diff --git a/include/crm/pengine/internal.h b/include/crm/pengine/internal.h index 87ae25f22b..c0c142fff7 100644 --- a/include/crm/pengine/internal.h +++ b/include/crm/pengine/internal.h @@ -1,723 +1,742 @@ /* * Copyright 2004-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PE_INTERNAL__H # define PE_INTERNAL__H +# include # include # include # include # include # include # include # include # include const char *pe__resource_description(const pe_resource_t *rsc, uint32_t show_opts); enum pe__clone_flags { // Whether instances should be started sequentially pe__clone_ordered = (1 << 0), // Whether promotion scores have been added pe__clone_promotion_added = (1 << 1), // Whether promotion constraints have been added pe__clone_promotion_constrained = (1 << 2), }; bool pe__clone_is_ordered(const pe_resource_t *clone); int pe__set_clone_flag(pe_resource_t *clone, enum pe__clone_flags flag); enum pe__group_flags { pe__group_ordered = (1 << 0), // Members start sequentially pe__group_colocated = (1 << 1), // Members must be on same node }; bool pe__group_flag_is_set(const pe_resource_t *group, uint32_t flags); pe_resource_t *pe__last_group_member(const pe_resource_t *group); # define pe_rsc_info(rsc, fmt, args...) crm_log_tag(LOG_INFO, rsc ? rsc->id : "", fmt, ##args) # define pe_rsc_debug(rsc, fmt, args...) crm_log_tag(LOG_DEBUG, rsc ? rsc->id : "", fmt, ##args) # define pe_rsc_trace(rsc, fmt, args...) crm_log_tag(LOG_TRACE, rsc ? rsc->id : "", fmt, ##args) # define pe_err(fmt...) do { \ was_processing_error = TRUE; \ pcmk__config_err(fmt); \ } while (0) # define pe_warn(fmt...) do { \ was_processing_warning = TRUE; \ pcmk__config_warn(fmt); \ } while (0) # define pe_proc_err(fmt...) { was_processing_error = TRUE; crm_err(fmt); } # define pe_proc_warn(fmt...) { was_processing_warning = TRUE; crm_warn(fmt); } #define pe__set_working_set_flags(working_set, flags_to_set) do { \ (working_set)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Working set", crm_system_name, \ (working_set)->flags, (flags_to_set), #flags_to_set); \ } while (0) #define pe__clear_working_set_flags(working_set, flags_to_clear) do { \ (working_set)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Working set", crm_system_name, \ (working_set)->flags, (flags_to_clear), #flags_to_clear); \ } while (0) #define pe__set_resource_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) #define pe__clear_resource_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) #define pe__set_action_flags(action, flags_to_set) do { \ (action)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, \ "Action", (action)->uuid, \ (action)->flags, \ (flags_to_set), \ #flags_to_set); \ } while (0) #define pe__clear_action_flags(action, flags_to_clear) do { \ (action)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, \ "Action", (action)->uuid, \ (action)->flags, \ (flags_to_clear), \ #flags_to_clear); \ } while (0) #define pe__set_raw_action_flags(action_flags, action_name, flags_to_set) do { \ action_flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Action", action_name, \ (action_flags), \ (flags_to_set), #flags_to_set); \ } while (0) #define pe__clear_raw_action_flags(action_flags, action_name, flags_to_clear) do { \ action_flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, \ "Action", action_name, \ (action_flags), \ (flags_to_clear), \ #flags_to_clear); \ } while (0) #define pe__set_action_flags_as(function, line, action, flags_to_set) do { \ (action)->flags = pcmk__set_flags_as((function), (line), \ LOG_TRACE, \ "Action", (action)->uuid, \ (action)->flags, \ (flags_to_set), \ #flags_to_set); \ } while (0) #define pe__clear_action_flags_as(function, line, action, flags_to_clear) do { \ (action)->flags = pcmk__clear_flags_as((function), (line), \ LOG_TRACE, \ "Action", (action)->uuid, \ (action)->flags, \ (flags_to_clear), \ #flags_to_clear); \ } while (0) #define pe__set_order_flags(order_flags, flags_to_set) do { \ order_flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Ordering", "constraint", \ order_flags, (flags_to_set), \ #flags_to_set); \ } while (0) #define pe__clear_order_flags(order_flags, flags_to_clear) do { \ order_flags = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Ordering", "constraint", \ order_flags, (flags_to_clear), \ #flags_to_clear); \ } while (0) // Some warnings we don't want to print every transition enum pe_warn_once_e { pe_wo_blind = (1 << 0), pe_wo_restart_type = (1 << 1), pe_wo_role_after = (1 << 2), pe_wo_poweroff = (1 << 3), pe_wo_require_all = (1 << 4), pe_wo_order_score = (1 << 5), pe_wo_neg_threshold = (1 << 6), pe_wo_remove_after = (1 << 7), pe_wo_ping_node = (1 << 8), pe_wo_order_inst = (1 << 9), pe_wo_coloc_inst = (1 << 10), pe_wo_group_order = (1 << 11), pe_wo_group_coloc = (1 << 12), pe_wo_upstart = (1 << 13), pe_wo_nagios = (1 << 14), }; extern uint32_t pe_wo; #define pe_warn_once(pe_wo_bit, fmt...) do { \ if (!pcmk_is_set(pe_wo, pe_wo_bit)) { \ if (pe_wo_bit == pe_wo_blind) { \ crm_warn(fmt); \ } else { \ pe_warn(fmt); \ } \ pe_wo = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Warn-once", "logging", pe_wo, \ (pe_wo_bit), #pe_wo_bit); \ } \ } while (0); typedef struct pe__location_constraint_s { char *id; // Constraint XML ID pe_resource_t *rsc_lh; // Resource being located enum rsc_role_e role_filter; // Role to locate enum pe_discover_e discover_mode; // Resource discovery GList *node_list_rh; // List of pe_node_t* } pe__location_t; typedef struct pe__order_constraint_s { int id; uint32_t flags; // Group of enum pe_ordering flags void *lh_opaque; pe_resource_t *lh_rsc; pe_action_t *lh_action; char *lh_action_task; void *rh_opaque; pe_resource_t *rh_rsc; pe_action_t *rh_action; char *rh_action_task; } pe__ordering_t; const pe_resource_t *pe__const_top_resource(const pe_resource_t *rsc, bool include_bundle); int pe__clone_max(const pe_resource_t *clone); int pe__clone_node_max(const pe_resource_t *clone); int pe__clone_promoted_max(const pe_resource_t *clone); int pe__clone_promoted_node_max(const pe_resource_t *clone); void pe__create_clone_notifications(pe_resource_t *clone); void pe__free_clone_notification_data(pe_resource_t *clone); void pe__create_clone_notif_pseudo_ops(pe_resource_t *clone, pe_action_t *start, pe_action_t *started, pe_action_t *stop, pe_action_t *stopped); pe_action_t *pe__new_rsc_pseudo_action(pe_resource_t *rsc, const char *task, bool optional, bool runnable); void pe__create_promotable_pseudo_ops(pe_resource_t *clone, bool any_promoting, bool any_demoting); bool pe_can_fence(const pe_working_set_t *data_set, const pe_node_t *node); void add_hash_param(GHashTable * hash, const char *name, const char *value); char *native_parameter(pe_resource_t * rsc, pe_node_t * node, gboolean create, const char *name, pe_working_set_t * data_set); pe_node_t *native_location(const pe_resource_t *rsc, GList **list, int current); void pe_metadata(pcmk__output_t *out); void verify_pe_options(GHashTable * options); void native_add_running(pe_resource_t * rsc, pe_node_t * node, pe_working_set_t * data_set, gboolean failed); gboolean native_unpack(pe_resource_t * rsc, pe_working_set_t * data_set); gboolean group_unpack(pe_resource_t * rsc, pe_working_set_t * data_set); gboolean clone_unpack(pe_resource_t * rsc, pe_working_set_t * data_set); gboolean pe__unpack_bundle(pe_resource_t *rsc, pe_working_set_t *data_set); pe_resource_t *native_find_rsc(pe_resource_t *rsc, const char *id, const pe_node_t *node, int flags); gboolean native_active(pe_resource_t * rsc, gboolean all); gboolean group_active(pe_resource_t * rsc, gboolean all); gboolean clone_active(pe_resource_t * rsc, gboolean all); gboolean pe__bundle_active(pe_resource_t *rsc, gboolean all); //! \deprecated This function will be removed in a future release void native_print(pe_resource_t *rsc, const char *pre_text, long options, void *print_data); //! \deprecated This function will be removed in a future release void group_print(pe_resource_t *rsc, const char *pre_text, long options, void *print_data); //! \deprecated This function will be removed in a future release void clone_print(pe_resource_t *rsc, const char *pre_text, long options, void *print_data); //! \deprecated This function will be removed in a future release void pe__print_bundle(pe_resource_t *rsc, const char *pre_text, long options, void *print_data); gchar *pcmk__native_output_string(const pe_resource_t *rsc, const char *name, const pe_node_t *node, uint32_t show_opts, const char *target_role, bool show_nodes); int pe__name_and_nvpairs_xml(pcmk__output_t *out, bool is_list, const char *tag_name , size_t pairs_count, ...); char *pe__node_display_name(pe_node_t *node, bool print_detail); // Clone notifications (pe_notif.c) void pe__order_notifs_after_fencing(const pe_action_t *action, pe_resource_t *rsc, pe_action_t *stonith_op); static inline const char * pe__rsc_bool_str(const pe_resource_t *rsc, uint64_t rsc_flag) { return pcmk__btoa(pcmk_is_set(rsc->flags, rsc_flag)); } int pe__clone_xml(pcmk__output_t *out, va_list args); int pe__clone_default(pcmk__output_t *out, va_list args); int pe__group_xml(pcmk__output_t *out, va_list args); int pe__group_default(pcmk__output_t *out, va_list args); int pe__bundle_xml(pcmk__output_t *out, va_list args); int pe__bundle_html(pcmk__output_t *out, va_list args); int pe__bundle_text(pcmk__output_t *out, va_list args); int pe__node_html(pcmk__output_t *out, va_list args); int pe__node_text(pcmk__output_t *out, va_list args); int pe__node_xml(pcmk__output_t *out, va_list args); int pe__resource_xml(pcmk__output_t *out, va_list args); int pe__resource_html(pcmk__output_t *out, va_list args); int pe__resource_text(pcmk__output_t *out, va_list args); void native_free(pe_resource_t * rsc); void group_free(pe_resource_t * rsc); void clone_free(pe_resource_t * rsc); void pe__free_bundle(pe_resource_t *rsc); enum rsc_role_e native_resource_state(const pe_resource_t * rsc, gboolean current); enum rsc_role_e group_resource_state(const pe_resource_t * rsc, gboolean current); enum rsc_role_e clone_resource_state(const pe_resource_t * rsc, gboolean current); enum rsc_role_e pe__bundle_resource_state(const pe_resource_t *rsc, gboolean current); void pe__count_common(pe_resource_t *rsc); void pe__count_bundle(pe_resource_t *rsc); void common_free(pe_resource_t * rsc); pe_node_t *pe__copy_node(const pe_node_t *this_node); extern time_t get_effective_time(pe_working_set_t * data_set); /* Failure handling utilities (from failcounts.c) */ // bit flags for fail count handling options enum pe_fc_flags_e { pe_fc_default = (1 << 0), pe_fc_effective = (1 << 1), // don't count expired failures pe_fc_fillers = (1 << 2), // if container, include filler failures in count }; int pe_get_failcount(const pe_node_t *node, pe_resource_t *rsc, time_t *last_failure, uint32_t flags, const xmlNode *xml_op); pe_action_t *pe__clear_failcount(pe_resource_t *rsc, const pe_node_t *node, const char *reason, pe_working_set_t *data_set); /* Functions for finding/counting a resource's active nodes */ bool pe__count_active_node(const pe_resource_t *rsc, pe_node_t *node, pe_node_t **active, unsigned int *count_all, unsigned int *count_clean); pe_node_t *pe__find_active_requires(const pe_resource_t *rsc, unsigned int *count); static inline pe_node_t * pe__current_node(const pe_resource_t *rsc) { return (rsc == NULL)? NULL : rsc->fns->active_node(rsc, NULL, NULL); } /* Binary like operators for lists of nodes */ extern void node_list_exclude(GHashTable * list, GList *list2, gboolean merge_scores); GHashTable *pe__node_list2table(const GList *list); static inline gpointer pe_hash_table_lookup(GHashTable * hash, gconstpointer key) { if (hash) { return g_hash_table_lookup(hash, key); } return NULL; } extern pe_action_t *get_pseudo_op(const char *name, pe_working_set_t * data_set); extern gboolean order_actions(pe_action_t * lh_action, pe_action_t * rh_action, enum pe_ordering order); void pe__show_node_weights_as(const char *file, const char *function, int line, bool to_log, const pe_resource_t *rsc, const char *comment, GHashTable *nodes, pe_working_set_t *data_set); #define pe__show_node_weights(level, rsc, text, nodes, data_set) \ pe__show_node_weights_as(__FILE__, __func__, __LINE__, \ (level), (rsc), (text), (nodes), (data_set)) xmlNode *find_rsc_op_entry(const pe_resource_t *rsc, const char *key); pe_action_t *custom_action(pe_resource_t *rsc, char *key, const char *task, const pe_node_t *on_node, gboolean optional, gboolean foo, pe_working_set_t *data_set); # define delete_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_DELETE, 0) # define delete_action(rsc, node, optional) custom_action( \ rsc, delete_key(rsc), CRMD_ACTION_DELETE, node, \ optional, TRUE, rsc->cluster); # define stopped_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_STOPPED, 0) # define stopped_action(rsc, node, optional) custom_action( \ rsc, stopped_key(rsc), CRMD_ACTION_STOPPED, node, \ optional, TRUE, rsc->cluster); # define stop_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_STOP, 0) # define stop_action(rsc, node, optional) custom_action( \ rsc, stop_key(rsc), CRMD_ACTION_STOP, node, \ optional, TRUE, rsc->cluster); # define reload_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_RELOAD_AGENT, 0) # define start_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_START, 0) # define start_action(rsc, node, optional) custom_action( \ rsc, start_key(rsc), CRMD_ACTION_START, node, \ optional, TRUE, rsc->cluster) # define started_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_STARTED, 0) # define started_action(rsc, node, optional) custom_action( \ rsc, started_key(rsc), CRMD_ACTION_STARTED, node, \ optional, TRUE, rsc->cluster) # define promote_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_PROMOTE, 0) # define promote_action(rsc, node, optional) custom_action( \ rsc, promote_key(rsc), CRMD_ACTION_PROMOTE, node, \ optional, TRUE, rsc->cluster) # define promoted_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_PROMOTED, 0) # define promoted_action(rsc, node, optional) custom_action( \ rsc, promoted_key(rsc), CRMD_ACTION_PROMOTED, node, \ optional, TRUE, rsc->cluster) # define demote_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_DEMOTE, 0) # define demote_action(rsc, node, optional) custom_action( \ rsc, demote_key(rsc), CRMD_ACTION_DEMOTE, node, \ optional, TRUE, rsc->cluster) # define demoted_key(rsc) pcmk__op_key(rsc->id, CRMD_ACTION_DEMOTED, 0) # define demoted_action(rsc, node, optional) custom_action( \ rsc, demoted_key(rsc), CRMD_ACTION_DEMOTED, node, \ optional, TRUE, rsc->cluster) extern int pe_get_configured_timeout(pe_resource_t *rsc, const char *action, pe_working_set_t *data_set); pe_action_t *find_first_action(const GList *input, const char *uuid, const char *task, const pe_node_t *on_node); enum action_tasks get_complex_task(const pe_resource_t *rsc, const char *name); extern GList *find_actions(GList *input, const char *key, const pe_node_t *on_node); GList *find_actions_exact(GList *input, const char *key, const pe_node_t *on_node); GList *pe__resource_actions(const pe_resource_t *rsc, const pe_node_t *node, const char *task, bool require_node); extern void pe_free_action(pe_action_t * action); void resource_location(pe_resource_t *rsc, const pe_node_t *node, int score, const char *tag, pe_working_set_t *data_set); extern int pe__is_newer_op(const xmlNode *xml_a, const xmlNode *xml_b, bool same_node_default); extern gint sort_op_by_callid(gconstpointer a, gconstpointer b); gboolean get_target_role(const pe_resource_t *rsc, enum rsc_role_e *role); void pe__set_next_role(pe_resource_t *rsc, enum rsc_role_e role, const char *why); pe_resource_t *find_clone_instance(const pe_resource_t *rsc, const char *sub_id); extern void destroy_ticket(gpointer data); extern pe_ticket_t *ticket_new(const char *ticket_id, pe_working_set_t * data_set); // Resources for manipulating resource names const char *pe_base_name_end(const char *id); char *clone_strip(const char *last_rsc_id); char *clone_zero(const char *last_rsc_id); static inline bool pe_base_name_eq(const pe_resource_t *rsc, const char *id) { if (id && rsc && rsc->id) { // Number of characters in rsc->id before any clone suffix size_t base_len = pe_base_name_end(rsc->id) - rsc->id + 1; return (strlen(id) == base_len) && !strncmp(id, rsc->id, base_len); } return false; } int pe__target_rc_from_xml(const xmlNode *xml_op); gint pe__cmp_node_name(gconstpointer a, gconstpointer b); bool is_set_recursive(const pe_resource_t *rsc, long long flag, bool any); enum rsc_digest_cmp_val { /*! Digests are the same */ RSC_DIGEST_MATCH = 0, /*! Params that require a restart changed */ RSC_DIGEST_RESTART, /*! Some parameter changed. */ RSC_DIGEST_ALL, /*! rsc op didn't have a digest associated with it, so * it is unknown if parameters changed or not. */ RSC_DIGEST_UNKNOWN, }; typedef struct op_digest_cache_s { enum rsc_digest_cmp_val rc; xmlNode *params_all; xmlNode *params_secure; xmlNode *params_restart; char *digest_all_calc; char *digest_secure_calc; char *digest_restart_calc; } op_digest_cache_t; op_digest_cache_t *pe__calculate_digests(pe_resource_t *rsc, const char *task, guint *interval_ms, const pe_node_t *node, const xmlNode *xml_op, GHashTable *overrides, bool calc_secure, pe_working_set_t *data_set); void pe__free_digests(gpointer ptr); op_digest_cache_t *rsc_action_digest_cmp(pe_resource_t *rsc, const xmlNode *xml_op, pe_node_t *node, pe_working_set_t *data_set); pe_action_t *pe_fence_op(pe_node_t *node, const char *op, bool optional, const char *reason, bool priority_delay, pe_working_set_t *data_set); void trigger_unfencing(pe_resource_t *rsc, pe_node_t *node, const char *reason, pe_action_t *dependency, pe_working_set_t *data_set); char *pe__action2reason(const pe_action_t *action, enum pe_action_flags flag); void pe_action_set_reason(pe_action_t *action, const char *reason, bool overwrite); void pe__add_action_expected_result(pe_action_t *action, int expected_result); void pe__set_resource_flags_recursive(pe_resource_t *rsc, uint64_t flags); void pe__clear_resource_flags_recursive(pe_resource_t *rsc, uint64_t flags); void pe__clear_resource_flags_on_all(pe_working_set_t *data_set, uint64_t flag); gboolean add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref); //! \deprecated This function will be removed in a future release void print_rscs_brief(GList *rsc_list, const char * pre_text, long options, void * print_data, gboolean print_all); int pe__rscs_brief_output(pcmk__output_t *out, GList *rsc_list, unsigned int options); void pe_fence_node(pe_working_set_t * data_set, pe_node_t * node, const char *reason, bool priority_delay); pe_node_t *pe_create_node(const char *id, const char *uname, const char *type, const char *score, pe_working_set_t * data_set); //! \deprecated This function will be removed in a future release void common_print(pe_resource_t *rsc, const char *pre_text, const char *name, const pe_node_t *node, long options, void *print_data); int pe__common_output_text(pcmk__output_t *out, const pe_resource_t *rsc, const char *name, const pe_node_t *node, unsigned int options); int pe__common_output_html(pcmk__output_t *out, const pe_resource_t *rsc, const char *name, const pe_node_t *node, unsigned int options); +//! A single instance of a bundle +typedef struct { + int offset; //!