn_once(pcmk__wo_clone_master_max, "Support for the " PCMK__META_PROMOTED_MAX_LEGACY " meta-attribute (such as in %s) is deprecated " "and will be removed in a future release. Use the " PCMK_META_PROMOTED_MAX " meta-attribute instead.", rsc->id); } else if (pcmk__str_eq(deprecated_name, PCMK__META_PROMOTED_NODE_MAX_LEGACY, pcmk__str_none)) { pcmk__warn_once(pcmk__wo_clone_master_node_max, "Support for the " PCMK__META_PROMOTED_NODE_MAX_LEGACY " meta-attribute (such as in %s) is deprecated " "and will be removed in a future release. Use the " PCMK_META_PROMOTED_NODE_MAX " meta-attribute instead.", rsc->id); } } } if (value != NULL) { pcmk__scan_min_int(value, &integer, 0); } return integer; } gboolean clone_unpack(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { int lpc = 0; xmlNode *a_child = NULL; xmlNode *xml_obj = rsc->private->xml; clone_variant_data_t *clone_data = NULL; pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id); clone_data = pcmk__assert_alloc(1, sizeof(clone_variant_data_t)); rsc->private->variant_opaque = clone_data; if (pcmk_is_set(rsc->flags, pcmk__rsc_promotable)) { // Use 1 as default but 0 for minimum and invalid // @COMPAT PCMK__META_PROMOTED_MAX_LEGACY deprecated since 2.0.0 clone_data->promoted_max = unpack_meta_int(rsc, PCMK_META_PROMOTED_MAX, PCMK__META_PROMOTED_MAX_LEGACY, 1); // Use 1 as default but 0 for minimum and invalid // @COMPAT PCMK__META_PROMOTED_NODE_MAX_LEGACY deprecated since 2.0.0 clone_data->promoted_node_max = unpack_meta_int(rsc, PCMK_META_PROMOTED_NODE_MAX, PCMK__META_PROMOTED_NODE_MAX_LEGACY, 1); } // Use 1 as default but 0 for minimum and invalid clone_data->clone_node_max = unpack_meta_int(rsc, PCMK_META_CLONE_NODE_MAX, NULL, 1); /* Use number of nodes (but always at least 1, which is handy for crm_verify * for a CIB without nodes) as default, but 0 for minimum and invalid */ clone_data->clone_max = unpack_meta_int(rsc, PCMK_META_CLONE_MAX, NULL, QB_MAX(1, g_list_length(scheduler->nodes))); if (crm_is_true(g_hash_table_lookup(rsc->meta, PCMK_META_ORDERED))) { clone_data->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Clone", rsc->id, clone_data->flags, pcmk__clone_ordered, "pcmk__clone_ordered"); } if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique) && (clone_data->clone_node_max > 1)) { pcmk__config_err("Ignoring " PCMK_META_CLONE_NODE_MAX " of %d for %s " "because anonymous clones support only one instance " "per node", clone_data->clone_node_max, rsc->id); clone_data->clone_node_max = 1; } pcmk__rsc_trace(rsc, "Options for %s", rsc->id); pcmk__rsc_trace(rsc, "\tClone max: %d", clone_data->clone_max); pcmk__rsc_trace(rsc, "\tClone node max: %d", clone_data->clone_node_max); pcmk__rsc_trace(rsc, "\tClone is unique: %s", pcmk__flag_text(rsc->flags, pcmk__rsc_unique)); pcmk__rsc_trace(rsc, "\tClone is promotable: %s", pcmk__flag_text(rsc->flags, pcmk__rsc_promotable)); // Clones may contain a single group or primitive for (a_child = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL); a_child != NULL; a_child = pcmk__xe_next(a_child)) { if (pcmk__str_any_of((const char *) a_child->name, PCMK_XE_PRIMITIVE, PCMK_XE_GROUP, NULL)) { clone_data->xml_obj_child = a_child; break; } } if (clone_data->xml_obj_child == NULL) { pcmk__config_err("%s has nothing to clone", rsc->id); return FALSE; } /* * Make clones ever so slightly sticky by default * * This helps ensure clone instances are not shuffled around the cluster * for no benefit in situations when pre-allocation is not appropriate */ if (g_hash_table_lookup(rsc->meta, PCMK_META_RESOURCE_STICKINESS) == NULL) { pcmk__insert_meta(rsc, PCMK_META_RESOURCE_STICKINESS, "1"); } /* This ensures that the PCMK_META_GLOBALLY_UNIQUE value always exists for * children to inherit when being unpacked, as well as in resource agents' * environment. */ pcmk__insert_meta(rsc, PCMK_META_GLOBALLY_UNIQUE, pcmk__flag_text(rsc->flags, pcmk__rsc_unique)); if (clone_data->clone_max <= 0) { /* Create one child instance so that unpack_find_resource() will hook up * any orphans up to the parent correctly. */ if (pe__create_clone_child(rsc, scheduler) == NULL) { return FALSE; } } else { // Create a child instance for each available instance number for (lpc = 0; lpc < clone_data->clone_max; lpc++) { if (pe__create_clone_child(rsc, scheduler) == NULL) { return FALSE; } } } pcmk__rsc_trace(rsc, "Added %d children to resource %s...", clone_data->clone_max, rsc->id); return TRUE; } gboolean clone_active(pcmk_resource_t * rsc, gboolean all) { GList *gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; gboolean child_active = child_rsc->private->fns->active(child_rsc, all); if (all == FALSE && child_active) { return TRUE; } else if (all && child_active == FALSE) { return FALSE; } } if (all) { return TRUE; } else { return FALSE; } } static const char * configured_role_str(pcmk_resource_t * rsc) { const char *target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE); if ((target_role == NULL) && rsc->children && rsc->children->data) { pcmk_resource_t *instance = rsc->children->data; // Any instance will do target_role = g_hash_table_lookup(instance->meta, PCMK_META_TARGET_ROLE); } return target_role; } static enum rsc_role_e configured_role(pcmk_resource_t *rsc) { enum rsc_role_e role = pcmk_role_unknown; const char *target_role = configured_role_str(rsc); if (target_role != NULL) { role = pcmk_parse_role(target_role); if (role == pcmk_role_unknown) { pcmk__config_err("Invalid " PCMK_META_TARGET_ROLE " for resource %s", rsc->id); } } return role; } bool is_set_recursive(const pcmk_resource_t *rsc, long long flag, bool any) { GList *gIter; bool all = !any; if (pcmk_is_set(rsc->flags, flag)) { if(any) { return TRUE; } } else if(all) { return FALSE; } for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) { if(is_set_recursive(gIter->data, flag, any)) { if(any) { return TRUE; } } else if(all) { return FALSE; } } if(all) { return TRUE; } return FALSE; } PCMK__OUTPUT_ARGS("clone", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__clone_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 *); GList *gIter = rsc->children; GList *all = NULL; int rc = pcmk_rc_no_output; gboolean printed_header = FALSE; gboolean print_everything = TRUE; if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) || (strstr(rsc->id, ":") != NULL && pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches)); all = g_list_prepend(all, (gpointer) "*"); for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; if (pcmk__rsc_filtered_by_node(child_rsc, only_node)) { continue; } if (child_rsc->private->fns->is_filtered(child_rsc, only_rsc, print_everything)) { continue; } if (!printed_header) { const char *multi_state = pcmk__flag_text(rsc->flags, pcmk__rsc_promotable); 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 *disabled = pcmk__btoa(pe__resource_is_disabled(rsc)); const char *failed = pcmk__flag_text(rsc->flags, pcmk__rsc_failed); const char *ignored = pcmk__flag_text(rsc->flags, pcmk__rsc_ignore_failure); const char *target_role = configured_role_str(rsc); const char *desc = pe__resource_description(rsc, show_opts); printed_header = TRUE; rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_CLONE, PCMK_XA_ID, rsc->id, PCMK_XA_MULTI_STATE, multi_state, PCMK_XA_UNIQUE, unique, PCMK_XA_MAINTENANCE, maintenance, PCMK_XA_MANAGED, managed, PCMK_XA_DISABLED, disabled, PCMK_XA_FAILED, failed, PCMK_XA_FAILURE_IGNORED, ignored, PCMK_XA_TARGET_ROLE, target_role, PCMK_XA_DESCRIPTION, desc, NULL); CRM_ASSERT(rc == pcmk_rc_ok); } out->message(out, (const char *) child_rsc->private->xml->name, show_opts, child_rsc, only_node, all); } if (printed_header) { pcmk__output_xml_pop_parent(out); } g_list_free(all); return rc; } PCMK__OUTPUT_ARGS("clone", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__clone_default(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 *); GHashTable *stopped = NULL; GString *list_text = NULL; GList *promoted_list = NULL; GList *started_list = NULL; GList *gIter = rsc->children; const char *desc = NULL; clone_variant_data_t *clone_data = NULL; int active_instances = 0; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; desc = pe__resource_description(rsc, show_opts); get_clone_variant_data(clone_data, rsc); if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) || (strstr(rsc->id, ":") != NULL && pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches)); for (; gIter != NULL; gIter = gIter->next) { gboolean print_full = FALSE; pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; gboolean partially_active = child_rsc->private->fns->active(child_rsc, FALSE); if (pcmk__rsc_filtered_by_node(child_rsc, only_node)) { continue; } if (child_rsc->private->fns->is_filtered(child_rsc, only_rsc, print_everything)) { continue; } if (pcmk_is_set(show_opts, pcmk_show_clone_detail)) { print_full = TRUE; } if (pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { // Print individual instance when unique (except stopped orphans) if (partially_active || !pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { print_full = TRUE; } // Everything else in this block is for anonymous clones } else if (pcmk_is_set(show_opts, pcmk_show_pending) && (child_rsc->private->pending_action != NULL) && (strcmp(child_rsc->private->pending_action, "probe") != 0)) { // Print individual instance when non-probe action is pending print_full = TRUE; } else if (partially_active == FALSE) { // List stopped instances when requested (except orphans) if (!pcmk_is_set(child_rsc->flags, pcmk__rsc_removed) && !pcmk_is_set(show_opts, pcmk_show_clone_detail) && pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { if (stopped == NULL) { stopped = pcmk__strkey_table(free, free); } pcmk__insert_dup(stopped, child_rsc->id, "Stopped"); } } else if (is_set_recursive(child_rsc, pcmk__rsc_removed, TRUE) || !is_set_recursive(child_rsc, pcmk__rsc_managed, FALSE) || is_set_recursive(child_rsc, pcmk__rsc_failed, TRUE)) { // Print individual instance when active orphaned/unmanaged/failed print_full = TRUE; } else if (child_rsc->private->fns->active(child_rsc, TRUE)) { // Instance of fully active anonymous clone pcmk_node_t *location = NULL; location = child_rsc->private->fns->location(child_rsc, NULL, TRUE); if (location) { // Instance is active on a single node enum rsc_role_e a_role; a_role = child_rsc->private->fns->state(child_rsc, TRUE); if (location->details->online == FALSE && location->details->unclean) { print_full = TRUE; } else if (a_role > pcmk_role_unpromoted) { promoted_list = g_list_append(promoted_list, location); } else { started_list = g_list_append(started_list, location); } } else { /* uncolocated group - bleh */ print_full = TRUE; } } else { // Instance of partially active anonymous clone print_full = TRUE; } if (print_full) { GList *all = NULL; clone_header(out, &rc, rsc, clone_data, desc); /* Print every resource that's a child of this clone. */ all = g_list_prepend(all, (gpointer) "*"); out->message(out, (const char *) child_rsc->private->xml->name, show_opts, child_rsc, only_node, all); g_list_free(all); } } if (pcmk_is_set(show_opts, pcmk_show_clone_detail)) { PCMK__OUTPUT_LIST_FOOTER(out, rc); return pcmk_rc_ok; } /* Promoted */ promoted_list = g_list_sort(promoted_list, pe__cmp_node_name); for (gIter = promoted_list; gIter; gIter = gIter->next) { pcmk_node_t *host = gIter->data; if (!pcmk__str_in_list(host->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { continue; } pcmk__add_word(&list_text, 1024, host->details->uname); active_instances++; } g_list_free(promoted_list); if ((list_text != NULL) && (list_text->len > 0)) { clone_header(out, &rc, rsc, clone_data, desc); out->list_item(out, NULL, PCMK_ROLE_PROMOTED ": [ %s ]", (const char *) list_text->str); g_string_truncate(list_text, 0); } /* Started/Unpromoted */ started_list = g_list_sort(started_list, pe__cmp_node_name); for (gIter = started_list; gIter; gIter = gIter->next) { pcmk_node_t *host = gIter->data; if (!pcmk__str_in_list(host->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { continue; } pcmk__add_word(&list_text, 1024, host->details->uname); active_instances++; } g_list_free(started_list); if ((list_text != NULL) && (list_text->len > 0)) { clone_header(out, &rc, rsc, clone_data, desc); if (pcmk_is_set(rsc->flags, pcmk__rsc_promotable)) { enum rsc_role_e role = configured_role(rsc); if (role == pcmk_role_unpromoted) { out->list_item(out, NULL, PCMK_ROLE_UNPROMOTED " (" PCMK_META_TARGET_ROLE "): [ %s ]", (const char *) list_text->str); } else { out->list_item(out, NULL, PCMK_ROLE_UNPROMOTED ": [ %s ]", (const char *) list_text->str); } } else { out->list_item(out, NULL, "Started: [ %s ]", (const char *) list_text->str); } } if (list_text != NULL) { g_string_free(list_text, TRUE); } if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique) && (clone_data->clone_max > active_instances)) { GList *nIter; GList *list = g_hash_table_get_values(rsc->allowed_nodes); /* Custom stopped table for non-unique clones */ if (stopped != NULL) { g_hash_table_destroy(stopped); stopped = NULL; } if (list == NULL) { /* Clusters with PCMK_OPT_SYMMETRIC_CLUSTER=false haven't * calculated allowed_nodes yet. If we've not probed for them * yet, the Stopped list will be empty. */ list = g_hash_table_get_values(rsc->known_on); } list = g_list_sort(list, pe__cmp_node_name); for (nIter = list; nIter != NULL; nIter = nIter->next) { pcmk_node_t *node = (pcmk_node_t *) nIter->data; - if ((pcmk__find_node_in_list(rsc->running_on, + if ((pcmk__find_node_in_list(rsc->private->active_nodes, node->details->uname) == NULL) && pcmk__str_in_list(node->details->uname, only_node, pcmk__str_star_matches|pcmk__str_casei)) { xmlNode *probe_op = pe__failed_probe_for_rsc(rsc, node->details->uname); const char *state = "Stopped"; if (configured_role(rsc) == pcmk_role_stopped) { state = "Stopped (disabled)"; } if (stopped == NULL) { stopped = pcmk__strkey_table(free, free); } if (probe_op != NULL) { int rc; pcmk__scan_min_int(crm_element_value(probe_op, PCMK__XA_RC_CODE), &rc, 0); g_hash_table_insert(stopped, strdup(node->details->uname), crm_strdup_printf("Stopped (%s)", services_ocf_exitcode_str(rc))); } else { pcmk__insert_dup(stopped, node->details->uname, state); } } } g_list_free(list); } if (stopped != NULL) { GList *list = sorted_hash_table_values(stopped); clone_header(out, &rc, rsc, clone_data, desc); for (GList *status_iter = list; status_iter != NULL; status_iter = status_iter->next) { const char *status = status_iter->data; GList *nodes = nodes_with_status(stopped, status); GString *nodes_str = node_list_to_str(nodes); if (nodes_str != NULL) { if (nodes_str->len > 0) { out->list_item(out, NULL, "%s: [ %s ]", status, (const char *) nodes_str->str); } g_string_free(nodes_str, TRUE); } g_list_free(nodes); } g_list_free(list); g_hash_table_destroy(stopped); /* If there are no instances of this clone (perhaps because there are no * nodes configured), simply output the clone header by itself. This can * come up in PCS testing. */ } else if (active_instances == 0) { clone_header(out, &rc, rsc, clone_data, desc); PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } void clone_free(pcmk_resource_t * rsc) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); pcmk__rsc_trace(rsc, "Freeing %s", rsc->id); for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; CRM_ASSERT(child_rsc); pcmk__rsc_trace(child_rsc, "Freeing child %s", child_rsc->id); pcmk__xml_free(child_rsc->private->xml); child_rsc->private->xml = NULL; /* There could be a saved unexpanded xml */ pcmk__xml_free(child_rsc->private->orig_xml); child_rsc->private->orig_xml = NULL; child_rsc->private->fns->free(child_rsc); } g_list_free(rsc->children); if (clone_data) { CRM_ASSERT(clone_data->demote_notify == NULL); CRM_ASSERT(clone_data->stop_notify == NULL); CRM_ASSERT(clone_data->start_notify == NULL); CRM_ASSERT(clone_data->promote_notify == NULL); } common_free(rsc); } enum rsc_role_e clone_resource_state(const pcmk_resource_t * rsc, gboolean current) { enum rsc_role_e clone_role = pcmk_role_unknown; GList *gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; enum rsc_role_e a_role = child_rsc->private->fns->state(child_rsc, current); if (a_role > clone_role) { clone_role = a_role; } } pcmk__rsc_trace(rsc, "%s role: %s", rsc->id, pcmk_role_text(clone_role)); return clone_role; } /*! * \internal * \brief Check whether a clone has an instance for every node * * \param[in] rsc Clone to check * \param[in] scheduler Scheduler data */ bool pe__is_universal_clone(const pcmk_resource_t *rsc, const pcmk_scheduler_t *scheduler) { if (pcmk__is_clone(rsc)) { clone_variant_data_t *clone_data = rsc->private->variant_opaque; if (clone_data->clone_max == g_list_length(scheduler->nodes)) { return TRUE; } } return FALSE; } gboolean pe__clone_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent) { gboolean passes = FALSE; clone_variant_data_t *clone_data = NULL; if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches)) { passes = TRUE; } else { get_clone_variant_data(clone_data, rsc); passes = pcmk__str_in_list(pcmk__xe_id(clone_data->xml_obj_child), only_rsc, pcmk__str_star_matches); if (!passes) { for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *child_rsc = NULL; child_rsc = (const pcmk_resource_t *) iter->data; if (!child_rsc->private->fns->is_filtered(child_rsc, only_rsc, FALSE)) { passes = TRUE; break; } } } } return !passes; } const char * pe__clone_child_id(const pcmk_resource_t *rsc) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); return pcmk__xe_id(clone_data->xml_obj_child); } /*! * \internal * \brief Check whether a clone is ordered * * \param[in] clone Clone resource to check * * \return true if clone is ordered, otherwise false */ bool pe__clone_is_ordered(const pcmk_resource_t *clone) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); return pcmk_is_set(clone_data->flags, pcmk__clone_ordered); } /*! * \internal * \brief Set a clone flag * * \param[in,out] clone Clone resource to set flag for * \param[in] flag Clone flag to set * * \return Standard Pacemaker return code (either pcmk_rc_ok if flag was not * already set or pcmk_rc_already if it was) */ int pe__set_clone_flag(pcmk_resource_t *clone, enum pcmk__clone_flags flag) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); if (pcmk_is_set(clone_data->flags, flag)) { return pcmk_rc_already; } clone_data->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Clone", clone->id, clone_data->flags, flag, "flag"); return pcmk_rc_ok; } /*! * \internal * \brief Check whether a clone flag is set * * \param[in] group Clone resource to check * \param[in] flags Flag or flags to check * * \return \c true if all \p flags are set for \p clone, otherwise \c false */ bool pe__clone_flag_is_set(const pcmk_resource_t *clone, uint32_t flags) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); CRM_ASSERT(clone_data != NULL); return pcmk_all_flags_set(clone_data->flags, flags); } /*! * \internal * \brief Create pseudo-actions needed for promotable clones * * \param[in,out] clone Promotable clone to create actions for * \param[in] any_promoting Whether any instances will be promoted * \param[in] any_demoting Whether any instance will be demoted */ void pe__create_promotable_pseudo_ops(pcmk_resource_t *clone, bool any_promoting, bool any_demoting) { pcmk_action_t *action = NULL; pcmk_action_t *action_complete = NULL; clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); // Create a "promote" action for the clone itself action = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_PROMOTE, !any_promoting, true); // Create a "promoted" action for when all promotions are done action_complete = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_PROMOTED, !any_promoting, true); action_complete->priority = PCMK_SCORE_INFINITY; // Create notification pseudo-actions for promotion if (clone_data->promote_notify == NULL) { clone_data->promote_notify = pe__action_notif_pseudo_ops(clone, PCMK_ACTION_PROMOTE, action, action_complete); } // Create a "demote" action for the clone itself action = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_DEMOTE, !any_demoting, true); // Create a "demoted" action for when all demotions are done action_complete = pe__new_rsc_pseudo_action(clone, PCMK_ACTION_DEMOTED, !any_demoting, true); action_complete->priority = PCMK_SCORE_INFINITY; // Create notification pseudo-actions for demotion if (clone_data->demote_notify == NULL) { clone_data->demote_notify = pe__action_notif_pseudo_ops(clone, PCMK_ACTION_DEMOTE, action, action_complete); if (clone_data->promote_notify != NULL) { order_actions(clone_data->stop_notify->post_done, clone_data->promote_notify->pre, pcmk__ar_ordered); order_actions(clone_data->start_notify->post_done, clone_data->promote_notify->pre, pcmk__ar_ordered); order_actions(clone_data->demote_notify->post_done, clone_data->promote_notify->pre, pcmk__ar_ordered); order_actions(clone_data->demote_notify->post_done, clone_data->start_notify->pre, pcmk__ar_ordered); order_actions(clone_data->demote_notify->post_done, clone_data->stop_notify->pre, pcmk__ar_ordered); } } } /*! * \internal * \brief Create all notification data and actions for a clone * * \param[in,out] clone Clone to create notifications for */ void pe__create_clone_notifications(pcmk_resource_t *clone) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); pe__create_action_notifications(clone, clone_data->start_notify); pe__create_action_notifications(clone, clone_data->stop_notify); pe__create_action_notifications(clone, clone_data->promote_notify); pe__create_action_notifications(clone, clone_data->demote_notify); } /*! * \internal * \brief Free all notification data for a clone * * \param[in,out] clone Clone to free notification data for */ void pe__free_clone_notification_data(pcmk_resource_t *clone) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); pe__free_action_notification_data(clone_data->demote_notify); clone_data->demote_notify = NULL; pe__free_action_notification_data(clone_data->stop_notify); clone_data->stop_notify = NULL; pe__free_action_notification_data(clone_data->start_notify); clone_data->start_notify = NULL; pe__free_action_notification_data(clone_data->promote_notify); clone_data->promote_notify = NULL; } /*! * \internal * \brief Create pseudo-actions for clone start/stop notifications * * \param[in,out] clone Clone to create pseudo-actions for * \param[in,out] start Start action for \p clone * \param[in,out] stop Stop action for \p clone * \param[in,out] started Started action for \p clone * \param[in,out] stopped Stopped action for \p clone */ void pe__create_clone_notif_pseudo_ops(pcmk_resource_t *clone, pcmk_action_t *start, pcmk_action_t *started, pcmk_action_t *stop, pcmk_action_t *stopped) { clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, clone); if (clone_data->start_notify == NULL) { clone_data->start_notify = pe__action_notif_pseudo_ops(clone, PCMK_ACTION_START, start, started); } if (clone_data->stop_notify == NULL) { clone_data->stop_notify = pe__action_notif_pseudo_ops(clone, PCMK_ACTION_STOP, stop, stopped); if ((clone_data->start_notify != NULL) && (clone_data->stop_notify != NULL)) { order_actions(clone_data->stop_notify->post_done, clone_data->start_notify->pre, pcmk__ar_ordered); } } } /*! * \internal * \brief Get maximum clone resource instances per node * * \param[in] rsc Clone resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int pe__clone_max_per_node(const pcmk_resource_t *rsc) { const clone_variant_data_t *clone_data = NULL; get_clone_variant_data(clone_data, rsc); return clone_data->clone_node_max; } diff --git a/lib/pengine/complex.c b/lib/pengine/complex.c index 0ede841b82..633e379a66 100644 --- a/lib/pengine/complex.c +++ b/lib/pengine/complex.c @@ -1,1268 +1,1267 @@ /* * 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->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->active_nodes); 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->private->partial_migration_source != NULL) { if (pcmk__same_node(node, rsc->private->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) { + for (GList *iter = rsc->private->active_nodes; + 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 b8c9da7c2b..aafb51001f 100644 --- a/lib/pengine/native.c +++ b/lib/pengine/native.c @@ -1,1170 +1,1179 @@ /* * 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; + const pcmk_resource_t *container = node->details->remote_rsc->container; + + for (GList *gIter = container->private->active_nodes; + gIter != NULL; gIter = gIter->next) { - 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) { + + for (GList *gIter = rsc->private->active_nodes; + 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); + rsc->private->active_nodes = g_list_append(rsc->private->active_nodes, + 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->private->active_nodes = g_list_append(p->private->active_nodes, + 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)) { + && (rsc->private->active_nodes != NULL)) { + + for (GList *iter = rsc->private->active_nodes; + iter != NULL; iter = iter->next) { - 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)) { + && (rsc->private->active_nodes == NULL)) { return true; } else if (!pcmk_is_set(flags, pcmk_rsc_match_current_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) { + for (GList *gIter = rsc->private->active_nodes; + 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)) { + || pcmk__list_of_multiple(rsc->private->active_nodes)) { 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)) { + || pcmk__list_of_multiple(rsc->private->active_nodes)) { 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)) { + && pcmk__list_of_multiple(rsc->private->active_nodes)) { bool have_nodes = false; - for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { + for (GList *iter = rsc->private->active_nodes; + 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)) { + } else if (pcmk__is_primitive(rsc) + && (rsc->private->active_nodes == NULL)) { cl = PCMK__VALUE_RSC_FAILED; - } else if (pcmk__list_of_multiple(rsc->running_on)) { + } else if (pcmk__list_of_multiple(rsc->private->active_nodes)) { 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)); + nodes_running_on = pcmk__itoa(g_list_length(rsc->private->active_nodes)); 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 (GList *gIter = rsc->private->active_nodes; + gIter != NULL; gIter = gIter->next) { - 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_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); - } + result = g_list_copy(rsc->private->active_nodes); 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 && (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 (GList *gIter2 = rsc->private->active_nodes; + gIter2 != NULL; gIter2 = gIter2->next) { - 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 e7b3ded4e1..2af702b8fb 100644 --- a/lib/pengine/pe_notif.c +++ b/lib/pengine/pe_notif.c @@ -1,1006 +1,1012 @@ /* * 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; + if (pcmk__is_guest_or_bundle_node(node)) { + const pcmk_resource_t *container = NULL; + + container = node->details->remote_rsc->container; + if (container->private->active_nodes != NULL) { + node = pcmk__current_node(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 + if (rsc->private->active_nodes != NULL) { + node = rsc->private->active_nodes->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) { + for (iter = rsc->private->active_nodes; + 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->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->private->assigned_node, n_data->pre, n_data->pre_done, 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/unpack.c b/lib/pengine/unpack.c index 592efb4787..1dcef46801 100644 --- a/lib/pengine/unpack.c +++ b/lib/pengine/unpack.c @@ -1,5172 +1,5172 @@ /* * 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 #include #include #include CRM_TRACE_INIT_DATA(pe_status); // A (parsed) resource action history entry struct action_history { pcmk_resource_t *rsc; // Resource that history is for pcmk_node_t *node; // Node that history is for xmlNode *xml; // History entry XML // Parsed from entry XML const char *id; // XML ID of history entry const char *key; // Operation key of action const char *task; // Action name const char *exit_reason; // Exit reason given for result guint interval_ms; // Action interval int call_id; // Call ID of action int expected_exit_status; // Expected exit status of action int exit_status; // Actual exit status of action int execution_status; // Execution status of action }; /* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than * use pcmk__set_scheduler_flags()/pcmk__clear_scheduler_flags() so that the * flag is stringified more readably in log messages. */ #define set_config_flag(scheduler, option, flag) do { \ GHashTable *config_hash = (scheduler)->config_hash; \ const char *scf_value = pcmk__cluster_option(config_hash, (option)); \ \ if (scf_value != NULL) { \ if (crm_is_true(scf_value)) { \ (scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", \ crm_system_name, (scheduler)->flags, \ (flag), #flag); \ } else { \ (scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", \ crm_system_name, (scheduler)->flags, \ (flag), #flag); \ } \ } \ } while(0) static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *failed); static void determine_remote_online_status(pcmk_scheduler_t *scheduler, pcmk_node_t *this_node); static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite, pcmk_scheduler_t *scheduler); static void determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node, pcmk_scheduler_t *scheduler); static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml, pcmk_scheduler_t *scheduler); static gboolean is_dangling_guest_node(pcmk_node_t *node) { /* we are looking for a remote-node that was supposed to be mapped to a * container resource, but all traces of that container have disappeared * from