diff --git a/include/crm/common/scheduler.h b/include/crm/common/scheduler.h index 53d7a0c585..53e1218f3d 100644 --- a/include/crm/common/scheduler.h +++ b/include/crm/common/scheduler.h @@ -1,127 +1,126 @@ /* * 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_SCHEDULER__H #define PCMK__CRM_COMMON_SCHEDULER__H #include // time_t #include // xmlNode #include // guint, GList, GHashTable #include // crm_time_t #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /*! * \file * \brief Scheduler API * \ingroup core */ // NOTE: sbd (as of at least 1.5.2) uses this enum //! Possible responses to loss of quorum enum pe_quorum_policy { pcmk_no_quorum_freeze, //!< Do not recover resources from outside partition pcmk_no_quorum_stop, //!< Stop all resources in partition pcmk_no_quorum_ignore, //!< Act as if partition still holds quorum pcmk_no_quorum_fence, //!< Fence all nodes in partition pcmk_no_quorum_demote, //!< Demote promotable resources and stop all others #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) // NOTE: sbd (as of at least 1.5.2) uses this value //! \deprecated Use pcmk_no_quorum_freeze instead no_quorum_freeze = pcmk_no_quorum_freeze, // NOTE: sbd (as of at least 1.5.2) uses this value //! \deprecated Use pcmk_no_quorum_stop instead no_quorum_stop = pcmk_no_quorum_stop, // NOTE: sbd (as of at least 1.5.2) uses this value //! \deprecated Use pcmk_no_quorum_ignore instead no_quorum_ignore = pcmk_no_quorum_ignore, //! \deprecated Use pcmk_no_quorum_fence instead no_quorum_suicide = pcmk_no_quorum_fence, // NOTE: sbd (as of at least 1.5.2) uses this value //! \deprecated Use pcmk_no_quorum_demote instead no_quorum_demote = pcmk_no_quorum_demote, #endif }; //! \internal Do not use typedef struct pcmk__scheduler_private pcmk__scheduler_private_t; /* Implementation of pcmk_scheduler_t * * @COMPAT Drop this struct once all members are moved to * pcmk__scheduler_private_t, and repoint pcmk_scheduler_t to that */ //!@{ //! \deprecated Do not use (public access will be removed in a future release) struct pcmk__scheduler { // Be careful about when each piece of information is available and final // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Set scheduler input with pcmk_set_scheduler_cib() instead xmlNode *input; // CIB XML // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_get_dc() instead pcmk_node_t *dc_node; // Node object for DC // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_has_quorum() to check quorum uint64_t flags; // Group of enum pcmk__scheduler_flags // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_get_no_quorum_policy() to get no-quorum policy enum pe_quorum_policy no_quorum_policy; // Response to loss of quorum // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_find_node() to find a node instead GList *nodes; // Nodes in cluster - GList *param_check; // History entries that need to be checked GList *stop_needed; // Containers that need stop actions time_t recheck_by; // Hint to controller when to reschedule int ninstances; // Total number of resource instances guint shutdown_lock; // How long to lock resources (seconds) int priority_fencing_delay; // Priority fencing delay pcmk__scheduler_private_t *priv; // For Pacemaker use only guint node_pending_timeout; // Pending join times out after this (ms) }; //!