< 0-origin index of this instance in bundle + char *ipaddr; //!< IP address associated with this instance + pe_node_t *node; //!< Node created for this instance + pe_resource_t *ip; //!< IP address resource for ipaddr + pe_resource_t *child; //!< Instance of bundled resource + pe_resource_t *container; //!< Container associated with this instance + pe_resource_t *remote; //!< Pacemaker Remote connection into container +} pe__bundle_replica_t; + GList *pe__bundle_containers(const pe_resource_t *bundle); int pe__bundle_max(const pe_resource_t *rsc); - +bool pe__node_is_bundle_instance(const pe_resource_t *bundle, + const pe_node_t *node); +pe_resource_t *pe__bundled_resource(const pe_resource_t *rsc); +const pe_resource_t *pe__get_rsc_in_container(const pe_resource_t *instance); +pe_resource_t *pe__first_container(const pe_resource_t *bundle); +void pe__foreach_bundle_replica(pe_resource_t *bundle, + bool (*fn)(pe__bundle_replica_t *, void *), + void *user_data); pe_resource_t *pe__find_bundle_replica(const pe_resource_t *bundle, const pe_node_t *node); bool pe__bundle_needs_remote_name(pe_resource_t *rsc); const char *pe__add_bundle_remote_name(pe_resource_t *rsc, pe_working_set_t *data_set, xmlNode *xml, const char *field); const char *pe_node_attribute_calculated(const pe_node_t *node, const char *name, const pe_resource_t *rsc); const char *pe_node_attribute_raw(const pe_node_t *node, const char *name); bool pe__is_universal_clone(const pe_resource_t *rsc, const pe_working_set_t *data_set); void pe__add_param_check(const xmlNode *rsc_op, pe_resource_t *rsc, pe_node_t *node, enum pe_check_parameters, pe_working_set_t *data_set); void pe__foreach_param_check(pe_working_set_t *data_set, void (*cb)(pe_resource_t*, pe_node_t*, const xmlNode*, enum pe_check_parameters)); void pe__free_param_checks(pe_working_set_t *data_set); bool pe__shutdown_requested(const pe_node_t *node); void pe__update_recheck_time(time_t recheck, pe_working_set_t *data_set); /*! * \internal * \brief Register xml formatting message functions. * * \param[in,out] out Output object to register messages with */ void pe__register_messages(pcmk__output_t *out); 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, pe_working_set_t *data_set); bool pe__resource_is_disabled(const pe_resource_t *rsc); pe_action_t *pe__clear_resource_history(pe_resource_t *rsc, const pe_node_t *node, pe_working_set_t *data_set); GList *pe__rscs_with_tag(pe_working_set_t *data_set, const char *tag_name); GList *pe__unames_with_tag(pe_working_set_t *data_set, const char *tag_name); bool pe__rsc_has_tag(pe_working_set_t *data_set, const char *rsc, const char *tag); bool pe__uname_has_tag(pe_working_set_t *data_set, const char *node, const char *tag); bool pe__rsc_running_on_only(const pe_resource_t *rsc, const pe_node_t *node); bool pe__rsc_running_on_any(pe_resource_t *rsc, GList *node_list); GList *pe__filter_rsc_list(GList *rscs, GList *filter); GList * pe__build_node_name_list(pe_working_set_t *data_set, const char *s); GList * pe__build_rsc_list(pe_working_set_t *data_set, const char *s); bool pcmk__rsc_filtered_by_node(pe_resource_t *rsc, GList *only_node); gboolean pe__bundle_is_filtered(const pe_resource_t *rsc, GList *only_rsc, gboolean check_parent); gboolean pe__clone_is_filtered(const pe_resource_t *rsc, GList *only_rsc, gboolean check_parent); gboolean pe__group_is_filtered(const pe_resource_t *rsc, GList *only_rsc, gboolean check_parent); gboolean pe__native_is_filtered(const pe_resource_t *rsc, GList *only_rsc, gboolean check_parent); xmlNode *pe__failed_probe_for_rsc(const pe_resource_t *rsc, const char *name); const char *pe__clone_child_id(const pe_resource_t *rsc); int pe__sum_node_health_scores(const pe_node_t *node, int base_health); int pe__node_health(pe_node_t *node); static inline enum pcmk__health_strategy pe__health_strategy(pe_working_set_t *data_set) { return pcmk__parse_health_strategy(pe_pref(data_set->config_hash, PCMK__OPT_NODE_HEALTH_STRATEGY)); } static inline int pe__health_score(const char *option, pe_working_set_t *data_set) { return char2score(pe_pref(data_set->config_hash, option)); } /*! * \internal * \brief Return a string suitable for logging as a node name * * \param[in] node Node to return a node name string for * * \return Node name if available, otherwise node ID if available, * otherwise "unspecified node" if node is NULL or "unidentified node" * if node has neither a name nor ID. */ static inline const char * pe__node_name(const pe_node_t *node) { if (node == NULL) { return "unspecified node"; } else if (node->details->uname != NULL) { return node->details->uname; } else if (node->details->id != NULL) { return node->details->id; } else { return "unidentified node"; } } /*! * \internal * \brief Check whether two node objects refer to the same node * * \param[in] node1 First node object to compare * \param[in] node2 Second node object to compare * * \return true if \p node1 and \p node2 refer to the same node */ static inline bool pe__same_node(const pe_node_t *node1, const pe_node_t *node2) { return (node1 != NULL) && (node2 != NULL) && (node1->details == node2->details); } /*! * \internal * \brief Get the operation key from an action history entry * * \param[in] xml Action history entry * * \return Entry's operation key */ static inline const char * pe__xe_history_key(const xmlNode *xml) { if (xml == NULL) { return NULL; } else { /* @COMPAT Pacemaker <= 1.1.5 did not add the key, and used the ID * instead. Checking for that allows us to process old saved CIBs, * including some regression tests. */ const char *key = crm_element_value(xml, XML_LRM_ATTR_TASK_KEY); return pcmk__str_empty(key)? ID(xml) : key; } } #endif diff --git a/lib/pacemaker/libpacemaker_private.h b/lib/pacemaker/libpacemaker_private.h index 8af35958fb..88823b1045 100644 --- a/lib/pacemaker/libpacemaker_private.h +++ b/lib/pacemaker/libpacemaker_private.h @@ -1,1016 +1,1012 @@ /* * Copyright 2021-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #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 // pe_action_t, pe_node_t, pe_working_set_t // 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 allocation methods struct resource_alloc_functions_s { /*! * \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 * * \return Node that \p rsc is assigned to, if assigned entirely to one node */ pe_node_t *(*assign)(pe_resource_t *rsc, const pe_node_t *prefer); /*! * \internal * \brief Create all actions needed for a given resource * * \param[in,out] rsc Resource to create actions for */ void (*create_actions)(pe_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)(pe_resource_t *rsc, pe_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)(pe_resource_t *rsc); /*! * \internal * \brief Apply a colocation's score to node weights or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node weights (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,out] 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) (pe_resource_t *dependent, - const pe_resource_t *primary, - const pcmk__colocation_t *colocation, - bool for_dependent); + void (*apply_coloc_score)(pe_resource_t *dependent, pe_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 indirectly 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 pe_resource_t *rsc, const pe_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 pe_resource_t *rsc, const pe_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 pe_resource_t *rsc, const pe_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] rsc Resource to check colocations for * \param[in] log_id Resource ID to use in logs (if NULL, use \p rsc ID) * \param[in,out] nodes Nodes to update * \param[in] attr Colocation attribute (NULL to use default) * \param[in] factor Incorporate scores multiplied by this factor * \param[in] flags Bitmask of enum pcmk__coloc_select values * * \note The caller remains responsible for freeing \p *nodes. */ void (*add_colocated_node_scores)(pe_resource_t *rsc, const char *log_id, GHashTable **nodes, const char *attr, 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)(pe_resource_t *rsc, pe__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. */ enum pe_action_flags (*action_flags)(pe_action_t *action, const pe_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 pe_action_optional to affect only * mandatory actions, and pe_action_runnable to * affect only runnable actions) * \param[in] type Group of enum pe_ordering flags to apply * \param[in,out] data_set Cluster working set * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t (*update_ordered_actions)(pe_action_t *first, pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pe_working_set_t *data_set); void (*output_actions)(pe_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)(pe_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 pe_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 allocated to a node. * * \param[in] rsc Resource with utilization to add * \param[in] orig_rsc Resource being allocated (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 pe_resource_t *rsc, const pe_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)(pe_resource_t *rsc); }; // Actions (pcmk_sched_actions.c) G_GNUC_INTERNAL void pcmk__update_action_for_orderings(pe_action_t *action, pe_working_set_t *data_set); G_GNUC_INTERNAL uint32_t pcmk__update_ordered_actions(pe_action_t *first, pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__log_action(const char *pre_text, const pe_action_t *action, bool details); G_GNUC_INTERNAL pe_action_t *pcmk__new_cancel_action(pe_resource_t *rsc, const char *name, guint interval_ms, const pe_node_t *node); G_GNUC_INTERNAL pe_action_t *pcmk__new_shutdown_action(pe_node_t *node); G_GNUC_INTERNAL bool pcmk__action_locks_rsc_to_node(const pe_action_t *action); G_GNUC_INTERNAL void pcmk__deduplicate_action_inputs(pe_action_t *action); G_GNUC_INTERNAL void pcmk__output_actions(pe_working_set_t *data_set); G_GNUC_INTERNAL bool pcmk__check_action_config(pe_resource_t *rsc, pe_node_t *node, const xmlNode *xml_op); G_GNUC_INTERNAL void pcmk__handle_rsc_config_changes(pe_working_set_t *data_set); // Recurring actions (pcmk_sched_recurring.c) G_GNUC_INTERNAL void pcmk__create_recurring_actions(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__schedule_cancel(pe_resource_t *rsc, const char *call_id, const char *task, guint interval_ms, const pe_node_t *node, const char *reason); G_GNUC_INTERNAL void pcmk__reschedule_recurring(pe_resource_t *rsc, const char *task, guint interval_ms, pe_node_t *node); G_GNUC_INTERNAL bool pcmk__action_is_recurring(const pe_action_t *action); // Producing transition graphs (pcmk_graph_producer.c) G_GNUC_INTERNAL bool pcmk__graph_has_loop(const pe_action_t *init_action, const pe_action_t *action, pe_action_wrapper_t *input); G_GNUC_INTERNAL void pcmk__add_rsc_actions_to_graph(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_graph(pe_working_set_t *data_set); // Fencing (pcmk_sched_fencing.c) G_GNUC_INTERNAL void pcmk__order_vs_fence(pe_action_t *stonith_op, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__order_vs_unfence(const pe_resource_t *rsc, pe_node_t *node, pe_action_t *action, enum pe_ordering order); G_GNUC_INTERNAL void pcmk__fence_guest(pe_node_t *node); G_GNUC_INTERNAL bool pcmk__node_unfenced(const pe_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(pe_working_set_t *data_set, cib_t *cib, const pcmk_injections_t *injections); // Constraints of any type (pcmk_sched_constraints.c) G_GNUC_INTERNAL pe_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 pe_working_set_t *data_set); G_GNUC_INTERNAL bool pcmk__valid_resource_or_tag(const pe_working_set_t *data_set, const char *id, pe_resource_t **rsc, pe_tag_t **tag); G_GNUC_INTERNAL bool pcmk__tag_to_set(xmlNode *xml_obj, xmlNode **rsc_set, const char *attr, bool convert_rsc, const pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__create_internal_constraints(pe_working_set_t *data_set); // Location constraints G_GNUC_INTERNAL void pcmk__unpack_location(xmlNode *xml_obj, pe_working_set_t *data_set); G_GNUC_INTERNAL pe__location_t *pcmk__new_location(const char *id, pe_resource_t *rsc, int node_weight, const char *discover_mode, pe_node_t *foo_node, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__apply_locations(pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__apply_location(pe_resource_t *rsc, pe__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, }; G_GNUC_INTERNAL enum pcmk__coloc_affects pcmk__colocation_affects(const pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation, bool preview); G_GNUC_INTERNAL void pcmk__apply_coloc_to_weights(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__apply_coloc_to_priority(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__add_colocated_node_scores(pe_resource_t *rsc, const char *log_id, GHashTable **nodes, const char *attr, float factor, uint32_t flags); G_GNUC_INTERNAL void pcmk__add_dependent_scores(gpointer data, gpointer user_data); G_GNUC_INTERNAL void pcmk__unpack_colocation(xmlNode *xml_obj, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__add_this_with(GList **list, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__add_this_with_list(GList **list, GList *addition); G_GNUC_INTERNAL void pcmk__add_with_this(GList **list, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__add_with_this_list(GList **list, GList *addition); G_GNUC_INTERNAL void pcmk__new_colocation(const char *id, const char *node_attr, int score, pe_resource_t *dependent, pe_resource_t *primary, const char *dependent_role, const char *primary_role, bool influence, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__block_colocation_dependents(pe_action_t *action, pe_working_set_t *data_set); /*! * \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 pe_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, pe_rsc_allow_remote_remotes) && !pcmk_is_set(rsc->flags, pe_rsc_failed) && pcmk__list_of_1(rsc->running_on)) { return false; } /* The dependent in a colocation influences the primary's location * if the influence option is true or the primary is not yet active. */ return colocation->influence || (rsc->running_on == NULL); } // Ordering constraints (pcmk_sched_ordering.c) G_GNUC_INTERNAL void pcmk__new_ordering(pe_resource_t *first_rsc, char *first_task, pe_action_t *first_action, pe_resource_t *then_rsc, char *then_task, pe_action_t *then_action, uint32_t flags, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__unpack_ordering(xmlNode *xml_obj, pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__disable_invalid_orderings(pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__order_stops_before_shutdown(pe_node_t *node, pe_action_t *shutdown_op); G_GNUC_INTERNAL void pcmk__apply_orderings(pe_working_set_t *data_set); G_GNUC_INTERNAL void pcmk__order_after_each(pe_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 Bitmask of enum pe_ordering 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)->cluster) #define pcmk__order_starts(rsc1, rsc2, flags) \ pcmk__order_resource_actions((rsc1), CRMD_ACTION_START, \ (rsc2), CRMD_ACTION_START, (flags)) #define pcmk__order_stops(rsc1, rsc2, flags) \ pcmk__order_resource_actions((rsc1), CRMD_ACTION_STOP, \ (rsc2), CRMD_ACTION_STOP, (flags)) // Ticket constraints (pcmk_sched_tickets.c) G_GNUC_INTERNAL void pcmk__unpack_rsc_ticket(xmlNode *xml_obj, pe_working_set_t *data_set); // Promotable clone resources (pcmk_sched_promotable.c) G_GNUC_INTERNAL void pcmk__add_promotion_scores(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__require_promotion_tickets(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__set_instance_roles(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_promotable_actions(pe_resource_t *clone); G_GNUC_INTERNAL void pcmk__promotable_restart_ordering(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__order_promotable_instances(pe_resource_t *clone); G_GNUC_INTERNAL void pcmk__update_dependent_with_promotable(const pe_resource_t *primary, pe_resource_t *dependent, const pcmk__colocation_t *colocation); G_GNUC_INTERNAL void pcmk__update_promotable_dependent_priority(const pe_resource_t *primary, pe_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 pe_node_t *node); G_GNUC_INTERNAL void pcmk__order_remote_connection_actions(pe_working_set_t *data_set); G_GNUC_INTERNAL bool pcmk__rsc_corresponds_to_guest(const pe_resource_t *rsc, const pe_node_t *node); G_GNUC_INTERNAL pe_node_t *pcmk__connection_host_for_action(const pe_action_t *action); G_GNUC_INTERNAL void pcmk__substitute_remote_addr(pe_resource_t *rsc, GHashTable *params); G_GNUC_INTERNAL void pcmk__add_bundle_meta_to_xml(xmlNode *args_xml, const pe_action_t *action); // Primitives (pcmk_sched_primitive.c) G_GNUC_INTERNAL pe_node_t *pcmk__primitive_assign(pe_resource_t *rsc, const pe_node_t *prefer); G_GNUC_INTERNAL void pcmk__primitive_create_actions(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__primitive_internal_constraints(pe_resource_t *rsc); G_GNUC_INTERNAL enum pe_action_flags pcmk__primitive_action_flags(pe_action_t *action, const pe_node_t *node); G_GNUC_INTERNAL void pcmk__primitive_apply_coloc_score(pe_resource_t *dependent, - const pe_resource_t *primary, + pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_primitive_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__primitive_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__schedule_cleanup(pe_resource_t *rsc, const pe_node_t *node, bool optional); G_GNUC_INTERNAL void pcmk__primitive_add_graph_meta(const pe_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__primitive_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__primitive_shutdown_lock(pe_resource_t *rsc); // Groups (pcmk_sched_group.c) G_GNUC_INTERNAL pe_node_t *pcmk__group_assign(pe_resource_t *rsc, const pe_node_t *prefer); G_GNUC_INTERNAL void pcmk__group_create_actions(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__group_internal_constraints(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__group_apply_coloc_score(pe_resource_t *dependent, - const pe_resource_t *primary, + pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_group_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__group_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__group_add_colocated_node_scores(pe_resource_t *rsc, const char *log_id, GHashTable **nodes, const char *attr, float factor, uint32_t flags); G_GNUC_INTERNAL void pcmk__group_apply_location(pe_resource_t *rsc, pe__location_t *location); G_GNUC_INTERNAL enum pe_action_flags pcmk__group_action_flags(pe_action_t *action, const pe_node_t *node); G_GNUC_INTERNAL uint32_t pcmk__group_update_ordered_actions(pe_action_t *first, pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pe_working_set_t *data_set); G_GNUC_INTERNAL GList *pcmk__group_colocated_resources(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *colocated_rscs); G_GNUC_INTERNAL void pcmk__group_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__group_shutdown_lock(pe_resource_t *rsc); // Clones (pcmk_sched_clone.c) G_GNUC_INTERNAL pe_node_t *pcmk__clone_assign(pe_resource_t *rsc, const pe_node_t *prefer); G_GNUC_INTERNAL void pcmk__clone_create_actions(pe_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__clone_create_probe(pe_resource_t *rsc, pe_node_t *node); G_GNUC_INTERNAL void pcmk__clone_internal_constraints(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__clone_apply_coloc_score(pe_resource_t *dependent, - const pe_resource_t *primary, + pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_clone_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__clone_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__clone_apply_location(pe_resource_t *rsc, pe__location_t *constraint); G_GNUC_INTERNAL enum pe_action_flags pcmk__clone_action_flags(pe_action_t *action, const pe_node_t *node); G_GNUC_INTERNAL void pcmk__clone_add_actions_to_graph(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__clone_add_graph_meta(const pe_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__clone_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization); G_GNUC_INTERNAL void pcmk__clone_shutdown_lock(pe_resource_t *rsc); // Bundles (pcmk_sched_bundle.c) -G_GNUC_INTERNAL -const pe_resource_t *pcmk__get_rsc_in_container(const pe_resource_t *instance); - G_GNUC_INTERNAL void pcmk__bundle_apply_coloc_score(pe_resource_t *dependent, - const pe_resource_t *primary, + pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent); G_GNUC_INTERNAL void pcmk__with_bundle_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__bundle_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list); G_GNUC_INTERNAL void pcmk__output_bundle_actions(pe_resource_t *rsc); // Clone instances or bundle replica containers (pcmk_sched_instances.c) G_GNUC_INTERNAL void pcmk__assign_instances(pe_resource_t *collective, GList *instances, int max_total, int max_per_node); G_GNUC_INTERNAL void pcmk__create_instance_actions(pe_resource_t *rsc, GList *instances); G_GNUC_INTERNAL bool pcmk__instance_matches(const pe_resource_t *instance, const pe_node_t *node, enum rsc_role_e role, bool current); G_GNUC_INTERNAL pe_resource_t *pcmk__find_compatible_instance(const pe_resource_t *match_rsc, const pe_resource_t *rsc, enum rsc_role_e role, bool current); G_GNUC_INTERNAL uint32_t pcmk__instance_update_ordered_actions(pe_action_t *first, pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pe_working_set_t *data_set); G_GNUC_INTERNAL enum pe_action_flags pcmk__collective_action_flags(pe_action_t *action, const GList *instances, const pe_node_t *node); G_GNUC_INTERNAL void pcmk__add_collective_constraints(GList **list, const pe_resource_t *instance, const pe_resource_t *collective, bool with_this); // 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, 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 pe_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 GList *pcmk__sort_nodes(GList *nodes, pe_node_t *active_node); G_GNUC_INTERNAL void pcmk__apply_node_health(pe_working_set_t *data_set); G_GNUC_INTERNAL pe_node_t *pcmk__top_allowed_node(const pe_resource_t *rsc, const pe_node_t *node); // Functions applying to more than one variant (pcmk_sched_resource.c) G_GNUC_INTERNAL void pcmk__set_allocation_methods(pe_working_set_t *data_set); G_GNUC_INTERNAL bool pcmk__rsc_agent_changed(pe_resource_t *rsc, pe_node_t *node, const xmlNode *rsc_entry, bool active_on_node); G_GNUC_INTERNAL GList *pcmk__rscs_matching_id(const char *id, const pe_working_set_t *data_set); G_GNUC_INTERNAL GList *pcmk__colocated_resources(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *colocated_rscs); G_GNUC_INTERNAL void pcmk__noop_add_graph_meta(const pe_resource_t *rsc, xmlNode *xml); G_GNUC_INTERNAL void pcmk__output_resource_actions(pe_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__finalize_assignment(pe_resource_t *rsc, pe_node_t *chosen, bool force); G_GNUC_INTERNAL bool pcmk__assign_resource(pe_resource_t *rsc, pe_node_t *node, bool force); G_GNUC_INTERNAL void pcmk__unassign_resource(pe_resource_t *rsc); G_GNUC_INTERNAL bool pcmk__threshold_reached(pe_resource_t *rsc, const pe_node_t *node, pe_resource_t **failed); G_GNUC_INTERNAL void pcmk__sort_resources(pe_working_set_t *data_set); 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(pe_resource_t *rsc, pe_node_t *node); G_GNUC_INTERNAL void pcmk__order_probes(pe_working_set_t *data_set); G_GNUC_INTERNAL bool pcmk__probe_resource_list(GList *rscs, pe_node_t *node); G_GNUC_INTERNAL void pcmk__schedule_probes(pe_working_set_t *data_set); // Functions related to live migration (pcmk_sched_migration.c) void pcmk__create_migration_actions(pe_resource_t *rsc, const pe_node_t *current); void pcmk__abort_dangling_migration(void *data, void *user_data); bool pcmk__rsc_can_migrate(const pe_resource_t *rsc, const pe_node_t *current); void pcmk__order_migration_equivalents(pe__ordering_t *order); // Functions related to node utilization (pcmk_sched_utilization.c) G_GNUC_INTERNAL int pcmk__compare_node_capacities(const pe_node_t *node1, const pe_node_t *node2); G_GNUC_INTERNAL void pcmk__consume_node_capacity(GHashTable *current_utilization, const pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__release_node_capacity(GHashTable *current_utilization, const pe_resource_t *rsc); G_GNUC_INTERNAL const pe_node_t *pcmk__ban_insufficient_capacity(pe_resource_t *rsc); G_GNUC_INTERNAL void pcmk__create_utilization_constraints(pe_resource_t *rsc, const GList *allowed_nodes); G_GNUC_INTERNAL void pcmk__show_node_capacities(const char *desc, pe_working_set_t *data_set); #endif // PCMK__LIBPACEMAKER_PRIVATE__H diff --git a/lib/pacemaker/pcmk_sched_bundle.c b/lib/pacemaker/pcmk_sched_bundle.c index d1b76baafd..775c535257 100644 --- a/lib/pacemaker/pcmk_sched_bundle.c +++ b/lib/pacemaker/pcmk_sched_bundle.c @@ -1,835 +1,908 @@ /* * Copyright 2004-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "libpacemaker_private.h" -#define PE__VARIANT_BUNDLE 1 -#include - +/*! + * \internal + * \brief Assign a single bundle replica's resources (other than container) + * + * \param[in,out] replica Replica to assign + * \param[in] user_data Preferred node, if any + * + * \return true (to indicate that any further replicas should be processed) + */ static bool -is_bundle_node(pe__bundle_variant_data_t *data, pe_node_t *node) +assign_replica(pe__bundle_replica_t *replica, void *user_data) { - for (GList *gIter = data->replicas; gIter != NULL; gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; + pe_node_t *container_host = NULL; + const pe_node_t *prefer = user_data; + const pe_resource_t *bundle = pe__const_top_resource(replica->container, + true); + + if (replica->ip != NULL) { + pe_rsc_trace(bundle, "Assigning bundle %s IP %s", + bundle->id, replica->ip->id); + replica->ip->cmds->assign(replica->ip, prefer); + } + + container_host = replica->container->allocated_to; + if (replica->remote != NULL) { + if (pe__is_guest_or_remote_node(container_host)) { + /* REMOTE_CONTAINER_HACK: "Nested" connection resources must be on + * the same host because Pacemaker Remote only supports a single + * active connection. + */ + pcmk__new_colocation("replica-remote-with-host-remote", NULL, + INFINITY, replica->remote, + container_host->details->remote_rsc, NULL, + NULL, true, bundle->cluster); + } + pe_rsc_trace(bundle, "Assigning bundle %s connection %s", + bundle->id, replica->remote->id); + replica->remote->cmds->assign(replica->remote, prefer); + } + + if (replica->child != NULL) { + pe_node_t *node = NULL; + GHashTableIter iter; - if (node->details == replica->node->details) { - return TRUE; + g_hash_table_iter_init(&iter, replica->child->allowed_nodes); + while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { + if (!pe__same_node(node, replica->node)) { + node->weight = -INFINITY; + } else if (!pcmk__threshold_reached(replica->child, node, NULL)) { + node->weight = INFINITY; + } } + + pe__set_resource_flags(replica->child->parent, pe_rsc_allocating); + pe_rsc_trace(bundle, "Assigning bundle %s replica child %s", + bundle->id, replica->child->id); + replica->child->cmds->assign(replica->child, replica->node); + pe__clear_resource_flags(replica->child->parent, pe_rsc_allocating); } - return FALSE; + return true; } /*! * \internal * \brief Assign a bundle resource to a node * * \param[in,out] rsc Resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * * \return Node that \p rsc is assigned to, if assigned entirely to one node */ pe_node_t * pcmk__bundle_allocate(pe_resource_t *rsc, const pe_node_t *prefer) { GList *containers = NULL; - pe__bundle_variant_data_t *bundle_data = NULL; + pe_resource_t *bundled_resource = NULL; CRM_CHECK(rsc != NULL, return NULL); - get_bundle_variant_data(bundle_data, rsc); - pe__set_resource_flags(rsc, pe_rsc_allocating); containers = pe__bundle_containers(rsc); pe__show_node_weights(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); containers = g_list_sort(containers, pcmk__cmp_instance); - pcmk__assign_instances(rsc, containers, bundle_data->nreplicas, - bundle_data->nreplicas_per_host); + pcmk__assign_instances(rsc, containers, pe__bundle_max(rsc), + rsc->fns->max_per_node(rsc)); g_list_free(containers); - for (GList *gIter = bundle_data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; - pe_node_t *container_host = NULL; - - CRM_ASSERT(replica); - if (replica->ip) { - pe_rsc_trace(rsc, "Allocating bundle %s IP %s", - rsc->id, replica->ip->id); - replica->ip->cmds->assign(replica->ip, prefer); - } - - container_host = replica->container->allocated_to; - if (replica->remote && pe__is_guest_or_remote_node(container_host)) { - /* We need 'nested' connection resources to be on the same - * host because pacemaker-remoted only supports a single - * active connection - */ - pcmk__new_colocation("child-remote-with-docker-remote", NULL, - INFINITY, replica->remote, - container_host->details->remote_rsc, NULL, - NULL, true, rsc->cluster); - } - - if (replica->remote) { - pe_rsc_trace(rsc, "Allocating bundle %s connection %s", - rsc->id, replica->remote->id); - replica->remote->cmds->assign(replica->remote, prefer); - } - - // Explicitly allocate replicas' children before bundle child - if (replica->child) { - pe_node_t *node = NULL; - GHashTableIter iter; - - g_hash_table_iter_init(&iter, replica->child->allowed_nodes); - while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { - if (node->details != replica->node->details) { - node->weight = -INFINITY; - } else if (!pcmk__threshold_reached(replica->child, node, - NULL)) { - node->weight = INFINITY; - } - } + pe__foreach_bundle_replica(rsc, assign_replica, (void *) prefer); - pe__set_resource_flags(replica->child->parent, pe_rsc_allocating); - pe_rsc_trace(rsc, "Allocating bundle %s replica child %s", - rsc->id, replica->child->id); - replica->child->cmds->assign(replica->child, replica->node); - pe__clear_resource_flags(replica->child->parent, - pe_rsc_allocating); - } - } - - if (bundle_data->child) { + bundled_resource = pe__bundled_resource(rsc); + if (bundled_resource != NULL) { pe_node_t *node = NULL; GHashTableIter iter; - g_hash_table_iter_init(&iter, bundle_data->child->allowed_nodes); + g_hash_table_iter_init(&iter, bundled_resource->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { - if (is_bundle_node(bundle_data, node)) { + if (pe__node_is_bundle_instance(rsc, node)) { node->weight = 0; } else { node->weight = -INFINITY; } } pe_rsc_trace(rsc, "Allocating bundle %s child %s", - rsc->id, bundle_data->child->id); - bundle_data->child->cmds->assign(bundle_data->child, prefer); + rsc->id, bundled_resource->id); + bundled_resource->cmds->assign(bundled_resource, prefer); } pe__clear_resource_flags(rsc, pe_rsc_allocating|pe_rsc_provisional); return NULL; } +/*! + * \internal + * \brief Create actions for a bundle replica's resources (other than child) + * + * \param[in,out] replica Replica to create actions for + * \param[in] user_data Unused + * + * \return true (to indicate that any further replicas should be processed) + */ +static bool +create_replica_actions(pe__bundle_replica_t *replica, void *user_data) +{ + if (replica->ip != NULL) { + replica->ip->cmds->create_actions(replica->ip); + } + if (replica->container != NULL) { + replica->container->cmds->create_actions(replica->container); + } + if (replica->remote != NULL) { + replica->remote->cmds->create_actions(replica->remote); + } + return true; +} void pcmk__bundle_create_actions(pe_resource_t *rsc) { pe_action_t *action = NULL; GList *containers = NULL; - pe__bundle_variant_data_t *bundle_data = NULL; + pe_resource_t *bundled_resource = NULL; CRM_CHECK(rsc != NULL, return); - containers = pe__bundle_containers(rsc); - get_bundle_variant_data(bundle_data, rsc); - for (GList *gIter = bundle_data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; - - CRM_ASSERT(replica); - if (replica->ip) { - replica->ip->cmds->create_actions(replica->ip); - } - if (replica->container) { - replica->container->cmds->create_actions(replica->container); - } - if (replica->remote) { - replica->remote->cmds->create_actions(replica->remote); - } - } + pe__foreach_bundle_replica(rsc, create_replica_actions, NULL); + containers = pe__bundle_containers(rsc); pcmk__create_instance_actions(rsc, containers); - if (bundle_data->child) { - bundle_data->child->cmds->create_actions(bundle_data->child); + bundled_resource = pe__bundled_resource(rsc); + if (bundled_resource != NULL) { + bundled_resource->cmds->create_actions(bundled_resource); - if (pcmk_is_set(bundle_data->child->flags, pe_rsc_promotable)) { + if (pcmk_is_set(bundled_resource->flags, pe_rsc_promotable)) { /* promote */ pe__new_rsc_pseudo_action(rsc, RSC_PROMOTE, true, true); action = pe__new_rsc_pseudo_action(rsc, RSC_PROMOTED, true, true); action->priority = INFINITY; /* demote */ pe__new_rsc_pseudo_action(rsc, RSC_DEMOTE, true, true); action = pe__new_rsc_pseudo_action(rsc, RSC_DEMOTED, true, true); action->priority = INFINITY; } } g_list_free(containers); } +/*! + * \internal + * \brief Create internal constraints for a bundle replica's resources + * + * \param[in,out] replica Replica to create internal constraints for + * \param[in,out] user_data Replica's parent bundle + * + * \return true (to indicate that any further replicas should be processed) + */ +static bool +replica_internal_constraints(pe__bundle_replica_t *replica, void *user_data) +{ + pe_resource_t *bundle = user_data; + + replica->container->cmds->internal_constraints(replica->container); + + // Start bundle -> start replica container + pcmk__order_starts(bundle, replica->container, + pe_order_runnable_left|pe_order_implies_first_printed); + + // Stop bundle -> stop replica child and container + if (replica->child != NULL) { + pcmk__order_stops(bundle, replica->child, + pe_order_implies_first_printed); + } + pcmk__order_stops(bundle, replica->container, + pe_order_implies_first_printed); + + // Start replica container -> bundle is started + pcmk__order_resource_actions(replica->container, RSC_START, bundle, + RSC_STARTED, + pe_order_implies_then_printed); + + // Stop replica container -> bundle is stopped + pcmk__order_resource_actions(replica->container, RSC_STOP, bundle, + RSC_STOPPED, + pe_order_implies_then_printed); + + if (replica->ip != NULL) { + replica->ip->cmds->internal_constraints(replica->ip); + + // Replica IP address -> replica container (symmetric) + pcmk__order_starts(replica->ip, replica->container, + pe_order_runnable_left|pe_order_preserve); + pcmk__order_stops(replica->container, replica->ip, + pe_order_implies_first|pe_order_preserve); + + pcmk__new_colocation("ip-with-container", NULL, INFINITY, replica->ip, + replica->container, NULL, NULL, true, + bundle->cluster); + } + + if (replica->remote != NULL) { + /* This handles ordering and colocating remote relative to container + * (via "resource-with-container"). Since IP is also ordered and + * colocated relative to the container, we don't need to do anything + * explicit here with IP. + */ + replica->remote->cmds->internal_constraints(replica->remote); + } + + if (replica->child != NULL) { + CRM_ASSERT(replica->remote != NULL); + // "Start remote then child" is implicit in scheduler's remote logic + } + return true; +} + void pcmk__bundle_internal_constraints(pe_resource_t *rsc) { - pe__bundle_variant_data_t *bundle_data = NULL; + pe_resource_t *bundled_resource = NULL; CRM_CHECK(rsc != NULL, return); - get_bundle_variant_data(bundle_data, rsc); - - if (bundle_data->child) { - pcmk__order_resource_actions(rsc, RSC_START, bundle_data->child, + bundled_resource = pe__bundled_resource(rsc); + if (bundled_resource != NULL) { + pcmk__order_resource_actions(rsc, RSC_START, bundled_resource, RSC_START, pe_order_implies_first_printed); - pcmk__order_resource_actions(rsc, RSC_STOP, bundle_data->child, - RSC_STOP, pe_order_implies_first_printed); + pcmk__order_resource_actions(rsc, RSC_STOP, bundled_resource, RSC_STOP, + pe_order_implies_first_printed); - if (bundle_data->child->children) { - pcmk__order_resource_actions(bundle_data->child, RSC_STARTED, rsc, + if (bundled_resource->children != NULL) { + pcmk__order_resource_actions(bundled_resource, RSC_STARTED, rsc, RSC_STARTED, pe_order_implies_then_printed); - pcmk__order_resource_actions(bundle_data->child, RSC_STOPPED, rsc, + pcmk__order_resource_actions(bundled_resource, RSC_STOPPED, rsc, RSC_STOPPED, pe_order_implies_then_printed); } else { - pcmk__order_resource_actions(bundle_data->child, RSC_START, rsc, + pcmk__order_resource_actions(bundled_resource, RSC_START, rsc, RSC_STARTED, pe_order_implies_then_printed); - pcmk__order_resource_actions(bundle_data->child, RSC_STOP, rsc, + pcmk__order_resource_actions(bundled_resource, RSC_STOP, rsc, RSC_STOPPED, pe_order_implies_then_printed); } } - for (GList *gIter = bundle_data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; - - CRM_ASSERT(replica); - CRM_ASSERT(replica->container); - - replica->container->cmds->internal_constraints(replica->container); - - pcmk__order_starts(rsc, replica->container, - pe_order_runnable_left|pe_order_implies_first_printed); - - if (replica->child) { - pcmk__order_stops(rsc, replica->child, - pe_order_implies_first_printed); - } - pcmk__order_stops(rsc, replica->container, - pe_order_implies_first_printed); - pcmk__order_resource_actions(replica->container, RSC_START, rsc, - RSC_STARTED, - pe_order_implies_then_printed); - pcmk__order_resource_actions(replica->container, RSC_STOP, rsc, - RSC_STOPPED, - pe_order_implies_then_printed); - - if (replica->ip) { - replica->ip->cmds->internal_constraints(replica->ip); - - // Start IP then container - pcmk__order_starts(replica->ip, replica->container, - pe_order_runnable_left|pe_order_preserve); - pcmk__order_stops(replica->container, replica->ip, - pe_order_implies_first|pe_order_preserve); - - pcmk__new_colocation("ip-with-docker", NULL, INFINITY, replica->ip, - replica->container, NULL, NULL, true, - rsc->cluster); - } - - if (replica->remote) { - /* This handles ordering and colocating remote relative to container - * (via "resource-with-container"). Since IP is also ordered and - * colocated relative to the container, we don't need to do anything - * explicit here with IP. - */ - replica->remote->cmds->internal_constraints(replica->remote); - } - - if (replica->child) { - CRM_ASSERT(replica->remote); + pe__foreach_bundle_replica(rsc, replica_internal_constraints, rsc); - // "Start remote then child" is implicit in scheduler's remote logic - } - - } - - if (bundle_data->child) { - bundle_data->child->cmds->internal_constraints(bundle_data->child); - if (pcmk_is_set(bundle_data->child->flags, pe_rsc_promotable)) { + if (bundled_resource != NULL) { + bundled_resource->cmds->internal_constraints(bundled_resource); + if (pcmk_is_set(bundled_resource->flags, pe_rsc_promotable)) { pcmk__promotable_restart_ordering(rsc); /* child demoted before global demoted */ - pcmk__order_resource_actions(bundle_data->child, RSC_DEMOTED, rsc, + pcmk__order_resource_actions(bundled_resource, RSC_DEMOTED, rsc, RSC_DEMOTED, pe_order_implies_then_printed); /* global demote before child demote */ - pcmk__order_resource_actions(rsc, RSC_DEMOTE, bundle_data->child, + pcmk__order_resource_actions(rsc, RSC_DEMOTE, bundled_resource, RSC_DEMOTE, pe_order_implies_first_printed); /* child promoted before global promoted */ - pcmk__order_resource_actions(bundle_data->child, RSC_PROMOTED, rsc, + pcmk__order_resource_actions(bundled_resource, RSC_PROMOTED, rsc, RSC_PROMOTED, pe_order_implies_then_printed); /* global promote before child promote */ - pcmk__order_resource_actions(rsc, RSC_PROMOTE, bundle_data->child, + pcmk__order_resource_actions(rsc, RSC_PROMOTE, bundled_resource, RSC_PROMOTE, pe_order_implies_first_printed); } } } +struct match_data { + const pe_node_t *node; // Node to compare against replica + pe_resource_t *container; // Replica container corresponding to node +}; + +/*! + * \internal + * \brief Check whether a replica container is assigned to a given node + * + * \param[in,out] replica Replica to check + * \param[in,out] user_data struct match_data with node to compare against + * + * \return true if the replica does not match (to indicate further replicas + * should be processed), otherwise false + */ +static bool +match_replica_container(pe__bundle_replica_t *replica, void *user_data) +{ + struct match_data *match_data = user_data; + + if (pcmk__instance_matches(replica->container, match_data->node, + RSC_ROLE_UNKNOWN, false)) { + match_data->container = replica->container; + return false; // Match found, don't bother searching further replicas + } + return true; // No match, keep searching +} + static pe_resource_t * compatible_replica_for_node(const pe_resource_t *rsc_lh, - const pe_node_t *candidate, - const pe_resource_t *rsc, enum rsc_role_e filter, - gboolean current) + const pe_node_t *candidate, pe_resource_t *rsc) { - pe__bundle_variant_data_t *bundle_data = NULL; + struct match_data match_data = { candidate, NULL }; CRM_CHECK(candidate != NULL, return NULL); - get_bundle_variant_data(bundle_data, rsc); crm_trace("Looking for compatible child from %s for %s on %s", rsc_lh->id, rsc->id, pe__node_name(candidate)); - - for (GList *gIter = bundle_data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; - - if (pcmk__instance_matches(replica->container, candidate, filter, - current)) { - crm_trace("Pairing %s with %s on %s", - rsc_lh->id, replica->container->id, - pe__node_name(candidate)); - return replica->container; - } + pe__foreach_bundle_replica(rsc, match_replica_container, &match_data); + if (match_data.container == NULL) { + pe_rsc_trace(rsc, "Can't pair %s with %s", rsc_lh->id, rsc->id); + } else { + pe_rsc_trace(rsc, "Pairing %s with %s on %s", + rsc_lh->id, match_data.container->id, + pe__node_name(candidate)); } - - crm_trace("Can't pair %s with %s", rsc_lh->id, rsc->id); - return NULL; + return match_data.container; } static pe_resource_t * -compatible_replica(const pe_resource_t *rsc_lh, const pe_resource_t *rsc, - enum rsc_role_e filter, gboolean current, +compatible_replica(const pe_resource_t *rsc_lh, pe_resource_t *rsc, pe_working_set_t *data_set) { GList *scratch = NULL; pe_resource_t *pair = NULL; pe_node_t *active_node_lh = NULL; - active_node_lh = rsc_lh->fns->location(rsc_lh, NULL, current); + active_node_lh = rsc_lh->fns->location(rsc_lh, NULL, 0); if (active_node_lh) { - return compatible_replica_for_node(rsc_lh, active_node_lh, rsc, filter, - current); + return compatible_replica_for_node(rsc_lh, active_node_lh, rsc); } scratch = g_hash_table_get_values(rsc_lh->allowed_nodes); scratch = pcmk__sort_nodes(scratch, NULL); for (GList *gIter = scratch; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; - pair = compatible_replica_for_node(rsc_lh, node, rsc, filter, current); + pair = compatible_replica_for_node(rsc_lh, node, rsc); if (pair) { goto done; } } pe_rsc_debug(rsc, "Can't pair %s with %s", rsc_lh->id, (rsc? rsc->id : "none")); done: g_list_free(scratch); return pair; } +struct coloc_data { + const pcmk__colocation_t *colocation; + pe_resource_t *dependent; + GList *container_hosts; +}; + +/*! + * \internal + * \brief Apply a colocation score to replica node weights or resource priority + * + * \param[in,out] replica Replica to apply colocation score to + * \param[in,out] user_data struct coloc_data for colocation being applied + * + * \return true (to indicate that any further replicas should be processed) + */ +static bool +replica_apply_coloc_score(pe__bundle_replica_t *replica, void *user_data) +{ + struct coloc_data *coloc_data = user_data; + pe_node_t *chosen = NULL; + + if (coloc_data->colocation->score < INFINITY) { + replica->container->cmds->apply_coloc_score(coloc_data->dependent, + replica->container, + coloc_data->colocation, + false); + return true; + } + + chosen = replica->container->fns->location(replica->container, NULL, 0); + if ((chosen == NULL) + || is_set_recursive(replica->container, pe_rsc_block, true)) { + return true; + } + + if ((coloc_data->colocation->primary_role >= RSC_ROLE_PROMOTED) + && ((replica->child == NULL) + || (replica->child->next_role < RSC_ROLE_PROMOTED))) { + return true; + } + + pe_rsc_trace(pe__const_top_resource(replica->container, true), + "Allowing mandatory colocation %s using %s @%d", + coloc_data->colocation->id, pe__node_name(chosen), + chosen->weight); + coloc_data->container_hosts = g_list_prepend(coloc_data->container_hosts, + chosen); + return true; +} + /*! * \internal * \brief Apply a colocation's score to node weights or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node weights (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,out] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void -pcmk__bundle_apply_coloc_score(pe_resource_t *dependent, - const pe_resource_t *primary, +pcmk__bundle_apply_coloc_score(pe_resource_t *dependent, pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { - GList *allocated_primaries = NULL; - pe__bundle_variant_data_t *bundle_data = NULL; + struct coloc_data coloc_data = { colocation, dependent, NULL }; /* This should never be called for the bundle itself as a dependent. * Instead, we add its colocation constraints to its replicas and call the * apply_coloc_score() for the replicas as dependents. */ CRM_ASSERT(!for_dependent); CRM_CHECK((colocation != NULL) && (dependent != NULL) && (primary != NULL), return); CRM_ASSERT(dependent->variant == pe_native); if (pcmk_is_set(primary->flags, pe_rsc_provisional)) { pe_rsc_trace(primary, "%s is still provisional", primary->id); return; } else if (colocation->dependent->variant > pe_group) { pe_resource_t *primary_replica = compatible_replica(dependent, primary, - RSC_ROLE_UNKNOWN, - FALSE, dependent->cluster); if (primary_replica) { pe_rsc_debug(primary, "Pairing %s with %s", dependent->id, primary_replica->id); dependent->cmds->apply_coloc_score(dependent, primary_replica, colocation, true); } else if (colocation->score >= INFINITY) { crm_notice("Cannot pair %s with instance of %s", dependent->id, primary->id); pcmk__assign_resource(dependent, NULL, true); } else { pe_rsc_debug(primary, "Cannot pair %s with instance of %s", dependent->id, primary->id); } return; } - get_bundle_variant_data(bundle_data, primary); pe_rsc_trace(primary, "Processing constraint %s: %s -> %s %d", colocation->id, dependent->id, primary->id, colocation->score); - for (GList *gIter = bundle_data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; - - if (colocation->score < INFINITY) { - replica->container->cmds->apply_coloc_score(dependent, - replica->container, - colocation, false); - - } else { - pe_node_t *chosen = replica->container->fns->location(replica->container, - NULL, FALSE); - - if ((chosen == NULL) - || is_set_recursive(replica->container, pe_rsc_block, TRUE)) { - continue; - } - if ((colocation->primary_role >= RSC_ROLE_PROMOTED) - && (replica->child == NULL)) { - continue; - } - if ((colocation->primary_role >= RSC_ROLE_PROMOTED) - && (replica->child->next_role < RSC_ROLE_PROMOTED)) { - continue; - } - - pe_rsc_trace(primary, "Allowing %s: %s %d", - colocation->id, pe__node_name(chosen), chosen->weight); - allocated_primaries = g_list_prepend(allocated_primaries, chosen); - } - } + pe__foreach_bundle_replica(primary, replica_apply_coloc_score, &coloc_data); if (colocation->score >= INFINITY) { - node_list_exclude(dependent->allowed_nodes, allocated_primaries, FALSE); + node_list_exclude(dependent->allowed_nodes, coloc_data.container_hosts, + FALSE); } - g_list_free(allocated_primaries); + g_list_free(coloc_data.container_hosts); } // Bundle implementation of resource_alloc_functions_t:with_this_colocations() void pcmk__with_bundle_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_CHECK((rsc != NULL) && (rsc->variant == pe_container) && (orig_rsc != NULL) && (list != NULL), return); if (rsc == orig_rsc) { // Colocations are wanted for bundle itself pcmk__add_with_this_list(list, rsc->rsc_cons_lhs); // Only the bundle replicas' containers get the bundle's constraints } else if (pcmk_is_set(orig_rsc->flags, pe_rsc_replica_container)) { pcmk__add_collective_constraints(list, orig_rsc, rsc, true); } } // Bundle implementation of resource_alloc_functions_t:this_with_colocations() void pcmk__bundle_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_CHECK((rsc != NULL) && (rsc->variant == pe_container) && (orig_rsc != NULL) && (list != NULL), return); if (rsc == orig_rsc) { // Colocations are wanted for bundle itself pcmk__add_this_with_list(list, rsc->rsc_cons); // Only the bundle replicas' containers get the bundle's constraints } else if (pcmk_is_set(orig_rsc->flags, pe_rsc_replica_container)) { pcmk__add_collective_constraints(list, orig_rsc, rsc, false); } } enum pe_action_flags pcmk__bundle_action_flags(pe_action_t *action, const pe_node_t *node) { GList *containers = NULL; enum pe_action_flags flags = 0; - pe__bundle_variant_data_t *data = NULL; + pe_resource_t *bundled_resource = pe__bundled_resource(action->rsc); + + if (bundled_resource != NULL) { + enum action_tasks task = get_complex_task(bundled_resource, + action->task); - get_bundle_variant_data(data, action->rsc); - if(data->child) { - enum action_tasks task = get_complex_task(data->child, action->task); switch(task) { case no_action: case action_notify: case action_notified: case action_promote: case action_promoted: case action_demote: case action_demoted: return pcmk__collective_action_flags(action, - data->child->children, + bundled_resource->children, node); default: break; } } containers = pe__bundle_containers(action->rsc); flags = pcmk__collective_action_flags(action, containers, node); g_list_free(containers); return flags; } /*! * \internal - * \brief Get containerized resource corresponding to a given bundle container + * \brief Apply a location constraint to a bundle replica * - * \param[in] instance Collective instance that might be a bundle container + * \param[in,out] replica Replica to apply constraint to + * \param[in,out] user_data Location constraint to apply * - * \return Bundled resource instance inside \p instance if it is a bundle - * container instance, otherwise NULL + * \return true (to indicate that any further