both the config and the status section. */ if (pcmk__is_pacemaker_remote_node(node) && (node->details->remote_rsc != NULL) && (node->details->remote_rsc->container == NULL) && pcmk_is_set(node->details->remote_rsc->flags, pcmk__rsc_removed_filler)) { return TRUE; } return FALSE; } /*! * \brief Schedule a fence action for a node * * \param[in,out] scheduler Scheduler data * \param[in,out] node Node to fence * \param[in] reason Text description of why fencing is needed * \param[in] priority_delay Whether to consider * \c PCMK_OPT_PRIORITY_FENCING_DELAY */ void pe_fence_node(pcmk_scheduler_t *scheduler, pcmk_node_t *node, const char *reason, bool priority_delay) { CRM_CHECK(node, return); /* A guest node is fenced by marking its container as failed */ if (pcmk__is_guest_or_bundle_node(node)) { pcmk_resource_t *rsc = node->details->remote_rsc->container; if (!pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { crm_notice("Not fencing guest node %s " "(otherwise would because %s): " "its guest resource %s is unmanaged", pcmk__node_name(node), reason, rsc->id); } else { pcmk__sched_warn("Guest node %s will be fenced " "(by recovering its guest resource %s): %s", pcmk__node_name(node), rsc->id, reason); /* We don't mark the node as unclean because that would prevent the * node from running resources. We want to allow it to run resources * in this transition if the recovery succeeds. */ node->details->remote_requires_reset = TRUE; pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); } } } else if (is_dangling_guest_node(node)) { crm_info("Cleaning up dangling connection for guest node %s: " "fencing was already done because %s, " "and guest resource no longer exists", pcmk__node_name(node), reason); pcmk__set_rsc_flags(node->details->remote_rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); } else if (pcmk__is_remote_node(node)) { pcmk_resource_t *rsc = node->details->remote_rsc; if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { crm_notice("Not fencing remote node %s " "(otherwise would because %s): connection is unmanaged", pcmk__node_name(node), reason); } else if(node->details->remote_requires_reset == FALSE) { node->details->remote_requires_reset = TRUE; pcmk__sched_warn("Remote node %s %s: %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "will be fenced" : "is unclean", reason); } node->details->unclean = TRUE; // No need to apply PCMK_OPT_PRIORITY_FENCING_DELAY for remote nodes pe_fence_op(node, NULL, TRUE, reason, FALSE, scheduler); } else if (node->details->unclean) { crm_trace("Cluster node %s %s because %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean", reason); } else { pcmk__sched_warn("Cluster node %s %s: %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "will be fenced" : "is unclean", reason); node->details->unclean = TRUE; pe_fence_op(node, NULL, TRUE, reason, priority_delay, scheduler); } } // @TODO xpaths can't handle templates, rules, or id-refs // nvpair with provides or requires set to unfencing #define XPATH_UNFENCING_NVPAIR PCMK_XE_NVPAIR \ "[(@" PCMK_XA_NAME "='" PCMK_STONITH_PROVIDES "'" \ "or @" PCMK_XA_NAME "='" PCMK_META_REQUIRES "') " \ "and @" PCMK_XA_VALUE "='" PCMK_VALUE_UNFENCING "']" // unfencing in rsc_defaults or any resource #define XPATH_ENABLE_UNFENCING \ "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \ "//" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR \ "|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RSC_DEFAULTS \ "/" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR static void set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler) { xmlXPathObjectPtr result = NULL; if (!pcmk_is_set(scheduler->flags, flag)) { result = xpath_search(scheduler->input, xpath); if (result && (numXpathResults(result) > 0)) { pcmk__set_scheduler_flags(scheduler, flag); } freeXpathObject(result); } } gboolean unpack_config(xmlNode *config, pcmk_scheduler_t *scheduler) { const char *value = NULL; guint interval_ms = 0U; GHashTable *config_hash = pcmk__strkey_table(free, free); pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; scheduler->config_hash = config_hash; pe__unpack_dataset_nvpairs(config, PCMK_XE_CLUSTER_PROPERTY_SET, &rule_data, config_hash, PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, FALSE, scheduler); pcmk__validate_cluster_options(config_hash); set_config_flag(scheduler, PCMK_OPT_ENABLE_STARTUP_PROBES, pcmk_sched_probe_resources); if (!pcmk_is_set(scheduler->flags, pcmk_sched_probe_resources)) { crm_info("Startup probes: disabled (dangerous)"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_HAVE_WATCHDOG); if (value && crm_is_true(value)) { crm_info("Watchdog-based self-fencing will be performed via SBD if " "fencing is required and " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " is nonzero"); pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_fencing); } /* Set certain flags via xpath here, so they can be used before the relevant * configuration sections are unpacked. */ set_if_xpath(pcmk_sched_enable_unfencing, XPATH_ENABLE_UNFENCING, scheduler); value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT); pcmk_parse_interval_spec(value, &interval_ms); if (interval_ms >= INT_MAX) { scheduler->stonith_timeout = INT_MAX; } else { scheduler->stonith_timeout = (int) interval_ms; } crm_debug("STONITH timeout: %d", scheduler->stonith_timeout); set_config_flag(scheduler, PCMK_OPT_STONITH_ENABLED, pcmk_sched_fencing_enabled); if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { crm_debug("STONITH of failed nodes is enabled"); } else { crm_debug("STONITH of failed nodes is disabled"); } scheduler->stonith_action = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_ACTION); if (!strcmp(scheduler->stonith_action, PCMK__ACTION_POWEROFF)) { pcmk__warn_once(pcmk__wo_poweroff, "Support for " PCMK_OPT_STONITH_ACTION " of " "'" PCMK__ACTION_POWEROFF "' is deprecated and will be " "removed in a future release " "(use '" PCMK_ACTION_OFF "' instead)"); scheduler->stonith_action = PCMK_ACTION_OFF; } crm_trace("STONITH will %s nodes", scheduler->stonith_action); set_config_flag(scheduler, PCMK_OPT_CONCURRENT_FENCING, pcmk_sched_concurrent_fencing); if (pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing)) { crm_debug("Concurrent fencing is enabled"); } else { crm_debug("Concurrent fencing is disabled"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_PRIORITY_FENCING_DELAY); if (value) { pcmk_parse_interval_spec(value, &interval_ms); scheduler->priority_fencing_delay = (int) (interval_ms / 1000); crm_trace("Priority fencing delay is %ds", scheduler->priority_fencing_delay); } set_config_flag(scheduler, PCMK_OPT_STOP_ALL_RESOURCES, pcmk_sched_stop_all); crm_debug("Stop all active resources: %s", pcmk__flag_text(scheduler->flags, pcmk_sched_stop_all)); set_config_flag(scheduler, PCMK_OPT_SYMMETRIC_CLUSTER, pcmk_sched_symmetric_cluster); if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) { crm_debug("Cluster is symmetric" " - resources can run anywhere by default"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_NO_QUORUM_POLICY); if (pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_ignore; } else if (pcmk__str_eq(value, PCMK_VALUE_FREEZE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_freeze; } else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_demote; } else if (pcmk__str_eq(value, PCMK_VALUE_FENCE_LEGACY, pcmk__str_casei)) { if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { int do_panic = 0; crm_element_value_int(scheduler->input, PCMK_XA_NO_QUORUM_PANIC, &do_panic); if (do_panic || pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) { scheduler->no_quorum_policy = pcmk_no_quorum_fence; } else { crm_notice("Resetting " PCMK_OPT_NO_QUORUM_POLICY " to 'stop': cluster has never had quorum"); scheduler->no_quorum_policy = pcmk_no_quorum_stop; } } else { pcmk__config_err("Resetting " PCMK_OPT_NO_QUORUM_POLICY " to 'stop' because fencing is disabled"); scheduler->no_quorum_policy = pcmk_no_quorum_stop; } } else { scheduler->no_quorum_policy = pcmk_no_quorum_stop; } switch (scheduler->no_quorum_policy) { case pcmk_no_quorum_freeze: crm_debug("On loss of quorum: Freeze resources"); break; case pcmk_no_quorum_stop: crm_debug("On loss of quorum: Stop ALL resources"); break; case pcmk_no_quorum_demote: crm_debug("On loss of quorum: " "Demote promotable resources and stop other resources"); break; case pcmk_no_quorum_fence: crm_notice("On loss of quorum: Fence all remaining nodes"); break; case pcmk_no_quorum_ignore: crm_notice("On loss of quorum: Ignore"); break; } set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_RESOURCES, pcmk_sched_stop_removed_resources); if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { crm_trace("Orphan resources are stopped"); } else { crm_trace("Orphan resources are ignored"); } set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_ACTIONS, pcmk_sched_cancel_removed_actions); if (pcmk_is_set(scheduler->flags, pcmk_sched_cancel_removed_actions)) { crm_trace("Orphan resource actions are stopped"); } else { crm_trace("Orphan resource actions are ignored"); } value = pcmk__cluster_option(config_hash, PCMK__OPT_REMOVE_AFTER_STOP); if (value != NULL) { if (crm_is_true(value)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_remove_after_stop); pcmk__warn_once(pcmk__wo_remove_after, "Support for the " PCMK__OPT_REMOVE_AFTER_STOP " cluster property is deprecated and will be " "removed in a future release"); } else { pcmk__clear_scheduler_flags(scheduler, pcmk_sched_remove_after_stop); } } set_config_flag(scheduler, PCMK_OPT_MAINTENANCE_MODE, pcmk_sched_in_maintenance); crm_trace("Maintenance mode: %s", pcmk__flag_text(scheduler->flags, pcmk_sched_in_maintenance)); set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL, pcmk_sched_start_failure_fatal); if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) { crm_trace("Start failures are always fatal"); } else { crm_trace("Start failures are handled by failcount"); } if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING, pcmk_sched_startup_fencing); } if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) { crm_trace("Unseen nodes will be fenced"); } else { pcmk__warn_once(pcmk__wo_blind, "Blind faith: not fencing unseen nodes"); } pe__unpack_node_health_scores(scheduler); scheduler->placement_strategy = pcmk__cluster_option(config_hash, PCMK_OPT_PLACEMENT_STRATEGY); crm_trace("Placement strategy: %s", scheduler->placement_strategy); set_config_flag(scheduler, PCMK_OPT_SHUTDOWN_LOCK, pcmk_sched_shutdown_lock); if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { value = pcmk__cluster_option(config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT); pcmk_parse_interval_spec(value, &(scheduler->shutdown_lock)); scheduler->shutdown_lock /= 1000; crm_trace("Resources will be locked to nodes that were cleanly " "shut down (locks expire after %s)", pcmk__readable_interval(scheduler->shutdown_lock)); } else { crm_trace("Resources will not be locked to nodes that were cleanly " "shut down"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT); pcmk_parse_interval_spec(value, &(scheduler->node_pending_timeout)); scheduler->node_pending_timeout /= 1000; if (scheduler->node_pending_timeout == 0) { crm_trace("Do not fence pending nodes"); } else { crm_trace("Fence pending nodes after %s", pcmk__readable_interval(scheduler->node_pending_timeout * 1000)); } return TRUE; } pcmk_node_t * pe_create_node(const char *id, const char *uname, const char *type, const char *score, pcmk_scheduler_t *scheduler) { pcmk_node_t *new_node = NULL; if (pcmk_find_node(scheduler, uname) != NULL) { pcmk__config_warn("More than one node entry has name '%s'", uname); } new_node = calloc(1, sizeof(pcmk_node_t)); if (new_node == NULL) { pcmk__sched_err("Could not allocate memory for node %s", uname); return NULL; } new_node->weight = char2score(score); new_node->details = calloc(1, sizeof(struct pe_node_shared_s)); if (new_node->details == NULL) { free(new_node); pcmk__sched_err("Could not allocate memory for node %s", uname); return NULL; } crm_trace("Creating node for entry %s/%s", uname, id); new_node->details->id = id; new_node->details->uname = uname; new_node->details->online = FALSE; new_node->details->shutdown = FALSE; new_node->details->rsc_discovery_enabled = TRUE; new_node->details->running_rsc = NULL; new_node->details->data_set = scheduler; if (pcmk__str_eq(type, PCMK_VALUE_MEMBER, pcmk__str_null_matches|pcmk__str_casei)) { new_node->details->type = pcmk_node_variant_cluster; } else if (pcmk__str_eq(type, PCMK_VALUE_REMOTE, pcmk__str_casei)) { new_node->details->type = pcmk_node_variant_remote; pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_remote_nodes); } else { /* @COMPAT 'ping' is the default for backward compatibility, but it * should be changed to 'member' at a compatibility break */ if (!pcmk__str_eq(type, PCMK__VALUE_PING, pcmk__str_casei)) { pcmk__config_warn("Node %s has unrecognized type '%s', " "assuming '" PCMK__VALUE_PING "'", pcmk__s(uname, "without name"), type); } pcmk__warn_once(pcmk__wo_ping_node, "Support for nodes of type '" PCMK__VALUE_PING "' " "(such as %s) is deprecated and will be removed in a " "future release", pcmk__s(uname, "unnamed node")); new_node->details->type = node_ping; } new_node->details->attrs = pcmk__strkey_table(free, free); if (pcmk__is_pacemaker_remote_node(new_node)) { pcmk__insert_dup(new_node->details->attrs, CRM_ATTR_KIND, "remote"); } else { pcmk__insert_dup(new_node->details->attrs, CRM_ATTR_KIND, "cluster"); } new_node->details->utilization = pcmk__strkey_table(free, free); new_node->details->digest_cache = pcmk__strkey_table(free, pe__free_digests); scheduler->nodes = g_list_insert_sorted(scheduler->nodes, new_node, pe__cmp_node_name); return new_node; } static const char * expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pcmk_scheduler_t *data) { xmlNode *attr_set = NULL; xmlNode *attr = NULL; const char *container_id = pcmk__xe_id(xml_obj); const char *remote_name = NULL; const char *remote_server = NULL; const char *remote_port = NULL; const char *connect_timeout = "60s"; const char *remote_allow_migrate=NULL; const char *is_managed = NULL; for (attr_set = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL); attr_set != NULL; attr_set = pcmk__xe_next(attr_set)) { if (!pcmk__xe_is(attr_set, PCMK_XE_META_ATTRIBUTES)) { continue; } for (attr = pcmk__xe_first_child(attr_set, NULL, NULL, NULL); attr != NULL; attr = pcmk__xe_next(attr)) { const char *value = crm_element_value(attr, PCMK_XA_VALUE); const char *name = crm_element_value(attr, PCMK_XA_NAME); if (name == NULL) { // Sanity continue; } if (strcmp(name, PCMK_META_REMOTE_NODE) == 0) { remote_name = value; } else if (strcmp(name, PCMK_META_REMOTE_ADDR) == 0) { remote_server = value; } else if (strcmp(name, PCMK_META_REMOTE_PORT) == 0) { remote_port = value; } else if (strcmp(name, PCMK_META_REMOTE_CONNECT_TIMEOUT) == 0) { connect_timeout = value; } else if (strcmp(name, PCMK_META_REMOTE_ALLOW_MIGRATE) == 0) { remote_allow_migrate = value; } else if (strcmp(name, PCMK_META_IS_MANAGED) == 0) { is_managed = value; } } } if (remote_name == NULL) { return NULL; } if (pe_find_resource(data->resources, remote_name) != NULL) { return NULL; } pe_create_remote_xml(parent, remote_name, container_id, remote_allow_migrate, is_managed, connect_timeout, remote_server, remote_port); return remote_name; } static void handle_startup_fencing(pcmk_scheduler_t *scheduler, pcmk_node_t *new_node) { if ((new_node->details->type == pcmk_node_variant_remote) && (new_node->details->remote_rsc == NULL)) { /* Ignore fencing for remote nodes that don't have a connection resource * associated with them. This happens when remote node entries get left * in the nodes section after the connection resource is removed. */ return; } if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) { // All nodes are unclean until we've seen their status entry new_node->details->unclean = TRUE; } else { // Blind faith ... new_node->details->unclean = FALSE; } /* We need to be able to determine if a node's status section * exists or not separate from whether the node is unclean. */ new_node->details->unseen = TRUE; } gboolean unpack_nodes(xmlNode *xml_nodes, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; pcmk_node_t *new_node = NULL; const char *id = NULL; const char *uname = NULL; const char *type = NULL; const char *score = NULL; for (xml_obj = pcmk__xe_first_child(xml_nodes, NULL, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (pcmk__xe_is(xml_obj, PCMK_XE_NODE)) { new_node = NULL; id = crm_element_value(xml_obj, PCMK_XA_ID); uname = crm_element_value(xml_obj, PCMK_XA_UNAME); type = crm_element_value(xml_obj, PCMK_XA_TYPE); score = crm_element_value(xml_obj, PCMK_XA_SCORE); crm_trace("Processing node %s/%s", uname, id); if (id == NULL) { pcmk__config_err("Ignoring <" PCMK_XE_NODE "> entry in configuration without id"); continue; } new_node = pe_create_node(id, uname, type, score, scheduler); if (new_node == NULL) { return FALSE; } handle_startup_fencing(scheduler, new_node); add_node_attrs(xml_obj, new_node, FALSE, scheduler); crm_trace("Done with node %s", crm_element_value(xml_obj, PCMK_XA_UNAME)); } } if (scheduler->localhost && (pcmk_find_node(scheduler, scheduler->localhost) == NULL)) { crm_info("Creating a fake local node"); pe_create_node(scheduler->localhost, scheduler->localhost, NULL, 0, scheduler); } return TRUE; } static void setup_container(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { const char *container_id = NULL; if (rsc->children) { g_list_foreach(rsc->children, (GFunc) setup_container, scheduler); return; } container_id = g_hash_table_lookup(rsc->meta, PCMK__META_CONTAINER); if (container_id && !pcmk__str_eq(container_id, rsc->id, pcmk__str_casei)) { pcmk_resource_t *container = pe_find_resource(scheduler->resources, container_id); if (container) { rsc->container = container; container->fillers = g_list_append(container->fillers, rsc); pcmk__rsc_trace(rsc, "Resource %s's container is %s", rsc->id, container_id); } else { pcmk__config_err("Resource %s: Unknown resource container (%s)", rsc->id, container_id); } } } gboolean unpack_remote_nodes(xmlNode *xml_resources, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; /* Create remote nodes and guest nodes from the resource configuration * before unpacking resources. */ for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { const char *new_node_id = NULL; /* Check for remote nodes, which are defined by ocf:pacemaker:remote * primitives. */ if (xml_contains_remote_node(xml_obj)) { new_node_id = pcmk__xe_id(xml_obj); /* The pcmk_find_node() check ensures we don't iterate over an * expanded node that has already been added to the node list */ if (new_node_id && (pcmk_find_node(scheduler, new_node_id) == NULL)) { crm_trace("Found remote node %s defined by resource %s", new_node_id, pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } continue; } /* Check for guest nodes, which are defined by special meta-attributes * of a primitive of any type (for example, VirtualDomain or Xen). */ if (pcmk__xe_is(xml_obj, PCMK_XE_PRIMITIVE)) { /* This will add an ocf:pacemaker:remote primitive to the * configuration for the guest node's connection, to be unpacked * later. */ new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources, scheduler); if (new_node_id && (pcmk_find_node(scheduler, new_node_id) == NULL)) { crm_trace("Found guest node %s in resource %s", new_node_id, pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } continue; } /* Check for guest nodes inside a group. Clones are currently not * supported as guest nodes. */ if (pcmk__xe_is(xml_obj, PCMK_XE_GROUP)) { xmlNode *xml_obj2 = NULL; for (xml_obj2 = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL); xml_obj2 != NULL; xml_obj2 = pcmk__xe_next(xml_obj2)) { new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources, scheduler); if (new_node_id && (pcmk_find_node(scheduler, new_node_id) == NULL)) { crm_trace("Found guest node %s in resource %s inside group %s", new_node_id, pcmk__xe_id(xml_obj2), pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } } } } return TRUE; } /* Call this after all the nodes and resources have been * unpacked, but before the status section is read. * * A remote node's online status is reflected by the state * of the remote node's connection resource. We need to link * the remote node to this connection resource so we can have * easy access to the connection resource during the scheduler calculations. */ static void link_rsc2remotenode(pcmk_scheduler_t *scheduler, pcmk_resource_t *new_rsc) { pcmk_node_t *remote_node = NULL; if (!pcmk_is_set(new_rsc->flags, pcmk__rsc_is_remote_connection)) { return; } if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { /* remote_nodes and remote_resources are not linked in quick location calculations */ return; } remote_node = pcmk_find_node(scheduler, new_rsc->id); CRM_CHECK(remote_node != NULL, return); pcmk__rsc_trace(new_rsc, "Linking remote connection resource %s to %s", new_rsc->id, pcmk__node_name(remote_node)); remote_node->details->remote_rsc = new_rsc; if (new_rsc->container == NULL) { /* Handle start-up fencing for remote nodes (as opposed to guest nodes) * the same as is done for cluster nodes. */ handle_startup_fencing(scheduler, remote_node); } else { /* pe_create_node() marks the new node as "remote" or "cluster"; now * that we know the node is a guest node, update it correctly. */ pcmk__insert_dup(remote_node->details->attrs, CRM_ATTR_KIND, "container"); } } static void destroy_tag(gpointer data) { pcmk_tag_t *tag = data; if (tag) { free(tag->id); g_list_free_full(tag->refs, free); free(tag); } } /*! * \internal * \brief Parse configuration XML for resource information * * \param[in] xml_resources Top of resource configuration XML * \param[in,out] scheduler Scheduler data * * \return TRUE * * \note unpack_remote_nodes() MUST be called before this, so that the nodes can * be used when pe__unpack_resource() calls resource_location() */ gboolean unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; GList *gIter = NULL; scheduler->template_rsc_sets = pcmk__strkey_table(free, destroy_tag); for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { pcmk_resource_t *new_rsc = NULL; const char *id = pcmk__xe_id(xml_obj); if (pcmk__str_empty(id)) { pcmk__config_err("Ignoring <%s> resource without ID", xml_obj->name); continue; } if (pcmk__xe_is(xml_obj, PCMK_XE_TEMPLATE)) { if (g_hash_table_lookup_extended(scheduler->template_rsc_sets, id, NULL, NULL) == FALSE) { /* Record the template's ID for the knowledge of its existence anyway. */ pcmk__insert_dup(scheduler->template_rsc_sets, id, NULL); } continue; } crm_trace("Unpacking <%s " PCMK_XA_ID "='%s'>", xml_obj->name, id); if (pe__unpack_resource(xml_obj, &new_rsc, NULL, scheduler) == pcmk_rc_ok) { scheduler->resources = g_list_append(scheduler->resources, new_rsc); pcmk__rsc_trace(new_rsc, "Added resource %s", new_rsc->id); } else { pcmk__config_err("Ignoring <%s> resource '%s' " "because configuration is invalid", xml_obj->name, id); } } for (gIter = scheduler->resources; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; setup_container(rsc, scheduler); link_rsc2remotenode(scheduler, rsc); } scheduler->resources = g_list_sort(scheduler->resources, pe__cmp_rsc_priority); if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { /* Ignore */ } else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled) && !pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) { pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined"); pcmk__config_err("Either configure some or disable STONITH with the " PCMK_OPT_STONITH_ENABLED " option"); pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity"); } return TRUE; } gboolean unpack_tags(xmlNode *xml_tags, pcmk_scheduler_t *scheduler) { xmlNode *xml_tag = NULL; scheduler->tags = pcmk__strkey_table(free, destroy_tag); for (xml_tag = pcmk__xe_first_child(xml_tags, NULL, NULL, NULL); xml_tag != NULL; xml_tag = pcmk__xe_next(xml_tag)) { xmlNode *xml_obj_ref = NULL; const char *tag_id = pcmk__xe_id(xml_tag); if (!pcmk__xe_is(xml_tag, PCMK_XE_TAG)) { continue; } if (tag_id == NULL) { pcmk__config_err("Ignoring <%s> without " PCMK_XA_ID, (const char *) xml_tag->name); continue; } for (xml_obj_ref = pcmk__xe_first_child(xml_tag, NULL, NULL, NULL); xml_obj_ref != NULL; xml_obj_ref = pcmk__xe_next(xml_obj_ref)) { const char *obj_ref = pcmk__xe_id(xml_obj_ref); if (!pcmk__xe_is(xml_obj_ref, PCMK_XE_OBJ_REF)) { continue; } if (obj_ref == NULL) { pcmk__config_err("Ignoring <%s> for tag '%s' without " PCMK_XA_ID, xml_obj_ref->name, tag_id); continue; } if (add_tag_ref(scheduler->tags, tag_id, obj_ref) == FALSE) { return FALSE; } } } return TRUE; } /* The ticket state section: * "/cib/status/tickets/ticket_state" */ static gboolean unpack_ticket_state(xmlNode *xml_ticket, pcmk_scheduler_t *scheduler) { const char *ticket_id = NULL; const char *granted = NULL; const char *last_granted = NULL; const char *standby = NULL; xmlAttrPtr xIter = NULL; pcmk_ticket_t *ticket = NULL; ticket_id = pcmk__xe_id(xml_ticket); if (pcmk__str_empty(ticket_id)) { return FALSE; } crm_trace("Processing ticket state for %s", ticket_id); ticket = g_hash_table_lookup(scheduler->tickets, ticket_id); if (ticket == NULL) { ticket = ticket_new(ticket_id, scheduler); if (ticket == NULL) { return FALSE; } } for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = pcmk__xml_attr_value(xIter); if (pcmk__str_eq(prop_name, PCMK_XA_ID, pcmk__str_none)) { continue; } pcmk__insert_dup(ticket->state, prop_name, prop_value); } granted = g_hash_table_lookup(ticket->state, PCMK__XA_GRANTED); if (granted && crm_is_true(granted)) { ticket->granted = TRUE; crm_info("We have ticket '%s'", ticket->id); } else { ticket->granted = FALSE; crm_info("We do not have ticket '%s'", ticket->id); } last_granted = g_hash_table_lookup(ticket->state, PCMK_XA_LAST_GRANTED); if (last_granted) { long long last_granted_ll; pcmk__scan_ll(last_granted, &last_granted_ll, 0LL); ticket->last_granted = (time_t) last_granted_ll; } standby = g_hash_table_lookup(ticket->state, PCMK_XA_STANDBY); if (standby && crm_is_true(standby)) { ticket->standby = TRUE; if (ticket->granted) { crm_info("Granted ticket '%s' is in standby-mode", ticket->id); } } else { ticket->standby = FALSE; } crm_trace("Done with ticket state for %s", ticket_id); return TRUE; } static gboolean unpack_tickets_state(xmlNode *xml_tickets, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; for (xml_obj = pcmk__xe_first_child(xml_tickets, NULL, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (!pcmk__xe_is(xml_obj, PCMK__XE_TICKET_STATE)) { continue; } unpack_ticket_state(xml_obj, scheduler); } return TRUE; } static void unpack_handle_remote_attrs(pcmk_node_t *this_node, const xmlNode *state, pcmk_scheduler_t *scheduler) { const char *discovery = NULL; const xmlNode *attrs = NULL; pcmk_resource_t *rsc = NULL; if (!pcmk__xe_is(state, PCMK__XE_NODE_STATE)) { return; } if ((this_node == NULL) || !pcmk__is_pacemaker_remote_node(this_node)) { return; } crm_trace("Processing Pacemaker Remote node %s", pcmk__node_name(this_node)); pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_IN_MAINTENANCE), &(this_node->details->remote_maintenance), 0); rsc = this_node->details->remote_rsc; if (this_node->details->remote_requires_reset == FALSE) { this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; } attrs = pcmk__xe_first_child(state, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL, NULL); add_node_attrs(attrs, this_node, TRUE, scheduler); if (pe__shutdown_requested(this_node)) { crm_info("%s is shutting down", pcmk__node_name(this_node)); this_node->details->shutdown = TRUE; } if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_STANDBY, NULL, pcmk__rsc_node_current))) { crm_info("%s is in standby mode", pcmk__node_name(this_node)); this_node->details->standby = TRUE; } if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_MAINTENANCE, NULL, pcmk__rsc_node_current)) || ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed))) { crm_info("%s is in maintenance mode", pcmk__node_name(this_node)); this_node->details->maintenance = TRUE; } discovery = pcmk__node_attr(this_node, PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED, NULL, pcmk__rsc_node_current); if ((discovery != NULL) && !crm_is_true(discovery)) { pcmk__warn_once(pcmk__wo_rdisc_enabled, "Support for the " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " node attribute is deprecated and will be removed" " (and behave as 'true') in a future release."); if (pcmk__is_remote_node(this_node) && !pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("Ignoring " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " attribute on Pacemaker Remote node %s" " because fencing is disabled", pcmk__node_name(this_node)); } else { /* This is either a remote node with fencing enabled, or a guest * node. We don't care whether fencing is enabled when fencing guest * nodes, because they are "fenced" by recovering their containing * resource. */ crm_info("%s has resource discovery disabled", pcmk__node_name(this_node)); this_node->details->rsc_discovery_enabled = FALSE; } } } /*! * \internal * \brief Unpack a cluster node's transient attributes * * \param[in] state CIB node state XML * \param[in,out] node Cluster node whose attributes are being unpacked * \param[in,out] scheduler Scheduler data */ static void unpack_transient_attributes(const xmlNode *state, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { const char *discovery = NULL; const xmlNode *attrs = pcmk__xe_first_child(state, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL, NULL); add_node_attrs(attrs, node, TRUE, scheduler); if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_STANDBY, NULL, pcmk__rsc_node_current))) { crm_info("%s is in standby mode", pcmk__node_name(node)); node->details->standby = TRUE; } if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_MAINTENANCE, NULL, pcmk__rsc_node_current))) { crm_info("%s is in maintenance mode", pcmk__node_name(node)); node->details->maintenance = TRUE; } discovery = pcmk__node_attr(node, PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED, NULL, pcmk__rsc_node_current); if ((discovery != NULL) && !crm_is_true(discovery)) { pcmk__config_warn("Ignoring " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " attribute for %s because disabling resource" " discovery is not allowed for cluster nodes", pcmk__node_name(node)); } } /*! * \internal * \brief Unpack a node state entry (first pass) * * Unpack one node state entry from status. This unpacks information from the * \C PCMK__XE_NODE_STATE element itself and node attributes inside it, but not * the resource history inside it. Multiple passes through the status are needed * to fully unpack everything. * * \param[in] state CIB node state XML * \param[in,out] scheduler Scheduler data */ static void unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *uname = NULL; pcmk_node_t *this_node = NULL; id = crm_element_value(state, PCMK_XA_ID); if (id == NULL) { pcmk__config_err("Ignoring invalid " PCMK__XE_NODE_STATE " entry without " PCMK_XA_ID); crm_log_xml_info(state, "missing-id"); return; } uname = crm_element_value(state, PCMK_XA_UNAME); if (uname == NULL) { /* If a joining peer makes the cluster acquire the quorum from corosync * meanwhile it has not joined CPG membership of pacemaker-controld yet, * it's possible that the created PCMK__XE_NODE_STATE entry doesn't have * a PCMK_XA_UNAME yet. We should recognize the node as `pending` and * wait for it to join CPG. */ crm_trace("Handling " PCMK__XE_NODE_STATE " entry with id=\"%s\" " "without " PCMK_XA_UNAME, id); } this_node = pe_find_node_any(scheduler->nodes, id, uname); if (this_node == NULL) { crm_notice("Ignoring recorded state for removed node with name %s and " PCMK_XA_ID " %s", pcmk__s(uname, "unknown"), id); return; } if (pcmk__is_pacemaker_remote_node(this_node)) { /* We can't determine the online status of Pacemaker Remote nodes until * after all resource history has been unpacked. In this first pass, we * do need to mark whether the node has been fenced, as this plays a * role during unpacking cluster node resource state. */ pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_FENCED), &(this_node->details->remote_was_fenced), 0); return; } unpack_transient_attributes(state, this_node, scheduler); /* Provisionally mark this cluster node as clean. We have at least seen it * in the current cluster's lifetime. */ this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; crm_trace("Determining online status of cluster node %s (id %s)", pcmk__node_name(this_node), id); determine_online_status(state, this_node, scheduler); if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate) && this_node->details->online && (scheduler->no_quorum_policy == pcmk_no_quorum_fence)) { /* Everything else should flow from this automatically * (at least until the scheduler becomes able to migrate off * healthy resources) */ pe_fence_node(scheduler, this_node, "cluster does not have quorum", FALSE); } } /*! * \internal * \brief Unpack nodes' resource history as much as possible * * Unpack as many nodes' resource history as possible in one pass through the * status. We need to process Pacemaker Remote nodes' connections/containers * before unpacking their history; the connection/container history will be * in another node's history, so it might take multiple passes to unpack * everything. * * \param[in] status CIB XML status section * \param[in] fence If true, treat any not-yet-unpacked nodes as unseen * \param[in,out] scheduler Scheduler data * * \return Standard Pacemaker return code (specifically pcmk_rc_ok if done, * or EAGAIN if more unpacking remains to be done) */ static int unpack_node_history(const xmlNode *status, bool fence, pcmk_scheduler_t *scheduler) { int rc = pcmk_rc_ok; // Loop through all PCMK__XE_NODE_STATE entries in CIB status for (const xmlNode *state = pcmk__xe_first_child(status, PCMK__XE_NODE_STATE, NULL, NULL); state != NULL; state = pcmk__xe_next_same(state)) { const char *id = pcmk__xe_id(state); const char *uname = crm_element_value(state, PCMK_XA_UNAME); pcmk_node_t *this_node = NULL; if ((id == NULL) || (uname == NULL)) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history from malformed " PCMK__XE_NODE_STATE " without id and/or uname"); continue; } this_node = pe_find_node_any(scheduler->nodes, id, uname); if (this_node == NULL) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history for node %s because " "no longer in configuration", id); continue; } if (this_node->details->unpacked) { crm_trace("Not unpacking resource history for node %s because " "already unpacked", id); continue; } if (fence) { // We're processing all remaining nodes } else if (pcmk__is_guest_or_bundle_node(this_node)) { /* We can unpack a guest node's history only after we've unpacked * other resource history to the point that we know that the node's * connection and containing resource are both up. */ pcmk_resource_t *rsc = this_node->details->remote_rsc; if ((rsc == NULL) || (rsc->role != pcmk_role_started) || (rsc->container->role != pcmk_role_started)) { crm_trace("Not unpacking resource history for guest node %s " "because container and connection are not known to " "be up", id); continue; } } else if (pcmk__is_remote_node(this_node)) { /* We can unpack a remote node's history only after we've unpacked * other resource history to the point that we know that the node's * connection is up, with the exception of when shutdown locks are * in use. */ pcmk_resource_t *rsc = this_node->details->remote_rsc; if ((rsc == NULL) || (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock) && (rsc->role != pcmk_role_started))) { crm_trace("Not unpacking resource history for remote node %s " "because connection is not known to be up", id); continue; } /* If fencing and shutdown locks are disabled and we're not processing * unseen nodes, then we don't want to unpack offline nodes until online * nodes have been unpacked. This allows us to number active clone * instances first. */ } else if (!pcmk_any_flags_set(scheduler->flags, pcmk_sched_fencing_enabled |pcmk_sched_shutdown_lock) && !this_node->details->online) { crm_trace("Not unpacking resource history for offline " "cluster node %s", id); continue; } if (pcmk__is_pacemaker_remote_node(this_node)) { determine_remote_online_status(scheduler, this_node); unpack_handle_remote_attrs(this_node, state, scheduler); } crm_trace("Unpacking resource history for %snode %s", (fence? "unseen " : ""), id); this_node->details->unpacked = TRUE; unpack_node_lrm(this_node, state, scheduler); rc = EAGAIN; // Other node histories might depend on this one } return rc; } /* remove nodes that are down, stopping */ /* create positive rsc_to_node constraints between resources and the nodes they are running on */ /* anything else? */ gboolean unpack_status(xmlNode *status, pcmk_scheduler_t *scheduler) { xmlNode *state = NULL; crm_trace("Beginning unpack"); if (scheduler->tickets == NULL) { scheduler->tickets = pcmk__strkey_table(free, destroy_ticket); } for (state = pcmk__xe_first_child(status, NULL, NULL, NULL); state != NULL; state = pcmk__xe_next(state)) { if (pcmk__xe_is(state, PCMK_XE_TICKETS)) { unpack_tickets_state((xmlNode *) state, scheduler); } else if (pcmk__xe_is(state, PCMK__XE_NODE_STATE)) { unpack_node_state(state, scheduler); } } while (unpack_node_history(status, FALSE, scheduler) == EAGAIN) { crm_trace("Another pass through node resource histories is needed"); } // Now catch any nodes we didn't see unpack_node_history(status, pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled), scheduler); /* Now that we know where resources are, we can schedule stops of containers * with failed bundle connections */ if (scheduler->stop_needed != NULL) { for (GList *item = scheduler->stop_needed; item; item = item->next) { pcmk_resource_t *container = item->data; pcmk_node_t *node = pcmk__current_node(container); if (node) { stop_action(container, node, FALSE); } } g_list_free(scheduler->stop_needed); scheduler->stop_needed = NULL; } /* Now that we know status of all Pacemaker Remote connections and nodes, * we can stop connections for node shutdowns, and check the online status * of remote/guest nodes that didn't have any node history to unpack. */ for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *this_node = gIter->data; if (!pcmk__is_pacemaker_remote_node(this_node)) { continue; } if (this_node->details->shutdown && (this_node->details->remote_rsc != NULL)) { pe__set_next_role(this_node->details->remote_rsc, pcmk_role_stopped, "remote shutdown"); } if (!this_node->details->unpacked) { determine_remote_online_status(scheduler, this_node); } } return TRUE; } /*! * \internal * \brief Unpack node's time when it became a member at the cluster layer * * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry * \param[in,out] scheduler Scheduler data * * \return Epoch time when node became a cluster member * (or scheduler effective time for legacy entries) if a member, * 0 if not a member, or -1 if no valid information available */ static long long unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler) { const char *member_time = crm_element_value(node_state, PCMK__XA_IN_CCM); int member = 0; if (member_time == NULL) { return -1LL; } else if (crm_str_to_boolean(member_time, &member) == 1) { /* If in_ccm=0, we'll return 0 here. If in_ccm=1, either the entry was * recorded as a boolean for a DC < 2.1.7, or the node is pending * shutdown and has left the CPG, in which case it was set to 1 to avoid * fencing for PCMK_OPT_NODE_PENDING_TIMEOUT. * * We return the effective time for in_ccm=1 because what's important to * avoid fencing is that effective time minus this value is less than * the pending node timeout. */ return member? (long long) get_effective_time(scheduler) : 0LL; } else { long long when_member = 0LL; if ((pcmk__scan_ll(member_time, &when_member, 0LL) != pcmk_rc_ok) || (when_member < 0LL)) { crm_warn("Unrecognized value '%s' for " PCMK__XA_IN_CCM " in " PCMK__XE_NODE_STATE " entry", member_time); return -1LL; } return when_member; } } /*! * \internal * \brief Unpack node's time when it became online in process group * * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry * * \return Epoch time when node became online in process group (or 0 if not * online, or 1 for legacy online entries) */ static long long unpack_node_online(const xmlNode *node_state) { const char *peer_time = crm_element_value(node_state, PCMK_XA_CRMD); // @COMPAT Entries recorded for DCs < 2.1.7 have "online" or "offline" if (pcmk__str_eq(peer_time, PCMK_VALUE_OFFLINE, pcmk__str_casei|pcmk__str_null_matches)) { return 0LL; } else if (pcmk__str_eq(peer_time, PCMK_VALUE_ONLINE, pcmk__str_casei)) { return 1LL; } else { long long when_online = 0LL; if ((pcmk__scan_ll(peer_time, &when_online, 0LL) != pcmk_rc_ok) || (when_online < 0)) { crm_warn("Unrecognized value '%s' for " PCMK_XA_CRMD " in " PCMK__XE_NODE_STATE " entry, assuming offline", peer_time); return 0LL; } return when_online; } } /*! * \internal * \brief Unpack node attribute for user-requested fencing * * \param[in] node Node to check * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry in CIB status * * \return \c true if fencing has been requested for \p node, otherwise \c false */ static bool unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state) { long long value = 0LL; int value_i = 0; const char *value_s = pcmk__node_attr(node, PCMK_NODE_ATTR_TERMINATE, NULL, pcmk__rsc_node_current); // Value may be boolean or an epoch time if (crm_str_to_boolean(value_s, &value_i) == 1) { return (value_i != 0); } if (pcmk__scan_ll(value_s, &value, 0LL) == pcmk_rc_ok) { return (value > 0); } crm_warn("Ignoring unrecognized value '%s' for " PCMK_NODE_ATTR_TERMINATE "node attribute for %s", value_s, pcmk__node_name(node)); return false; } static gboolean determine_online_status_no_fencing(pcmk_scheduler_t *scheduler, const xmlNode *node_state, pcmk_node_t *this_node) { gboolean online = FALSE; const char *join = crm_element_value(node_state, PCMK__XA_JOIN); const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); long long when_member = unpack_node_member(node_state, scheduler); long long when_online = unpack_node_online(node_state); if (when_member <= 0) { crm_trace("Node %s is %sdown", pcmk__node_name(this_node), ((when_member < 0)? "presumed " : "")); } else if (when_online > 0) { if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { online = TRUE; } else { crm_debug("Node %s is not ready to run resources: %s", pcmk__node_name(this_node), join); } } else if (this_node->details->expected_up == FALSE) { crm_trace("Node %s controller is down: " "member@%lld online@%lld join=%s expected=%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, "")); } else { /* mark it unclean */ pe_fence_node(scheduler, this_node, "peer is unexpectedly down", FALSE); crm_info("Node %s member@%lld online@%lld join=%s expected=%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, "")); } return online; } /*! * \internal * \brief Check whether a node has taken too long to join controller group * * \param[in,out] scheduler Scheduler data * \param[in] node Node to check * \param[in] when_member Epoch time when node became a cluster member * \param[in] when_online Epoch time when node joined controller group * * \return true if node has been pending (on the way up) longer than * \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false * \note This will also update the cluster's recheck time if appropriate. */ static inline bool pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, long long when_member, long long when_online) { if ((scheduler->node_pending_timeout > 0) && (when_member > 0) && (when_online <= 0)) { // There is a timeout on pending nodes, and node is pending time_t timeout = when_member + scheduler->node_pending_timeout; if (get_effective_time(node->details->data_set) >= timeout) { return true; // Node has timed out } // Node is pending, but still has time pe__update_recheck_time(timeout, scheduler, "pending node timeout"); } return false; } static bool determine_online_status_fencing(pcmk_scheduler_t *scheduler, const xmlNode *node_state, pcmk_node_t *this_node) { bool termination_requested = unpack_node_terminate(this_node, node_state); const char *join = crm_element_value(node_state, PCMK__XA_JOIN); const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); long long when_member = unpack_node_member(node_state, scheduler); long long when_online = unpack_node_online(node_state); /* - PCMK__XA_JOIN ::= member|down|pending|banned - PCMK_XA_EXPECTED ::= member|down @COMPAT with entries recorded for DCs < 2.1.7 - PCMK__XA_IN_CCM ::= true|false - PCMK_XA_CRMD ::= online|offline Since crm_feature_set 3.18.0 (pacemaker-2.1.7): - PCMK__XA_IN_CCM ::= |0 Since when node has been a cluster member. A value 0 of means the node is not a cluster member. - PCMK_XA_CRMD ::= |0 Since when peer has been online in CPG. A value 0 means the peer is offline in CPG. */ crm_trace("Node %s member@%lld online@%lld join=%s expected=%s%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, ""), (termination_requested? " (termination requested)" : "")); if (this_node->details->shutdown) { crm_debug("%s is shutting down", pcmk__node_name(this_node)); /* Slightly different criteria since we can't shut down a dead peer */ return (when_online > 0); } if (when_member < 0) { pe_fence_node(scheduler, this_node, "peer has not been seen by the cluster", FALSE); return false; } if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_none)) { pe_fence_node(scheduler, this_node, "peer failed Pacemaker membership criteria", FALSE); } else if (termination_requested) { if ((when_member <= 0) && (when_online <= 0) && pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_none)) { crm_info("%s was fenced as requested", pcmk__node_name(this_node)); return false; } pe_fence_node(scheduler, this_node, "fencing was requested", false); } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN, pcmk__str_null_matches)) { if (pending_too_long(scheduler, this_node, when_member, when_online)) { pe_fence_node(scheduler, this_node, "peer pending timed out on joining the process group", FALSE); } else if ((when_member > 0) || (when_online > 0)) { crm_info("- %s is not ready to run resources", pcmk__node_name(this_node)); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { crm_trace("%s is down or still coming up", pcmk__node_name(this_node)); } } else if (when_member <= 0) { // Consider PCMK_OPT_PRIORITY_FENCING_DELAY for lost nodes pe_fence_node(scheduler, this_node, "peer is no longer part of the cluster", TRUE); } else if (when_online <= 0) { pe_fence_node(scheduler, this_node, "peer process is no longer available", FALSE); /* Everything is running at this point, now check join state */ } else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_none)) { crm_info("%s is active", pcmk__node_name(this_node)); } else if (pcmk__str_any_of(join, CRMD_JOINSTATE_PENDING, CRMD_JOINSTATE_DOWN, NULL)) { crm_info("%s is not ready to run resources", pcmk__node_name(this_node)); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { pe_fence_node(scheduler, this_node, "peer was in an unknown state", FALSE); } return (when_member > 0); } static void determine_remote_online_status(pcmk_scheduler_t *scheduler, pcmk_node_t *this_node) { pcmk_resource_t *rsc = this_node->details->remote_rsc; pcmk_resource_t *container = NULL; pcmk_node_t *host = NULL; /* If there is a node state entry for a (former) Pacemaker Remote node * but no resource creating that node, the node's connection resource will * be NULL. Consider it an offline remote node in that case. */ if (rsc == NULL) { this_node->details->online = FALSE; goto remote_online_done; } container = rsc->container; - if (container && pcmk__list_of_1(rsc->running_on)) { - host = rsc->running_on->data; + if ((container != NULL) && pcmk__list_of_1(rsc->private->active_nodes)) { + host = rsc->private->active_nodes->data; } /* If the resource is currently started, mark it online. */ if (rsc->role == pcmk_role_started) { crm_trace("%s node %s presumed ONLINE because connection resource is started", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = TRUE; } /* consider this node shutting down if transitioning start->stop */ if ((rsc->role == pcmk_role_started) && (rsc->next_role == pcmk_role_stopped)) { crm_trace("%s node %s shutting down because connection resource is stopping", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->shutdown = TRUE; } /* Now check all the failure conditions. */ if(container && pcmk_is_set(container->flags, pcmk__rsc_failed)) { crm_trace("Guest node %s UNCLEAN because guest resource failed", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { crm_trace("%s node %s OFFLINE because connection resource failed", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; } else if ((rsc->role == pcmk_role_stopped) || ((container != NULL) && (container->role == pcmk_role_stopped))) { crm_trace("%s node %s OFFLINE because its resource is stopped", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = FALSE; } else if (host && (host->details->online == FALSE) && host->details->unclean) { crm_trace("Guest node %s UNCLEAN because host is unclean", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } remote_online_done: crm_trace("Remote node %s online=%s", this_node->details->id, this_node->details->online ? "TRUE" : "FALSE"); } static void determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node, pcmk_scheduler_t *scheduler) { gboolean online = FALSE; const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); CRM_CHECK(this_node != NULL, return); this_node->details->shutdown = FALSE; this_node->details->expected_up = FALSE; if (pe__shutdown_requested(this_node)) { this_node->details->shutdown = TRUE; } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { this_node->details->expected_up = TRUE; } if (this_node->details->type == node_ping) { this_node->details->unclean = FALSE; online = FALSE; /* As far as resource management is concerned, * the node is safely offline. * Anyone caught abusing this logic will be shot */ } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { online = determine_online_status_no_fencing(scheduler, node_state, this_node); } else { online = determine_online_status_fencing(scheduler, node_state, this_node); } if (online) { this_node->details->online = TRUE; } else { /* remove node from contention */ this_node->fixed = TRUE; // @COMPAT deprecated and unused this_node->weight = -PCMK_SCORE_INFINITY; } if (online && this_node->details->shutdown) { /* don't run resources here */ this_node->fixed = TRUE; // @COMPAT deprecated and unused this_node->weight = -PCMK_SCORE_INFINITY; } if (this_node->details->type == node_ping) { crm_info("%s is not a Pacemaker node", pcmk__node_name(this_node)); } else if (this_node->details->unclean) { pcmk__sched_warn("%s is unclean", pcmk__node_name(this_node)); } else if (this_node->details->online) { crm_info("%s is %s", pcmk__node_name(this_node), this_node->details->shutdown ? "shutting down" : this_node->details->pending ? "pending" : this_node->details->standby ? "standby" : this_node->details->maintenance ? "maintenance" : "online"); } else { crm_trace("%s is offline", pcmk__node_name(this_node)); } } /*! * \internal * \brief Find the end of a resource's name, excluding any clone suffix * * \param[in] id Resource ID to check * * \return Pointer to last character of resource's base name */ const char * pe_base_name_end(const char *id) { if (!pcmk__str_empty(id)) { const char *end = id + strlen(id) - 1; for (const char *s = end; s > id; --s) { switch (*s) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': break; case ':': return (s == end)? s : (s - 1); default: return end; } } return end; } return NULL; } /*! * \internal * \brief Get a resource name excluding any clone suffix * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_strip(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); char *basename = NULL; CRM_ASSERT(end); basename = strndup(last_rsc_id, end - last_rsc_id + 1); CRM_ASSERT(basename); return basename; } /*! * \internal * \brief Get the name of the first instance of a cloned resource * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name plus :0 * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_zero(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); size_t base_name_len = end - last_rsc_id + 1; char *zero = NULL; CRM_ASSERT(end); zero = pcmk__assert_alloc(base_name_len + 3, sizeof(char)); memcpy(zero, last_rsc_id, base_name_len); zero[base_name_len] = ':'; zero[base_name_len + 1] = '0'; return zero; } static pcmk_resource_t * create_fake_resource(const char *rsc_id, const xmlNode *rsc_entry, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = NULL; xmlNode *xml_rsc = pcmk__xe_create(NULL, PCMK_XE_PRIMITIVE); pcmk__xe_copy_attrs(xml_rsc, rsc_entry, pcmk__xaf_none); crm_xml_add(xml_rsc, PCMK_XA_ID, rsc_id); crm_log_xml_debug(xml_rsc, "Orphan resource"); if (pe__unpack_resource(xml_rsc, &rsc, NULL, scheduler) != pcmk_rc_ok) { return NULL; } if (xml_contains_remote_node(xml_rsc)) { pcmk_node_t *node; crm_debug("Detected orphaned remote node %s", rsc_id); node = pcmk_find_node(scheduler, rsc_id); if (node == NULL) { node = pe_create_node(rsc_id, rsc_id, PCMK_VALUE_REMOTE, NULL, scheduler); } link_rsc2remotenode(scheduler, rsc); if (node) { crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id); node->details->shutdown = TRUE; } } if (crm_element_value(rsc_entry, PCMK__META_CONTAINER)) { /* This orphaned rsc needs to be mapped to a container. */ crm_trace("Detected orphaned container filler %s", rsc_id); pcmk__set_rsc_flags(rsc, pcmk__rsc_removed_filler); } pcmk__set_rsc_flags(rsc, pcmk__rsc_removed); scheduler->resources = g_list_append(scheduler->resources, rsc); return rsc; } /*! * \internal * \brief Create orphan instance for anonymous clone resource history * * \param[in,out] parent Clone resource that orphan will be added to * \param[in] rsc_id Orphan's resource ID * \param[in] node Where orphan is active (for logging only) * \param[in,out] scheduler Scheduler data * * \return Newly added orphaned instance of \p parent */ static pcmk_resource_t * create_anonymous_orphan(pcmk_resource_t *parent, const char *rsc_id, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk_resource_t *top = pe__create_clone_child(parent, scheduler); pcmk_resource_t *orphan = NULL; // find_rsc() because we might be a cloned group orphan = top->private->fns->find_rsc(top, rsc_id, NULL, pcmk_rsc_match_clone_only); pcmk__rsc_debug(parent, "Created orphan %s for %s: %s on %s", top->id, parent->id, rsc_id, pcmk__node_name(node)); return orphan; } /*! * \internal * \brief Check a node for an instance of an anonymous clone * * Return a child instance of the specified anonymous clone, in order of * preference: (1) the instance running on the specified node, if any; * (2) an inactive instance (i.e. within the total of \c PCMK_META_CLONE_MAX * instances); (3) a newly created orphan (that is, \c PCMK_META_CLONE_MAX * instances are already active). * * \param[in,out] scheduler Scheduler data * \param[in] node Node on which to check for instance * \param[in,out] parent Clone to check * \param[in] rsc_id Name of cloned resource in history (no instance) */ static pcmk_resource_t * find_anonymous_clone(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, pcmk_resource_t *parent, const char *rsc_id) { GList *rIter = NULL; pcmk_resource_t *rsc = NULL; pcmk_resource_t *inactive_instance = NULL; gboolean skip_inactive = FALSE; CRM_ASSERT(pcmk__is_anonymous_clone(parent)); // Check for active (or partially active, for cloned groups) instance pcmk__rsc_trace(parent, "Looking for %s on %s in %s", rsc_id, pcmk__node_name(node), parent->id); for (rIter = parent->children; rsc == NULL && rIter; rIter = rIter->next) { GList *locations = NULL; pcmk_resource_t *child = rIter->data; /* Check whether this instance is already known to be active or pending * anywhere, at this stage of unpacking. Because this function is called * for a resource before the resource's individual operation history * entries are unpacked, locations will generally not contain the * desired node. * * However, there are three exceptions: * (1) when child is a cloned group and we have already unpacked the * history of another member of the group on the same node; * (2) when we've already unpacked the history of another numbered * instance on the same node (which can happen if * PCMK_META_GLOBALLY_UNIQUE was flipped from true to false); and * (3) when we re-run calculations on the same scheduler data as part of * a simulation. */ child->private->fns->location(child, &locations, 2); if (locations) { /* We should never associate the same numbered anonymous clone * instance with multiple nodes, and clone instances can't migrate, * so there must be only one location, regardless of history. */ CRM_LOG_ASSERT(locations->next == NULL); if (pcmk__same_node((pcmk_node_t *) locations->data, node)) { /* This child instance is active on the requested node, so check * for a corresponding configured resource. We use find_rsc() * instead of child because child may be a cloned group, and we * need the particular member corresponding to rsc_id. * * If the history entry is orphaned, rsc will be NULL. */ rsc = parent->private->fns->find_rsc(child, rsc_id, NULL, pcmk_rsc_match_clone_only); if (rsc) { /* If there are multiple instance history entries for an * anonymous clone in a single node's history (which can * happen if PCMK_META_GLOBALLY_UNIQUE is switched from true * to false), we want to consider the instances beyond the * first as orphans, even if there are inactive instance * numbers available. */ - if (rsc->running_on) { + if (rsc->private->active_nodes != NULL) { crm_notice("Active (now-)anonymous clone %s has " "multiple (orphan) instance histories on %s", parent->id, pcmk__node_name(node)); skip_inactive = TRUE; rsc = NULL; } else { pcmk__rsc_trace(parent, "Resource %s, active", rsc->id); } } } g_list_free(locations); } else { pcmk__rsc_trace(parent, "Resource %s, skip inactive", child->id); if (!skip_inactive && !inactive_instance && !pcmk_is_set(child->flags, pcmk__rsc_blocked)) { // Remember one inactive instance in case we don't find active inactive_instance = parent->private->fns->find_rsc(child, rsc_id, NULL, pcmk_rsc_match_clone_only); /* ... but don't use it if it was already associated with a * pending action on another node */ if ((inactive_instance != NULL) && (inactive_instance->pending_node != NULL) && !pcmk__same_node(inactive_instance->pending_node, node)) { inactive_instance = NULL; } } } } if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) { pcmk__rsc_trace(parent, "Resource %s, empty slot", inactive_instance->id); rsc = inactive_instance; } /* If the resource has PCMK_META_REQUIRES set to PCMK_VALUE_QUORUM or * PCMK_VALUE_NOTHING, and we don't have a clone instance for every node, we * don't want to consume a valid instance number for unclean nodes. Such * instances may appear to be active according to the history, but should be * considered inactive, so we can start an instance elsewhere. Treat such * instances as orphans. * * An exception is instances running on guest nodes -- since guest node * "fencing" is actually just a resource stop, requires shouldn't apply. * * @TODO Ideally, we'd use an inactive instance number if it is not needed * for any clean instances. However, we don't know that at this point. */ if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing) && (!node->details->online || node->details->unclean) && !pcmk__is_guest_or_bundle_node(node) && !pe__is_universal_clone(parent, scheduler)) { rsc = NULL; } if (rsc == NULL) { rsc = create_anonymous_orphan(parent, rsc_id, node, scheduler); pcmk__rsc_trace(parent, "Resource %s, orphan", rsc->id); } return rsc; } static pcmk_resource_t * unpack_find_resource(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, const char *rsc_id) { pcmk_resource_t *rsc = NULL; pcmk_resource_t *parent = NULL; crm_trace("looking for %s", rsc_id); rsc = pe_find_resource(scheduler->resources, rsc_id); if (rsc == NULL) { /* If we didn't find the resource by its name in the operation history, * check it again as a clone instance. Even when PCMK_META_CLONE_MAX=0, * we create a single :0 orphan to match against here. */ char *clone0_id = clone_zero(rsc_id); pcmk_resource_t *clone0 = pe_find_resource(scheduler->resources, clone0_id); if (clone0 && !pcmk_is_set(clone0->flags, pcmk__rsc_unique)) { rsc = clone0; parent = uber_parent(clone0); crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id); } else { crm_trace("%s is not known as %s either (orphan)", rsc_id, clone0_id); } free(clone0_id); } else if (rsc->private->variant > pcmk__rsc_variant_primitive) { crm_trace("Resource history for %s is orphaned " "because it is no longer primitive", rsc_id); return NULL; } else { parent = uber_parent(rsc); } if (pcmk__is_anonymous_clone(parent)) { if (pcmk__is_bundled(parent)) { rsc = pe__find_bundle_replica(parent->private->parent, node); } else { char *base = clone_strip(rsc_id); rsc = find_anonymous_clone(scheduler, node, parent, base); free(base); CRM_ASSERT(rsc != NULL); } } if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_none) && !pcmk__str_eq(rsc_id, rsc->private->history_id, pcmk__str_none)) { pcmk__str_update(&(rsc->private->history_id), rsc_id); pcmk__rsc_debug(rsc, "Internally renamed %s on %s to %s%s", rsc_id, pcmk__node_name(node), rsc->id, pcmk_is_set(rsc->flags, pcmk__rsc_removed)? " (ORPHAN)" : ""); } return rsc; } static pcmk_resource_t * process_orphan_resource(const xmlNode *rsc_entry, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = NULL; const char *rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID); crm_debug("Detected orphan resource %s on %s", rsc_id, pcmk__node_name(node)); rsc = create_fake_resource(rsc_id, rsc_entry, scheduler); if (rsc == NULL) { return NULL; } if (!pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); } else { CRM_CHECK(rsc != NULL, return NULL); pcmk__rsc_trace(rsc, "Added orphan %s", rsc->id); resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__orphan_do_not_run__", scheduler); } return rsc; } static void process_rsc_state(pcmk_resource_t *rsc, pcmk_node_t *node, enum action_fail_response on_fail) { pcmk_node_t *tmpnode = NULL; char *reason = NULL; enum action_fail_response save_on_fail = pcmk_on_fail_ignore; pcmk_scheduler_t *scheduler = NULL; CRM_ASSERT(rsc); scheduler = rsc->private->scheduler; pcmk__rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s", rsc->id, pcmk_role_text(rsc->role), pcmk__node_name(node), pcmk_on_fail_text(on_fail)); /* process current state */ if (rsc->role != pcmk_role_unknown) { pcmk_resource_t *iter = rsc; while (iter) { if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) { pcmk_node_t *n = pe__copy_node(node); pcmk__rsc_trace(rsc, "%s (%s in history) known on %s", rsc->id, pcmk__s(rsc->private->history_id, "the same"), pcmk__node_name(n)); g_hash_table_insert(iter->known_on, (gpointer) n->details->id, n); } if (pcmk_is_set(iter->flags, pcmk__rsc_unique)) { break; } iter = iter->private->parent; } } /* If a managed resource is believed to be running, but node is down ... */ if ((rsc->role > pcmk_role_stopped) && node->details->online == FALSE && node->details->maintenance == FALSE && pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { gboolean should_fence = FALSE; /* If this is a guest node, fence it (regardless of whether fencing is * enabled, because guest node fencing is done by recovery of the * container resource rather than by the fencer). Mark the resource * we're processing as failed. When the guest comes back up, its * operation history in the CIB will be cleared, freeing the affected * resource to run again once we are sure we know its state. */ if (pcmk__is_guest_or_bundle_node(node)) { pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); should_fence = TRUE; } else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { if (pcmk__is_remote_node(node) && (node->details->remote_rsc != NULL) && !pcmk_is_set(node->details->remote_rsc->flags, pcmk__rsc_failed)) { /* Setting unseen means that fencing of the remote node will * occur only if the connection resource is not going to start * somewhere. This allows connection resources on a failed * cluster node to move to another node without requiring the * remote nodes to be fenced as well. */ node->details->unseen = TRUE; reason = crm_strdup_printf("%s is active there (fencing will be" " revoked if remote connection can " "be re-established elsewhere)", rsc->id); } should_fence = TRUE; } if (should_fence) { if (reason == NULL) { reason = crm_strdup_printf("%s is thought to be active there", rsc->id); } pe_fence_node(scheduler, node, reason, FALSE); } free(reason); } /* In order to calculate priority_fencing_delay correctly, save the failure information and pass it to native_add_running(). */ save_on_fail = on_fail; if (node->details->unclean) { /* No extra processing needed * Also allows resources to be started again after a node is shot */ on_fail = pcmk_on_fail_ignore; } switch (on_fail) { case pcmk_on_fail_ignore: /* nothing to do */ break; case pcmk_on_fail_demote: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed); demote_action(rsc, node, FALSE); break; case pcmk_on_fail_fence_node: /* treat it as if it is still running * but also mark the node as unclean */ reason = crm_strdup_printf("%s failed there", rsc->id); pe_fence_node(scheduler, node, reason, FALSE); free(reason); break; case pcmk_on_fail_standby_node: node->details->standby = TRUE; node->details->standby_onfail = TRUE; break; case pcmk_on_fail_block: /* is_managed == FALSE will prevent any * actions being sent for the resource */ pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(rsc, pcmk__rsc_blocked); break; case pcmk_on_fail_ban: /* make sure it comes up somewhere else * or not at all */ resource_location(rsc, node, -PCMK_SCORE_INFINITY, "__action_migration_auto__", scheduler); break; case pcmk_on_fail_stop: pe__set_next_role(rsc, pcmk_role_stopped, PCMK_META_ON_FAIL "=" PCMK_VALUE_STOP); break; case pcmk_on_fail_restart: if ((rsc->role != pcmk_role_stopped) && (rsc->role != pcmk_role_unknown)) { pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); stop_action(rsc, node, FALSE); } break; case pcmk_on_fail_restart_container: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); if ((rsc->container != NULL) && pcmk__is_bundled(rsc)) { /* A bundle's remote connection can run on a different node than * the bundle's container. We don't necessarily know where the * container is running yet, so remember it and add a stop * action for it later. */ scheduler->stop_needed = g_list_prepend(scheduler->stop_needed, rsc->container); } else if (rsc->container) { stop_action(rsc->container, node, FALSE); } else if ((rsc->role != pcmk_role_stopped) && (rsc->role != pcmk_role_unknown)) { stop_action(rsc, node, FALSE); } break; case pcmk_on_fail_reset_remote: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { tmpnode = NULL; if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { tmpnode = pcmk_find_node(scheduler, rsc->id); } if (pcmk__is_remote_node(tmpnode) && !(tmpnode->details->remote_was_fenced)) { /* The remote connection resource failed in a way that * should result in fencing the remote node. */ pe_fence_node(scheduler, tmpnode, "remote connection is unrecoverable", FALSE); } } /* require the stop action regardless if fencing is occurring or not. */ if (rsc->role > pcmk_role_stopped) { stop_action(rsc, node, FALSE); } /* if reconnect delay is in use, prevent the connection from exiting the * "STOPPED" role until the failure is cleared by the delay timeout. */ if (rsc->private->remote_reconnect_ms > 0U) { pe__set_next_role(rsc, pcmk_role_stopped, "remote reset"); } break; } /* ensure a remote-node connection failure forces an unclean remote-node * to be fenced. By setting unseen = FALSE, the remote-node failure will * result in a fencing operation regardless if we're going to attempt to * reconnect to the remote-node in this transition or not. */ if (pcmk_all_flags_set(rsc->flags, pcmk__rsc_failed|pcmk__rsc_is_remote_connection)) { tmpnode = pcmk_find_node(scheduler, rsc->id); if (tmpnode && tmpnode->details->unclean) { tmpnode->details->unseen = FALSE; } } if ((rsc->role != pcmk_role_stopped) && (rsc->role != pcmk_role_unknown)) { if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { crm_notice("Removed resource %s is active on %s and will be " "stopped when possible", rsc->id, pcmk__node_name(node)); } else { crm_notice("Removed resource %s must be stopped manually on %s " "because " PCMK_OPT_STOP_ORPHAN_RESOURCES " is set to false", rsc->id, pcmk__node_name(node)); } } native_add_running(rsc, node, scheduler, (save_on_fail != pcmk_on_fail_ignore)); switch (on_fail) { case pcmk_on_fail_ignore: break; case pcmk_on_fail_demote: case pcmk_on_fail_block: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed); break; default: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); break; } } else if ((rsc->private->history_id != NULL) && (strchr(rsc->private->history_id, ':') != NULL)) { /* Only do this for older status sections that included instance numbers * Otherwise stopped instances will appear as orphans */ pcmk__rsc_trace(rsc, "Clearing history ID %s for %s (stopped)", rsc->private->history_id, rsc->id); free(rsc->private->history_id); rsc->private->history_id = NULL; } else { GList *possible_matches = pe__resource_actions(rsc, node, PCMK_ACTION_STOP, FALSE); GList *gIter = possible_matches; for (; gIter != NULL; gIter = gIter->next) { pcmk_action_t *stop = (pcmk_action_t *) gIter->data; pcmk__set_action_flags(stop, pcmk_action_optional); } g_list_free(possible_matches); } /* A successful stop after migrate_to on the migration source doesn't make * the partially migrated resource stopped on the migration target. */ if ((rsc->role == pcmk_role_stopped) + && (rsc->private->active_nodes != NULL) && (rsc->private->partial_migration_target != NULL) - && pcmk__same_node(rsc->private->partial_migration_source, node) - && rsc->running_on) { + && pcmk__same_node(rsc->private->partial_migration_source, node)) { rsc->role = pcmk_role_started; } } /* create active recurring operations as optional */ static void process_recurring(pcmk_node_t *node, pcmk_resource_t *rsc, int start_index, int stop_index, GList *sorted_op_list, pcmk_scheduler_t *scheduler) { int counter = -1; const char *task = NULL; const char *status = NULL; GList *gIter = sorted_op_list; CRM_ASSERT(rsc); pcmk__rsc_trace(rsc, "%s: Start index %d, stop index = %d", rsc->id, start_index, stop_index); for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; guint interval_ms = 0; char *key = NULL; const char *id = pcmk__xe_id(rsc_op); counter++; if (node->details->online == FALSE) { pcmk__rsc_trace(rsc, "Skipping %s on %s: node is offline", rsc->id, pcmk__node_name(node)); break; /* Need to check if there's a monitor for role="Stopped" */ } else if (start_index < stop_index && counter <= stop_index) { pcmk__rsc_trace(rsc, "Skipping %s on %s: resource is not active", id, pcmk__node_name(node)); continue; } else if (counter < start_index) { pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d", id, pcmk__node_name(node), counter); continue; } crm_element_value_ms(rsc_op, PCMK_META_INTERVAL, &interval_ms); if (interval_ms == 0) { pcmk__rsc_trace(rsc, "Skipping %s on %s: non-recurring", id, pcmk__node_name(node)); continue; } status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS); if (pcmk__str_eq(status, "-1", pcmk__str_casei)) { pcmk__rsc_trace(rsc, "Skipping %s on %s: status", id, pcmk__node_name(node)); continue; } task = crm_element_value(rsc_op, PCMK_XA_OPERATION); /* create the action */ key = pcmk__op_key(rsc->id, task, interval_ms); pcmk__rsc_trace(rsc, "Creating %s on %s", key, pcmk__node_name(node)); custom_action(rsc, key, task, node, TRUE, scheduler); } } void calculate_active_ops(const GList *sorted_op_list, int *start_index, int *stop_index) { int counter = -1; int implied_monitor_start = -1; int implied_clone_start = -1; const char *task = NULL; const char *status = NULL; *stop_index = -1; *start_index = -1; for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) { const xmlNode *rsc_op = (const xmlNode *) iter->data; counter++; task = crm_element_value(rsc_op, PCMK_XA_OPERATION); status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS); if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_casei) && pcmk__str_eq(status, "0", pcmk__str_casei)) { *stop_index = counter; } else if (pcmk__strcase_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { *start_index = counter; } else if ((implied_monitor_start <= *stop_index) && pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { const char *rc = crm_element_value(rsc_op, PCMK__XA_RC_CODE); if (pcmk__strcase_any_of(rc, "0", "8", NULL)) { implied_monitor_start = counter; } } else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, NULL)) { implied_clone_start = counter; } } if (*start_index == -1) { if (implied_clone_start != -1) { *start_index = implied_clone_start; } else if (implied_monitor_start != -1) { *start_index = implied_monitor_start; } } } // If resource history entry has shutdown lock, remember lock node and time static void unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { time_t lock_time = 0; // When lock started (i.e. node shutdown time) if ((crm_element_value_epoch(rsc_entry, PCMK_OPT_SHUTDOWN_LOCK, &lock_time) == pcmk_ok) && (lock_time != 0)) { if ((scheduler->shutdown_lock > 0) && (get_effective_time(scheduler) > (lock_time + scheduler->shutdown_lock))) { pcmk__rsc_info(rsc, "Shutdown lock for %s on %s expired", rsc->id, pcmk__node_name(node)); pe__clear_resource_history(rsc, node); } else { /* @COMPAT I don't like breaking const signatures, but * rsc->lock_node should really be const -- we just can't change it * until the next API compatibility break. */ rsc->lock_node = (pcmk_node_t *) node; rsc->lock_time = lock_time; } } } /*! * \internal * \brief Unpack one \c PCMK__XE_LRM_RESOURCE entry from a node's CIB status * * \param[in,out] node Node whose status is being unpacked * \param[in] rsc_entry \c PCMK__XE_LRM_RESOURCE XML being unpacked * \param[in,out] scheduler Scheduler data * * \return Resource corresponding to the entry, or NULL if no operation history */ static pcmk_resource_t * unpack_lrm_resource(pcmk_node_t *node, const xmlNode *lrm_resource, pcmk_scheduler_t *scheduler) { GList *gIter = NULL; int stop_index = -1; int start_index = -1; enum rsc_role_e req_role = pcmk_role_unknown; const char *rsc_id = pcmk__xe_id(lrm_resource); pcmk_resource_t *rsc = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; xmlNode *rsc_op = NULL; xmlNode *last_failure = NULL; enum action_fail_response on_fail = pcmk_on_fail_ignore; enum rsc_role_e saved_role = pcmk_role_unknown; if (rsc_id == NULL) { pcmk__config_err("Ignoring invalid " PCMK__XE_LRM_RESOURCE " entry: No " PCMK_XA_ID); crm_log_xml_info(lrm_resource, "missing-id"); return NULL; } crm_trace("Unpacking " PCMK__XE_LRM_RESOURCE " for %s on %s", rsc_id, pcmk__node_name(node)); /* Build a list of individual PCMK__XE_LRM_RSC_OP entries, so we can sort * them */ for (rsc_op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP, NULL, NULL); rsc_op != NULL; rsc_op = pcmk__xe_next_same(rsc_op)) { op_list = g_list_prepend(op_list, rsc_op); } if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } } /* find the resource */ rsc = unpack_find_resource(scheduler, node, rsc_id); if (rsc == NULL) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } else { rsc = process_orphan_resource(lrm_resource, node, scheduler); } } CRM_ASSERT(rsc != NULL); // Check whether the resource is "shutdown-locked" to this node if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { unpack_shutdown_lock(lrm_resource, rsc, node, scheduler); } /* process operations */ saved_role = rsc->role; rsc->role = pcmk_role_unknown; sorted_op_list = g_list_sort(op_list, sort_op_by_callid); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail); } /* create active recurring operations as optional */ calculate_active_ops(sorted_op_list, &start_index, &stop_index); process_recurring(node, rsc, start_index, stop_index, sorted_op_list, scheduler); /* no need to free the contents */ g_list_free(sorted_op_list); process_rsc_state(rsc, node, on_fail); if (get_target_role(rsc, &req_role)) { if ((rsc->next_role == pcmk_role_unknown) || (req_role < rsc->next_role)) { pe__set_next_role(rsc, req_role, PCMK_META_TARGET_ROLE); } else if (req_role > rsc->next_role) { pcmk__rsc_info(rsc, "%s: Not overwriting calculated next role %s" " with requested next role %s", rsc->id, pcmk_role_text(rsc->next_role), pcmk_role_text(req_role)); } } if (saved_role > rsc->role) { rsc->role = saved_role; } return rsc; } static void handle_orphaned_container_fillers(const xmlNode *lrm_rsc_list, pcmk_scheduler_t *scheduler) { for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list, NULL, NULL, NULL); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) { pcmk_resource_t *rsc; pcmk_resource_t *container; const char *rsc_id; const char *container_id; if (!pcmk__xe_is(rsc_entry, PCMK__XE_LRM_RESOURCE)) { continue; } container_id = crm_element_value(rsc_entry, PCMK__META_CONTAINER); rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID); if (container_id == NULL || rsc_id == NULL) { continue; } container = pe_find_resource(scheduler->resources, container_id); if (container == NULL) { continue; } rsc = pe_find_resource(scheduler->resources, rsc_id); if ((rsc == NULL) || (rsc->container != NULL) || !pcmk_is_set(rsc->flags, pcmk__rsc_removed_filler)) { continue; } pcmk__rsc_trace(rsc, "Mapped container of orphaned resource %s to %s", rsc->id, container_id); rsc->container = container; container->fillers = g_list_append(container->fillers, rsc); } } /*! * \internal * \brief Unpack one node's lrm status section * * \param[in,out] node Node whose status is being unpacked * \param[in] xml CIB node state XML * \param[in,out] scheduler Scheduler data */ static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml, pcmk_scheduler_t *scheduler) { bool found_orphaned_container_filler = false; // Drill down to PCMK__XE_LRM_RESOURCES section xml = pcmk__xe_first_child(xml, PCMK__XE_LRM, NULL, NULL); if (xml == NULL) { return; } xml = pcmk__xe_first_child(xml, PCMK__XE_LRM_RESOURCES, NULL, NULL); if (xml == NULL) { return; } // Unpack each PCMK__XE_LRM_RESOURCE entry for (const xmlNode *rsc_entry = pcmk__xe_first_child(xml, PCMK__XE_LRM_RESOURCE, NULL, NULL); rsc_entry != NULL; rsc_entry = pcmk__xe_next_same(rsc_entry)) { pcmk_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, scheduler); if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed_filler)) { found_orphaned_container_filler = true; } } /* Now that all resource state has been unpacked for this node, map any * orphaned container fillers to their container resource. */ if (found_orphaned_container_filler) { handle_orphaned_container_fillers(xml, scheduler); } } static void set_active(pcmk_resource_t *rsc) { const pcmk_resource_t *top = pe__const_top_resource(rsc, false); if (top && pcmk_is_set(top->flags, pcmk__rsc_promotable)) { rsc->role = pcmk_role_unpromoted; } else { rsc->role = pcmk_role_started; } } static void set_node_score(gpointer key, gpointer value, gpointer user_data) { pcmk_node_t *node = value; int *score = user_data; node->weight = *score; } #define XPATH_NODE_STATE "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE #define SUB_XPATH_LRM_RESOURCE "/" PCMK__XE_LRM \ "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE #define SUB_XPATH_LRM_RSC_OP "/" PCMK__XE_LRM_RSC_OP static xmlNode * find_lrm_op(const char *resource, const char *op, const char *node, const char *source, int target_rc, pcmk_scheduler_t *scheduler) { GString *xpath = NULL; xmlNode *xml = NULL; CRM_CHECK((resource != NULL) && (op != NULL) && (node != NULL), return NULL); xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node, "']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", resource, "']" SUB_XPATH_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='", op, "'", NULL); /* Need to check against transition_magic too? */ if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_TO) == 0)) { pcmk__g_strcat(xpath, " and @" PCMK__META_MIGRATE_TARGET "='", source, "']", NULL); } else if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) { pcmk__g_strcat(xpath, " and @" PCMK__META_MIGRATE_SOURCE "='", source, "']", NULL); } else { g_string_append_c(xpath, ']'); } xml = get_xpath_object((const char *) xpath->str, scheduler->input, LOG_DEBUG); g_string_free(xpath, TRUE); if (xml && target_rc >= 0) { int rc = PCMK_OCF_UNKNOWN_ERROR; int status = PCMK_EXEC_ERROR; crm_element_value_int(xml, PCMK__XA_RC_CODE, &rc); crm_element_value_int(xml, PCMK__XA_OP_STATUS, &status); if ((rc != target_rc) || (status != PCMK_EXEC_DONE)) { return NULL; } } return xml; } static xmlNode * find_lrm_resource(const char *rsc_id, const char *node_name, pcmk_scheduler_t *scheduler) { GString *xpath = NULL; xmlNode *xml = NULL; CRM_CHECK((rsc_id != NULL) && (node_name != NULL), return NULL); xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node_name, "']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", rsc_id, "']", NULL); xml = get_xpath_object((const char *) xpath->str, scheduler->input, LOG_DEBUG); g_string_free(xpath, TRUE); return xml; } /*! * \internal * \brief Check whether a resource has no completed action history on a node * * \param[in,out] rsc Resource to check * \param[in] node_name Node to check * * \return true if \p rsc_id is unknown on \p node_name, otherwise false */ static bool unknown_on_node(pcmk_resource_t *rsc, const char *node_name) { bool result = false; xmlXPathObjectPtr search; char *xpath = NULL; xpath = crm_strdup_printf(XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" SUB_XPATH_LRM_RSC_OP "[@" PCMK__XA_RC_CODE "!='%d']", node_name, rsc->id, PCMK_OCF_UNKNOWN); search = xpath_search(rsc->private->scheduler->input, xpath); result = (numXpathResults(search) == 0); freeXpathObject(search); free(xpath); return result; } /*! * \brief Check whether a probe/monitor indicating the resource was not running * on a node happened after some event * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] xml_op Event that monitor is being compared to * \param[in] same_node Whether the operations are on the same node * \param[in,out] scheduler Scheduler data * * \return true if such a monitor happened after event, false otherwise */ static bool monitor_not_running_after(const char *rsc_id, const char *node_name, const xmlNode *xml_op, bool same_node, pcmk_scheduler_t *scheduler) { /* Any probe/monitor operation on the node indicating it was not running * there */ xmlNode *monitor = find_lrm_op(rsc_id, PCMK_ACTION_MONITOR, node_name, NULL, PCMK_OCF_NOT_RUNNING, scheduler); return (monitor && pe__is_newer_op(monitor, xml_op, same_node) > 0); } /*! * \brief Check whether any non-monitor operation on a node happened after some * event * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] xml_op Event that non-monitor is being compared to * \param[in] same_node Whether the operations are on the same node * \param[in,out] scheduler Scheduler data * * \return true if such a operation happened after event, false otherwise */ static bool non_monitor_after(const char *rsc_id, const char *node_name, const xmlNode *xml_op, bool same_node, pcmk_scheduler_t *scheduler) { xmlNode *lrm_resource = NULL; lrm_resource = find_lrm_resource(rsc_id, node_name, scheduler); if (lrm_resource == NULL) { return false; } for (xmlNode *op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP, NULL, NULL); op != NULL; op = pcmk__xe_next_same(op)) { const char * task = NULL; if (op == xml_op) { continue; } task = crm_element_value(op, PCMK_XA_OPERATION); if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL) && pe__is_newer_op(op, xml_op, same_node) > 0) { return true; } } return false; } /*! * \brief Check whether the resource has newer state on a node after a migration * attempt * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] migrate_to Any migrate_to event that is being compared to * \param[in] migrate_from Any migrate_from event that is being compared to * \param[in,out] scheduler Scheduler data * * \return true if such a operation happened after event, false otherwise */ static bool newer_state_after_migrate(const char *rsc_id, const char *node_name, const xmlNode *migrate_to, const xmlNode *migrate_from, pcmk_scheduler_t *scheduler) { const xmlNode *xml_op = migrate_to; const char *source = NULL; const char *target = NULL; bool same_node = false; if (migrate_from) { xml_op = migrate_from; } source = crm_element_value(xml_op, PCMK__META_MIGRATE_SOURCE); target = crm_element_value(xml_op, PCMK__META_MIGRATE_TARGET); /* It's preferred to compare to the migrate event on the same node if * existing, since call ids are more reliable. */ if (pcmk__str_eq(node_name, target, pcmk__str_casei)) { if (migrate_from) { xml_op = migrate_from; same_node = true; } else { xml_op = migrate_to; } } else if (pcmk__str_eq(node_name, source, pcmk__str_casei)) { if (migrate_to) { xml_op = migrate_to; same_node = true; } else { xml_op = migrate_from; } } /* If there's any newer non-monitor operation on the node, or any newer * probe/monitor operation on the node indicating it was not running there, * the migration events potentially no longer matter for the node. */ return non_monitor_after(rsc_id, node_name, xml_op, same_node, scheduler) || monitor_not_running_after(rsc_id, node_name, xml_op, same_node, scheduler); } /*! * \internal * \brief Parse migration source and target node names from history entry * * \param[in] entry Resource history entry for a migration action * \param[in] source_node If not NULL, source must match this node * \param[in] target_node If not NULL, target must match this node * \param[out] source_name Where to store migration source node name * \param[out] target_name Where to store migration target node name * * \return Standard Pacemaker return code */ static int get_migration_node_names(const xmlNode *entry, const pcmk_node_t *source_node, const pcmk_node_t *target_node, const char **source_name, const char **target_name) { *source_name = crm_element_value(entry, PCMK__META_MIGRATE_SOURCE); *target_name = crm_element_value(entry, PCMK__META_MIGRATE_TARGET); if ((*source_name == NULL) || (*target_name == NULL)) { pcmk__config_err("Ignoring resource history entry %s without " PCMK__META_MIGRATE_SOURCE " and " PCMK__META_MIGRATE_TARGET, pcmk__xe_id(entry)); return pcmk_rc_unpack_error; } if ((source_node != NULL) && !pcmk__str_eq(*source_name, source_node->details->uname, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_err("Ignoring resource history entry %s because " PCMK__META_MIGRATE_SOURCE "='%s' does not match %s", pcmk__xe_id(entry), *source_name, pcmk__node_name(source_node)); return pcmk_rc_unpack_error; } if ((target_node != NULL) && !pcmk__str_eq(*target_name, target_node->details->uname, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_err("Ignoring resource history entry %s because " PCMK__META_MIGRATE_TARGET "='%s' does not match %s", pcmk__xe_id(entry), *target_name, pcmk__node_name(target_node)); return pcmk_rc_unpack_error; } return pcmk_rc_ok; } /* * \internal * \brief Add a migration source to a resource's list of dangling migrations * * If the migrate_to and migrate_from actions in a live migration both * succeeded, but there is no stop on the source, the migration is considered * "dangling." Add the source to the resource's dangling migration list, which * will be used to schedule a stop on the source without affecting the target. * * \param[in,out] rsc Resource involved in migration * \param[in] node Migration source */ static void add_dangling_migration(pcmk_resource_t *rsc, const pcmk_node_t *node) { pcmk__rsc_trace(rsc, "Dangling migration of %s requires stop on %s", rsc->id, pcmk__node_name(node)); rsc->role = pcmk_role_stopped; rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, (gpointer) node); } /*! * \internal * \brief Update resource role etc. after a successful migrate_to action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_to_success(struct action_history *history) { /* A complete migration sequence is: * 1. migrate_to on source node (which succeeded if we get to this function) * 2. migrate_from on target node * 3. stop on source node * * If no migrate_from has happened, the migration is considered to be * "partial". If the migrate_from succeeded but no stop has happened, the * migration is considered to be "dangling". * * If a successful migrate_to and stop have happened on the source node, we * still need to check for a partial migration, due to scenarios (easier to * produce with batch-limit=1) like: * * - A resource is migrating from node1 to node2, and a migrate_to is * initiated for it on node1. * * - node2 goes into standby mode while the migrate_to is pending, which * aborts the transition. * * - Upon completion of the migrate_to, a new transition schedules a stop * on both nodes and a start on node1. * * - If the new transition is aborted for any reason while the resource is * stopping on node1, the transition after that stop completes will see * the migrate_to and stop on the source, but it's still a partial * migration, and the resource must be stopped on node2 because it is * potentially active there due to the