@} pcmk_node_t *pcmk_get_dc(const pcmk_scheduler_t *scheduler); enum pe_quorum_policy pcmk_get_no_quorum_policy(const pcmk_scheduler_t *scheduler); int pcmk_set_scheduler_cib(pcmk_scheduler_t *scheduler, xmlNode *cib); bool pcmk_has_quorum(const pcmk_scheduler_t *scheduler); pcmk_node_t *pcmk_find_node(const pcmk_scheduler_t *scheduler, const char *node_name); #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_SCHEDULER__H diff --git a/include/crm/common/scheduler_internal.h b/include/crm/common/scheduler_internal.h index 4af7e2f55c..2eb170c580 100644 --- a/include/crm/common/scheduler_internal.h +++ b/include/crm/common/scheduler_internal.h @@ -1,282 +1,283 @@ /* * 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_SCHEDULER_INTERNAL__H #define PCMK__CRM_COMMON_SCHEDULER_INTERNAL__H #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif enum pcmk__check_parameters { /* Clear fail count if parameters changed for un-expired start or monitor * last_failure. */ pcmk__check_last_failure, /* Clear fail count if parameters changed for start, monitor, promote, or * migrate_from actions for active resources. */ pcmk__check_active, }; // Scheduling options and conditions enum pcmk__scheduler_flags { // No scheduler flags set (compare with equality rather than bit set) pcmk__sched_none = 0ULL, /* These flags are dynamically determined conditions */ // Whether partition has quorum (via \c PCMK_XA_HAVE_QUORUM attribute) //! \deprecated Call pcmk_has_quorum() to check quorum instead pcmk__sched_quorate = (1ULL << 0), // Whether cluster is symmetric (via symmetric-cluster property) pcmk__sched_symmetric_cluster = (1ULL << 1), // Whether scheduling encountered a non-configuration error pcmk__sched_processing_error = (1ULL << 2), // Whether cluster is in maintenance mode (via maintenance-mode property) pcmk__sched_in_maintenance = (1ULL << 3), // Whether fencing is enabled (via stonith-enabled property) pcmk__sched_fencing_enabled = (1ULL << 4), // Whether cluster has a fencing resource (via CIB resources) /*! \deprecated To indicate the cluster has a fencing resource, add either a * fencing resource configuration or the have-watchdog cluster option to the * input CIB */ pcmk__sched_have_fencing = (1ULL << 5), // Whether any resource provides or requires unfencing (via CIB resources) pcmk__sched_enable_unfencing = (1ULL << 6), // Whether concurrent fencing is allowed (via concurrent-fencing property) pcmk__sched_concurrent_fencing = (1ULL << 7), /* * Whether resources removed from the configuration should be stopped (via * stop-orphan-resources property) */ pcmk__sched_stop_removed_resources = (1ULL << 8), /* * Whether recurring actions removed from the configuration should be * cancelled (via stop-orphan-actions property) */ pcmk__sched_cancel_removed_actions = (1ULL << 9), // Whether to stop all resources (via stop-all-resources property) pcmk__sched_stop_all = (1ULL << 10), // Whether scheduler processing encountered a warning pcmk__sched_processing_warning = (1ULL << 11), /* * Whether start failure should be treated as if * \c PCMK_META_MIGRATION_THRESHOLD is 1 (via * \c PCMK_OPT_START_FAILURE_IS_FATAL property) */ pcmk__sched_start_failure_fatal = (1ULL << 12), // Unused pcmk__sched_remove_after_stop = (1ULL << 13), // Whether unseen nodes should be fenced (via startup-fencing property) pcmk__sched_startup_fencing = (1ULL << 14), /* * Whether resources should be left stopped when their node shuts down * cleanly (via shutdown-lock property) */ pcmk__sched_shutdown_lock = (1ULL << 15), /* * Whether resources' current state should be probed (when unknown) before * scheduling any other actions (via the enable-startup-probes property) */ pcmk__sched_probe_resources = (1ULL << 16), // Whether the CIB status section has been parsed yet pcmk__sched_have_status = (1ULL << 17), // Whether the cluster includes any Pacemaker Remote nodes (via CIB) pcmk__sched_have_remote_nodes = (1ULL << 18), /* The remaining flags are scheduling options that must be set explicitly */ /* * Whether to skip unpacking the CIB status section and stop the scheduling * sequence after applying node-specific location criteria (skipping * assignment, ordering, actions, etc.). */ pcmk__sched_location_only = (1ULL << 20), // Whether sensitive resource attributes have been