replicas should be processed) */ -const pe_resource_t * -pcmk__get_rsc_in_container(const pe_resource_t *instance) +static bool +apply_location_to_replica(pe__bundle_replica_t *replica, void *user_data) { - const pe__bundle_variant_data_t *data = NULL; - const pe_resource_t *top = pe__const_top_resource(instance, true); + pe__location_t *location = user_data; - if ((top == NULL) || (top->variant != pe_container)) { - return NULL; + if (replica->container != NULL) { + replica->container->cmds->apply_location(replica->container, location); } - get_bundle_variant_data(data, top); - - for (const GList *iter = data->replicas; iter != NULL; iter = iter->next) { - const pe__bundle_replica_t *replica = iter->data; - - if (instance == replica->container) { - return replica->child; - } + if (replica->ip != NULL) { + replica->ip->cmds->apply_location(replica->ip, location); } - return NULL; + return true; } void pcmk__bundle_rsc_location(pe_resource_t *rsc, pe__location_t *constraint) { - pe__bundle_variant_data_t *bundle_data = NULL; - get_bundle_variant_data(bundle_data, rsc); + pe_resource_t *bundled_resource = NULL; pcmk__apply_location(rsc, constraint); + pe__foreach_bundle_replica(rsc, apply_location_to_replica, constraint); + + bundled_resource = pe__bundled_resource(rsc); + if ((bundled_resource != NULL) + && ((constraint->role_filter == RSC_ROLE_UNPROMOTED) + || (constraint->role_filter == RSC_ROLE_PROMOTED))) { + bundled_resource->cmds->apply_location(bundled_resource, + constraint); + bundled_resource->rsc_location = g_list_prepend(bundled_resource->rsc_location, + constraint); + } +} - for (GList *gIter = bundle_data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; +#define XPATH_REMOTE "//nvpair[@name='" XML_RSC_ATTR_REMOTE_RA_ADDR "']" - if (replica->container) { - replica->container->cmds->apply_location(replica->container, - constraint); - } - if (replica->ip) { - replica->ip->cmds->apply_location(replica->ip, constraint); +/*! + * \internal + * \brief Add a bundle replica's actions to transition graph + * + * \param[in,out] replica Replica to add to graph + * \param[in] user_data Bundle that replica belongs to (for logging only) + * + * \return true (to indicate that any further replicas should be processed) + */ +static bool +add_replica_actions_to_graph(pe__bundle_replica_t *replica, void *user_data) +{ + if ((replica->remote != NULL) && (replica->container != NULL) + && pe__bundle_needs_remote_name(replica->remote)) { + + /* REMOTE_CONTAINER_HACK: Allow remote nodes to run containers that + * run pacemaker-remoted inside, without needing a separate IP for + * the container. This is done by configuring the inner remote's + * connection host as the magic string "#uname", then + * replacing it with the underlying host when needed. + */ + xmlNode *nvpair = get_xpath_object(XPATH_REMOTE, replica->remote->xml, + LOG_ERR); + const char *calculated_addr = NULL; + + // Replace the value in replica->remote->xml (if appropriate) + calculated_addr = pe__add_bundle_remote_name(replica->remote, + replica->remote->cluster, + nvpair, "value"); + if (calculated_addr != NULL) { + /* Since this is for the bundle as a resource, and not any + * particular action, replace the value in the default + * parameters (not evaluated for node). create_graph_action() + * will grab it from there to replace it in node-evaluated + * parameters. + */ + GHashTable *params = pe_rsc_params(replica->remote, + NULL, replica->remote->cluster); + + g_hash_table_replace(params, + strdup(XML_RSC_ATTR_REMOTE_RA_ADDR), + strdup(calculated_addr)); + } else { + pe_resource_t *bundle = user_data; + + /* The only way to get here is if the remote connection is + * neither currently running nor scheduled to run. That means we + * won't be doing any operations that require addr (only start + * requires it; we additionally use it to compare digests when + * unpacking status, promote, and migrate_from history, but + * that's already happened by this point). + */ + pe_rsc_info(bundle, + "Unable to determine address for bundle %s " + "remote connection", bundle->id); } } - - if (bundle_data->child - && ((constraint->role_filter == RSC_ROLE_UNPROMOTED) - || (constraint->role_filter == RSC_ROLE_PROMOTED))) { - bundle_data->child->cmds->apply_location(bundle_data->child, - constraint); - bundle_data->child->rsc_location = g_list_prepend(bundle_data->child->rsc_location, - constraint); + if (replica->ip != NULL) { + replica->ip->cmds->add_actions_to_graph(replica->ip); + } + if (replica->container != NULL) { + replica->container->cmds->add_actions_to_graph(replica->container); + } + if (replica->remote != NULL) { + replica->remote->cmds->add_actions_to_graph(replica->remote); } + return true; } /*! * \internal * \brief Add a resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void pcmk__bundle_expand(pe_resource_t *rsc) { - pe__bundle_variant_data_t *bundle_data = NULL; + pe_resource_t *bundled_resource = NULL; CRM_CHECK(rsc != NULL, return); - get_bundle_variant_data(bundle_data, rsc); - - if (bundle_data->child) { - bundle_data->child->cmds->add_actions_to_graph(bundle_data->child); + bundled_resource = pe__bundled_resource(rsc); + if (bundled_resource != NULL) { + bundled_resource->cmds->add_actions_to_graph(bundled_resource); } + pe__foreach_bundle_replica(rsc, add_replica_actions_to_graph, rsc); +} - for (GList *gIter = bundle_data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; +struct probe_data { + pe_resource_t *bundle; // Bundle being probed + pe_node_t *node; // Node to create probes on + bool any_created; // Whether any probes have been created +}; - CRM_ASSERT(replica); - if (replica->remote && replica->container - && pe__bundle_needs_remote_name(replica->remote)) { +/*! + * \internal + * \brief Order a bundle replica's start after another replica's probe + * + * \param[in,out] replica Replica to order start for + * \param[in,out] user_data Replica with probe to order after + * + * \return true (to indicate that any further replicas should be processed) + */ +static bool +order_replica_start_after(pe__bundle_replica_t *replica, void *user_data) +{ + pe__bundle_replica_t *probed_replica = user_data; - /* REMOTE_CONTAINER_HACK: Allow remote nodes to run containers that - * run pacemaker-remoted inside, without needing a separate IP for - * the container. This is done by configuring the inner remote's - * connection host as the magic string "#uname", then - * replacing it with the underlying host when needed. - */ - xmlNode *nvpair = get_xpath_object("//nvpair[@name='" XML_RSC_ATTR_REMOTE_RA_ADDR "']", - replica->remote->xml, LOG_ERR); - const char *calculated_addr = NULL; - - // Replace the value in replica->remote->xml (if appropriate) - calculated_addr = pe__add_bundle_remote_name(replica->remote, - rsc->cluster, - nvpair, "value"); - if (calculated_addr) { - /* Since this is for the bundle as a resource, and not any - * particular action, replace the value in the default - * parameters (not evaluated for node). create_graph_action() - * will grab it from there to replace it in node-evaluated - * parameters. - */ - GHashTable *params = pe_rsc_params(replica->remote, - NULL, rsc->cluster); - - g_hash_table_replace(params, - strdup(XML_RSC_ATTR_REMOTE_RA_ADDR), - strdup(calculated_addr)); - } else { - /* The only way to get here is if the remote connection is - * neither currently running nor scheduled to run. That means we - * won't be doing any operations that require addr (only start - * requires it; we additionally use it to compare digests when - * unpacking status, promote, and migrate_from history, but - * that's already happened by this point). - */ - crm_info("Unable to determine address for bundle %s remote connection", - rsc->id); - } - } - if (replica->ip) { - replica->ip->cmds->add_actions_to_graph(replica->ip); - } - if (replica->container) { - replica->container->cmds->add_actions_to_graph(replica->container); + if ((replica == probed_replica) || (replica->container == NULL)) { + return true; + } + pcmk__new_ordering(probed_replica->container, + pcmk__op_key(probed_replica->container->id, RSC_STATUS, + 0), + NULL, replica->container, + pcmk__op_key(replica->container->id, RSC_START, 0), NULL, + pe_order_optional|pe_order_same_node, + replica->container->cluster); + return true; +} + +/*! + * \internal + * \brief Create probes for a bundle replica's resources + * + * \param[in,out] replica Replica to create probes for + * \param[in,out] user_data struct probe_data + * + * \return true (to indicate that any further replicas should be processed) + */ +static bool +create_replica_probes(pe__bundle_replica_t *replica, void *user_data) +{ + struct probe_data *probe_data = user_data; + + if ((replica->ip != NULL) + && replica->ip->cmds->create_probe(replica->ip, probe_data->node)) { + probe_data->any_created = true; + } + if ((replica->child != NULL) + && pe__same_node(probe_data->node, replica->node) + && replica->child->cmds->create_probe(replica->child, probe_data->node)) { + probe_data->any_created = true; + } + if ((replica->container != NULL) + && replica->container->cmds->create_probe(replica->container, + probe_data->node)) { + probe_data->any_created = true; + + /* If we're limited to one replica per host (due to + * the lack of an IP range probably), then we don't + * want any of our peer containers starting until + * we've established that no other copies are already + * running. + * + * Partly this is to ensure that the maximum replicas per host is + * observed, but also to ensure that the containers + * don't fail to start because the necessary port + * mappings (which won't include an IP for uniqueness) + * are already taken + */ + if (probe_data->bundle->fns->max_per_node(probe_data->bundle) == 1) { + pe__foreach_bundle_replica(probe_data->bundle, + order_replica_start_after, replica); } - if (replica->remote) { - replica->remote->cmds->add_actions_to_graph(replica->remote); + } + if ((replica->container != NULL) && (replica->remote != NULL) + && replica->remote->cmds->create_probe(replica->remote, + probe_data->node)) { + /* Do not probe the remote resource until we know where the container is + * running. This is required for REMOTE_CONTAINER_HACK to correctly + * probe remote resources. + */ + char *probe_uuid = pcmk__op_key(replica->remote->id, RSC_STATUS, 0); + pe_action_t *probe = find_first_action(replica->remote->actions, + probe_uuid, NULL, + probe_data->node); + + free(probe_uuid); + if (probe != NULL) { + probe_data->any_created = true; + pe_rsc_trace(probe_data->bundle, "Ordering %s probe on %s", + replica->remote->id, pe__node_name(probe_data->node)); + pcmk__new_ordering(replica->container, + pcmk__op_key(replica->container->id, RSC_START, + 0), + NULL, replica->remote, NULL, probe, + pe_order_probe, probe_data->bundle->cluster); } } + return true; } /*! * \internal * * \brief Schedule any probes needed for a resource on a node * * \param[in,out] rsc Resource to create probe for * \param[in,out] node Node to create probe on * * \return true if any probe was created, otherwise false */ bool pcmk__bundle_create_probe(pe_resource_t *rsc, pe_node_t *node) { - bool any_created = false; - pe__bundle_variant_data_t *bundle_data = NULL; + struct probe_data probe_data = { rsc, node, false }; CRM_CHECK(rsc != NULL, return false); + pe__foreach_bundle_replica(rsc, create_replica_probes, &probe_data); + return probe_data.any_created; +} - get_bundle_variant_data(bundle_data, rsc); - for (GList *gIter = bundle_data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; - - CRM_ASSERT(replica); - if ((replica->ip != NULL) - && replica->ip->cmds->create_probe(replica->ip, node)) { - any_created = true; - } - if ((replica->child != NULL) && (node->details == replica->node->details) - && replica->child->cmds->create_probe(replica->child, node)) { - any_created = true; - } - if ((replica->container != NULL) - && replica->container->cmds->create_probe(replica->container, - node)) { - any_created = true; - - /* If we're limited to one replica per host (due to - * the lack of an IP range probably), then we don't - * want any of our peer containers starting until - * we've established that no other copies are already - * running. - * - * Partly this is to ensure that nreplicas_per_host is - * observed, but also to ensure that the containers - * don't fail to start because the necessary port - * mappings (which won't include an IP for uniqueness) - * are already taken - */ - - for (GList *tIter = bundle_data->replicas; - tIter && (bundle_data->nreplicas_per_host == 1); - tIter = tIter->next) { - pe__bundle_replica_t *other = tIter->data; - - if ((other != replica) && (other != NULL) - && (other->container != NULL)) { - - pcmk__new_ordering(replica->container, - pcmk__op_key(replica->container->id, RSC_STATUS, 0), - NULL, other->container, - pcmk__op_key(other->container->id, RSC_START, 0), - NULL, - pe_order_optional|pe_order_same_node, - rsc->cluster); - } - } - } - if ((replica->container != NULL) && (replica->remote != NULL) - && replica->remote->cmds->create_probe(replica->remote, node)) { - - /* Do not probe the remote resource until we know where the - * container is running. This is required for REMOTE_CONTAINER_HACK - * to correctly probe remote resources. - */ - char *probe_uuid = pcmk__op_key(replica->remote->id, RSC_STATUS, - 0); - pe_action_t *probe = find_first_action(replica->remote->actions, - probe_uuid, NULL, node); - - free(probe_uuid); - if (probe != NULL) { - any_created = true; - crm_trace("Ordering %s probe on %s", - replica->remote->id, pe__node_name(node)); - pcmk__new_ordering(replica->container, - pcmk__op_key(replica->container->id, RSC_START, 0), - NULL, replica->remote, NULL, probe, - pe_order_probe, rsc->cluster); - } - } +/*! + * \internal + * \brief Output actions for one bundle replica + * + * \param[in,out] replica Replica to output actions for + * \param[in] user_data Unused + * + * \return true (to indicate that any further replicas should be processed) + */ +static bool +output_replica_actions(pe__bundle_replica_t *replica, void *user_data) +{ + if (replica->ip != NULL) { + replica->ip->cmds->output_actions(replica->ip); + } + if (replica->container != NULL) { + replica->container->cmds->output_actions(replica->container); } - return any_created; + if (replica->remote != NULL) { + replica->remote->cmds->output_actions(replica->remote); + } + if (replica->child != NULL) { + replica->child->cmds->output_actions(replica->child); + } + return true; } void pcmk__output_bundle_actions(pe_resource_t *rsc) { - pe__bundle_variant_data_t *bundle_data = NULL; - CRM_CHECK(rsc != NULL, return); - get_bundle_variant_data(bundle_data, rsc); - for (GList *gIter = bundle_data->replicas; gIter != NULL; - gIter = gIter->next) { - pe__bundle_replica_t *replica = gIter->data; - - CRM_ASSERT(replica); - if (replica->ip != NULL) { - replica->ip->cmds->output_actions(replica->ip); - } - if (replica->container != NULL) { - replica->container->cmds->output_actions(replica->container); - } - if (replica->remote != NULL) { - replica->remote->cmds->output_actions(replica->remote); - } - if (replica->child != NULL) { - replica->child->cmds->output_actions(replica->child); - } - } + pe__foreach_bundle_replica(rsc, output_replica_actions, NULL); } // Bundle implementation of resource_alloc_functions_t:add_utilization() void pcmk__bundle_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { - pe__bundle_variant_data_t *bundle_data = NULL; - pe__bundle_replica_t *replica = NULL; + pe_resource_t *container = NULL; if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return; } - get_bundle_variant_data(bundle_data, rsc); - if (bundle_data->replicas == NULL) { - return; - } - /* All bundle replicas are identical, so using the utilization of the first * is sufficient for any. Only the implicit container resource can have * utilization values. */ - replica = (pe__bundle_replica_t *) bundle_data->replicas->data; - if (replica->container != NULL) { - replica->container->cmds->add_utilization(replica->container, orig_rsc, - all_rscs, utilization); + container = pe__first_container(rsc); + if (container != NULL) { + container->cmds->add_utilization(container, orig_rsc, all_rscs, + utilization); } } // Bundle implementation of resource_alloc_functions_t:shutdown_lock() void pcmk__bundle_shutdown_lock(pe_resource_t *rsc) { return; // Bundles currently don't support shutdown locks } diff --git a/lib/pacemaker/pcmk_sched_clone.c b/lib/pacemaker/pcmk_sched_clone.c index 9e4a982f10..9afa5997c3 100644 --- a/lib/pacemaker/pcmk_sched_clone.c +++ b/lib/pacemaker/pcmk_sched_clone.c @@ -1,673 +1,672 @@ /* * Copyright 2004-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Assign a clone resource's instances to nodes * * \param[in,out] rsc Clone resource to assign * \param[in] prefer Node to prefer, if all else is equal * * \return NULL (clones are not assigned to a single node) */ pe_node_t * pcmk__clone_assign(pe_resource_t *rsc, const pe_node_t *prefer) { CRM_ASSERT(pe_rsc_is_clone(rsc)); if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return NULL; // Assignment has already been done } // Detect assignment loops if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { pe_rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id); return NULL; } pe__set_resource_flags(rsc, pe_rsc_allocating); // If this clone is promotable, consider