migrate_to. * * We also need to take into account that either node's history may be * cleared at any point in the migration process. */ int from_rc = PCMK_OCF_OK; int from_status = PCMK_EXEC_PENDING; pcmk_node_t *target_node = NULL; xmlNode *migrate_from = NULL; const char *source = NULL; const char *target = NULL; bool source_newer_op = false; bool target_newer_state = false; bool active_on_target = false; pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; // Get source and target node names from XML if (get_migration_node_names(history->xml, history->node, NULL, &source, &target) != pcmk_rc_ok) { return; } // Check for newer state on the source source_newer_op = non_monitor_after(history->rsc->id, source, history->xml, true, scheduler); // Check for a migrate_from action from this source on the target migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM, target, source, -1, scheduler); if (migrate_from != NULL) { if (source_newer_op) { /* There's a newer non-monitor operation on the source and a * migrate_from on the target, so this migrate_to is irrelevant to * the resource's state. */ return; } crm_element_value_int(migrate_from, PCMK__XA_RC_CODE, &from_rc); crm_element_value_int(migrate_from, PCMK__XA_OP_STATUS, &from_status); } /* If the resource has newer state on both the source and target after the * migration events, this migrate_to is irrelevant to the resource's state. */ target_newer_state = newer_state_after_migrate(history->rsc->id, target, history->xml, migrate_from, scheduler); if (source_newer_op && target_newer_state) { return; } /* Check for dangling migration (migrate_from succeeded but stop not done). * We know there's no stop because we already returned if the target has a * migrate_from and the source has any newer non-monitor operation. */ if ((from_rc == PCMK_OCF_OK) && (from_status == PCMK_EXEC_DONE)) { add_dangling_migration(history->rsc, history->node); return; } /* Without newer state, this migrate_to implies the resource is active. * (Clones are not allowed to migrate, so role can't be promoted.) */ history->rsc->role = pcmk_role_started; target_node = pcmk_find_node(scheduler, target); active_on_target = !target_newer_state && (target_node != NULL) && target_node->details->online; if (from_status != PCMK_EXEC_PENDING) { // migrate_from failed on target if (active_on_target) { native_add_running(history->rsc, target_node, scheduler, TRUE); } else { // Mark resource as failed, require recovery, and prevent migration pcmk__set_rsc_flags(history->rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable); } return; } // The migrate_from is pending, complete but erased, or to be scheduled /* If there is no history at all for the resource on an online target, then * it was likely cleaned. Just return, and we'll schedule a probe. Once we * have the probe result, it will be reflected in target_newer_state. */ if ((target_node != NULL) && target_node->details->online && unknown_on_node(history->rsc, target)) { return; } if (active_on_target) { pcmk_node_t *source_node = pcmk_find_node(scheduler, source); native_add_running(history->rsc, target_node, scheduler, FALSE); if ((source_node != NULL) && source_node->details->online) { /* This is a partial migration: the migrate_to completed * successfully on the source, but the migrate_from has not * completed. Remember the source and target; if the newly * chosen target remains the same when we schedule actions * later, we may continue with the migration. */ history->rsc->private->partial_migration_target = target_node; history->rsc->private->partial_migration_source = source_node; } } else if (!source_newer_op) { // Mark resource as failed, require recovery, and prevent migration pcmk__set_rsc_flags(history->rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable); } } /*! * \internal * \brief Update resource role etc. after a failed migrate_to action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_to_failure(struct action_history *history) { xmlNode *target_migrate_from = NULL; const char *source = NULL; const char *target = NULL; pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; // Get source and target node names from XML if (get_migration_node_names(history->xml, history->node, NULL, &source, &target) != pcmk_rc_ok) { return; } /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ history->rsc->role = pcmk_role_started; // Check for migrate_from on the target target_migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM, target, source, PCMK_OCF_OK, scheduler); if (/* If the resource state is unknown on the target, it will likely be * probed there. * Don't just consider it running there. We will get back here anyway in * case the probe detects it's running there. */ !unknown_on_node(history->rsc, target) /* If the resource has newer state on the target after the migration * events, this migrate_to no longer matters for the target. */ && !newer_state_after_migrate(history->rsc->id, target, history->xml, target_migrate_from, scheduler)) { /* The resource has no newer state on the target, so assume it's still * active there. * (if it is up). */ pcmk_node_t *target_node = pcmk_find_node(scheduler, target); if (target_node && target_node->details->online) { native_add_running(history->rsc, target_node, scheduler, FALSE); } } else if (!non_monitor_after(history->rsc->id, source, history->xml, true, scheduler)) { /* We know the resource has newer state on the target, but this * migrate_to still matters for the source as long as there's no newer * non-monitor operation there. */ // Mark node as having dangling migration so we can force a stop later history->rsc->dangling_migrations = g_list_prepend(history->rsc->dangling_migrations, (gpointer) history->node); } } /*! * \internal * \brief Update resource role etc. after a failed migrate_from action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_from_failure(struct action_history *history) { xmlNode *source_migrate_to = NULL; const char *source = NULL; const char *target = NULL; pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; // Get source and target node names from XML if (get_migration_node_names(history->xml, NULL, history->node, &source, &target) != pcmk_rc_ok) { return; } /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ history->rsc->role = pcmk_role_started; // Check for a migrate_to on the source source_migrate_to = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_TO, source, target, PCMK_OCF_OK, scheduler); if (/* If the resource state is unknown on the source, it will likely be * probed there. * Don't just consider it running there. We will get back here anyway in * case the probe detects it's running there. */ !unknown_on_node(history->rsc, source) /* If the resource has newer state on the source after the migration * events, this migrate_from no longer matters for the source. */ && !newer_state_after_migrate(history->rsc->id, source, source_migrate_to, history->xml, scheduler)) { /* The resource has no newer state on the source, so assume it's still * active there (if it is up). */ pcmk_node_t *source_node = pcmk_find_node(scheduler, source); if (source_node && source_node->details->online) { native_add_running(history->rsc, source_node, scheduler, TRUE); } } } /*! * \internal * \brief Add an action to cluster's list of failed actions * * \param[in,out] history Parsed action result history */ static void record_failed_op(struct action_history *history) { const pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; if (!(history->node->details->online)) { return; } for (const xmlNode *xIter = scheduler->failed->children; xIter != NULL; xIter = xIter->next) { const char *key = pcmk__xe_history_key(xIter); const char *uname = crm_element_value(xIter, PCMK_XA_UNAME); if (pcmk__str_eq(history->key, key, pcmk__str_none) && pcmk__str_eq(uname, history->node->details->uname, pcmk__str_casei)) { crm_trace("Skipping duplicate entry %s on %s", history->key, pcmk__node_name(history->node)); return; } } crm_trace("Adding entry for %s on %s to failed action list", history->key, pcmk__node_name(history->node)); crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->details->uname); crm_xml_add(history->xml, PCMK__XA_RSC_ID, history->rsc->id); pcmk__xml_copy(scheduler->failed, history->xml); } static char * last_change_str(const xmlNode *xml_op) { time_t when; char *result = NULL; if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &when) == pcmk_ok) { char *when_s = pcmk__epoch2str(&when, 0); const char *p = strchr(when_s, ' '); // Skip day of week to make message shorter if ((p != NULL) && (*(++p) != '\0')) { result = pcmk__str_copy(p); } free(when_s); } if (result == NULL) { result = pcmk__str_copy("unknown_time"); } return result; } /*! * \internal * \brief Compare two on-fail values * * \param[in] first One on-fail value to compare * \param[in] second The other on-fail value to compare * * \return A negative number if second is more severe than first, zero if they * are equal, or a positive number if first is more severe than second. * \note This is only needed until the action_fail_response values can be * renumbered at the next API compatibility break. */ static int cmp_on_fail(enum action_fail_response first, enum action_fail_response second) { switch (first) { case pcmk_on_fail_demote: switch (second) { case pcmk_on_fail_ignore: return 1; case pcmk_on_fail_demote: return 0; default: return -1; } break; case pcmk_on_fail_reset_remote: switch (second) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: return 1; case pcmk_on_fail_reset_remote: return 0; default: return -1; } break; case pcmk_on_fail_restart_container: switch (second) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: case pcmk_on_fail_reset_remote: return 1; case pcmk_on_fail_restart_container: return 0; default: return -1; } break; default: break; } switch (second) { case pcmk_on_fail_demote: return (first == pcmk_on_fail_ignore)? -1 : 1; case pcmk_on_fail_reset_remote: switch (first) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: return -1; default: return 1; } break; case pcmk_on_fail_restart_container: switch (first) { case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: case pcmk_on_fail_reset_remote: return -1; default: return 1; } break; default: break; } return first - second; } /*! * \internal * \brief Ban a resource (or its clone if an anonymous instance) from all nodes * * \param[in,out] rsc Resource to ban */ static void ban_from_all_nodes(pcmk_resource_t *rsc) { int score = -PCMK_SCORE_INFINITY; pcmk_resource_t *fail_rsc = rsc; const pcmk_scheduler_t *scheduler = rsc->private->scheduler; if (fail_rsc->private->parent != NULL) { pcmk_resource_t *parent = uber_parent(fail_rsc); if (pcmk__is_anonymous_clone(parent)) { /* For anonymous clones, if an operation with * PCMK_META_ON_FAIL=PCMK_VALUE_STOP fails for any instance, the * entire clone must stop. */ fail_rsc = parent; } } // Ban the resource from all nodes crm_notice("%s will not be started under current conditions", fail_rsc->id); if (fail_rsc->allowed_nodes != NULL) { g_hash_table_destroy(fail_rsc->allowed_nodes); } fail_rsc->allowed_nodes = pe__node_list2table(scheduler->nodes); g_hash_table_foreach(fail_rsc->allowed_nodes, set_node_score, &score); } /*! * \internal * \brief Get configured failure handling and role after failure for an action * * \param[in,out] history Unpacked action history entry * \param[out] on_fail Where to set configured failure handling * \param[out] fail_role Where to set to role after failure */ static void unpack_failure_handling(struct action_history *history, enum action_fail_response *on_fail, enum rsc_role_e *fail_role) { xmlNode *config = pcmk__find_action_config(history->rsc, history->task, history->interval_ms, true); GHashTable *meta = pcmk__unpack_action_meta(history->rsc, history->node, history->task, history->interval_ms, config); const char *on_fail_str = g_hash_table_lookup(meta, PCMK_META_ON_FAIL); *on_fail = pcmk__parse_on_fail(history->rsc, history->task, history->interval_ms, on_fail_str); *fail_role = pcmk__role_after_failure(history->rsc, history->task, *on_fail, meta); g_hash_table_destroy(meta); } /*! * \internal * \brief Update resource role, failure handling, etc., after a failed action * * \param[in,out] history Parsed action result history * \param[in] config_on_fail Action failure handling from configuration * \param[in] fail_role Resource's role after failure of this action * \param[out] last_failure This will be set to the history XML * \param[in,out] on_fail Actual handling of action result */ static void unpack_rsc_op_failure(struct action_history *history, enum action_fail_response config_on_fail, enum rsc_role_e fail_role, xmlNode **last_failure, enum action_fail_response *on_fail) { bool is_probe = false; char *last_change_s = NULL; pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; *last_failure = history->xml; is_probe = pcmk_xe_is_probe(history->xml); last_change_s = last_change_str(history->xml); if (!pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster) && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) { crm_trace("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " QB_XS " exit-status=%d id=%s", services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), (is_probe? "probe" : history->task), history->rsc->id, pcmk__node_name(history->node), last_change_s, history->exit_status, history->id); } else { pcmk__sched_warn("Unexpected result (%s%s%s) was recorded for %s of " "%s on %s at %s " QB_XS " exit-status=%d id=%s", services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), (is_probe? "probe" : history->task), history->rsc->id, pcmk__node_name(history->node), last_change_s, history->exit_status, history->id); if (is_probe && (history->exit_status != PCMK_OCF_OK) && (history->exit_status != PCMK_OCF_NOT_RUNNING) && (history->exit_status != PCMK_OCF_RUNNING_PROMOTED)) { /* A failed (not just unexpected) probe result could mean the user * didn't know resources will be probed even where they can't run. */ crm_notice("If it is not possible for %s to run on %s, see " "the " PCMK_XA_RESOURCE_DISCOVERY " option for location " "constraints", history->rsc->id, pcmk__node_name(history->node)); } record_failed_op(history); } free(last_change_s); if (cmp_on_fail(*on_fail, config_on_fail) < 0) { pcmk__rsc_trace(history->rsc, "on-fail %s -> %s for %s", pcmk_on_fail_text(*on_fail), pcmk_on_fail_text(config_on_fail), history->key); *on_fail = config_on_fail; } if (strcmp(history->task, PCMK_ACTION_STOP) == 0) { resource_location(history->rsc, history->node, -PCMK_SCORE_INFINITY, "__stop_fail__", scheduler); } else if (strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) { unpack_migrate_to_failure(history); } else if (strcmp(history->task, PCMK_ACTION_MIGRATE_FROM) == 0) { unpack_migrate_from_failure(history); } else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) { history->rsc->role = pcmk_role_promoted; } else if (strcmp(history->task, PCMK_ACTION_DEMOTE) == 0) { if (config_on_fail == pcmk_on_fail_block) { history->rsc->role = pcmk_role_promoted; pe__set_next_role(history->rsc, pcmk_role_stopped, "demote with " PCMK_META_ON_FAIL "=block"); } else if (history->exit_status == PCMK_OCF_NOT_RUNNING) { history->rsc->role = pcmk_role_stopped; } else { /* Staying in the promoted role would put the scheduler and * controller into a loop. Setting the role to unpromoted is not * dangerous because the resource will be stopped as part of * recovery, and any promotion will be ordered after that stop. */ history->rsc->role = pcmk_role_unpromoted; } } if (is_probe && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) { /* leave stopped */ pcmk__rsc_trace(history->rsc, "Leaving %s stopped", history->rsc->id); history->rsc->role = pcmk_role_stopped; } else if (history->rsc->role < pcmk_role_started) { pcmk__rsc_trace(history->rsc, "Setting %s active", history->rsc->id); set_active(history->rsc); } pcmk__rsc_trace(history->rsc, "Resource %s: role=%s unclean=%s on_fail=%s fail_role=%s", history->rsc->id, pcmk_role_text(history->rsc->role), pcmk__btoa(history->node->details->unclean), pcmk_on_fail_text(config_on_fail), pcmk_role_text(fail_role)); if ((fail_role != pcmk_role_started) && (history->rsc->next_role < fail_role)) { pe__set_next_role(history->rsc, fail_role, "failure"); } if (fail_role == pcmk_role_stopped) { ban_from_all_nodes(history->rsc); } } /*! * \internal * \brief Block a resource with a failed action if it cannot be recovered * * If resource action is a failed stop and fencing is not possible, mark the * resource as unmanaged and blocked, since recovery cannot be done. * * \param[in,out] history Parsed action history entry */ static void block_if_unrecoverable(struct action_history *history) { char *last_change_s = NULL; if (strcmp(history->task, PCMK_ACTION_STOP) != 0) { return; // All actions besides stop are always recoverable } if (pe_can_fence(history->node->details->data_set, history->node)) { return; // Failed stops are recoverable via fencing } last_change_s = last_change_str(history->xml); pcmk__sched_err("No further recovery can be attempted for %s " "because %s on %s failed (%s%s%s) at %s " QB_XS " rc=%d id=%s", history->rsc->id, history->task, pcmk__node_name(history->node), services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), last_change_s, history->exit_status, history->id); free(last_change_s); pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(history->rsc, pcmk__rsc_blocked); } /*! * \internal * \brief Update action history's execution status and why * * \param[in,out] history Parsed action history entry * \param[out] why Where to store reason for update * \param[in] value New value * \param[in] reason Description of why value was changed */ static inline void remap_because(struct action_history *history, const char **why, int value, const char *reason) { if (history->execution_status != value) { history->execution_status = value; *why = reason; } } /*! * \internal * \brief Remap informational monitor results and operation status * * For the monitor results, certain OCF codes are for providing extended information * to the user about services that aren't yet failed but not entirely healthy either. * These must be treated as the "normal" result by Pacemaker. * * For operation status, the action result can be used to determine an appropriate * status for the purposes of responding to the action. The status provided by the * executor is not directly usable since the executor does not know what was expected. * * \param[in,out] history Parsed action history entry * \param[in,out] on_fail What should be done about the result * \param[in] expired Whether result is expired * * \note If the result is remapped and the node is not shutting down or failed, * the operation will be recorded in the scheduler data's list of failed * operations to highlight it for the user. * * \note This may update the resource's current and next role. */ static void remap_operation(struct action_history *history, enum action_fail_response *on_fail, bool expired) { bool is_probe = false; int orig_exit_status = history->exit_status; int orig_exec_status = history->execution_status; const char *why = NULL; const char *task = history->task; // Remap degraded results to their successful counterparts history->exit_status = pcmk__effective_rc(history->exit_status); if (history->exit_status != orig_exit_status) { why = "degraded result"; if (!expired && (!history->node->details->shutdown || history->node->details->online)) { record_failed_op(history); } } if (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml) && ((history->execution_status != PCMK_EXEC_DONE) || (history->exit_status != PCMK_OCF_NOT_RUNNING))) { history->execution_status = PCMK_EXEC_DONE; history->exit_status = PCMK_OCF_NOT_RUNNING; why = "equivalent probe result"; } /* If the executor reported an execution status of anything but done or * error, consider that final. But for done or error, we know better whether * it should be treated as a failure or not, because we know the expected * result. */ switch (history->execution_status) { case PCMK_EXEC_DONE: case PCMK_EXEC_ERROR: break; // These should be treated as node-fatal case PCMK_EXEC_NO_FENCE_DEVICE: case PCMK_EXEC_NO_SECRETS: remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "node-fatal error"); goto remap_done; default: goto remap_done; } is_probe = pcmk_xe_is_probe(history->xml); if (is_probe) { task = "probe"; } if (history->expected_exit_status < 0) { /* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with * Heartbeat 2.0.7 or earlier as the cluster layer, did not include the * expected exit status in the transition key, which (along with the * similar case of a corrupted transition key in the CIB) will be * reported to this function as -1. Pacemaker 2.0+ does not support * rolling upgrades from those versions or processing of saved CIB files * from those versions, so we do not need to care much about this case. */ remap_because(history, &why, PCMK_EXEC_ERROR, "obsolete history format"); pcmk__config_warn("Expected result not found for %s on %s " "(corrupt or obsolete CIB?)", history->key, pcmk__node_name(history->node)); } else if (history->exit_status == history->expected_exit_status) { remap_because(history, &why, PCMK_EXEC_DONE, "expected result"); } else { remap_because(history, &why, PCMK_EXEC_ERROR, "unexpected result"); pcmk__rsc_debug(history->rsc, "%s on %s: expected %d (%s), got %d (%s%s%s)", history->key, pcmk__node_name(history->node), history->expected_exit_status, services_ocf_exitcode_str(history->expected_exit_status), history->exit_status, services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, "")); } switch (history->exit_status) { case PCMK_OCF_OK: if (is_probe && (history->expected_exit_status == PCMK_OCF_NOT_RUNNING)) { char *last_change_s = last_change_str(history->xml); remap_because(history, &why, PCMK_EXEC_DONE, "probe"); pcmk__rsc_info(history->rsc, "Probe found %s active on %s at %s", history->rsc->id, pcmk__node_name(history->node), last_change_s); free(last_change_s); } break; case PCMK_OCF_NOT_RUNNING: if (is_probe || (history->expected_exit_status == history->exit_status) || !pcmk_is_set(history->rsc->flags, pcmk__rsc_managed)) { /* For probes, recurring monitors for the Stopped role, and * unmanaged resources, "not running" is not considered a * failure. */ remap_because(history, &why, PCMK_EXEC_DONE, "exit status"); history->rsc->role = pcmk_role_stopped; *on_fail = pcmk_on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "not running"); } break; case PCMK_OCF_RUNNING_PROMOTED: if (is_probe && (history->exit_status != history->expected_exit_status)) { char *last_change_s = last_change_str(history->xml); remap_because(history, &why, PCMK_EXEC_DONE, "probe"); pcmk__rsc_info(history->rsc, "Probe found %s active and promoted on %s at %s", history->rsc->id, pcmk__node_name(history->node), last_change_s); free(last_change_s); } if (!expired || (history->exit_status == history->expected_exit_status)) { history->rsc->role = pcmk_role_promoted; } break; case PCMK_OCF_FAILED_PROMOTED: if (!expired) { history->rsc->role = pcmk_role_promoted; } remap_because(history, &why, PCMK_EXEC_ERROR, "exit status"); break; case PCMK_OCF_NOT_CONFIGURED: remap_because(history, &why, PCMK_EXEC_ERROR_FATAL, "exit status"); break; case PCMK_OCF_UNIMPLEMENT_FEATURE: { guint interval_ms = 0; crm_element_value_ms(history->xml, PCMK_META_INTERVAL, &interval_ms); if (interval_ms == 0) { if (!expired) { block_if_unrecoverable(history); } remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status"); } else { remap_because(history, &why, PCMK_EXEC_NOT_SUPPORTED, "exit status"); } } break; case PCMK_OCF_NOT_INSTALLED: case PCMK_OCF_INVALID_PARAM: case PCMK_OCF_INSUFFICIENT_PRIV: if (!expired) { block_if_unrecoverable(history); } remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status"); break; default: if (history->execution_status == PCMK_EXEC_DONE) { char *last_change_s = last_change_str(history->xml); crm_info("Treating unknown exit status %d from %s of %s " "on %s at %s as failure", history->exit_status, task, history->rsc->id, pcmk__node_name(history->node), last_change_s); remap_because(history, &why, PCMK_EXEC_ERROR, "unknown exit status"); free(last_change_s); } break; } remap_done: if (why != NULL) { pcmk__rsc_trace(history->rsc, "Remapped %s result from [%s: %s] to [%s: %s] " "because of %s", history->key, pcmk_exec_status_str(orig_exec_status), crm_exit_str(orig_exit_status), pcmk_exec_status_str(history->execution_status), crm_exit_str(history->exit_status), why); } } // return TRUE if start or monitor last failure but parameters changed static bool should_clear_for_param_change(const xmlNode *xml_op, const char *task, pcmk_resource_t *rsc, pcmk_node_t *node) { if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) { if (pe__bundle_needs_remote_name(rsc)) { /* We haven't allocated resources yet, so we can't reliably * substitute addr parameters for the REMOTE_CONTAINER_HACK. * When that's needed, defer the check until later. */ pe__add_param_check(xml_op, rsc, node, pcmk__check_last_failure, rsc->private->scheduler); } else { pcmk__op_digest_t *digest_data = NULL; digest_data = rsc_action_digest_cmp(rsc, xml_op, node, rsc->private->scheduler); switch (digest_data->rc) { case pcmk__digest_unknown: crm_trace("Resource %s history entry %s on %s" " has no digest to compare", rsc->id, pcmk__xe_history_key(xml_op), node->details->id); break; case pcmk__digest_match: break; default: return TRUE; } } } return FALSE; } // Order action after fencing of remote node, given connection rsc static void order_after_remote_fencing(pcmk_action_t *action, pcmk_resource_t *remote_conn, pcmk_scheduler_t *scheduler) { pcmk_node_t *remote_node = pcmk_find_node(scheduler, remote_conn->id); if (remote_node) { pcmk_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL, FALSE, scheduler); order_actions(fence, action, pcmk__ar_first_implies_then); } } static bool should_ignore_failure_timeout(const pcmk_resource_t *rsc, const char *task, guint interval_ms, bool is_last_failure) { /* Clearing failures of recurring monitors has special concerns. The * executor reports only changes in the monitor result, so if the * monitor is still active and still getting the same failure result, * that will go undetected after the failure is cleared. * * Also, the operation history will have the time when the recurring * monitor result changed to the given code, not the time when the * result last happened. * * @TODO We probably should clear such failures only when the failure * timeout has passed since the last occurrence of the failed result. * However we don't record that information. We could maybe approximate * that by clearing only if there is a more recent successful monitor or * stop result, but we don't even have that information at this point * since we are still unpacking the resource's operation history. * * This is especially important for remote connection resources with a * reconnect interval, so in that case, we skip clearing failures * if the remote node hasn't been fenced. */ if ((rsc->private->remote_reconnect_ms > 0U) && pcmk_is_set(rsc->private->scheduler->flags, pcmk_sched_fencing_enabled) && (interval_ms != 0) && pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { pcmk_node_t *remote_node = pcmk_find_node(rsc->private->scheduler, rsc->id); if (remote_node && !remote_node->details->remote_was_fenced) { if (is_last_failure) { crm_info("Waiting to clear monitor failure for remote node %s" " until fencing has occurred", rsc->id); } return TRUE; } } return FALSE; } /*! * \internal * \brief Check operation age and schedule failure clearing when appropriate * * This function has two distinct purposes. The first is to check whether an * operation history entry is expired (i.e. the resource has a failure timeout, * the entry is older than the timeout, and the resource either has no fail * count or its fail count is entirely older than the timeout). The second is to * schedule fail count clearing when appropriate (i.e. the operation is expired * and either the resource has an expired fail count or the operation is a * last_failure for a remote connection resource with a reconnect interval, * or the operation is a last_failure for a start or monitor operation and the * resource's parameters have changed since the operation). * * \param[in,out] history Parsed action result history * * \return true if operation history entry is expired, otherwise false */ static bool check_operation_expiry(struct action_history *history) { bool expired = false; bool is_last_failure = pcmk__ends_with(history->id, "_last_failure_0"); time_t last_run = 0; int unexpired_fail_count = 0; const char *clear_reason = NULL; const guint expiration_sec = history->rsc->private->failure_expiration_ms / 1000; pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; if (history->execution_status == PCMK_EXEC_NOT_INSTALLED) { pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not expired: " "Not Installed does not expire", history->id, pcmk__node_name(history->node)); return false; // "Not installed" must always be cleared manually } if ((expiration_sec > 0) && (crm_element_value_epoch(history->xml, PCMK_XA_LAST_RC_CHANGE, &last_run) == 0)) { /* Resource has a PCMK_META_FAILURE_TIMEOUT and history entry has a * timestamp */ time_t now = get_effective_time(scheduler); time_t last_failure = 0; // Is this particular operation history older than the failure timeout? if ((now >= (last_run + expiration_sec)) && !should_ignore_failure_timeout(history->rsc, history->task, history->interval_ms, is_last_failure)) { expired = true; } // Does the resource as a whole have an unexpired fail count? unexpired_fail_count = pe_get_failcount(history->node, history->rsc, &last_failure, pcmk__fc_effective, history->xml); // Update scheduler recheck time according to *last* failure crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d " "expiration=%s last-failure@%lld", history->id, (long long) last_run, (expired? "" : "not "), (long long) now, unexpired_fail_count, pcmk__readable_interval(expiration_sec * 1000), (long long) last_failure); last_failure += expiration_sec + 1; if (unexpired_fail_count && (now < last_failure)) { pe__update_recheck_time(last_failure, scheduler, "fail count expiration"); } } if (expired) { if (pe_get_failcount(history->node, history->rsc, NULL, pcmk__fc_default, history->xml)) { // There is a fail count ignoring timeout if (unexpired_fail_count == 0) { // There is no fail count considering timeout clear_reason = "it expired"; } else { /* This operation is old, but there is an unexpired fail count. * In a properly functioning cluster, this should only be * possible if this operation is not a failure (otherwise the * fail count should be expired too), so this is really just a * failsafe. */ pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not " "expired: Unexpired fail count", history->id, pcmk__node_name(history->node)); expired = false; } } else if (is_last_failure && (history->rsc->private->remote_reconnect_ms > 0U)) { /* Clear any expired last failure when reconnect interval is set, * even if there is no fail count. */ clear_reason = "reconnect interval is set"; } } if (!expired && is_last_failure && should_clear_for_param_change(history->xml, history->task, history->rsc, history->node)) { clear_reason = "resource parameters have changed"; } if (clear_reason != NULL) { pcmk_action_t *clear_op = NULL; // Schedule clearing of the fail count clear_op = pe__clear_failcount(history->rsc, history->node, clear_reason, scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled) && (history->rsc->private->remote_reconnect_ms > 0)) { /* If we're clearing a remote connection due to a reconnect * interval, we want to wait until any scheduled fencing * completes. * * We could limit this to remote_node->details->unclean, but at * this point, that's always true (it won't be reliable until * after unpack_node_history() is done). */ crm_info("Clearing %s failure will wait until any scheduled " "fencing of %s completes", history->task, history->rsc->id); order_after_remote_fencing(clear_op, history->rsc, scheduler); } } if (expired && (history->interval_ms == 0) && pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { switch (history->exit_status) { case PCMK_OCF_OK: case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_PROMOTED: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: // Don't expire probes that return these values pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not " "expired: Probe result", history->id, pcmk__node_name(history->node)); expired = false; break; } } return expired; } int pe__target_rc_from_xml(const xmlNode *xml_op) { int target_rc = 0; const char *key = crm_element_value(xml_op, PCMK__XA_TRANSITION_KEY); if (key == NULL) { return -1; } decode_transition_key(key, NULL, NULL, NULL, &target_rc); return target_rc; } /*! * \internal * \brief Update a resource's state for an action result * * \param[in,out] history Parsed action history entry * \param[in] exit_status Exit status to base new state on * \param[in] last_failure Resource's last_failure entry, if known * \param[in,out] on_fail Resource's current failure handling */ static void update_resource_state(struct action_history *history, int exit_status, const xmlNode *last_failure, enum action_fail_response *on_fail) { bool clear_past_failure = false; if ((exit_status == PCMK_OCF_NOT_INSTALLED) || (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml))) { history->rsc->role = pcmk_role_stopped; } else if (exit_status == PCMK_OCF_NOT_RUNNING) { clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { if ((last_failure != NULL) && pcmk__str_eq(history->key, pcmk__xe_history_key(last_failure), pcmk__str_none)) { clear_past_failure = true; } if (history->rsc->role < pcmk_role_started) { set_active(history->rsc); } } else if (pcmk__str_eq(history->task, PCMK_ACTION_START, pcmk__str_none)) { history->rsc->role = pcmk_role_started; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_STOP, pcmk__str_none)) { history->rsc->role = pcmk_role_stopped; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_PROMOTE, pcmk__str_none)) { history->rsc->role = pcmk_role_promoted; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_DEMOTE, pcmk__str_none)) { if (*on_fail == pcmk_on_fail_demote) { /* Demote clears an error only if * PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE */ clear_past_failure = true; } history->rsc->role = pcmk_role_unpromoted; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_FROM, pcmk__str_none)) { history->rsc->role = pcmk_role_started; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)) { unpack_migrate_to_success(history); } else if (history->rsc->role < pcmk_role_started) { pcmk__rsc_trace(history->rsc, "%s active on %s", history->rsc->id, pcmk__node_name(history->node)); set_active(history->rsc); } if (!clear_past_failure) { return; } switch (*on_fail) { case pcmk_on_fail_stop: case pcmk_on_fail_ban: case pcmk_on_fail_standby_node: case pcmk_on_fail_fence_node: pcmk__rsc_trace(history->rsc, "%s (%s) is not cleared by a completed %s", history->rsc->id, pcmk_on_fail_text(*on_fail), history->task); break; case pcmk_on_fail_block: case pcmk_on_fail_ignore: case pcmk_on_fail_demote: case pcmk_on_fail_restart: case pcmk_on_fail_restart_container: *on_fail = pcmk_on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "clear past failures"); break; case pcmk_on_fail_reset_remote: if (history->rsc->private->remote_reconnect_ms == 0U) { /* With no reconnect interval, the connection is allowed to * start again after the remote node is fenced and * completely stopped. (With a reconnect interval, we wait * for the failure to be cleared entirely before attempting * to reconnect.) */ *on_fail = pcmk_on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "clear past failures and reset remote"); } break; } } /*! * \internal * \brief Check whether a given history entry matters for resource state * * \param[in] history Parsed action history entry * * \return true if action can affect resource state, otherwise false */ static inline bool can_affect_state(struct action_history *history) { #if 0 /* @COMPAT It might be better to parse only actions we know we're interested * in, rather than exclude a couple we don't. However that would be a * behavioral change that should be done at a major or minor series release. * Currently, unknown operations can affect whether a resource is considered * active and/or failed. */ return pcmk__str_any_of(history->task, PCMK_ACTION_MONITOR, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, "asyncmon", NULL); #else return !pcmk__str_any_of(history->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_META_DATA, NULL); #endif } /*! * \internal * \brief Unpack execution/exit status and exit reason from a history entry * * \param[in,out] history Action history entry to unpack * * \return Standard Pacemaker return code */ static int unpack_action_result(struct action_history *history) { if ((crm_element_value_int(history->xml, PCMK__XA_OP_STATUS, &(history->execution_status)) < 0) || (history->execution_status < PCMK_EXEC_PENDING) || (history->execution_status > PCMK_EXEC_MAX) || (history->execution_status == PCMK_EXEC_CANCELLED)) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "with invalid " PCMK__XA_OP_STATUS " '%s'", history->id, history->rsc->id, pcmk__node_name(history->node), pcmk__s(crm_element_value(history->xml, PCMK__XA_OP_STATUS), "")); return pcmk_rc_unpack_error; } if ((crm_element_value_int(history->xml, PCMK__XA_RC_CODE, &(history->exit_status)) < 0) || (history->exit_status < 0) || (history->exit_status > CRM_EX_MAX)) { #if 0 /* @COMPAT We should ignore malformed entries, but since that would * change behavior, it should be done at a major or minor series * release. */ pcmk__config_err("Ignoring resource history entry %s for %s on %s " "with invalid " PCMK__XA_RC_CODE " '%s'", history->id, history->rsc->id, pcmk__node_name(history->node), pcmk__s(crm_element_value(history->xml, PCMK__XA_RC_CODE), "")); return pcmk_rc_unpack_error; #else history->exit_status = CRM_EX_ERROR; #endif } history->exit_reason = crm_element_value(history->xml, PCMK_XA_EXIT_REASON); return pcmk_rc_ok; } /*! * \internal * \brief Process an action history entry whose result expired * * \param[in,out] history Parsed action history entry * \param[in] orig_exit_status Action exit status before remapping * * \return Standard Pacemaker return code (in particular, pcmk_rc_ok means the * entry needs no further processing) */ static int process_expired_result(struct action_history *history, int orig_exit_status) { if (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml) && (orig_exit_status != history->expected_exit_status)) { if (history->rsc->role <= pcmk_role_stopped) { history->rsc->role = pcmk_role_unknown; } crm_trace("Ignoring resource history entry %s for probe of %s on %s: " "Masked failure expired", history->id, history->rsc->id, pcmk__node_name(history->node)); return pcmk_rc_ok; } if (history->exit_status == history->expected_exit_status) { return pcmk_rc_undetermined; // Only failures expire } if (history->interval_ms == 0) { crm_notice("Ignoring resource history entry %s for %s of %s on %s: " "Expired failure", history->id, history->task, history->rsc->id, pcmk__node_name(history->node)); return pcmk_rc_ok; } if (history->node->details->online && !history->node->details->unclean) { /* Reschedule the recurring action. schedule_cancel() won't work at * this stage, so as a hacky workaround, forcibly change the restart * digest so pcmk__check_action_config() does what we want later. * * @TODO We should skip this if there is a newer successful monitor. * Also, this causes rescheduling only if the history entry * has a PCMK__XA_OP_DIGEST (which the expire-non-blocked-failure * scheduler regression test doesn't, but that may not be a * realistic scenario in production). */ crm_notice("Rescheduling %s-interval %s of %s on %s " "after failure expired", pcmk__readable_interval(history->interval_ms), history->task, history->rsc->id, pcmk__node_name(history->node)); crm_xml_add(history->xml, PCMK__XA_OP_RESTART_DIGEST, "calculated-failure-timeout"); return pcmk_rc_ok; } return pcmk_rc_undetermined; } /*! * \internal * \brief Process a masked probe failure * * \param[in,out] history Parsed action history entry * \param[in] orig_exit_status Action exit status before remapping * \param[in] last_failure Resource's last_failure entry, if known * \param[in,out] on_fail Resource's current failure handling */ static void mask_probe_failure(struct action_history *history, int orig_exit_status, const xmlNode *last_failure, enum action_fail_response *on_fail) { pcmk_resource_t *ban_rsc = history->rsc; if (!pcmk_is_set(history->rsc->flags, pcmk__rsc_unique)) { ban_rsc = uber_parent(history->rsc); } crm_notice("Treating probe result '%s' for %s on %s as 'not running'", services_ocf_exitcode_str(orig_exit_status), history->rsc->id, pcmk__node_name(history->node)); update_resource_state(history, history->expected_exit_status, last_failure, on_fail); crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->details->uname); record_failed_op(history); resource_location(ban_rsc, history->node, -PCMK_SCORE_INFINITY, "masked-probe-failure", ban_rsc->private->scheduler); } /*! * \internal Check whether a given failure is for a given pending action * * \param[in] history Parsed history entry for pending action * \param[in] last_failure Resource's last_failure entry, if known * * \return true if \p last_failure is failure of pending action in \p history, * otherwise false * \note Both \p history and \p last_failure must come from the same * \c PCMK__XE_LRM_RESOURCE block, as node and resource are assumed to be * the same. */ static bool failure_is_newer(const struct action_history *history, const xmlNode *last_failure) { guint failure_interval_ms = 0U; long long failure_change = 0LL; long long this_change = 0LL; if (last_failure == NULL) { return false; // Resource has no last_failure entry } if (!pcmk__str_eq(history->task, crm_element_value(last_failure, PCMK_XA_OPERATION), pcmk__str_none)) { return false; // last_failure is for different action } if ((crm_element_value_ms(last_failure, PCMK_META_INTERVAL, &failure_interval_ms) != pcmk_ok) || (history->interval_ms != failure_interval_ms)) { return false; // last_failure is for action with different interval } if ((pcmk__scan_ll(crm_element_value(history->xml, PCMK_XA_LAST_RC_CHANGE), &this_change, 0LL) != pcmk_rc_ok) || (pcmk__scan_ll(crm_element_value(last_failure, PCMK_XA_LAST_RC_CHANGE), &failure_change, 0LL) != pcmk_rc_ok) || (failure_change < this_change)) { return false; // Failure is not known to be newer } return true; } /*! * \internal * \brief Update a resource's role etc. for a pending action * * \param[in,out] history Parsed history entry for pending action * \param[in] last_failure Resource's last_failure entry, if known */ static void process_pending_action(struct action_history *history, const xmlNode *last_failure) { /* For recurring monitors, a failure is recorded only in RSC_last_failure_0, * and there might be a RSC_monitor_INTERVAL entry with the last successful * or pending result. * * If last_failure contains the failure of the pending recurring monitor * we're processing here, and is newer, the action is no longer pending. * (Pending results have call ID -1, which sorts last, so the last failure * if any should be known.) */ if (failure_is_newer(history, last_failure)) { return; } if (strcmp(history->task, PCMK_ACTION_START) == 0) { pcmk__set_rsc_flags(history->rsc, pcmk__rsc_start_pending); set_active(history->rsc); } else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) { history->rsc->role = pcmk_role_promoted; } else if ((strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) && history->node->details->unclean) { /* A migrate_to action is pending on a unclean source, so force a stop * on the target. */ const char *migrate_target = NULL; pcmk_node_t *target = NULL; migrate_target = crm_element_value(history->xml, PCMK__META_MIGRATE_TARGET); target = pcmk_find_node(history->rsc->private->scheduler, migrate_target); if (target != NULL) { stop_action(history->rsc, target, FALSE); } } if (history->rsc->private->pending_action != NULL) { /* There should never be multiple pending actions, but as a failsafe, * just remember the first one processed for display purposes. */ return; } if (pcmk_is_probe(history->task, history->interval_ms)) { /* Pending probes are currently never displayed, even if pending * operations are requested. If we ever want to change that, * enable the below and the corresponding part of * native.c:native_pending_action(). */ #if 0 history->rsc->private->pending_action = strdup("probe"); history->rsc->pending_node = history->node; #endif } else { history->rsc->private->pending_action = strdup(history->task); history->rsc->pending_node = history->node; } } static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *on_fail) { int old_rc = 0; bool expired = false; pcmk_resource_t *parent = rsc; enum rsc_role_e fail_role = pcmk_role_unknown; enum action_fail_response failure_strategy = pcmk_on_fail_restart; struct action_history history = { .rsc = rsc, .node = node, .xml = xml_op, .execution_status = PCMK_EXEC_UNKNOWN, }; CRM_CHECK(rsc && node && xml_op, return); history.id = pcmk__xe_id(xml_op); if (history.id == NULL) { pcmk__config_err("Ignoring resource history entry for %s on %s " "without ID", rsc->id, pcmk__node_name(node)); return; } // Task and interval history.task = crm_element_value(xml_op, PCMK_XA_OPERATION); if (history.task == NULL) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "without " PCMK_XA_OPERATION, history.id, rsc->id, pcmk__node_name(node)); return; } crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &(history.interval_ms)); if (!can_affect_state(&history)) { pcmk__rsc_trace(rsc, "Ignoring resource history entry %s for %s on %s " "with irrelevant action '%s'", history.id, rsc->id, pcmk__node_name(node), history.task); return; } if (unpack_action_result(&history) != pcmk_rc_ok) { return; // Error already logged } history.expected_exit_status = pe__target_rc_from_xml(xml_op); history.key = pcmk__xe_history_key(xml_op); crm_element_value_int(xml_op, PCMK__XA_CALL_ID, &(history.call_id)); pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)", history.id, history.task, history.call_id, pcmk__node_name(node), pcmk_exec_status_str(history.execution_status), crm_exit_str(history.exit_status)); if (node->details->unclean) { pcmk__rsc_trace(rsc, "%s is running on %s, which is unclean (further action " "depends on value of stop's on-fail attribute)", rsc->id, pcmk__node_name(node)); } expired = check_operation_expiry(&history); old_rc = history.exit_status; remap_operation(&history, on_fail, expired); if (expired && (process_expired_result(&history, old_rc) == pcmk_rc_ok)) { goto done; } if (!pcmk__is_bundled(rsc) && pcmk_xe_mask_probe_failure(xml_op)) { mask_probe_failure(&history, old_rc, *last_failure, on_fail); goto done; } if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { parent = uber_parent(rsc); } switch (history.execution_status) { case PCMK_EXEC_PENDING: process_pending_action(&history, *last_failure); goto done; case PCMK_EXEC_DONE: update_resource_state(&history, history.exit_status, *last_failure, on_fail); goto done; case PCMK_EXEC_NOT_INSTALLED: unpack_failure_handling(&history, &failure_strategy, &fail_role); if (failure_strategy == pcmk_on_fail_ignore) { crm_warn("Cannot ignore failed %s of %s on %s: " "Resource agent doesn't exist " QB_XS " status=%d rc=%d id=%s", history.task, rsc->id, pcmk__node_name(node), history.execution_status, history.exit_status, history.id); /* Also for printing it as "FAILED" by marking it as * pcmk__rsc_failed later */ *on_fail = pcmk_on_fail_ban; } resource_location(parent, node, -PCMK_SCORE_INFINITY, "hard-error", rsc->private->scheduler); unpack_rsc_op_failure(&history, failure_strategy, fail_role, last_failure, on_fail); goto done; case PCMK_EXEC_NOT_CONNECTED: if (pcmk__is_pacemaker_remote_node(node) && pcmk_is_set(node->details->remote_rsc->flags, pcmk__rsc_managed)) { /* We should never get into a situation where a managed remote * connection resource is considered OK but a resource action * behind the connection gets a "not connected" status. But as a * fail-safe in case a bug or unusual circumstances do lead to * that, ensure the remote connection is considered failed. */ pcmk__set_rsc_flags(node->details->remote_rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); } break; // Not done, do error handling case PCMK_EXEC_ERROR: case PCMK_EXEC_ERROR_HARD: case PCMK_EXEC_ERROR_FATAL: case PCMK_EXEC_TIMEOUT: case PCMK_EXEC_NOT_SUPPORTED: case PCMK_EXEC_INVALID: break; // Not done, do error handling default: // No other value should be possible at this point break; } unpack_failure_handling(&history, &failure_strategy, &fail_role); if ((failure_strategy == pcmk_on_fail_ignore) || ((failure_strategy == pcmk_on_fail_restart_container) && (strcmp(history.task, PCMK_ACTION_STOP) == 0))) { char *last_change_s = last_change_str(xml_op); crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s succeeded " QB_XS " %s", history.task, services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), rsc->id, pcmk__node_name(node), last_change_s, history.id); free(last_change_s); update_resource_state(&history, history.expected_exit_status, *last_failure, on_fail); crm_xml_add(xml_op, PCMK_XA_UNAME, node->details->uname); pcmk__set_rsc_flags(rsc, pcmk__rsc_ignore_failure); record_failed_op(&history); if ((failure_strategy == pcmk_on_fail_restart_container) && cmp_on_fail(*on_fail, pcmk_on_fail_restart) <= 0) { *on_fail = failure_strategy; } } else { unpack_rsc_op_failure(&history, failure_strategy, fail_role, last_failure, on_fail); if (history.execution_status == PCMK_EXEC_ERROR_HARD) { uint8_t log_level = LOG_ERR; if (history.exit_status == PCMK_OCF_NOT_INSTALLED) { log_level = LOG_NOTICE; } do_crm_log(log_level, "Preventing %s from restarting on %s because " "of hard failure (%s%s%s) " QB_XS " %s", parent->id, pcmk__node_name(node), services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), history.id); resource_location(parent, node, -PCMK_SCORE_INFINITY, "hard-error", rsc->private->scheduler); } else if (history.execution_status == PCMK_EXEC_ERROR_FATAL) { pcmk__sched_err("Preventing %s from restarting anywhere because " "of fatal failure (%s%s%s) " QB_XS " %s", parent->id, services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), history.id); resource_location(parent, NULL, -PCMK_SCORE_INFINITY, "fatal-error", rsc->private->scheduler); } } done: pcmk__rsc_trace(rsc, "%s role on %s after %s is %s (next %s)", rsc->id, pcmk__node_name(node), history.id, pcmk_role_text(rsc->role), pcmk_role_text(rsc->next_role)); } static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite, pcmk_scheduler_t *scheduler) { const char *cluster_name = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; pcmk__insert_dup(node->details->attrs, CRM_ATTR_UNAME, node->details->uname); pcmk__insert_dup(node->details->attrs, CRM_ATTR_ID, node->details->id); if (pcmk__str_eq(node->details->id, scheduler->dc_uuid, pcmk__str_casei)) { scheduler->dc_node = node; node->details->is_dc = TRUE; pcmk__insert_dup(node->details->attrs, CRM_ATTR_IS_DC, PCMK_VALUE_TRUE); } else { pcmk__insert_dup(node->details->attrs, CRM_ATTR_IS_DC, PCMK_VALUE_FALSE); } cluster_name = g_hash_table_lookup(scheduler->config_hash, PCMK_OPT_CLUSTER_NAME); if (cluster_name) { pcmk__insert_dup(node->details->attrs, CRM_ATTR_CLUSTER_NAME, cluster_name); } pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, node->details->attrs, NULL, overwrite, scheduler); pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_UTILIZATION, &rule_data, node->details->utilization, NULL, FALSE, scheduler); if (pcmk__node_attr(node, CRM_ATTR_SITE_NAME, NULL, pcmk__rsc_node_current) == NULL) { const char *site_name = pcmk__node_attr(node, "site-name", NULL, pcmk__rsc_node_current); if (site_name) { pcmk__insert_dup(node->details->attrs, CRM_ATTR_SITE_NAME, site_name); } else if (cluster_name) { /* Default to cluster-name if unset */ pcmk__insert_dup(node->details->attrs, CRM_ATTR_SITE_NAME, cluster_name); } } } static GList * extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter) { int counter = -1; int stop_index = -1; int start_index = -1; xmlNode *rsc_op = NULL; GList *gIter = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; /* extract operations */ op_list = NULL; sorted_op_list = NULL; for (rsc_op = pcmk__xe_first_child(rsc_entry, NULL, NULL, NULL); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) { if (pcmk__xe_is(rsc_op, PCMK__XE_LRM_RSC_OP)) { crm_xml_add(rsc_op, PCMK_XA_RESOURCE, rsc); crm_xml_add(rsc_op, PCMK_XA_UNAME, node); op_list = g_list_prepend(op_list, rsc_op); } } if (op_list == NULL) { /* if there are no operations, there is nothing to do */ return NULL; } sorted_op_list = g_list_sort(op_list, sort_op_by_callid); /* create active recurring operations as optional */ if (active_filter == FALSE) { return sorted_op_list; } op_list = NULL; calculate_active_ops(sorted_op_list, &start_index, &stop_index); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; if (start_index < stop_index) { crm_trace("Skipping %s: not active", pcmk__xe_id(rsc_entry)); break; } else if (counter < start_index) { crm_trace("Skipping %s: old", pcmk__xe_id(rsc_op)); continue; } op_list = g_list_append(op_list, rsc_op); } g_list_free(sorted_op_list); return op_list; } GList * find_operations(const char *rsc, const char *node, gboolean active_filter, pcmk_scheduler_t *scheduler) { GList *output = NULL; GList *intermediate = NULL; xmlNode *tmp = NULL; xmlNode *status = pcmk__xe_first_child(scheduler->input, PCMK_XE_STATUS, NULL, NULL); pcmk_node_t *this_node = NULL; xmlNode *node_state = NULL; CRM_CHECK(status != NULL, return NULL); for (node_state = pcmk__xe_first_child(status, NULL, NULL, NULL); node_state != NULL; node_state = pcmk__xe_next(node_state)) { if (pcmk__xe_is(node_state, PCMK__XE_NODE_STATE)) { const char *uname = crm_element_value(node_state, PCMK_XA_UNAME); if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) { continue; } this_node = pcmk_find_node(scheduler, uname); if(this_node == NULL) { CRM_LOG_ASSERT(this_node != NULL); continue; } else if (pcmk__is_pacemaker_remote_node(this_node)) { determine_remote_online_status(scheduler, this_node); } else { determine_online_status(node_state, this_node, scheduler); } if (this_node->details->online || pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { /* offline nodes run no resources... * unless stonith is enabled in which case we need to * make sure rsc start events happen after the stonith */ xmlNode *lrm_rsc = NULL; tmp = pcmk__xe_first_child(node_state, PCMK__XE_LRM, NULL, NULL); tmp = pcmk__xe_first_child(tmp, PCMK__XE_LRM_RESOURCES, NULL, NULL); for (lrm_rsc = pcmk__xe_first_child(tmp, NULL, NULL, NULL); lrm_rsc != NULL; lrm_rsc = pcmk__xe_next(lrm_rsc)) { if (pcmk__xe_is(lrm_rsc, PCMK__XE_LRM_RESOURCE)) { const char *rsc_id = crm_element_value(lrm_rsc, PCMK_XA_ID); if (rsc != NULL && !pcmk__str_eq(rsc_id, rsc, pcmk__str_casei)) { continue; } intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter); output = g_list_concat(output, intermediate); } } } } } return output; } diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c index 1acc2b00ad..3656e2f728 100644 --- a/lib/pengine/utils.c +++ b/lib/pengine/utils.c @@ -1,928 +1,929 @@ /* * 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) { + for (GList *n = rsc->private->active_nodes; 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) && (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); + return (rsc != NULL) && pcmk__list_of_1(rsc->private->active_nodes) + && pcmk__same_node((const pcmk_node_t *) + rsc->private->active_nodes->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; + if (rsc != NULL) { + for (GList *ele = rsc->private->active_nodes; 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; } diff --git a/tools/crm_resource_runtime.c b/tools/crm_resource_runtime.c index 84d6e41e4d..c9b9346afc 100644 --- a/tools/crm_resource_runtime.c +++ b/tools/crm_resource_runtime.c @@ -1,2415 +1,2417 @@ /* * 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 static GList * build_node_info_list(const pcmk_resource_t *rsc) { GList *retval = NULL; for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { const pcmk_resource_t *child = (const pcmk_resource_t *) iter->data; - for (const GList *iter2 = child->running_on; + for (const GList *iter2 = child->private->active_nodes; iter2 != NULL; iter2 = iter2->next) { const pcmk_node_t *node = (const pcmk_node_t *) iter2->data; node_info_t *ni = pcmk__assert_alloc(1, sizeof(node_info_t)); ni->node_name = node->details->uname; if (pcmk_is_set(rsc->flags, pcmk__rsc_promotable) && (child->private->fns->state(child, TRUE) == pcmk_role_promoted)) { ni->promoted = true; } retval = g_list_prepend(retval, ni); } } return retval; } GList * cli_resource_search(pcmk_resource_t *rsc, const char *requested_name, pcmk_scheduler_t *scheduler) { GList *retval = NULL; const pcmk_resource_t *parent = pe__const_top_resource(rsc, false); if (pcmk__is_clone(rsc)) { retval = build_node_info_list(rsc); /* The anonymous clone children's common ID is supplied */ } else if (pcmk__is_clone(parent) && !pcmk_is_set(rsc->flags, pcmk__rsc_unique) && (rsc->private->history_id != NULL) && pcmk__str_eq(requested_name, rsc->private->history_id, pcmk__str_none) && !pcmk__str_eq(requested_name, rsc->id, pcmk__str_none)) { retval = build_node_info_list(parent); - } else if (rsc->running_on != NULL) { - for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { + } else { + for (GList *iter = rsc->private->active_nodes; + iter != NULL; iter = iter->next) { + pcmk_node_t *node = (pcmk_node_t *) iter->data; node_info_t *ni = pcmk__assert_alloc(1, sizeof(node_info_t)); ni->node_name = node->details->uname; if (rsc->private->fns->state(rsc, TRUE) == pcmk_role_promoted) { ni->promoted = true; } retval = g_list_prepend(retval, ni); } } return retval; } // \return Standard Pacemaker return code static int find_resource_attr(pcmk__output_t *out, cib_t * the_cib, const char *attr, const char *rsc, const char *attr_set_type, const char *set_name, const char *attr_id, const char *attr_name, xmlNode **result) { xmlNode *xml_search; int rc = pcmk_rc_ok; GString *xpath = NULL; const char *xpath_base = NULL; if (result) { *result = NULL; } if(the_cib == NULL) { return ENOTCONN; } xpath_base = pcmk_cib_xpath_for(PCMK_XE_RESOURCES); if (xpath_base == NULL) { crm_err(PCMK_XE_RESOURCES " CIB element not known (bug?)"); return ENOMSG; } xpath = g_string_sized_new(1024); pcmk__g_strcat(xpath, xpath_base, "//*[@" PCMK_XA_ID "=\"", rsc, "\"]", NULL); if (attr_set_type != NULL) { pcmk__g_strcat(xpath, "/", attr_set_type, NULL); if (set_name != NULL) { pcmk__g_strcat(xpath, "[@" PCMK_XA_ID "=\"", set_name, "\"]", NULL); } } g_string_append(xpath, "//" PCMK_XE_NVPAIR); if (attr_id != NULL && attr_name!