masked pcmk__sched_sanitized = (1ULL << 21), // Skip counting of total, disabled, and blocked resource instances pcmk__sched_no_counts = (1ULL << 23), // Whether node scores should be output instead of logged pcmk__sched_output_scores = (1ULL << 25), // Whether to show node and resource utilization (in log or output) pcmk__sched_show_utilization = (1ULL << 26), /* * Whether to stop the scheduling sequence after unpacking the CIB, * calculating cluster status, and applying node health (skipping * applying node-specific location criteria, assignment, etc.) */ pcmk__sched_validate_only = (1ULL << 27), }; // Implementation of pcmk__scheduler_private_t struct pcmk__scheduler_private { // Be careful about when each piece of information is available and final char *local_node_name; // Name of node running scheduler (if known) crm_time_t *now; // Time to use when evaluating rules pcmk__output_t *out; // Output object for displaying messages GHashTable *options; // Cluster options const char *fence_action; // Default fencing action int fence_timeout_ms; // Value of stonith-timeout property in ms const char *placement_strategy; // Value of placement-strategy property xmlNode *rsc_defaults; // Configured resource defaults xmlNode *op_defaults; // Configured operation defaults GList *resources; // Resources in cluster GHashTable *templates; // Key = template ID, value = resource list GHashTable *tags; // Key = tag ID, value = element list GList *actions; // All scheduled actions GHashTable *singletons; // Scheduled non-resource actions int next_action_id; // Counter used as ID for actions xmlNode *failed; // History entries of failed actions + GList *param_check; // History entries that need to be checked GList *location_constraints; // Location constraints GList *colocation_constraints; // Colocation constraints GList *ordering_constraints; // Ordering constraints GHashTable *ticket_constraints; // Key = ticket ID, value = pcmk__ticket_t int next_ordering_id; // Counter used as ID for orderings int blocked_resources; // Number of blocked resources in cluster int disabled_resources; // Number of disabled resources in cluster xmlNode *graph; // Transition graph int synapse_count; // Number of transition graph synapses }; // Group of enum pcmk__warnings flags for warnings we want to log once extern uint32_t pcmk__warnings; /*! * \internal * \brief Log a resource-tagged message at info severity * * \param[in] rsc Tag message with this resource's ID * \param[in] fmt... printf(3)-style format and arguments */ #define pcmk__rsc_info(rsc, fmt, args...) \ crm_log_tag(LOG_INFO, ((rsc) == NULL)? "" : (rsc)->id, (fmt), ##args) /*! * \internal * \brief Log a resource-tagged message at debug severity * * \param[in] rsc Tag message with this resource's ID * \param[in] fmt... printf(3)-style format and arguments */ #define pcmk__rsc_debug(rsc, fmt, args...) \ crm_log_tag(LOG_DEBUG, ((rsc) == NULL)? "" : (rsc)->id, (fmt), ##args) /*! * \internal * \brief Log a resource-tagged message at trace severity * * \param[in] rsc Tag message with this resource's ID * \param[in] fmt... printf(3)-style format and arguments */ #define pcmk__rsc_trace(rsc, fmt, args...) \ crm_log_tag(LOG_TRACE, ((rsc) == NULL)? "" : (rsc)->id, (fmt), ##args) /*! * \internal * \brief Log an error and remember that current scheduler input has errors * * \param[in,out] scheduler Scheduler data * \param[in] fmt... printf(3)-style format and arguments */ #define pcmk__sched_err(scheduler, fmt...) do { \ pcmk__set_scheduler_flags((scheduler), \ pcmk__sched_processing_error); \ crm_err(fmt); \ } while (0) /*! * \internal * \brief Log a warning and remember that current scheduler input has warnings * * \param[in,out] scheduler Scheduler data * \param[in] fmt... printf(3)-style format and arguments */ #define