nodes' promotion scores if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__add_promotion_scores(rsc); } /* If this clone is colocated with any other resources, assign those first. * Since the this_with_colocations() method boils down to a copy of rsc_cons * for clones, we can use that here directly for efficiency. */ for (GList *iter = rsc->rsc_cons; iter != NULL; iter = iter->next) { pcmk__colocation_t *constraint = (pcmk__colocation_t *) iter->data; pe_rsc_trace(rsc, "%s: Assigning colocation %s primary %s first", rsc->id, constraint->id, constraint->primary->id); constraint->primary->cmds->assign(constraint->primary, prefer); } /* If any resources are colocated with this one, consider their preferences. * Because the with_this_colocations() method boils down to a copy of * rsc_cons_lhs for clones, we can use that here directly for efficiency. */ g_list_foreach(rsc->rsc_cons_lhs, pcmk__add_dependent_scores, rsc); pe__show_node_weights(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance); pcmk__assign_instances(rsc, rsc->children, pe__clone_max(rsc), pe__clone_node_max(rsc)); if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__set_instance_roles(rsc); } pe__clear_resource_flags(rsc, pe_rsc_provisional|pe_rsc_allocating); pe_rsc_trace(rsc, "Assigned clone %s", rsc->id); return NULL; } /*! * \internal * \brief Create all actions needed for a given clone resource * * \param[in,out] rsc Clone resource to create actions for */ void pcmk__clone_create_actions(pe_resource_t *rsc) { CRM_ASSERT(pe_rsc_is_clone(rsc)); pe_rsc_trace(rsc, "Creating actions for clone %s", rsc->id); pcmk__create_instance_actions(rsc, rsc->children); if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__create_promotable_actions(rsc); } } /*! * \internal * \brief Create implicit constraints needed for a clone resource * * \param[in,out] rsc Clone resource to create implicit constraints for */ void pcmk__clone_internal_constraints(pe_resource_t *rsc) { bool ordered = pe__clone_is_ordered(rsc); CRM_ASSERT(pe_rsc_is_clone(rsc)); pe_rsc_trace(rsc, "Creating internal constraints for clone %s", rsc->id); // Restart ordering: Stop -> stopped -> start -> started pcmk__order_resource_actions(rsc, RSC_STOPPED, rsc, RSC_START, pe_order_optional); pcmk__order_resource_actions(rsc, RSC_START, rsc, RSC_STARTED, pe_order_runnable_left); pcmk__order_resource_actions(rsc, RSC_STOP, rsc, RSC_STOPPED, pe_order_runnable_left); // Demoted -> stop and started -> promote if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__order_resource_actions(rsc, RSC_DEMOTED, rsc, RSC_STOP, pe_order_optional); pcmk__order_resource_actions(rsc, RSC_STARTED, rsc, RSC_PROMOTE, pe_order_runnable_left); } if (ordered) { /* Ordered clone instances must start and stop by instance number. The * instances might have been previously shuffled for assignment or * promotion purposes, so re-sort them. */ rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance_number); } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; instance->cmds->internal_constraints(instance); // Start clone -> start instance -> clone started pcmk__order_starts(rsc, instance, pe_order_runnable_left |pe_order_implies_first_printed); pcmk__order_resource_actions(instance, RSC_START, rsc, RSC_STARTED, pe_order_implies_then_printed); // Stop clone -> stop instance -> clone stopped pcmk__order_stops(rsc, instance, pe_order_implies_first_printed); pcmk__order_resource_actions(instance, RSC_STOP, rsc, RSC_STOPPED, pe_order_implies_then_printed); /* Instances of ordered clones must be started and stopped by instance * number. Since only some instances may be starting or stopping, order * each instance relative to every later instance. */ if (ordered) { for (GList *later = iter->next; later != NULL; later = later->next) { pcmk__order_starts(instance, (pe_resource_t *) later->data, pe_order_optional); pcmk__order_stops((pe_resource_t *) later->data, instance, pe_order_optional); } } } if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { pcmk__order_promotable_instances(rsc); } } /*! * \internal * \brief Check whether colocated resources can be interleaved * * \param[in] colocation Colocation constraint with clone as primary * * \return true if colocated resources can be interleaved, otherwise false */ static bool can_interleave(const pcmk__colocation_t *colocation) { const pe_resource_t *dependent = colocation->dependent; // Only colocations between clone or bundle resources use interleaving if (dependent->variant <= pe_group) { return false; } // Only the dependent needs to be marked for interleaving if (!crm_is_true(g_hash_table_lookup(dependent->meta, XML_RSC_ATTR_INTERLEAVE))) { return false; } /* @TODO Do we actually care about multiple primary instances sharing a * dependent instance? */ if (dependent->fns->max_per_node(dependent) != colocation->primary->fns->max_per_node(colocation->primary)) { pcmk__config_err("Cannot interleave %s and %s because they do not " "support the same number of instances per node", dependent->id, colocation->primary->id); return false; } return true; } /*! * \internal * \brief Apply a colocation's score to node weights or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node weights (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,out] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void -pcmk__clone_apply_coloc_score(pe_resource_t *dependent, - const pe_resource_t *primary, +pcmk__clone_apply_coloc_score(pe_resource_t *dependent, pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { GList *iter = NULL; /* This should never be called for the clone itself as a dependent. Instead, * we add its colocation constraints to its instances and call the * apply_coloc_score() method for the instances as dependents. */ CRM_ASSERT(!for_dependent); CRM_ASSERT((colocation != NULL) && pe_rsc_is_clone(primary) && (dependent != NULL) && (dependent->variant == pe_native)); if (pcmk_is_set(primary->flags, pe_rsc_provisional)) { pe_rsc_trace(primary, "Delaying processing colocation %s " "because cloned primary %s is still provisional", colocation->id, primary->id); return; } pe_rsc_trace(primary, "Processing colocation %s (%s with clone %s @%s)", colocation->id, dependent->id, primary->id, pcmk_readable_score(colocation->score)); // Apply role-specific colocations if (pcmk_is_set(primary->flags, pe_rsc_promotable) && (colocation->primary_role != RSC_ROLE_UNKNOWN)) { if (pcmk_is_set(dependent->flags, pe_rsc_provisional)) { // We're assigning the dependent to a node pcmk__update_dependent_with_promotable(primary, dependent, colocation); return; } if (colocation->dependent_role == RSC_ROLE_PROMOTED) { // We're choosing a role for the dependent pcmk__update_promotable_dependent_priority(primary, dependent, colocation); return; } } // Apply interleaved colocations if (can_interleave(colocation)) { pe_resource_t *primary_instance = NULL; primary_instance = pcmk__find_compatible_instance(dependent, primary, RSC_ROLE_UNKNOWN, false); if (primary_instance != NULL) { pe_rsc_debug(primary, "Interleaving %s with %s", dependent->id, primary_instance->id); dependent->cmds->apply_coloc_score(dependent, primary_instance, colocation, true); } else if (colocation->score >= INFINITY) { crm_notice("%s cannot run because it cannot interleave with " "any instance of %s", dependent->id, primary->id); pcmk__assign_resource(dependent, NULL, true); } else { pe_rsc_debug(primary, "%s will not colocate with %s " "because no instance can interleave with it", dependent->id, primary->id); } return; } // Apply mandatory colocations if (colocation->score >= INFINITY) { GList *primary_nodes = NULL; // Dependent can run only where primary will have unblocked instances for (iter = primary->children; iter != NULL; iter = iter->next) { const pe_resource_t *instance = iter->data; pe_node_t *chosen = instance->fns->location(instance, NULL, 0); if ((chosen != NULL) && !is_set_recursive(instance, pe_rsc_block, TRUE)) { pe_rsc_trace(primary, "Allowing %s: %s %d", colocation->id, pe__node_name(chosen), chosen->weight); primary_nodes = g_list_prepend(primary_nodes, chosen); } } node_list_exclude(dependent->allowed_nodes, primary_nodes, FALSE); g_list_free(primary_nodes); return; } // Apply optional colocations for (iter = primary->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; instance->cmds->apply_coloc_score(dependent, instance, colocation, false); } } // Clone implementation of resource_alloc_functions_t:with_this_colocations() void pcmk__with_clone_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_CHECK((rsc != NULL) && (orig_rsc != NULL) && (list != NULL), return); if (rsc == orig_rsc) { // Colocations are wanted for clone itself pcmk__add_with_this_list(list, rsc->rsc_cons_lhs); } else { pcmk__add_collective_constraints(list, orig_rsc, rsc, true); } } // Clone implementation of resource_alloc_functions_t:this_with_colocations() void pcmk__clone_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_CHECK((rsc != NULL) && (orig_rsc != NULL) && (list != NULL), return); if (rsc == orig_rsc) { // Colocations are wanted for clone itself pcmk__add_this_with_list(list, rsc->rsc_cons); } else { pcmk__add_collective_constraints(list, orig_rsc, rsc, false); } } /*! * \internal * \brief Return action flags for a given clone resource action * * \param[in,out] action Action to get flags for * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node */ enum pe_action_flags pcmk__clone_action_flags(pe_action_t *action, const pe_node_t *node) { CRM_ASSERT((action != NULL) && pe_rsc_is_clone(action->rsc)); return pcmk__collective_action_flags(action, action->rsc->children, node); } /*! * \internal * \brief Apply a location constraint to a clone resource's allowed node scores * * \param[in,out] rsc Clone resource to apply constraint to * \param[in,out] location Location constraint to apply */ void pcmk__clone_apply_location(pe_resource_t *rsc, pe__location_t *location) { CRM_CHECK((location != NULL) && pe_rsc_is_clone(rsc), return); pcmk__apply_location(rsc, location); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; instance->cmds->apply_location(instance, location); } } /*! * \internal * \brief Add a clone resource's actions to the transition graph * * \param[in,out] rsc Resource whose actions should be added */ void pcmk__clone_add_actions_to_graph(pe_resource_t *rsc) { CRM_CHECK(pe_rsc_is_clone(rsc), return); g_list_foreach(rsc->actions, (GFunc) rsc->cmds->action_flags, NULL); pe__create_clone_notifications(rsc); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *child_rsc = (pe_resource_t *) iter->data; child_rsc->cmds->add_actions_to_graph(child_rsc); } pcmk__add_rsc_actions_to_graph(rsc); pe__free_clone_notification_data(rsc); } /*! * \internal * \brief Check whether a resource or any children have been probed on a node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p node is in the known_on table of \p rsc or any of its * children, otherwise false */ static bool rsc_probed_on(const pe_resource_t *rsc, const pe_node_t *node) { if (rsc->children != NULL) { for (GList *child_iter = rsc->children; child_iter != NULL; child_iter = child_iter->next) { pe_resource_t *child = (pe_resource_t *) child_iter->data; if (rsc_probed_on(child, node)) { return true; } } return false; } if (rsc->known_on != NULL) { GHashTableIter iter; pe_node_t *known_node = NULL; g_hash_table_iter_init(&iter, rsc->known_on); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &known_node)) { if (pe__same_node(node, known_node)) { return true; } } } return false; } /*! * \internal * \brief Find clone instance that has been probed on given node * * \param[in] clone Clone resource to check * \param[in] node Node to check * * \return Instance of \p clone that has been probed on \p node if any, * otherwise NULL */ static pe_resource_t * find_probed_instance_on(const pe_resource_t *clone, const pe_node_t *node) { for (GList *iter = clone->children; iter != NULL; iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; if (rsc_probed_on(instance, node)) { return instance; } } return NULL; } /*! * \internal * \brief Probe an anonymous clone on a node * * \param[in,out] clone Anonymous clone to probe * \param[in,out] node Node to probe \p clone on */ static bool probe_anonymous_clone(pe_resource_t *clone, pe_node_t *node) { // Check whether we already probed an instance on this node pe_resource_t *child = find_probed_instance_on(clone, node); // Otherwise, check if we plan to start an instance on this node for (GList *iter = clone->children; (iter != NULL) && (child == NULL); iter = iter->next) { pe_resource_t *instance = (pe_resource_t *) iter->data; const pe_node_t *instance_node = NULL; instance_node = instance->fns->location(instance, NULL, 0); if (pe__same_node(instance_node, node)) { child = instance; } } // Otherwise, use the first clone instance if (child == NULL) { child = clone->children->data; } // Anonymous clones only need to probe a single instance return child->cmds->create_probe(child, node); } /*! * \internal * \brief Schedule any probes needed for a resource on a node * * \param[in,out] rsc Resource to create probe for * \param[in,out] node Node to create probe on * * \return true if any probe was created, otherwise false */ bool pcmk__clone_create_probe(pe_resource_t *rsc, pe_node_t *node) { CRM_CHECK((node != NULL) && pe_rsc_is_clone(rsc), return false); if (rsc->exclusive_discover) { /* The clone is configured to be probed only where a location constraint * exists with resource-discovery set to exclusive. * * This check is not strictly necessary here since the instance's * create_probe() method would also check, but doing it here is more * efficient (especially for unique clones with a large number of * instances), and affects the CRM_meta_notify_available_uname variable * passed with notify actions. */ pe_node_t *allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id); if ((allowed == NULL) || (allowed->rsc_discover_mode != pe_discover_exclusive)) { /* This node is not marked for resource discovery. Remove it from * allowed_nodes so that notifications contain only nodes that the * clone can possibly run on. */ pe_rsc_trace(rsc, "Skipping probe for %s on %s because resource has " "exclusive discovery but is not allowed on node", rsc->id, pe__node_name(node)); g_hash_table_remove(rsc->allowed_nodes, node->details->id); return false; } } rsc->children = g_list_sort(rsc->children, pcmk__cmp_instance_number); if (pcmk_is_set(rsc->flags, pe_rsc_unique)) { return pcmk__probe_resource_list(rsc->children, node); } else { return probe_anonymous_clone(rsc, node); } } /*! * \internal * \brief Add meta-attributes relevant to transition graph actions to XML * * Add clone-specific meta-attributes needed for transition graph actions. * * \param[in] rsc Clone resource whose meta-attributes should be added * \param[in,out] xml Transition graph action attributes XML to add to */ void pcmk__clone_add_graph_meta(const pe_resource_t *rsc, xmlNode *xml) { char *name = NULL; CRM_ASSERT(pe_rsc_is_clone(rsc) && (xml != NULL)); name = crm_meta_name(XML_RSC_ATTR_UNIQUE); crm_xml_add(xml, name, pe__rsc_bool_str(rsc, pe_rsc_unique)); free(name); name = crm_meta_name(XML_RSC_ATTR_NOTIFY); crm_xml_add(xml, name, pe__rsc_bool_str(rsc, pe_rsc_notify)); free(name); name = crm_meta_name(XML_RSC_ATTR_INCARNATION_MAX); crm_xml_add_int(xml, name, pe__clone_max(rsc)); free(name); name = crm_meta_name(XML_RSC_ATTR_INCARNATION_NODEMAX); crm_xml_add_int(xml, name, pe__clone_node_max(rsc)); free(name); if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) { int promoted_max = pe__clone_promoted_max(rsc); int promoted_node_max = pe__clone_promoted_node_max(rsc); name = crm_meta_name(XML_RSC_ATTR_PROMOTED_MAX); crm_xml_add_int(xml, name, promoted_max); free(name); name = crm_meta_name(XML_RSC_ATTR_PROMOTED_NODEMAX); crm_xml_add_int(xml, name, promoted_node_max); free(name); /* @COMPAT Maintain backward compatibility with resource agents that * expect the old names (deprecated since 2.0.0). */ name = crm_meta_name(PCMK_XA_PROMOTED_MAX_LEGACY); crm_xml_add_int(xml, name, promoted_max); free(name); name = crm_meta_name(PCMK_XA_PROMOTED_NODE_MAX_LEGACY); crm_xml_add_int(xml, name, promoted_node_max); free(name); } } // Clone implementation of resource_alloc_functions_t:add_utilization() void pcmk__clone_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { bool existing = false; pe_resource_t *child = NULL; if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return; } // Look for any child already existing in the list for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { child = (pe_resource_t *) iter->data; if (g_list_find(all_rscs, child)) { existing = true; // Keep checking remaining children } else { // If this is a clone of a group, look for group's members for (GList *member_iter = child->children; member_iter != NULL; member_iter = member_iter->next) { pe_resource_t *member = (pe_resource_t *) member_iter->data; if (g_list_find(all_rscs, member) != NULL) { // Add *child's* utilization, not group member's child->cmds->add_utilization(child, orig_rsc, all_rscs, utilization); existing = true; break; } } } } if (!existing && (rsc->children != NULL)) { // If nothing was found, still add first child's utilization child = (pe_resource_t *) rsc->children->data; child->cmds->add_utilization(child, orig_rsc, all_rscs, utilization); } } // Clone implementation