= NULL) { pcmk__g_strcat(xpath, "[@" PCMK_XA_ID "='", attr_id, "' " "and @" PCMK_XA_NAME "='", attr_name, "']", NULL); } else if (attr_id != NULL) { pcmk__g_strcat(xpath, "[@" PCMK_XA_ID "='", attr_id, "']", NULL); } else if (attr_name != NULL) { pcmk__g_strcat(xpath, "[@" PCMK_XA_NAME "='", attr_name, "']", NULL); } rc = the_cib->cmds->query(the_cib, (const char *) xpath->str, &xml_search, cib_sync_call | cib_scope_local | cib_xpath); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { crm_log_xml_debug(xml_search, "Match"); if (xml_search->children != NULL) { rc = ENOTUNIQ; pcmk__warn_multiple_name_matches(out, xml_search, attr_name); out->spacer(out); } } if (result) { *result = xml_search; } else { pcmk__xml_free(xml_search); } g_string_free(xpath, TRUE); return rc; } /* PRIVATE. Use the find_matching_attr_resources instead. */ static void find_matching_attr_resources_recursive(pcmk__output_t *out, GList /* */ **result, pcmk_resource_t *rsc, const char * attr_set, const char * attr_set_type, const char * attr_id, const char * attr_name, cib_t * cib, int depth) { int rc = pcmk_rc_ok; char *lookup_id = clone_strip(rsc->id); /* visit the children */ for(GList *gIter = rsc->children; gIter; gIter = gIter->next) { find_matching_attr_resources_recursive(out, result, (pcmk_resource_t *) gIter->data, attr_set, attr_set_type, attr_id, attr_name, cib, depth+1); /* do it only once for clones */ if (pcmk__is_clone(rsc)) { break; } } rc = find_resource_attr(out, cib, PCMK_XA_ID, lookup_id, attr_set_type, attr_set, attr_id, attr_name, NULL); /* Post-order traversal. * The root is always on the list and it is the last item. */ if((0 == depth) || (pcmk_rc_ok == rc)) { /* push the head */ *result = g_list_append(*result, rsc); } free(lookup_id); } /* The result is a linearized pre-ordered tree of resources. */ static GList/**/ * find_matching_attr_resources(pcmk__output_t *out, pcmk_resource_t *rsc, const char * rsc_id, const char * attr_set, const char * attr_set_type, const char * attr_id, const char * attr_name, cib_t * cib, const char * cmd, gboolean force) { int rc = pcmk_rc_ok; char *lookup_id = NULL; GList * result = NULL; /* If --force is used, update only the requested resource (clone or primitive). * Otherwise, if the primitive has the attribute, use that. * Otherwise use the clone. */ if(force == TRUE) { return g_list_append(result, rsc); } if (pcmk__is_clone(rsc->private->parent)) { int rc = find_resource_attr(out, cib, PCMK_XA_ID, rsc_id, attr_set_type, attr_set, attr_id, attr_name, NULL); if(rc != pcmk_rc_ok) { rsc = rsc->private->parent; out->info(out, "Performing %s of '%s' on '%s', the parent of '%s'", cmd, attr_name, rsc->id, rsc_id); } return g_list_append(result, rsc); } else if ((rsc->private->parent == NULL) && (rsc->children != NULL) && pcmk__is_clone(rsc)) { pcmk_resource_t *child = rsc->children->data; if (pcmk__is_primitive(child)) { lookup_id = clone_strip(child->id); /* Could be a cloned group! */ rc = find_resource_attr(out, cib, PCMK_XA_ID, lookup_id, attr_set_type, attr_set, attr_id, attr_name, NULL); if(rc == pcmk_rc_ok) { rsc = child; out->info(out, "A value for '%s' already exists in child '%s', performing %s on that instead of '%s'", attr_name, lookup_id, cmd, rsc_id); } free(lookup_id); } return g_list_append(result, rsc); } /* If the resource is a group ==> children inherit the attribute if defined. */ find_matching_attr_resources_recursive(out, &result, rsc, attr_set, attr_set_type, attr_id, attr_name, cib, 0); return result; } static int update_element_attribute(pcmk__output_t *out, pcmk_resource_t *rsc, cib_t *cib, const char *attr_name, const char *attr_value) { int rc = pcmk_rc_ok; if (cib == NULL) { return ENOTCONN; } crm_xml_add(rsc->private->xml, attr_name, attr_value); rc = cib->cmds->replace(cib, PCMK_XE_RESOURCES, rsc->private->xml, cib_sync_call); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { out->info(out, "Set attribute: " PCMK_XA_NAME "=%s value=%s", attr_name, attr_value); } return rc; } static int resources_with_attr(pcmk__output_t *out, cib_t *cib, pcmk_resource_t *rsc, const char *requested_name, const char *attr_set, const char *attr_set_type, const char *attr_id, const char *attr_name, const char *top_id, gboolean force, GList **resources) { if (pcmk__str_eq(attr_set_type, PCMK_XE_INSTANCE_ATTRIBUTES, pcmk__str_casei)) { if (!force) { xmlNode *xml_search = NULL; int rc = pcmk_rc_ok; rc = find_resource_attr(out, cib, PCMK_XA_ID, top_id, PCMK_XE_META_ATTRIBUTES, attr_set, attr_id, attr_name, &xml_search); if (rc == pcmk_rc_ok || rc == ENOTUNIQ) { char *found_attr_id = NULL; found_attr_id = crm_element_value_copy(xml_search, PCMK_XA_ID); if (!out->is_quiet(out)) { out->err(out, "WARNING: There is already a meta attribute " "for '%s' called '%s' (id=%s)", top_id, attr_name, found_attr_id); out->err(out, " Delete '%s' first or use the force option " "to override", found_attr_id); } free(found_attr_id); pcmk__xml_free(xml_search); return ENOTUNIQ; } pcmk__xml_free(xml_search); } *resources = g_list_append(*resources, rsc); } else { *resources = find_matching_attr_resources(out, rsc, requested_name, attr_set, attr_set_type, attr_id, attr_name, cib, "update", force); } /* If the user specified attr_set or attr_id, the intent is to modify a * single resource, which will be the last item in the list. */ if ((attr_set != NULL) || (attr_id != NULL)) { GList *last = g_list_last(*resources); *resources = g_list_remove_link(*resources, last); g_list_free(*resources); *resources = last; } return pcmk_rc_ok; } static void free_attr_update_data(gpointer data) { attr_update_data_t *ud = data; if (ud == NULL) { return; } free(ud->attr_set_type); free(ud->attr_set_id); free(ud->attr_name); free(ud->attr_value); free(ud->given_rsc_id); free(ud->found_attr_id); free(ud); } static int update_attribute(pcmk_resource_t *rsc, const char *requested_name, const char *attr_set, const char *attr_set_type, const char *attr_id, const char *attr_name, const char *attr_value, gboolean recursive, cib_t *cib, gboolean force, GList **results) { pcmk__output_t *out = rsc->private->scheduler->priv; int rc = pcmk_rc_ok; GList/**/ *resources = NULL; const char *top_id = pe__const_top_resource(rsc, false)->id; if ((attr_id == NULL) && !force) { find_resource_attr(out, cib, PCMK_XA_ID, top_id, NULL, NULL, NULL, attr_name, NULL); } rc = resources_with_attr(out, cib, rsc, requested_name, attr_set, attr_set_type, attr_id, attr_name, top_id, force, &resources); if (rc != pcmk_rc_ok) { return rc; } for (GList *iter = resources; iter != NULL; iter = iter->next) { char *lookup_id = NULL; char *local_attr_set = NULL; char *found_attr_id = NULL; const char *rsc_attr_id = attr_id; const char *rsc_attr_set = attr_set; xmlNode *xml_top = NULL; xmlNode *xml_obj = NULL; xmlNode *xml_search = NULL; rsc = (pcmk_resource_t *) iter->data; lookup_id = clone_strip(rsc->id); /* Could be a cloned group! */ rc = find_resource_attr(out, cib, PCMK_XA_ID, lookup_id, attr_set_type, attr_set, attr_id, attr_name, &xml_search); switch (rc) { case pcmk_rc_ok: found_attr_id = crm_element_value_copy(xml_search, PCMK_XA_ID); crm_debug("Found a match for " PCMK_XA_NAME "='%s': " PCMK_XA_ID "='%s'", attr_name, found_attr_id); rsc_attr_id = found_attr_id; break; case ENXIO: if (rsc_attr_set == NULL) { local_attr_set = crm_strdup_printf("%s-%s", lookup_id, attr_set_type); rsc_attr_set = local_attr_set; } if (rsc_attr_id == NULL) { found_attr_id = crm_strdup_printf("%s-%s", rsc_attr_set, attr_name); rsc_attr_id = found_attr_id; } xml_top = pcmk__xe_create(NULL, (const char *) rsc->private->xml->name); crm_xml_add(xml_top, PCMK_XA_ID, lookup_id); xml_obj = pcmk__xe_create(xml_top, attr_set_type); crm_xml_add(xml_obj, PCMK_XA_ID, rsc_attr_set); break; default: free(lookup_id); free(found_attr_id); pcmk__xml_free(xml_search); g_list_free(resources); return rc; } xml_obj = crm_create_nvpair_xml(xml_obj, rsc_attr_id, attr_name, attr_value); if (xml_top == NULL) { xml_top = xml_obj; } crm_log_xml_debug(xml_top, "Update"); rc = cib->cmds->modify(cib, PCMK_XE_RESOURCES, xml_top, cib_sync_call); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { attr_update_data_t *ud = pcmk__assert_alloc(1, sizeof(attr_update_data_t)); if (attr_set_type == NULL) { attr_set_type = (const char *) xml_search->parent->name; } if (rsc_attr_set == NULL) { rsc_attr_set = crm_element_value(xml_search->parent, PCMK_XA_ID); } ud->attr_set_type = pcmk__str_copy(attr_set_type); ud->attr_set_id = pcmk__str_copy(rsc_attr_set); ud->attr_name = pcmk__str_copy(attr_name); ud->attr_value = pcmk__str_copy(attr_value); ud->given_rsc_id = pcmk__str_copy(lookup_id); ud->found_attr_id = pcmk__str_copy(found_attr_id); ud->rsc = rsc; *results = g_list_append(*results, ud); } pcmk__xml_free(xml_top); pcmk__xml_free(xml_search); free(lookup_id); free(found_attr_id); free(local_attr_set); if (recursive && pcmk__str_eq(attr_set_type, PCMK_XE_META_ATTRIBUTES, pcmk__str_casei)) { /* We want to set the attribute only on resources explicitly * colocated with this one, so we use * rsc->private->with_this_colocations directly rather than the * with_this_colocations() method. */ pcmk__set_rsc_flags(rsc, pcmk__rsc_detect_loop); for (GList *lpc = rsc->private->with_this_colocations; lpc != NULL; lpc = lpc->next) { pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data; crm_debug("Checking %s %d", cons->id, cons->score); if (pcmk_is_set(cons->dependent->flags, pcmk__rsc_detect_loop) || (cons->score <= 0)) { continue; } crm_debug("Setting %s=%s for dependent resource %s", attr_name, attr_value, cons->dependent->id); update_attribute(cons->dependent, cons->dependent->id, NULL, attr_set_type, NULL, attr_name, attr_value, recursive, cib, force, results); } } } g_list_free(resources); return rc; } // \return Standard Pacemaker return code int cli_resource_update_attribute(pcmk_resource_t *rsc, const char *requested_name, const char *attr_set, const char *attr_set_type, const char *attr_id, const char *attr_name, const char *attr_value, gboolean recursive, cib_t *cib, gboolean force) { static bool need_init = true; int rc = pcmk_rc_ok; GList *results = NULL; pcmk__output_t *out = rsc->private->scheduler->priv; /* If we were asked to update the attribute in a resource element (for * instance, ) there's really not much we need to do. */ if (pcmk__str_eq(attr_set_type, ATTR_SET_ELEMENT, pcmk__str_none)) { return update_element_attribute(out, rsc, cib, attr_name, attr_value); } /* One time initialization - clear flags so we can detect loops */ if (need_init) { need_init = false; pcmk__unpack_constraints(rsc->private->scheduler); pe__clear_resource_flags_on_all(rsc->private->scheduler, pcmk__rsc_detect_loop); } rc = update_attribute(rsc, requested_name, attr_set, attr_set_type, attr_id, attr_name, attr_value, recursive, cib, force, &results); if (rc == pcmk_rc_ok) { if (results == NULL) { return rc; } out->message(out, "attribute-changed-list", results); g_list_free_full(results, free_attr_update_data); } return rc; } // \return Standard Pacemaker return code int cli_resource_delete_attribute(pcmk_resource_t *rsc, const char *requested_name, const char *attr_set, const char *attr_set_type, const char *attr_id, const char *attr_name, cib_t *cib, int cib_options, gboolean force) { pcmk__output_t *out = rsc->private->scheduler->priv; int rc = pcmk_rc_ok; GList/**/ *resources = NULL; if ((attr_id == NULL) && !force) { find_resource_attr(out, cib, PCMK_XA_ID, pe__const_top_resource(rsc, false)->id, NULL, NULL, NULL, attr_name, NULL); } if (pcmk__str_eq(attr_set_type, PCMK_XE_META_ATTRIBUTES, pcmk__str_casei)) { resources = find_matching_attr_resources(out, rsc, requested_name, attr_set, attr_set_type, attr_id, attr_name, cib, "delete", force); } else if (pcmk__str_eq(attr_set_type, ATTR_SET_ELEMENT, pcmk__str_none)) { pcmk__xe_remove_attr(rsc->private->xml, attr_name); CRM_ASSERT(cib != NULL); rc = cib->cmds->replace(cib, PCMK_XE_RESOURCES, rsc->private->xml, cib_options); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { out->info(out, "Deleted attribute: %s", attr_name); } return rc; } else { resources = g_list_append(resources, rsc); } for (GList *iter = resources; iter != NULL; iter = iter->next) { char *lookup_id = NULL; xmlNode *xml_obj = NULL; xmlNode *xml_search = NULL; char *found_attr_id = NULL; const char *rsc_attr_id = attr_id; rsc = (pcmk_resource_t *) iter->data; lookup_id = clone_strip(rsc->id); rc = find_resource_attr(out, cib, PCMK_XA_ID, lookup_id, attr_set_type, attr_set, attr_id, attr_name, &xml_search); switch (rc) { case pcmk_rc_ok: found_attr_id = crm_element_value_copy(xml_search, PCMK_XA_ID); pcmk__xml_free(xml_search); break; case ENXIO: free(lookup_id); pcmk__xml_free(xml_search); continue; default: free(lookup_id); pcmk__xml_free(xml_search); g_list_free(resources); return rc; } if (rsc_attr_id == NULL) { rsc_attr_id = found_attr_id; } xml_obj = crm_create_nvpair_xml(NULL, rsc_attr_id, attr_name, NULL); crm_log_xml_debug(xml_obj, "Delete"); CRM_ASSERT(cib); rc = cib->cmds->remove(cib, PCMK_XE_RESOURCES, xml_obj, cib_options); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { out->info(out, "Deleted '%s' option: " PCMK_XA_ID "=%s%s%s%s%s", lookup_id, found_attr_id, ((attr_set == NULL)? "" : " set="), pcmk__s(attr_set, ""), ((attr_name == NULL)? "" : " " PCMK_XA_NAME "="), pcmk__s(attr_name, "")); } free(lookup_id); pcmk__xml_free(xml_obj); free(found_attr_id); } g_list_free(resources); return rc; } // \return Standard Pacemaker return code static int send_lrm_rsc_op(pcmk_ipc_api_t *controld_api, bool do_fail_resource, const char *host_uname, const char *rsc_id, pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv; const char *router_node = host_uname; const char *rsc_api_id = NULL; const char *rsc_long_id = NULL; const char *rsc_class = NULL; const char *rsc_provider = NULL; const char *rsc_type = NULL; bool cib_only = false; pcmk_resource_t *rsc = pe_find_resource(scheduler->resources, rsc_id); if (rsc == NULL) { out->err(out, "Resource %s not found", rsc_id); return ENXIO; } else if (!pcmk__is_primitive(rsc)) { out->err(out, "We can only process primitive resources, not %s", rsc_id); return EINVAL; } rsc_class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); rsc_provider = crm_element_value(rsc->private->xml, PCMK_XA_PROVIDER); rsc_type = crm_element_value(rsc->private->xml, PCMK_XA_TYPE); if ((rsc_class == NULL) || (rsc_type == NULL)) { out->err(out, "Resource %s does not have a class and type", rsc_id); return EINVAL; } { pcmk_node_t *node = pcmk_find_node(scheduler, host_uname); if (node == NULL) { out->err(out, "Node %s not found", host_uname); return pcmk_rc_node_unknown; } if (!(node->details->online)) { if (do_fail_resource) { out->err(out, "Node %s is not online", host_uname); return ENOTCONN; } else { cib_only = true; } } if (!cib_only && pcmk__is_pacemaker_remote_node(node)) { node = pcmk__current_node(node->details->remote_rsc); if (node == NULL) { out->err(out, "No cluster connection to Pacemaker Remote node %s detected", host_uname); return ENOTCONN; } router_node = node->details->uname; } } if (rsc->private->history_id != NULL) { rsc_api_id = rsc->private->history_id; rsc_long_id = rsc->id; } else { rsc_api_id = rsc->id; } if (do_fail_resource) { return pcmk_controld_api_fail(controld_api, host_uname, router_node, rsc_api_id, rsc_long_id, rsc_class, rsc_provider, rsc_type); } else { return pcmk_controld_api_refresh(controld_api, host_uname, router_node, rsc_api_id, rsc_long_id, rsc_class, rsc_provider, rsc_type, cib_only); } } /*! * \internal * \brief Get resource name as used in failure-related node attributes * * \param[in] rsc Resource to check * * \return Newly allocated string containing resource's fail name * \note The caller is responsible for freeing the result. */ static inline char * rsc_fail_name(const pcmk_resource_t *rsc) { const char *name = pcmk__s(rsc->private->history_id, rsc->id); if (pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { return strdup(name); } return clone_strip(name); } // \return Standard Pacemaker return code static int clear_rsc_history(pcmk_ipc_api_t *controld_api, const char *host_uname, const char *rsc_id, pcmk_scheduler_t *scheduler) { int rc = pcmk_rc_ok; /* Erase the resource's entire LRM history in the CIB, even if we're only * clearing a single operation's fail count. If we erased only entries for a * single operation, we might wind up with a wrong idea of the current * resource state, and we might not re-probe the resource. */ rc = send_lrm_rsc_op(controld_api, false, host_uname, rsc_id, scheduler); if (rc != pcmk_rc_ok) { return rc; } crm_trace("Processing %d mainloop inputs", pcmk_controld_api_replies_expected(controld_api)); while (g_main_context_iteration(NULL, FALSE)) { crm_trace("Processed mainloop input, %d still remaining", pcmk_controld_api_replies_expected(controld_api)); } return rc; } // \return Standard Pacemaker return code static int clear_rsc_failures(pcmk__output_t *out, pcmk_ipc_api_t *controld_api, const char *node_name, const char *rsc_id, const char *operation, const char *interval_spec, pcmk_scheduler_t *scheduler) { int rc = pcmk_rc_ok; const char *failed_value = NULL; const char *failed_id = NULL; char *interval_ms_s = NULL; GHashTable *rscs = NULL; GHashTableIter iter; /* Create a hash table to use as a set of resources to clean. This lets us * clean each resource only once (per node) regardless of how many failed * operations it has. */ rscs = pcmk__strkey_table(NULL, NULL); // Normalize interval to milliseconds for comparison to history entry if (operation) { guint interval_ms = 0U; pcmk_parse_interval_spec(interval_spec, &interval_ms); interval_ms_s = crm_strdup_printf("%u", interval_ms); } for (xmlNode *xml_op = pcmk__xe_first_child(scheduler->failed, NULL, NULL, NULL); xml_op != NULL; xml_op = pcmk__xe_next(xml_op)) { failed_id = crm_element_value(xml_op, PCMK__XA_RSC_ID); if (failed_id == NULL) { // Malformed history entry, should never happen continue; } // No resource specified means all resources match if (rsc_id) { pcmk_resource_t *fail_rsc = NULL; fail_rsc = pe_find_resource_with_flags(scheduler->resources, failed_id, pcmk_rsc_match_history |pcmk_rsc_match_anon_basename); if (!fail_rsc || !pcmk__str_eq(rsc_id, fail_rsc->id, pcmk__str_casei)) { continue; } } // Host name should always have been provided by this point failed_value = crm_element_value(xml_op, PCMK_XA_UNAME); if (!pcmk__str_eq(node_name, failed_value, pcmk__str_casei)) { continue; } // No operation specified means all operations match if (operation) { failed_value = crm_element_value(xml_op, PCMK_XA_OPERATION); if (!pcmk__str_eq(operation, failed_value, pcmk__str_casei)) { continue; } // Interval (if operation was specified) defaults to 0 (not all) failed_value = crm_element_value(xml_op, PCMK_META_INTERVAL); if (!pcmk__str_eq(interval_ms_s, failed_value, pcmk__str_casei)) { continue; } } g_hash_table_add(rscs, (gpointer) failed_id); } free(interval_ms_s); g_hash_table_iter_init(&iter, rscs); while (g_hash_table_iter_next(&iter, (gpointer *) &failed_id, NULL)) { crm_debug("Erasing failures of %s on %s", failed_id, node_name); rc = clear_rsc_history(controld_api, node_name, failed_id, scheduler); if (rc != pcmk_rc_ok) { return rc; } } g_hash_table_destroy(rscs); return rc; } // \return Standard Pacemaker return code static int clear_rsc_fail_attrs(const pcmk_resource_t *rsc, const char *operation, const char *interval_spec, const pcmk_node_t *node) { int rc = pcmk_rc_ok; int attr_options = pcmk__node_attr_none; char *rsc_name = rsc_fail_name(rsc); if (pcmk__is_pacemaker_remote_node(node)) { attr_options |= pcmk__node_attr_remote; } rc = pcmk__attrd_api_clear_failures(NULL, node->details->uname, rsc_name, operation, interval_spec, NULL, attr_options); free(rsc_name); return rc; } // \return Standard Pacemaker return code int cli_resource_delete(pcmk_ipc_api_t *controld_api, const char *host_uname, const pcmk_resource_t *rsc, const char *operation, const char *interval_spec, bool just_failures, pcmk_scheduler_t *scheduler, gboolean force) { pcmk__output_t *out = scheduler->priv; int rc = pcmk_rc_ok; pcmk_node_t *node = NULL; if (rsc == NULL) { return ENXIO; } else if (rsc->children) { for (const GList *lpc = rsc->children; lpc != NULL; lpc = lpc->next) { const pcmk_resource_t *child = (const pcmk_resource_t *) lpc->data; rc = cli_resource_delete(controld_api, host_uname, child, operation, interval_spec, just_failures, scheduler, force); if (rc != pcmk_rc_ok) { return rc; } } return pcmk_rc_ok; } else if (host_uname == NULL) { GList *lpc = NULL; GList *nodes = g_hash_table_get_values(rsc->known_on); if(nodes == NULL && force) { nodes = pcmk__copy_node_list(scheduler->nodes, false); } else if ((nodes == NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_exclusive_probes)) { 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->weight >= 0) { nodes = g_list_prepend(nodes, node); } } } else if(nodes == NULL) { nodes = g_hash_table_get_values(rsc->allowed_nodes); } for (lpc = nodes; lpc != NULL; lpc = lpc->next) { node = (pcmk_node_t *) lpc->data; if (node->details->online) { rc = cli_resource_delete(controld_api, node->details->uname, rsc, operation, interval_spec, just_failures, scheduler, force); } if (rc != pcmk_rc_ok) { g_list_free(nodes); return rc; } } g_list_free(nodes); return pcmk_rc_ok; } node = pcmk_find_node(scheduler, host_uname); if (node == NULL) { out->err(out, "Unable to clean up %s because node %s not found", rsc->id, host_uname); return ENODEV; } if (!node->details->rsc_discovery_enabled) { out->err(out, "Unable to clean up %s because resource discovery disabled on %s", rsc->id, host_uname); return EOPNOTSUPP; } if (controld_api == NULL) { out->err(out, "Dry run: skipping clean-up of %s on %s due to CIB_file", rsc->id, host_uname); return pcmk_rc_ok; } rc = clear_rsc_fail_attrs(rsc, operation, interval_spec, node); if (rc != pcmk_rc_ok) { out->err(out, "Unable to clean up %s failures on %s: %s", rsc->id, host_uname, pcmk_rc_str(rc)); return rc; } if (just_failures) { rc = clear_rsc_failures(out, controld_api, host_uname, rsc->id, operation, interval_spec, scheduler); } else { rc = clear_rsc_history(controld_api, host_uname, rsc->id, scheduler); } if (rc != pcmk_rc_ok) { out->err(out, "Cleaned %s failures on %s, but unable to clean history: %s", rsc->id, host_uname, pcmk_rc_str(rc)); } else { out->info(out, "Cleaned up %s on %s", rsc->id, host_uname); } return rc; } // \return Standard Pacemaker return code int cli_cleanup_all(pcmk_ipc_api_t *controld_api, const char *node_name, const char *operation, const char *interval_spec, pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv; int rc = pcmk_rc_ok; int attr_options = pcmk__node_attr_none; const char *display_name = node_name? node_name : "all nodes"; if (controld_api == NULL) { out->info(out, "Dry run: skipping clean-up of %s due to CIB_file", display_name); return rc; } if (node_name) { pcmk_node_t *node = pcmk_find_node(scheduler, node_name); if (node == NULL) { out->err(out, "Unknown node: %s", node_name); return ENXIO; } if (pcmk__is_pacemaker_remote_node(node)) { attr_options |= pcmk__node_attr_remote; } } rc = pcmk__attrd_api_clear_failures(NULL, node_name, NULL, operation, interval_spec, NULL, attr_options); if (rc != pcmk_rc_ok) { out->err(out, "Unable to clean up all failures on %s: %s", display_name, pcmk_rc_str(rc)); return rc; } if (node_name) { rc = clear_rsc_failures(out, controld_api, node_name, NULL, operation, interval_spec, scheduler); if (rc != pcmk_rc_ok) { out->err(out, "Cleaned all resource failures on %s, but unable to clean history: %s", node_name, pcmk_rc_str(rc)); return rc; } } else { for (GList *iter = scheduler->nodes; iter; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; rc = clear_rsc_failures(out, controld_api, node->details->uname, NULL, operation, interval_spec, scheduler); if (rc != pcmk_rc_ok) { out->err(out, "Cleaned all resource failures on all nodes, but unable to clean history: %s", pcmk_rc_str(rc)); return rc; } } } out->info(out, "Cleaned up all resources on %s", display_name); return rc; } static void check_role(resource_checks_t *checks) { const char *role_s = g_hash_table_lookup(checks->rsc->meta, PCMK_META_TARGET_ROLE); if (role_s == NULL) { return; } switch (pcmk_parse_role(role_s)) { case pcmk_role_stopped: checks->flags |= rsc_remain_stopped; break; case pcmk_role_unpromoted: if (pcmk_is_set(pe__const_top_resource(checks->rsc, false)->flags, pcmk__rsc_promotable)) { checks->flags |= rsc_unpromotable; } break; default: break; } } static void check_managed(resource_checks_t *checks) { const char *managed_s = g_hash_table_lookup(checks->rsc->meta, PCMK_META_IS_MANAGED); if ((managed_s != NULL) && !crm_is_true(managed_s)) { checks->flags |= rsc_unmanaged; } } static void check_locked(resource_checks_t *checks) { if (checks->rsc->lock_node != NULL) { checks->flags |= rsc_locked; checks->lock_node = checks->rsc->lock_node->details->uname; } } static bool node_is_unhealthy(pcmk_node_t *node) { switch (pe__health_strategy(node->details->data_set)) { case pcmk__health_strategy_none: break; case pcmk__health_strategy_no_red: if (pe__node_health(node) < 0) { return true; } break; case pcmk__health_strategy_only_green: if (pe__node_health(node) <= 0) { return true; } break; case pcmk__health_strategy_progressive: case pcmk__health_strategy_custom: /* @TODO These are finite scores, possibly with rules, and possibly * combining with other scores, so attributing these as a cause is * nontrivial. */ break; } return false; } static void check_node_health(resource_checks_t *checks, pcmk_node_t *node) { if (node == NULL) { GHashTableIter iter; bool allowed = false; bool all_nodes_unhealthy = true; g_hash_table_iter_init(&iter, checks->rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { allowed = true; if (!node_is_unhealthy(node)) { all_nodes_unhealthy = false; break; } } if (allowed && all_nodes_unhealthy) { checks->flags |= rsc_node_health; } } else if (node_is_unhealthy(node)) { checks->flags |= rsc_node_health; } } int cli_resource_check(pcmk__output_t *out, pcmk_resource_t *rsc, pcmk_node_t *node) { resource_checks_t checks = { .rsc = rsc }; check_role(&checks); check_managed(&checks); check_locked(&checks); check_node_health(&checks, node); return out->message(out, "resource-check-list", &checks); } // \return Standard Pacemaker return code int cli_resource_fail(pcmk_ipc_api_t *controld_api, const char *host_uname, const char *rsc_id, pcmk_scheduler_t *scheduler) { crm_notice("Failing %s on %s", rsc_id, host_uname); return send_lrm_rsc_op(controld_api, true, host_uname, rsc_id, scheduler); } static GHashTable * generate_resource_params(pcmk_resource_t *rsc, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { GHashTable *params = NULL; GHashTable *meta = NULL; GHashTable *combined = NULL; GHashTableIter iter; char *key = NULL; char *value = NULL; combined = pcmk__strkey_table(free, free); params = pe_rsc_params(rsc, node, scheduler); if (params != NULL) { g_hash_table_iter_init(&iter, params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { pcmk__insert_dup(combined, key, value); } } meta = pcmk__strkey_table(free, free); get_meta_attributes(meta, rsc, NULL, scheduler); if (meta != NULL) { g_hash_table_iter_init(&iter, meta); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { char *crm_name = crm_meta_name(key); g_hash_table_insert(combined, crm_name, strdup(value)); } g_hash_table_destroy(meta); } return combined; } bool resource_is_running_on(pcmk_resource_t *rsc, const char *host) { bool found = true; GList *hIter = NULL; GList *hosts = NULL; if (rsc == NULL) { return false; } rsc->private->fns->location(rsc, &hosts, TRUE); for (hIter = hosts; host != NULL && hIter != NULL; hIter = hIter->next) { pcmk_node_t *node = (pcmk_node_t *) hIter->data; if (pcmk__strcase_any_of(host, node->details->uname, node->details->id, NULL)) { crm_trace("Resource %s is running on %s\n", rsc->id, host); goto done; } } if (host != NULL) { crm_trace("Resource %s is not running on: %s\n", rsc->id, host); found = false; } else if(host == NULL && hosts == NULL) { crm_trace("Resource %s is not running\n", rsc->id); found = false; } done: g_list_free(hosts); return found; } /*! * \internal * \brief Create a list of all resources active on host from a given list * * \param[in] host Name of host to check whether resources are active * \param[in] rsc_list List of resources to check * * \return New list of resources from list that are active on host */ static GList * get_active_resources(const char *host, GList *rsc_list) { GList *rIter = NULL; GList *active = NULL; for (rIter = rsc_list; rIter != NULL; rIter = rIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) rIter->data; /* Expand groups to their members, because if we're restarting a member * other than the first, we can't otherwise tell which resources are * stopping and starting. */ if (pcmk__is_group(rsc)) { active = g_list_concat(active, get_active_resources(host, rsc->children)); } else if (resource_is_running_on(rsc, host)) { active = g_list_append(active, strdup(rsc->id)); } } return active; } static void dump_list(GList *items, const char *tag) { int lpc = 0; GList *item = NULL; for (item = items; item != NULL; item = item->next) { crm_trace("%s[%d]: %s", tag, lpc, (char*)item->data); lpc++; } } static void display_list(pcmk__output_t *out, GList *items, const char *tag) { GList *item = NULL; for (item = items; item != NULL; item = item->next) { out->info(out, "%s%s", tag, (const char *)item->data); } } /*! * \internal * \brief Upgrade XML to latest schema version and use it as scheduler input * * This also updates the scheduler timestamp to the current time. * * \param[in,out] scheduler Scheduler data to update * \param[in,out] xml XML to use as input * * \return Standard Pacemaker return code * \note On success, caller is responsible for freeing memory allocated for * scheduler->now. */ int update_scheduler_input(pcmk_scheduler_t *scheduler, xmlNode **xml) { int rc = pcmk_update_configured_schema(xml, false); if (rc == pcmk_rc_ok) { scheduler->input = *xml; scheduler->now = crm_time_new(NULL); } return pcmk_rc_ok; } /*! * \internal * \brief Update scheduler XML input based on a CIB query * * \param[in] scheduler Scheduler data to initialize * \param[in] cib Connection to the CIB manager * * \return Standard Pacemaker return code * \note On success, caller is responsible for freeing memory allocated for * scheduler->input and scheduler->now. */ static int