pcmk__sched_warn(scheduler, fmt...) do { \ pcmk__set_scheduler_flags((scheduler), \ pcmk__sched_processing_warning); \ crm_warn(fmt); \ } while (0) /*! * \internal * \brief Set scheduler flags * * \param[in,out] scheduler Scheduler data * \param[in] flags_to_set Group of enum pcmk__scheduler_flags to set */ #define pcmk__set_scheduler_flags(scheduler, flags_to_set) do { \ (scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", crm_system_name, \ (scheduler)->flags, (flags_to_set), #flags_to_set); \ } while (0) /*! * \internal * \brief Clear scheduler flags * * \param[in,out] scheduler Scheduler data * \param[in] flags_to_clear Group of enum pcmk__scheduler_flags to clear */ #define pcmk__clear_scheduler_flags(scheduler, flags_to_clear) do { \ (scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", crm_system_name, \ (scheduler)->flags, (flags_to_clear), #flags_to_clear); \ } while (0) #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_SCHEDULER_INTERNAL__H diff --git a/lib/pengine/remote.c b/lib/pengine/remote.c index b3ffcef4a1..395fa473e6 100644 --- a/lib/pengine/remote.c +++ b/lib/pengine/remote.c @@ -1,241 +1,242 @@ /* * Copyright 2013-2024 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 /*! * \internal * \brief Check whether a resource creates a guest node * * If a given resource contains a launched resource that is a remote connection, * return that launched resource (or NULL if none is found). * * \param[in] scheduler Scheduler data * \param[in] rsc Resource to check * * \return Launched remote connection, or NULL if none found */ pcmk_resource_t * pe__resource_contains_guest_node(const pcmk_scheduler_t *scheduler, const pcmk_resource_t *rsc) { if ((rsc != NULL) && (scheduler != NULL) && pcmk_is_set(scheduler->flags, pcmk__sched_have_remote_nodes)) { for (GList *gIter = rsc->priv->launched; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *launched = gIter->data; if (pcmk_is_set(launched->flags, pcmk__rsc_is_remote_connection)) { return launched; } } } return NULL; } bool xml_contains_remote_node(xmlNode *xml) { const char *value = NULL; if (xml == NULL) { return false; } value = crm_element_value(xml, PCMK_XA_TYPE); if (!pcmk__str_eq(value, "remote", pcmk__str_casei)) { return false; } value = crm_element_value(xml, PCMK_XA_CLASS); if (!pcmk__str_eq(value, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) { return false; } value = crm_element_value(xml, PCMK_XA_PROVIDER); if (!pcmk__str_eq(value, "pacemaker", pcmk__str_casei)) { return false; } return true; } /*! * \internal * \brief Execute a supplied function for each guest node running on a host * * \param[in] scheduler Scheduler data * \param[in] host Host node to check * \param[in] helper Function to call for each guest node * \param[in,out] user_data Pointer to pass to helper function */ void pe_foreach_guest_node(const pcmk_scheduler_t *scheduler, const pcmk_node_t *host, void (*helper)(const pcmk_node_t*, void*), void *user_data) { GList *iter; CRM_CHECK(scheduler && host && host->details && helper, return); if (!pcmk_is_set(scheduler->flags, pcmk__sched_have_remote_nodes)) { return; } for (iter = host->details->running_rsc; 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) && (rsc->priv->launcher != NULL)) { pcmk_node_t *guest_node = pcmk_find_node(scheduler, rsc->id); if (guest_node) { (*helper)(guest_node, user_data); } } } } /*! * \internal * \brief Create CIB XML for an implicit remote connection * * \param[in,out] parent If not \c NULL, use as parent XML element * \param[in] uname Name of Pacemaker Remote node * \param[in] container_id If not \c NULL, use this as connection container * \param[in] migrateable If not \c NULL, use as remote * \c PCMK_META_ALLOW_MIGRATE value * \param[in] is_managed If not \c NULL, use as remote * \c PCMK_META_IS_MANAGED