of resource_alloc_functions_t:shutdown_lock() void pcmk__clone_shutdown_lock(pe_resource_t *rsc) { return; // Clones currently don't support shutdown locks } diff --git a/lib/pacemaker/pcmk_sched_group.c b/lib/pacemaker/pcmk_sched_group.c index cb139f7ddf..05d09d0c71 100644 --- a/lib/pacemaker/pcmk_sched_group.c +++ b/lib/pacemaker/pcmk_sched_group.c @@ -1,865 +1,864 @@ /* * Copyright 2004-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Assign a group resource to a node * * \param[in,out] rsc Group resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * * \return Node that \p rsc is assigned to, if assigned entirely to one node */ pe_node_t * pcmk__group_assign(pe_resource_t *rsc, const pe_node_t *prefer) { pe_node_t *first_assigned_node = NULL; pe_resource_t *first_member = NULL; CRM_ASSERT(rsc != NULL); if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return rsc->allocated_to; // Assignment already done } if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { pe_rsc_debug(rsc, "Assignment dependency loop detected involving %s", rsc->id); return NULL; } if (rsc->children == NULL) { // No members to assign pe__clear_resource_flags(rsc, pe_rsc_provisional); return NULL; } pe__set_resource_flags(rsc, pe_rsc_allocating); first_member = (pe_resource_t *) rsc->children->data; rsc->role = first_member->role; pe__show_node_weights(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; pe_node_t *node = NULL; pe_rsc_trace(rsc, "Assigning group %s member %s", rsc->id, member->id); node = member->cmds->assign(member, prefer); if (first_assigned_node == NULL) { first_assigned_node = node; } } pe__set_next_role(rsc, first_member->next_role, "first group member"); pe__clear_resource_flags(rsc, pe_rsc_allocating|pe_rsc_provisional); if (!pe__group_flag_is_set(rsc, pe__group_colocated)) { return NULL; } return first_assigned_node; } /*! * \internal * \brief Create a pseudo-operation for a group as an ordering point * * \param[in,out] group Group resource to create action for * \param[in] action Action name * * \return Newly created pseudo-operation */ static pe_action_t * create_group_pseudo_op(pe_resource_t *group, const char *action) { pe_action_t *op = custom_action(group, pcmk__op_key(group->id, action, 0), action, NULL, TRUE, TRUE, group->cluster); pe__set_action_flags(op, pe_action_pseudo|pe_action_runnable); return op; } /*! * \internal * \brief Create all actions needed for a given group resource * * \param[in,out] rsc Group resource to create actions for */ void pcmk__group_create_actions(pe_resource_t *rsc) { CRM_ASSERT(rsc != NULL); pe_rsc_trace(rsc, "Creating actions for group %s", rsc->id); // Create actions for individual group members for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; member->cmds->create_actions(member); } // Create pseudo-actions for group itself to serve as ordering points create_group_pseudo_op(rsc, RSC_START); create_group_pseudo_op(rsc, RSC_STARTED); create_group_pseudo_op(rsc, RSC_STOP); create_group_pseudo_op(rsc, RSC_STOPPED); if (crm_is_true(g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_PROMOTABLE))) { create_group_pseudo_op(rsc, RSC_DEMOTE); create_group_pseudo_op(rsc, RSC_DEMOTED); create_group_pseudo_op(rsc, RSC_PROMOTE); create_group_pseudo_op(rsc, RSC_PROMOTED); } } // User data for member_internal_constraints() struct member_data { // These could be derived from member but this avoids some function calls bool ordered; bool colocated; bool promotable; pe_resource_t *last_active; pe_resource_t *previous_member; }; /*! * \internal * \brief Create implicit constraints needed for a group member * * \param[in,out] data Group member to create implicit constraints for * \param[in,out] user_data Member data (struct member_data *) */ static void member_internal_constraints(gpointer data, gpointer user_data) { pe_resource_t *member = (pe_resource_t *) data; struct member_data *member_data = (struct member_data *) user_data; // For ordering demote vs demote or stop vs stop uint32_t down_flags = pe_order_implies_first_printed; // For ordering demote vs demoted or stop vs stopped uint32_t post_down_flags = pe_order_implies_then_printed; // Create the individual member's implicit constraints member->cmds->internal_constraints(member); if (member_data->previous_member == NULL) { // This is first member if (member_data->ordered) { pe__set_order_flags(down_flags, pe_order_optional); post_down_flags = pe_order_implies_then; } } else if (member_data->colocated) { // Colocate this member with the previous one pcmk__new_colocation("group:internal_colocation", NULL, INFINITY, member, member_data->previous_member, NULL, NULL, pcmk_is_set(member->flags, pe_rsc_critical), member->cluster); } if (member_data->promotable) { // Demote group -> demote member -> group is demoted pcmk__order_resource_actions(member->parent, RSC_DEMOTE, member, RSC_DEMOTE, down_flags); pcmk__order_resource_actions(member, RSC_DEMOTE, member->parent, RSC_DEMOTED, post_down_flags); // Promote group -> promote member -> group is promoted pcmk__order_resource_actions(member, RSC_PROMOTE, member->parent, RSC_PROMOTED, pe_order_runnable_left |pe_order_implies_then |pe_order_implies_then_printed); pcmk__order_resource_actions(member->parent, RSC_PROMOTE, member, RSC_PROMOTE, pe_order_implies_first_printed); } // Stop group -> stop member -> group is stopped pcmk__order_stops(member->parent, member, down_flags); pcmk__order_resource_actions(member, RSC_STOP, member->parent, RSC_STOPPED, post_down_flags); // Start group -> start member -> group is started pcmk__order_starts(member->parent, member, pe_order_implies_first_printed); pcmk__order_resource_actions(member, RSC_START, member->parent, RSC_STARTED, pe_order_runnable_left |pe_order_implies_then |pe_order_implies_then_printed); if (!member_data->ordered) { pcmk__order_starts(member->parent, member, pe_order_implies_then |pe_order_runnable_left |pe_order_implies_first_printed); if (member_data->promotable) { pcmk__order_resource_actions(member->parent, RSC_PROMOTE, member, RSC_PROMOTE, pe_order_implies_then |pe_order_runnable_left |pe_order_implies_first_printed); } } else if (member_data->previous_member == NULL) { pcmk__order_starts(member->parent, member, pe_order_none); if (member_data->promotable) { pcmk__order_resource_actions(member->parent, RSC_PROMOTE, member, RSC_PROMOTE, pe_order_none); } } else { // Order this member relative to the previous one pcmk__order_starts(member_data->previous_member, member, pe_order_implies_then|pe_order_runnable_left); pcmk__order_stops(member, member_data->previous_member, pe_order_optional|pe_order_restart); /* In unusual circumstances (such as adding a new member to the middle * of a group with unmanaged later members), this member may be active * while the previous (new) member is inactive. In this situation, the * usual restart orderings will be irrelevant, so we need to order this * member's stop before the previous member's start. */ if ((member->running_on != NULL) && (member_data->previous_member->running_on == NULL)) { pcmk__order_resource_actions(member, RSC_STOP, member_data->previous_member, RSC_START, pe_order_implies_first |pe_order_runnable_left); } if (member_data->promotable) { pcmk__order_resource_actions(member_data->previous_member, RSC_PROMOTE, member, RSC_PROMOTE, pe_order_implies_then |pe_order_runnable_left); pcmk__order_resource_actions(member, RSC_DEMOTE, member_data->previous_member, RSC_DEMOTE, pe_order_optional); } } // Make sure partially active groups shut down in sequence if (member->running_on != NULL) { if (member_data->ordered && (member_data->previous_member != NULL) && (member_data->previous_member->running_on == NULL) && (member_data->last_active != NULL) && (member_data->last_active->running_on != NULL)) { pcmk__order_stops(member, member_data->last_active, pe_order_optional); } member_data->last_active = member; } member_data->previous_member = member; } /*! * \internal * \brief Create implicit constraints needed for a group resource * * \param[in,out] rsc Group resource to create implicit constraints for */ void pcmk__group_internal_constraints(pe_resource_t *rsc) { struct member_data member_data = { false, }; CRM_ASSERT(rsc != NULL); /* 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, RSC_STOP, rsc, RSC_STOPPED, pe_order_runnable_left); pcmk__order_resource_actions(rsc, RSC_STOPPED, rsc, RSC_START, pe_order_optional); pcmk__order_resource_actions(rsc, RSC_START, rsc, RSC_STARTED, pe_order_runnable_left); member_data.ordered = pe__group_flag_is_set(rsc, pe__group_ordered); member_data.colocated = pe__group_flag_is_set(rsc, pe__group_colocated); member_data.promotable = pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pe_rsc_promotable); g_list_foreach(rsc->children, member_internal_constraints, &member_data); } /*! * \internal * \brief Apply a colocation's score to node weights or resource priority * * Given a colocation constraint for a group with some other resource, apply the * score to the dependent's allowed node weights (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,out] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply */ static void -colocate_group_with(pe_resource_t *dependent, const pe_resource_t *primary, +colocate_group_with(pe_resource_t *dependent, pe_resource_t *primary, const pcmk__colocation_t *colocation) { pe_resource_t *member = NULL; if (dependent->children == NULL) { return; } pe_rsc_trace(primary, "Processing %s (group %s with %s) for dependent", colocation->id, dependent->id, primary->id); if (pe__group_flag_is_set(dependent, pe__group_colocated)) { // Colocate first member (internal colocations will handle the rest) member = (pe_resource_t *) dependent->children->data; member->cmds->apply_coloc_score(member, primary, colocation, true); return; } if (colocation->score >= INFINITY) { pcmk__config_err("%s: Cannot perform mandatory colocation between " "non-colocated group and %s", dependent->id, primary->id); return; } // Colocate each member individually for (GList *iter = dependent->children; iter != NULL; iter = iter->next) { member = (pe_resource_t *) iter->data; member->cmds->apply_coloc_score(member, primary, colocation, true); } } /*! * \internal * \brief Apply a colocation's score to node weights or resource priority * * Given a colocation constraint for some other resource with a group, apply the * score to the dependent's allowed node weights (if we are still placing * resources) or priority (if we are choosing promotable clone instance roles). * * \param[in,out] dependent Dependent resource in colocation * \param[in] primary Primary group resource in colocation * \param[in] colocation Colocation constraint to apply */ static void colocate_with_group(pe_resource_t *dependent, const pe_resource_t *primary, const pcmk__colocation_t *colocation) { pe_resource_t *member = NULL; pe_rsc_trace(primary, "Processing colocation %s (%s with group %s) for primary", colocation->id, dependent->id, primary->id); if (pcmk_is_set(primary->flags, pe_rsc_provisional)) { return; } if (pe__group_flag_is_set(primary, pe__group_colocated)) { if (colocation->score >= INFINITY) { /* For mandatory colocations, the entire group must be assignable * (and in the specified role if any), so apply the colocation based * on the last member. */ member = pe__last_group_member(primary); } else if (primary->children != NULL) { /* For optional colocations, whether the group is partially or fully * up doesn't matter, so apply the colocation based on the first * member. */ member = (pe_resource_t *) primary->children->data; } if (member == NULL) { return; // Nothing to colocate with } member->cmds->apply_coloc_score(dependent, member, colocation, false); return; } if (colocation->score >= INFINITY) { pcmk__config_err("%s: Cannot perform mandatory colocation with" " non-colocated group %s", dependent->id, primary->id); return; } // Colocate dependent with each member individually for (GList *iter = primary->children; iter != NULL; iter = iter->next) { member = (pe_resource_t *) iter->data; member->cmds->apply_coloc_score(dependent, member, colocation, false); } } /*! * \internal * \brief Apply a colocation's score to node weights or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node weights (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,out] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply * \param[in] for_dependent true if called on behalf of dependent */ void -pcmk__group_apply_coloc_score(pe_resource_t *dependent, - const pe_resource_t *primary, +pcmk__group_apply_coloc_score(pe_resource_t *dependent, pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { CRM_ASSERT((dependent != NULL) && (primary != NULL) && (colocation != NULL)); if (for_dependent) { colocate_group_with(dependent, primary, colocation); } else { // Method should only be called for primitive dependents CRM_ASSERT(dependent->variant == pe_native); 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 */ enum pe_action_flags pcmk__group_action_flags(pe_action_t *action, const pe_node_t *node) { // Default flags for a group action enum pe_action_flags flags = pe_action_optional |pe_action_runnable |pe_action_pseudo; CRM_ASSERT(action != NULL); // Update flags considering each member's own flags for same action for (GList *iter = action->rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; // Check whether member has the same action enum action_tasks task = get_complex_task(member, action->task); const char *task_s = task2text(task); pe_action_t *member_action = find_first_action(member->actions, NULL, task_s, node); if (member_action != NULL) { enum pe_action_flags member_flags; member_flags = member->cmds->action_flags(member_action, node); // Group action is mandatory if any member action is if (pcmk_is_set(flags, pe_action_optional) && !pcmk_is_set(member_flags, pe_action_optional)) { pe_rsc_trace(action->rsc, "%s is mandatory because %s is", action->uuid, member_action->uuid); pe__clear_raw_action_flags(flags, "group action", pe_action_optional); pe__clear_action_flags(action, pe_action_optional); } // Group action is unrunnable if any member action is if (!pcmk__str_eq(task_s, action->task, pcmk__str_none) && pcmk_is_set(flags, pe_action_runnable) && !pcmk_is_set(member_flags, pe_action_runnable)) { pe_rsc_trace(action->rsc, "%s is unrunnable because %s is", action->uuid, member_action->uuid); pe__clear_raw_action_flags(flags, "group action", pe_action_runnable); pe__clear_action_flags(action, pe_action_runnable); } /* Group (pseudo-)actions other than stop or demote are unrunnable * unless every member will do it. */ } else if ((task != stop_rsc) && (task != action_demote)) { pe_rsc_trace(action->rsc, "%s is not runnable because %s will not %s", action->uuid, member->id, task_s); pe__clear_raw_action_flags(flags, "group action", pe_action_runnable); } } return flags; } /*! * \internal * \brief Update two actions according to an ordering between them * * Given information about an ordering of two actions, update the actions' flags * (and runnable_before members if appropriate) as appropriate for the ordering. * Effects may cascade to other orderings involving the actions as well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this node * (only used when interleaving instances) * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates (may * include pe_action_optional to affect only mandatory * actions, and pe_action_runnable to affect only * runnable actions) * \param[in] type Group of enum pe_ordering flags to apply * \param[in,out] data_set Cluster working set * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t pcmk__group_update_ordered_actions(pe_action_t *first, pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pe_working_set_t *data_set) { uint32_t changed = pcmk__updated_none; CRM_ASSERT((first != NULL) && (then != NULL) && (data_set != NULL)); // Group method can be called only for group action as "then" action CRM_ASSERT(then->rsc != NULL); // Update the actions for the group itself changed |= pcmk__update_ordered_actions(first, then, node, flags, filter, type, data_set); // Update the actions for each group member for (GList *iter = then->rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; pe_action_t *member_action = find_first_action(member->actions, NULL, then->task, node); if (member_action != NULL) { changed |= member->cmds->update_ordered_actions(first, member_action, node, flags, filter, type, data_set); } } return changed; } /*! * \internal * \brief Apply a location constraint to a group's allowed node scores * * \param[in,out] rsc Group resource to apply constraint to * \param[in,out] location Location constraint to apply */ void pcmk__group_apply_location(pe_resource_t *rsc, pe__location_t *location) { GList *node_list_orig = NULL; GList *node_list_copy = NULL; bool reset_scores = true; CRM_ASSERT((rsc != NULL) && (location != NULL)); node_list_orig = location->node_list_rh; node_list_copy = pcmk__copy_node_list(node_list_orig, true); reset_scores = pe__group_flag_is_set(rsc, pe__group_colocated); // Apply the constraint for the group itself (updates node scores) pcmk__apply_location(rsc, location); // Apply the constraint for each member for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; member->cmds->apply_location(member, location); if (reset_scores) { /* The first member of colocated groups needs to use the