update_scheduler_input_to_cib(pcmk__output_t *out, pcmk_scheduler_t *scheduler, cib_t *cib) { xmlNode *cib_xml_copy = NULL; int rc = pcmk_rc_ok; rc = cib->cmds->query(cib, NULL, &cib_xml_copy, cib_scope_local | cib_sync_call); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { out->err(out, "Could not obtain the current CIB: %s (%d)", pcmk_rc_str(rc), rc); return rc; } rc = update_scheduler_input(scheduler, &cib_xml_copy); if (rc != pcmk_rc_ok) { out->err(out, "Could not upgrade the current CIB XML"); pcmk__xml_free(cib_xml_copy); return rc; } return rc; } // \return Standard Pacemaker return code static int update_dataset(cib_t *cib, pcmk_scheduler_t *scheduler, bool simulate) { char *pid = NULL; char *shadow_file = NULL; cib_t *shadow_cib = NULL; int rc = pcmk_rc_ok; pcmk__output_t *out = scheduler->priv; pe_reset_working_set(scheduler); pcmk__set_scheduler_flags(scheduler, pcmk_sched_no_counts|pcmk_sched_no_compat); rc = update_scheduler_input_to_cib(out, scheduler, cib); if (rc != pcmk_rc_ok) { return rc; } if(simulate) { bool prev_quiet = false; pid = pcmk__getpid_s(); shadow_cib = cib_shadow_new(pid); shadow_file = get_shadow_file(pid); if (shadow_cib == NULL) { out->err(out, "Could not create shadow cib: '%s'", pid); rc = ENXIO; goto done; } rc = pcmk__xml_write_file(scheduler->input, shadow_file, false, NULL); if (rc != pcmk_rc_ok) { out->err(out, "Could not populate shadow cib: %s", pcmk_rc_str(rc)); goto done; } rc = shadow_cib->cmds->signon(shadow_cib, crm_system_name, cib_command); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { out->err(out, "Could not connect to shadow cib: %s", pcmk_rc_str(rc)); goto done; } pcmk__schedule_actions(scheduler->input, pcmk_sched_no_counts|pcmk_sched_no_compat, scheduler); prev_quiet = out->is_quiet(out); out->quiet = true; pcmk__simulate_transition(scheduler, shadow_cib, NULL); out->quiet = prev_quiet; rc = update_dataset(shadow_cib, scheduler, false); } else { cluster_status(scheduler); } done: // Do not free scheduler->input because rsc->private->xml must remain valid cib_delete(shadow_cib); free(pid); if(shadow_file) { unlink(shadow_file); free(shadow_file); } return rc; } /*! * \internal * \brief Find the maximum stop timeout of a resource and its children (if any) * * \param[in,out] rsc Resource to get timeout for * * \return Maximum stop timeout for \p rsc (in milliseconds) */ static guint max_rsc_stop_timeout(pcmk_resource_t *rsc) { long long result_ll; guint max_delay = 0; xmlNode *config = NULL; GHashTable *meta = NULL; if (rsc == NULL) { return 0; } // If resource is collective, use maximum of its children's stop timeouts if (rsc->children != NULL) { for (GList *iter = rsc->children; iter; iter = iter->next) { pcmk_resource_t *child = iter->data; guint delay = max_rsc_stop_timeout(child); if (delay > max_delay) { pcmk__rsc_trace(rsc, "Maximum stop timeout for %s is now %s " "due to %s", rsc->id, pcmk__readable_interval(delay), child->id); max_delay = delay; } } return max_delay; } // Get resource's stop action configuration from CIB config = pcmk__find_action_config(rsc, PCMK_ACTION_STOP, 0, true); /* Get configured timeout for stop action (fully evaluated for rules, * defaults, etc.). * * @TODO This currently ignores node (which might matter for rules) */ meta = pcmk__unpack_action_meta(rsc, NULL, PCMK_ACTION_STOP, 0, config); if ((pcmk__scan_ll(g_hash_table_lookup(meta, PCMK_META_TIMEOUT), &result_ll, -1LL) == pcmk_rc_ok) && (result_ll >= 0)) { max_delay = (guint) QB_MIN(result_ll, UINT_MAX); } g_hash_table_destroy(meta); return max_delay; } /*! * \internal * \brief Find a reasonable waiting time for stopping any one resource in a list * * \param[in,out] scheduler Scheduler data * \param[in] resources List of names of resources that will be stopped * * \return Rough estimate of a reasonable time to wait (in seconds) to stop any * one resource in \p resources * \note This estimate is very rough, simply the maximum stop timeout of all * given resources and their children, plus a small fudge factor. It does * not account for children that must be stopped in sequence, action * throttling, or any demotions needed. It checks the stop timeout, even * if the resources in question are actually being started. */ static guint wait_time_estimate(pcmk_scheduler_t *scheduler, const GList *resources) { guint max_delay = 0U; // Find maximum stop timeout in milliseconds for (const GList *item = resources; item != NULL; item = item->next) { pcmk_resource_t *rsc = pe_find_resource(scheduler->resources, (const char *) item->data); guint delay = max_rsc_stop_timeout(rsc); if (delay > max_delay) { pcmk__rsc_trace(rsc, "Wait time is now %s due to %s", pcmk__readable_interval(delay), rsc->id); max_delay = delay; } } return (max_delay / 1000U) + 5U; } #define waiting_for_starts(d, r, h) ((d != NULL) || \ (!resource_is_running_on((r), (h)))) /*! * \internal * \brief Restart a resource (on a particular host if requested). * * \param[in,out] out Output object * \param[in,out] rsc The resource to restart * \param[in] node Node to restart resource on (NULL for all) * \param[in] move_lifetime If not NULL, how long constraint should * remain in effect (as ISO 8601 string) * \param[in] timeout_ms Consider failed if actions do not complete * in this time (specified in milliseconds, * but a two-second granularity is actually * used; if 0, it will be calculated based on * the resource timeout) * \param[in,out] cib Connection to the CIB manager * \param[in] cib_options Group of enum cib_call_options flags to * use with CIB calls * \param[in] promoted_role_only If true, limit to promoted instances * \param[in] force If true, apply only to requested instance * if part of a collective resource * * \return Standard Pacemaker return code (exits on certain failures) */ int cli_resource_restart(pcmk__output_t *out, pcmk_resource_t *rsc, const pcmk_node_t *node, const char *move_lifetime, guint timeout_ms, cib_t *cib, int cib_options, gboolean promoted_role_only, gboolean force) { int rc = pcmk_rc_ok; int lpc = 0; int before = 0; guint step_timeout_s = 0; guint sleep_interval = 2U; guint timeout = timeout_ms / 1000U; bool stop_via_ban = false; char *rsc_id = NULL; char *lookup_id = NULL; char *orig_target_role = NULL; GList *list_delta = NULL; GList *target_active = NULL; GList *current_active = NULL; GList *restart_target_active = NULL; pcmk_scheduler_t *scheduler = NULL; pcmk_resource_t *parent = uber_parent(rsc); bool running = false; const char *id = pcmk__s(rsc->private->history_id, rsc->id); const char *host = node ? node->details->uname : NULL; /* If the implicit resource or primitive resource of a bundle is given, operate on the * bundle itself instead. */ if (pcmk__is_bundled(rsc)) { rsc = parent->private->parent; } running = resource_is_running_on(rsc, host); if (pcmk__is_clone(parent) && !running) { if (pcmk__is_unique_clone(parent)) { lookup_id = strdup(rsc->id); } else { lookup_id = clone_strip(rsc->id); } rsc = parent->private->fns->find_rsc(parent, lookup_id, node, pcmk_rsc_match_basename |pcmk_rsc_match_current_node); free(lookup_id); running = resource_is_running_on(rsc, host); } if (!running) { if (host) { out->err(out, "%s is not running on %s and so cannot be restarted", id, host); } else { out->err(out, "%s is not running anywhere and so cannot be restarted", id); } return ENXIO; } if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { out->err(out, "Unmanaged resources cannot be restarted."); return EAGAIN; } rsc_id = strdup(rsc->id); if (pcmk__is_unique_clone(parent)) { lookup_id = strdup(rsc->id); } else { lookup_id = clone_strip(rsc->id); } if (host) { if (pcmk__is_clone(rsc) || pe_bundle_replicas(rsc)) { stop_via_ban = true; } else if (pcmk__is_clone(parent)) { stop_via_ban = true; free(lookup_id); lookup_id = strdup(parent->id); } } /* grab full cib determine originally active resources disable or ban poll cib and watch for affected resources to get stopped without --timeout, calculate the stop timeout for each step and wait for that if we hit --timeout or the service timeout, re-enable or un-ban, report failure and indicate which resources we couldn't take down if everything stopped, re-enable or un-ban poll cib and watch for affected resources to get started without --timeout, calculate the start timeout for each step and wait for that if we hit --timeout or the service timeout, report (different) failure and indicate which resources we couldn't bring back up report success Optimizations: - use constraints to determine ordered list of affected resources - Allow a --no-deps option (aka. --force-restart) */ scheduler = pe_new_working_set(); if (scheduler == NULL) { rc = errno; out->err(out, "Could not allocate scheduler data: %s", pcmk_rc_str(rc)); goto done; } scheduler->priv = out; rc = update_dataset(cib, scheduler, false); if(rc != pcmk_rc_ok) { out->err(out, "Could not get new resource list: %s (%d)", pcmk_rc_str(rc), rc); goto done; } restart_target_active = get_active_resources(host, scheduler->resources); current_active = get_active_resources(host, scheduler->resources); dump_list(current_active, "Origin"); if (stop_via_ban) { /* Stop the clone or bundle instance by banning it from the host */ out->quiet = true; rc = cli_resource_ban(out, lookup_id, host, move_lifetime, cib, cib_options, promoted_role_only, PCMK_ROLE_PROMOTED); } else { xmlNode *xml_search = NULL; /* Stop the resource by setting PCMK_META_TARGET_ROLE to Stopped. * Remember any existing PCMK_META_TARGET_ROLE so we can restore it * later (though it only makes any difference if it's Unpromoted). */ rc = find_resource_attr(out, cib, PCMK_XA_VALUE, lookup_id, NULL, NULL, NULL, PCMK_META_TARGET_ROLE, &xml_search); if (rc == pcmk_rc_ok) { orig_target_role = crm_element_value_copy(xml_search, PCMK_XA_VALUE); } pcmk__xml_free(xml_search); rc = cli_resource_update_attribute(rsc, rsc_id, NULL, PCMK_XE_META_ATTRIBUTES, NULL, PCMK_META_TARGET_ROLE, PCMK_ACTION_STOPPED, FALSE, cib, force); } if(rc != pcmk_rc_ok) { out->err(out, "Could not set " PCMK_META_TARGET_ROLE " for %s: %s (%d)", rsc_id, pcmk_rc_str(rc), rc); if (current_active != NULL) { g_list_free_full(current_active, free); current_active = NULL; } if (restart_target_active != NULL) { g_list_free_full(restart_target_active, free); restart_target_active = NULL; } goto done; } rc = update_dataset(cib, scheduler, true); if(rc != pcmk_rc_ok) { out->err(out, "Could not determine which resources would be stopped"); goto failure; } target_active = get_active_resources(host, scheduler->resources); dump_list(target_active, "Target"); list_delta = pcmk__subtract_lists(current_active, target_active, (GCompareFunc) strcmp); out->info(out, "Waiting for %d resources to stop:", g_list_length(list_delta)); display_list(out, list_delta, " * "); step_timeout_s = timeout / sleep_interval; while (list_delta != NULL) { before = g_list_length(list_delta); if(timeout_ms == 0) { step_timeout_s = wait_time_estimate(scheduler, list_delta) / sleep_interval; } /* We probably don't need the entire step timeout */ for(lpc = 0; (lpc < step_timeout_s) && (list_delta != NULL); lpc++) { sleep(sleep_interval); if(timeout) { timeout -= sleep_interval; crm_trace("%us remaining", timeout); } rc = update_dataset(cib, scheduler, FALSE); if(rc != pcmk_rc_ok) { out->err(out, "Could not determine which resources were stopped"); goto failure; } if (current_active != NULL) { g_list_free_full(current_active, free); } current_active = get_active_resources(host, scheduler->resources); g_list_free(list_delta); list_delta = pcmk__subtract_lists(current_active, target_active, (GCompareFunc) strcmp); dump_list(current_active, "Current"); dump_list(list_delta, "Delta"); } crm_trace("%d (was %d) resources remaining", g_list_length(list_delta), before); if(before == g_list_length(list_delta)) { /* aborted during stop phase, print the contents of list_delta */ out->err(out, "Could not complete shutdown of %s, %d resources remaining", rsc_id, g_list_length(list_delta)); display_list(out, list_delta, " * "); rc = ETIME; goto failure; } } if (stop_via_ban) { rc = cli_resource_clear(lookup_id, host, NULL, cib, cib_options, true, force); } else if (orig_target_role) { rc = cli_resource_update_attribute(rsc, rsc_id, NULL, PCMK_XE_META_ATTRIBUTES, NULL, PCMK_META_TARGET_ROLE, orig_target_role, FALSE, cib, force); free(orig_target_role); orig_target_role = NULL; } else { rc = cli_resource_delete_attribute(rsc, rsc_id, NULL, PCMK_XE_META_ATTRIBUTES, NULL, PCMK_META_TARGET_ROLE, cib, cib_options, force); } if(rc != pcmk_rc_ok) { out->err(out, "Could not unset " PCMK_META_TARGET_ROLE " for %s: %s (%d)", rsc_id, pcmk_rc_str(rc), rc); goto done; } if (target_active != NULL) { g_list_free_full(target_active, free); } target_active = restart_target_active; list_delta = pcmk__subtract_lists(target_active, current_active, (GCompareFunc) strcmp); out->info(out, "Waiting for %d resources to start again:", g_list_length(list_delta)); display_list(out, list_delta, " * "); step_timeout_s = timeout / sleep_interval; while (waiting_for_starts(list_delta, rsc, host)) { before = g_list_length(list_delta); if(timeout_ms == 0) { step_timeout_s = wait_time_estimate(scheduler, list_delta) / sleep_interval; } /* We probably don't need the entire step timeout */ for (lpc = 0; (lpc < step_timeout_s) && waiting_for_starts(list_delta, rsc, host); lpc++) { sleep(sleep_interval); if(timeout) { timeout -= sleep_interval; crm_trace("%ds remaining", timeout); } rc = update_dataset(cib, scheduler, false); if(rc != pcmk_rc_ok) { out->err(out, "Could not determine which resources were started"); goto failure; } /* It's OK if dependent resources moved to a different node, * so we check active resources on all nodes. */ if (current_active != NULL) { g_list_free_full(current_active, free); } current_active = get_active_resources(NULL, scheduler->resources); g_list_free(list_delta); list_delta = pcmk__subtract_lists(target_active, current_active, (GCompareFunc) strcmp); dump_list(current_active, "Current"); dump_list(list_delta, "Delta"); } if(before == g_list_length(list_delta)) { /* aborted during start phase, print the contents of list_delta */ out->err(out, "Could not complete restart of %s, %d resources remaining", rsc_id, g_list_length(list_delta)); display_list(out, list_delta, " * "); rc = ETIME; goto failure; } } rc = pcmk_rc_ok; goto done; failure: if (stop_via_ban) { cli_resource_clear(lookup_id, host, NULL, cib, cib_options, true, force); } else if (orig_target_role) { cli_resource_update_attribute(rsc, rsc_id, NULL, PCMK_XE_META_ATTRIBUTES, NULL, PCMK_META_TARGET_ROLE, orig_target_role, FALSE, cib, force); free(orig_target_role); } else { cli_resource_delete_attribute(rsc, rsc_id, NULL, PCMK_XE_META_ATTRIBUTES, NULL, PCMK_META_TARGET_ROLE, cib, cib_options, force); } done: if (list_delta != NULL) { g_list_free(list_delta); } if (current_active != NULL) { g_list_free_full(current_active, free); } if (target_active != NULL && (target_active != restart_target_active)) { g_list_free_full(target_active, free); } if (restart_target_active != NULL) { g_list_free_full(restart_target_active, free); } free(rsc_id); free(lookup_id); pe_free_working_set(scheduler); return rc; } static inline bool action_is_pending(const pcmk_action_t *action) { if (pcmk_any_flags_set(action->flags, pcmk_action_optional|pcmk_action_pseudo) || !pcmk_is_set(action->flags, pcmk_action_runnable) || pcmk__str_eq(PCMK_ACTION_NOTIFY, action->task, pcmk__str_casei)) { return false; } return true; } /*! * \internal * \brief Check whether any actions in a list are pending * * \param[in] actions List of actions to check * * \return true if any actions in the list are pending, otherwise false */ static bool actions_are_pending(const GList *actions) { for (const GList *action = actions; action != NULL; action = action->next) { const pcmk_action_t *a = (const pcmk_action_t *) action->data; if (action_is_pending(a)) { crm_notice("Waiting for %s (flags=%#.8x)", a->uuid, a->flags); return true; } } return false; } static void print_pending_actions(pcmk__output_t *out, GList *actions) { GList *action; out->info(out, "Pending actions:"); for (action = actions; action != NULL; action = action->next) { pcmk_action_t *a = (pcmk_action_t *) action->data; if (!action_is_pending(a)) { continue; } if (a->node) { out->info(out, "\tAction %d: %s\ton %s", a->id, a->uuid, pcmk__node_name(a->node)); } else { out->info(out, "\tAction %d: %s", a->id, a->uuid); } } } /* For --wait, timeout (in seconds) to use if caller doesn't specify one */ #define WAIT_DEFAULT_TIMEOUT_S (60 * 60) /* For --wait, how long to sleep between cluster state checks */ #define WAIT_SLEEP_S (2) /*! * \internal * \brief Wait until all pending cluster actions are complete * * This waits until either the CIB's transition graph is idle or a timeout is * reached. * * \param[in,out] out Output object * \param[in] timeout_ms Consider failed if actions do not complete in * this time (specified in milliseconds, but * one-second granularity is actually used; if 0, a * default will be used) * \param[in,out] cib Connection to the CIB manager * * \return Standard Pacemaker return code */ int wait_till_stable(pcmk__output_t *out, guint timeout_ms, cib_t * cib) { pcmk_scheduler_t *scheduler = NULL; xmlXPathObjectPtr search; int rc = pcmk_rc_ok; bool pending_unknown_state_resources; time_t expire_time = time(NULL); time_t time_diff; bool printed_version_warning = out->is_quiet(out); // i.e. don't print if quiet char *xpath = NULL; if (timeout_ms == 0) { expire_time += WAIT_DEFAULT_TIMEOUT_S; } else { expire_time += (timeout_ms + 999) / 1000; } scheduler = pe_new_working_set(); if (scheduler == NULL) { return ENOMEM; } xpath = crm_strdup_printf("/" PCMK_XE_CIB "/" PCMK_XE_STATUS "/" PCMK__XE_NODE_STATE "/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES "/" PCMK__XE_LRM_RESOURCE "/" PCMK__XE_LRM_RSC_OP "[@" PCMK__XA_RC_CODE "='%d']", PCMK_OCF_UNKNOWN); do { /* Abort if timeout is reached */ time_diff = expire_time - time(NULL); if (time_diff <= 0) { print_pending_actions(out, scheduler->actions); rc = ETIME; break; } crm_info("Waiting up to %lld seconds for cluster actions to complete", (long long) time_diff); if (rc == pcmk_rc_ok) { /* this avoids sleep on first loop iteration */ sleep(WAIT_SLEEP_S); } /* Get latest transition graph */ pe_reset_working_set(scheduler); rc = update_scheduler_input_to_cib(out, scheduler, cib); if (rc != pcmk_rc_ok) { break; } pcmk__schedule_actions(scheduler->input, pcmk_sched_no_counts|pcmk_sched_no_compat, scheduler); if (!printed_version_warning) { /* If the DC has a different version than the local node, the two * could come to different conclusions about what actions need to be * done. Warn the user in this case. * * @TODO A possible long-term solution would be to reimplement the * wait as a new controller operation that would be forwarded to the * DC. However, that would have potential problems of its own. */ const char *dc_version = g_hash_table_lookup(scheduler->config_hash, PCMK_OPT_DC_VERSION); if (!pcmk__str_eq(dc_version, PACEMAKER_VERSION "-" BUILD_VERSION, pcmk__str_casei)) { out->info(out, "warning: wait option may not work properly in " "mixed-version cluster"); printed_version_warning = true; } } search = xpath_search(scheduler->input, xpath); pending_unknown_state_resources = (numXpathResults(search) > 0); freeXpathObject(search); } while (actions_are_pending(scheduler->actions) || pending_unknown_state_resources); pe_free_working_set(scheduler); free(xpath); return rc; } static const char * get_action(const char *rsc_action) { const char *action = NULL; if (pcmk__str_eq(rsc_action, "validate", pcmk__str_casei)) { action = PCMK_ACTION_VALIDATE_ALL; } else if (pcmk__str_eq(rsc_action, "force-check", pcmk__str_casei)) { action = PCMK_ACTION_MONITOR; } else if (pcmk__strcase_any_of(rsc_action, "force-start", "force-stop", "force-demote", "force-promote", NULL)) { action = rsc_action+6; } else { action = rsc_action; } return action; } /*! * \brief Set up environment variables as expected by resource agents * * When the cluster executes resource agents, it adds certain environment * variables (directly or via resource meta-attributes) expected by some * resource agents. Add the essential ones that many resource agents expect, so * the behavior is the same for command-line execution. * * \param[in,out] params Resource parameters that will be passed to agent * \param[in] timeout_ms Action timeout (in milliseconds) * \param[in] check_level OCF check level * \param[in] verbosity Verbosity level */ static void set_agent_environment(GHashTable *params, guint timeout_ms, int check_level, int verbosity) { g_hash_table_insert(params, crm_meta_name(PCMK_META_TIMEOUT), crm_strdup_printf("%u", timeout_ms)); pcmk__insert_dup(params, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); if (check_level >= 0) { char *level = crm_strdup_printf("%d", check_level); setenv("OCF_CHECK_LEVEL", level, 1); free(level); } pcmk__set_env_option(PCMK__ENV_DEBUG, ((verbosity > 0)? "1" : "0"), true); if (verbosity > 1) { setenv("OCF_TRACE_RA", "1", 1); } /* A resource agent using the standard ocf-shellfuncs library will not print * messages to stderr if it doesn't have a controlling terminal (e.g. if * crm_resource is called via script or ssh). This forces it to do so. */ setenv("OCF_TRACE_FILE", "/dev/stderr", 0); } /*! * \internal * \brief Apply command-line overrides to resource parameters * * \param[in,out] params Parameters to be passed to agent * \param[in] overrides Parameters to override (or NULL if none) */ static void apply_overrides(GHashTable *params, GHashTable *overrides) { if (overrides != NULL) { GHashTableIter iter; char *name = NULL; char *value = NULL; g_hash_table_iter_init(&iter, overrides); while (g_hash_table_iter_next(&iter, (gpointer *) &name, (gpointer *) &value)) { pcmk__insert_dup(params, name, value); } } } crm_exit_t cli_resource_execute_from_params(pcmk__output_t *out, const char *rsc_name, const char *rsc_class, const char *rsc_prov, const char *rsc_type, const char *rsc_action, GHashTable *params, GHashTable *override_hash, guint timeout_ms, int resource_verbose, gboolean force, int check_level) { const char *class = rsc_class; const char *action = get_action(rsc_action); crm_exit_t exit_code = CRM_EX_OK; svc_action_t *op = NULL; // If no timeout was provided, use the same default as the cluster if (timeout_ms == 0U) { timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS; } set_agent_environment(params, timeout_ms, check_level, resource_verbose); apply_overrides(params, override_hash); op = services__create_resource_action(rsc_name? rsc_name : "test", rsc_class, rsc_prov, rsc_type, action, 0, QB_MIN(timeout_ms, INT_MAX), params, 0); if (op == NULL) { out->err(out, "Could not execute %s using %s%s%s:%s: %s", action, rsc_class, (rsc_prov? ":" : ""), (rsc_prov? rsc_prov : ""), rsc_type, strerror(ENOMEM)); g_hash_table_destroy(params); return CRM_EX_OSERR; } if (pcmk__str_eq(rsc_class, PCMK_RESOURCE_CLASS_SERVICE, pcmk__str_casei)) { class = resources_find_service_class(rsc_type); } if (!pcmk__strcase_any_of(class, PCMK_RESOURCE_CLASS_OCF, PCMK_RESOURCE_CLASS_LSB, NULL)) { services__format_result(op, CRM_EX_UNIMPLEMENT_FEATURE, PCMK_EXEC_ERROR, "Manual execution of the %s standard is " "unsupported", pcmk__s(class, "unspecified")); } if (op->rc != PCMK_OCF_UNKNOWN) { exit_code = op->rc; goto done; } services_action_sync(op); // Map results to OCF codes for consistent reporting to user { enum ocf_exitcode ocf_code = services_result2ocf(class, action, op->rc); // Cast variable instead of function return to keep compilers happy exit_code = (crm_exit_t) ocf_code; } done: out->message(out, "resource-agent-action", resource_verbose, rsc_class, rsc_prov, rsc_type, rsc_name, rsc_action, override_hash, exit_code, op->status, services__exit_reason(op), op->stdout_data, op->stderr_data); services_action_free(op); return exit_code; } /*! * \internal * \brief Get the timeout the cluster would use for an action * * \param[in] rsc Resource that action is for * \param[in] action Name of action */ static guint get_action_timeout(pcmk_resource_t *rsc, const char *action) { long long timeout_ms = -1LL; xmlNode *op = pcmk__find_action_config(rsc, action, 0, true); GHashTable *meta = pcmk__unpack_action_meta(rsc, NULL, action, 0, op); if ((pcmk__scan_ll(g_hash_table_lookup(meta, PCMK_META_TIMEOUT), &timeout_ms, -1LL) != pcmk_rc_ok) || (timeout_ms <= 0LL)) { timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS; } g_hash_table_destroy(meta); return (guint) QB_MIN(timeout_ms, UINT_MAX); } crm_exit_t cli_resource_execute(pcmk_resource_t *rsc, const char *requested_name, const char *rsc_action, GHashTable *override_hash, guint timeout_ms, cib_t *cib, pcmk_scheduler_t *scheduler, int resource_verbose, gboolean force, int check_level) { pcmk__output_t *out = scheduler->priv; crm_exit_t exit_code = CRM_EX_OK; const char *rid = requested_name; const char *rtype = NULL; const char *rprov = NULL; const char *rclass = NULL; GHashTable *params = NULL; if (pcmk__strcase_any_of(rsc_action, "force-start", "force-demote", "force-promote", NULL)) { if (pcmk__is_clone(rsc)) { GList *nodes = cli_resource_search(rsc, requested_name, scheduler); if(nodes != NULL && force == FALSE) { out->err(out, "It is not safe to %s %s here: the cluster claims it is already active", rsc_action, rsc->id); out->err(out, "Try setting " PCMK_META_TARGET_ROLE "=" PCMK_ROLE_STOPPED " first or specifying the force option"); return CRM_EX_UNSAFE; } g_list_free_full(nodes, free); } } if (pcmk__is_clone(rsc)) { /* Grab the first child resource in the hope it's not a group */ rsc = rsc->children->data; } if (pcmk__is_group(rsc)) { out->err(out, "Sorry, the %s option doesn't support group resources", rsc_action); return CRM_EX_UNIMPLEMENT_FEATURE; } else if (pcmk__is_bundled(rsc)) { out->err(out, "Sorry, the %s option doesn't support bundled resources", rsc_action); return CRM_EX_UNIMPLEMENT_FEATURE; } rclass = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); rprov = crm_element_value(rsc->private->xml, PCMK_XA_PROVIDER); rtype = crm_element_value(rsc->private->xml, PCMK_XA_TYPE); params = generate_resource_params(rsc, NULL /* @TODO use local node */, scheduler); if (timeout_ms == 0U) { timeout_ms = get_action_timeout(rsc, get_action(rsc_action)); } if (!pcmk__is_anonymous_clone(rsc->private->parent)) { rid = rsc->id; } exit_code = cli_resource_execute_from_params(out, rid, rclass, rprov, rtype, rsc_action, params, override_hash, timeout_ms, resource_verbose, force, check_level); return exit_code; } // \return Standard Pacemaker return code int cli_resource_move(const pcmk_resource_t *rsc, const char *rsc_id, const char *host_name, const char *move_lifetime, cib_t *cib, int cib_options, pcmk_scheduler_t *scheduler, gboolean promoted_role_only, gboolean force) { pcmk__output_t *out = scheduler->priv; int rc = pcmk_rc_ok; unsigned int count = 0; pcmk_node_t *current = NULL; pcmk_node_t *dest = pcmk_find_node(scheduler, host_name); bool cur_is_dest = false; if (dest == NULL) { return pcmk_rc_node_unknown; } if (promoted_role_only && !pcmk_is_set(rsc->flags, pcmk__rsc_promotable)) { const pcmk_resource_t *p = pe__const_top_resource(rsc, false); if (pcmk_is_set(p->flags, pcmk__rsc_promotable)) { out->info(out, "Using parent '%s' for move instead of '%s'.", rsc->id, rsc_id); rsc_id = p->id; rsc = p; } else { out->info(out, "Ignoring --promoted option: %s is not promotable", rsc_id); promoted_role_only = FALSE; } } current = pe__find_active_requires(rsc, &count); if (pcmk_is_set(rsc->flags, pcmk__rsc_promotable)) { unsigned int promoted_count = 0; pcmk_node_t *promoted_node = NULL; for (const GList *iter = rsc->children; iter; iter = iter->next) { const pcmk_resource_t *child = (const pcmk_resource_t *) iter->data; enum rsc_role_e child_role = child->private->fns->state(child, TRUE); if (child_role == pcmk_role_promoted) { rsc = child; promoted_node = pcmk__current_node(child); promoted_count++; } } if (promoted_role_only || (promoted_count != 0)) { count = promoted_count; current = promoted_node; } } if (count > 1) { if (pcmk__is_clone(rsc)) { current = NULL; } else { return pcmk_rc_multiple; } } if (pcmk__same_node(current, dest)) { cur_is_dest = true; if (force) { crm_info("%s is already %s on %s, reinforcing placement with location constraint.", rsc_id, promoted_role_only?"promoted":"active", pcmk__node_name(dest)); } else { return pcmk_rc_already; } } /* Clear any previous prefer constraints across all nodes. */ cli_resource_clear(rsc_id, NULL, scheduler->nodes, cib, cib_options, false, force); /* Clear any previous ban constraints on 'dest'. */ cli_resource_clear(rsc_id, dest->details->uname, scheduler->nodes, cib, cib_options, TRUE, force); /* Record an explicit preference for 'dest' */ rc = cli_resource_prefer(out, rsc_id, dest->details->uname, move_lifetime, cib, cib_options, promoted_role_only, PCMK_ROLE_PROMOTED); crm_trace("%s%s now prefers %s%s", rsc->id, (promoted_role_only? " (promoted)" : ""), pcmk__node_name(dest), force?"(forced)":""); /* only ban the previous location if current location != destination location. * it is possible to use -M to enforce a location without regard of where the * resource is currently located */ if (force && !cur_is_dest) { /* Ban the original location if possible */ if(current) { (void)cli_resource_ban(out, rsc_id, current->details->uname, move_lifetime, cib, cib_options, promoted_role_only, PCMK_ROLE_PROMOTED); } else if(count > 1) { out->info(out, "Resource '%s' is currently %s in %d locations. " "One may now move to %s", rsc_id, (promoted_role_only? "promoted" : "active"), count, pcmk__node_name(dest)); out->info(out, "To prevent '%s' from being %s at a specific location, " "specify a node.", rsc_id, (promoted_role_only? "promoted" : "active")); } else { crm_trace("Not banning %s from its current location: not active", rsc_id); } } return rc; }