value * \param[in] start_timeout If not \c NULL, use as remote connect timeout * \param[in] server If not \c NULL, use as \c PCMK_REMOTE_RA_ADDR * \param[in] port If not \c NULL, use as \c PCMK_REMOTE_RA_PORT * * \return Newly created XML */ xmlNode * pe_create_remote_xml(xmlNode *parent, const char *uname, const char *container_id, const char *migrateable, const char *is_managed, const char *start_timeout, const char *server, const char *port) { xmlNode *remote; xmlNode *xml_sub; remote = pcmk__xe_create(parent, PCMK_XE_PRIMITIVE); // Add identity crm_xml_add(remote, PCMK_XA_ID, uname); crm_xml_add(remote, PCMK_XA_CLASS, PCMK_RESOURCE_CLASS_OCF); crm_xml_add(remote, PCMK_XA_PROVIDER, "pacemaker"); crm_xml_add(remote, PCMK_XA_TYPE, "remote"); // Add meta-attributes xml_sub = pcmk__xe_create(remote, PCMK_XE_META_ATTRIBUTES); pcmk__xe_set_id(xml_sub, "%s-%s", uname, PCMK_XE_META_ATTRIBUTES); crm_create_nvpair_xml(xml_sub, NULL, PCMK__META_INTERNAL_RSC, PCMK_VALUE_TRUE); if (container_id) { crm_create_nvpair_xml(xml_sub, NULL, PCMK__META_CONTAINER, container_id); } if (migrateable) { crm_create_nvpair_xml(xml_sub, NULL, PCMK_META_ALLOW_MIGRATE, migrateable); } if (is_managed) { crm_create_nvpair_xml(xml_sub, NULL, PCMK_META_IS_MANAGED, is_managed); } // Add instance attributes if (port || server) { xml_sub = pcmk__xe_create(remote, PCMK_XE_INSTANCE_ATTRIBUTES); pcmk__xe_set_id(xml_sub, "%s-%s", uname, PCMK_XE_INSTANCE_ATTRIBUTES); if (server) { crm_create_nvpair_xml(xml_sub, NULL, PCMK_REMOTE_RA_ADDR, server); } if (port) { crm_create_nvpair_xml(xml_sub, NULL, PCMK_REMOTE_RA_PORT, port); } } // Add operations xml_sub = pcmk__xe_create(remote, PCMK_XE_OPERATIONS); crm_create_op_xml(xml_sub, uname, PCMK_ACTION_MONITOR, "30s", "30s"); if (start_timeout) { crm_create_op_xml(xml_sub, uname, PCMK_ACTION_START, "0", start_timeout); } return remote; } // History entry to be checked for fail count clearing struct check_op { const xmlNode *rsc_op; // History entry XML pcmk_resource_t *rsc; // Known resource corresponding to history entry pcmk_node_t *node; // Known node corresponding to history entry enum pcmk__check_parameters check_type; // What needs checking }; void pe__add_param_check(const xmlNode *rsc_op, pcmk_resource_t *rsc, pcmk_node_t *node, enum pcmk__check_parameters flag, pcmk_scheduler_t *scheduler) { struct check_op *check_op = NULL; CRM_CHECK(scheduler && rsc_op && rsc && node, return); check_op = pcmk__assert_alloc(1, sizeof(struct check_op)); crm_trace("Deferring checks of %s until after allocation", pcmk__xe_id(rsc_op)); check_op->rsc_op = rsc_op; check_op->rsc = rsc; check_op->node = node; check_op->check_type = flag; - scheduler->param_check = g_list_prepend(scheduler->param_check, check_op); + scheduler->priv->param_check = g_list_prepend(scheduler->priv->param_check, + check_op); } /*! * \internal * \brief Call a function for each action to be checked for addr substitution * * \param[in,out] scheduler Scheduler data * \param[in] cb Function to be called */ void pe__foreach_param_check(pcmk_scheduler_t *scheduler, void (*cb)(pcmk_resource_t*, pcmk_node_t*, const xmlNode*, enum pcmk__check_parameters)) { CRM_CHECK(scheduler && cb, return); - for (GList *item = scheduler->param_check; + for (GList *item = scheduler->priv->param_check; item != NULL; item = item->next) { struct check_op *check_op = item->data; cb(check_op->rsc, check_op->node, check_op->rsc_op, check_op->check_type); } } void pe__free_param_checks(pcmk_scheduler_t *scheduler) { - if (scheduler && scheduler->param_check) { - g_list_free_full(scheduler->param_check, free); - scheduler->param_check = NULL; + if ((scheduler != NULL) && (scheduler->priv->param_check != NULL)) { + g_list_free_full(scheduler->priv->param_check, free); + scheduler->priv->param_check = NULL; } }