original * node scores, but subsequent members should work on a copy, since * the first member's scores already incorporate theirs. */ reset_scores = false; location->node_list_rh = node_list_copy; } } location->node_list_rh = node_list_orig; g_list_free_full(node_list_copy, free); } // Group implementation of resource_alloc_functions_t:colocated_resources() GList * pcmk__group_colocated_resources(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *colocated_rscs) { const pe_resource_t *member = NULL; CRM_ASSERT(rsc != NULL); if (orig_rsc == NULL) { orig_rsc = rsc; } if (pe__group_flag_is_set(rsc, pe__group_colocated) || pe_rsc_is_clone(rsc->parent)) { /* This group has colocated members and/or is cloned -- either way, * add every child's colocated resources to the list. The first and last * members will include the group's own colocations. */ colocated_rscs = g_list_prepend(colocated_rscs, (gpointer) rsc); for (const GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = (const pe_resource_t *) iter->data; colocated_rscs = member->cmds->colocated_resources(member, orig_rsc, colocated_rscs); } } else if (rsc->children != NULL) { /* This group's members are not colocated, and the group is not cloned, * so just add the group's own colocations to the list. */ colocated_rscs = pcmk__colocated_resources(rsc, orig_rsc, colocated_rscs); } return colocated_rscs; } // Group implementation of resource_alloc_functions_t:with_this_colocations() void pcmk__with_group_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_CHECK((rsc != NULL) && (rsc->variant == pe_group) && (orig_rsc != NULL) && (list != NULL), return); // Ignore empty groups if (rsc->children == NULL) { return; } /* "With this" colocations are needed only for the group itself and for its * last member. Add the group's colocations plus any relevant * parent colocations if cloned. */ if ((rsc == orig_rsc) || (orig_rsc == pe__last_group_member(rsc))) { crm_trace("Adding 'with %s' colocations to list for %s", rsc->id, orig_rsc->id); pcmk__add_with_this_list(list, rsc->rsc_cons_lhs); if (rsc->parent != NULL) { // Cloned group rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list); } } } // Group implementation of resource_alloc_functions_t:this_with_colocations() void pcmk__group_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { CRM_CHECK((rsc != NULL) && (rsc->variant == pe_group) && (orig_rsc != NULL) && (list != NULL), return); // Ignore empty groups if (rsc->children == NULL) { return; } /* Colocations for the group itself, or for its first member, consist of the * group's colocations plus any relevant parent colocations if cloned. */ if ((rsc == orig_rsc) || (orig_rsc == (const pe_resource_t *) rsc->children->data)) { crm_trace("Adding '%s with' colocations to list for %s", rsc->id, orig_rsc->id); pcmk__add_this_with_list(list, rsc->rsc_cons); if (rsc->parent != NULL) { // Cloned group rsc->parent->cmds->this_with_colocations(rsc->parent, 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 (GList *iter = rsc->children; iter != NULL; iter = iter->next) { const pe_resource_t *member = (const pe_resource_t *) iter->data; if (orig_rsc == member) { break; // We've seen all earlier members, and none are unmanaged } if (!pcmk_is_set(member->flags, pe_rsc_managed)) { crm_trace("Adding mandatory '%s with' colocations to list for " "member %s because earlier member %s is unmanaged", rsc->id, orig_rsc->id, member->id); for (const GList *cons_iter = rsc->rsc_cons; cons_iter != NULL; cons_iter = cons_iter->next) { const pcmk__colocation_t *colocation = NULL; colocation = (const pcmk__colocation_t *) cons_iter->data; if (colocation->score == INFINITY) { pcmk__add_this_with(list, colocation); } } // @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] rsc Resource to check colocations for * \param[in] log_id Resource ID to use in logs (if NULL, use \p rsc ID) * \param[in,out] nodes Nodes to update * \param[in] attr Colocation attribute (NULL to use default) * \param[in] factor Incorporate scores multiplied by this factor * \param[in] flags Bitmask of enum pcmk__coloc_select values * * \note The caller remains responsible for freeing \p *nodes. */ void pcmk__group_add_colocated_node_scores(pe_resource_t *rsc, const char *log_id, GHashTable **nodes, const char *attr, float factor, uint32_t flags) { pe_resource_t *member = NULL; CRM_CHECK((rsc != NULL) && (nodes != NULL), return); if (log_id == NULL) { log_id = rsc->id; } // Avoid infinite recursion if (pcmk_is_set(rsc->flags, pe_rsc_merging)) { pe_rsc_info(rsc, "%s: Breaking dependency loop at %s", log_id, rsc->id); return; } pe__set_resource_flags(rsc, pe_rsc_merging); // Ignore empty groups (only possible with schema validation disabled) if (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(rsc); } else { member = rsc->children->data; } pe_rsc_trace(rsc, "%s: Merging scores from group %s using member %s " "(at %.6f)", log_id, rsc->id, member->id, factor); member->cmds->add_colocated_node_scores(member, log_id, nodes, attr, factor, flags); pe__clear_resource_flags(rsc, pe_rsc_merging); } // Group implementation of resource_alloc_functions_t:add_utilization() void pcmk__group_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { pe_resource_t *member = NULL; CRM_ASSERT((rsc != NULL) && (orig_rsc != NULL) && (utilization != NULL)); if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return; } pe_rsc_trace(orig_rsc, "%s: Adding group %s as colocated utilization", orig_rsc->id, rsc->id); if (pe__group_flag_is_set(rsc, pe__group_colocated) || pe_rsc_is_clone(rsc->parent)) { // Every group member will be on same node, so sum all members for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { member = (pe_resource_t *) iter->data; if (pcmk_is_set(member->flags, pe_rsc_provisional) && (g_list_find(all_rscs, member) == NULL)) { member->cmds->add_utilization(member, orig_rsc, all_rscs, utilization); } } } else if (rsc->children != NULL) { // Just add first member's utilization member = (pe_resource_t *) rsc->children->data; if ((member != NULL) && pcmk_is_set(member->flags, pe_rsc_provisional) && (g_list_find(all_rscs, member) == NULL)) { member->cmds->add_utilization(member, orig_rsc, all_rscs, utilization); } } } // Group implementation of resource_alloc_functions_t:shutdown_lock() void pcmk__group_shutdown_lock(pe_resource_t *rsc) { CRM_ASSERT(rsc != NULL); for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { pe_resource_t *member = (pe_resource_t *) iter->data; member->cmds->shutdown_lock(member); } } diff --git a/lib/pacemaker/pcmk_sched_instances.c b/lib/pacemaker/pcmk_sched_instances.c index c880196f70..8ca421744f 100644 --- a/lib/pacemaker/pcmk_sched_instances.c +++ b/lib/pacemaker/pcmk_sched_instances.c @@ -1,1659 +1,1659 @@ /* * Copyright 2004-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ /* 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 clone or bundle has instances for all available nodes * * \param[in] collective Clone or bundle to check * * \return true if \p collective has enough instances for all of its available * allowed nodes, otherwise false */ static bool can_run_everywhere(const pe_resource_t *collective) { GHashTableIter iter; pe_node_t *node = NULL; int available_nodes = 0; int max_instances = 0; switch (collective->variant) { case pe_clone: max_instances = pe__clone_max(collective); break; case pe_container: max_instances = pe__bundle_max(collective); break; default: return false; // Not actually possible } g_hash_table_iter_init(&iter, collective->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (pcmk__node_available(node, false, false) && (max_instances < ++available_nodes)) { return false; } } return true; } /*! * \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 pe_resource_t *instance, const pe_node_t *node, int max_per_node) { pe_node_t *allowed_node = NULL; if (pcmk_is_set(instance->flags, pe_rsc_orphan)) { pe_rsc_trace(instance, "%s cannot run on %s: orphaned", instance->id, pe__node_name(node)); return false; } if (!pcmk__node_available(node, false, false)) { pe_rsc_trace(instance, "%s cannot run on %s: node cannot run resources", instance->id, pe__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, pe__node_name(node)); return false; } if (allowed_node->weight < 0) { pe_rsc_trace(instance, "%s cannot run on %s: parent score is %s there", instance->id, pe__node_name(node), pcmk_readable_score(allowed_node->weight)); return false; } if (allowed_node->count >= max_per_node) { pe_rsc_trace(instance, "%s cannot run on %s: node already has %d instance%s", instance->id, pe__node_name(node), max_per_node, pcmk__plural_s(max_per_node)); return false; } pe_rsc_trace(instance, "%s can run on %s (%d already running)", instance->id, pe__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(pe_resource_t *instance, int max_per_node) { if (instance->allowed_nodes != NULL) { GHashTableIter iter; pe_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)) { pe_rsc_trace(instance, "Banning %s from unavailable node %s", instance->id, pe__node_name(node)); node->weight = -INFINITY; for (GList *child_iter = instance->children; child_iter != NULL; child_iter = child_iter->next) { pe_resource_t *child = (pe_resource_t *) child_iter->data; pe_node_t *child_node = NULL; child_node = pe_hash_table_lookup(child->allowed_nodes, node->details->id); if (child_node != NULL) { pe_rsc_trace(instance, "Banning %s child %s " "from unavailable node %s", instance->id, child->id, pe__node_name(node)); child_node->weight = -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(pe_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 pe_resource_t *rsc, GHashTable **nodes) { GList *iter = NULL; pcmk__colocation_t *colocation = NULL; pe_resource_t *other = NULL; float factor = 0.0; /* Because the this_with_colocations() and with_this_colocations() methods * boil down to copies of rsc_cons and rsc_cons_lhs for clones and bundles, * we can use those here directly for efficiency. */ for (iter = rsc->parent->rsc_cons; iter != NULL; iter = iter->next) { colocation = (pcmk__colocation_t *) iter->data; other = colocation->primary; factor = colocation->score / (float) INFINITY, other->cmds->add_colocated_node_scores(other, rsc->id, nodes, colocation->node_attribute, factor, pcmk__coloc_select_default); } for (iter = rsc->parent->rsc_cons_lhs; iter != NULL; iter = iter->next) { colocation = (pcmk__colocation_t *) iter->data; if (!pcmk__colocation_has_influence(colocation, rsc)) { continue; } other = colocation->dependent; factor = colocation->score / (float) INFINITY, other->cmds->add_colocated_node_scores(other, rsc->id, nodes, colocation->node_attribute, factor, pcmk__coloc_select_nonnegative); } } /*! * \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 pe_resource_t *instance1, const pe_resource_t *instance2) { int rc = 0; pe_node_t *node1 = NULL; pe_node_t *node2 = NULL; pe_node_t *current_node1 = pe__current_node(instance1); pe_node_t *current_node2 = pe__current_node(instance2); GHashTable *colocated_scores1 = NULL; GHashTable *colocated_scores2 = NULL; CRM_ASSERT((instance1 != NULL) && (instance1->parent != NULL) && (instance2 != NULL) && (instance2->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, pe__node_name(node1), instance2->id, node2->weight, pe__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, pe__node_name(node1), instance2->id, node2->weight, pe__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 pe_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { return true; } for (GList *iter = rsc->children; iter != NULL; iter = iter->next) { if (did_fail((const pe_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 pe_resource_t *rsc, pe_node_t **node) { if (*node != NULL) { pe_node_t *allowed = pe_hash_table_lookup(rsc->allowed_nodes, (*node)->details->id); if ((allowed == NULL) || (allowed->weight < 0)) { pe_rsc_trace(rsc, "%s: current location (%s) is unavailable", rsc->id, pe__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 pe_resource_t *instance1 = (const pe_resource_t *) a; const pe_resource_t *instance2 = (const pe_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; pe_node_t *node1 = NULL; pe_node_t *node2 = NULL; unsigned int nnodes1 = 0; unsigned int nnodes2 = 0; bool can1 = true; bool can2 = true; const pe_resource_t *instance1 = (const pe_resource_t *) a; const pe_resource_t *instance2 = (const pe_resource_t *) b; CRM_ASSERT((instance1 != NULL) && (instance2 != NULL)); node1 = instance1->fns->active_node(instance1, &nnodes1, NULL); node2 = instance2->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->priority > instance2->priority) { crm_trace("Assign %s before %s: priority (%d > %d)", instance1->id, instance2->id, instance1->priority, instance2->priority); return -1; } else if (instance1->priority < instance2->priority) { crm_trace("Assign %s after %s: priority (%d < %d)", instance1->id, instance2->id, instance1->priority, instance2->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 Choose a node for an instance * * \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 true if \p instance could be assigned to a node, otherwise false */ static bool assign_instance(pe_resource_t *instance, const pe_node_t *prefer, int max_per_node) { pe_node_t *chosen = NULL; pe_node_t *allowed = NULL; CRM_ASSERT(instance != NULL); pe_rsc_trace(instance, "Assigning %s (preferring %s)", instance->id, ((prefer == NULL)? "no node" : prefer->details->uname)); if (!pcmk_is_set(instance->flags, pe_rsc_provisional)) { // Instance is already assigned return instance->fns->location(instance, NULL, FALSE) != NULL; } if (pcmk_is_set(instance->flags, pe_rsc_allocating)) { pe_rsc_debug(instance, "Assignment loop detected involving %s colocations", instance->id); return false; } if (prefer != NULL) { // Possible early assignment to preferred node // Get preferred node with instance's scores allowed = g_hash_table_lookup(instance->allowed_nodes, prefer->details->id); if ((allowed == NULL) || (allowed->weight < 0)) { pe_rsc_trace(instance, "Not assigning %s to preferred node %s: unavailable", instance->id, pe__node_name(prefer)); return false; } } ban_unavailable_allowed_nodes(instance, max_per_node); if (prefer == NULL) { // Final assignment chosen = instance->cmds->assign(instance, NULL); } else { // Possible early assignment to preferred node GHashTable *backup = pcmk__copy_node_table(instance->allowed_nodes); chosen = instance->cmds->assign(instance, prefer); // Revert nodes if preferred node won't be assigned if ((chosen != NULL) && (chosen->details != prefer->details)) { crm_info("Not assigning %s to preferred node %s: %s is better", instance->id, pe__node_name(prefer), pe__node_name(chosen)); g_hash_table_destroy(instance->allowed_nodes); instance->allowed_nodes = backup; pcmk__unassign_resource(instance); chosen = NULL; } else if (backup != NULL) { g_hash_table_destroy(backup); } } // The parent tracks how many instances have been assigned to each node if (chosen != NULL) { allowed = pcmk__top_allowed_node(instance, chosen); 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, pe_rsc_managed)); } else { allowed->count++; } } return chosen != NULL; } /*! * \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(pe_resource_t *rsc) { unsigned int available_nodes = 0; pe_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] rsc Clone or bundle being assigned (for logs only) * \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 pe_node_t * preferred_node(const pe_resource_t *rsc, const pe_resource_t *instance, int optimal_per_node) { const pe_node_t *node = NULL; const pe_node_t *parent_node = NULL; // Check whether instance is active, healthy, and not yet assigned if ((instance->running_on == NULL) || !pcmk_is_set(instance->flags, pe_rsc_provisional) || pcmk_is_set(instance->flags, pe_rsc_failed)) { return NULL; } // Check whether instance's current node can run resources node = pe__current_node(instance); if (!pcmk__node_available(node, true, false)) { pe_rsc_trace(rsc, "Not assigning %s to %s early (unavailable)", instance->id, pe__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)) { pe_rsc_trace(rsc, "Not assigning %s to %s early " "(optimal instances already assigned)", instance->id, pe__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(pe_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; pe_resource_t *instance = NULL; const pe_node_t *current = NULL; if (available_nodes > 0) { optimal_per_node = max_total / available_nodes; } if (optimal_per_node < 1) { optimal_per_node = 1; } pe_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) { instance = (pe_resource_t *) iter->data; current = preferred_node(collective, instance, optimal_per_node); if ((current != NULL) && assign_instance(instance, current, max_per_node)) { pe_rsc_trace(collective, "Assigned %s to current node %s", instance->id, pe__node_name(current)); assigned++; } } pe_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 = (pe_resource_t *) iter->data; if (!pcmk_is_set(instance->flags, pe_rsc_provisional)) { continue; // Already assigned } if (instance->running_on != NULL) { current = pe__current_node(instance); if (pcmk__top_allowed_node(instance, current) == NULL) { const char *unmanaged = ""; if (!pcmk_is_set(instance->flags, pe_rsc_managed)) { unmanaged = "Unmanaged resource "; } crm_notice("%s%s is running on %s which is no longer allowed", unmanaged, instance->id, pe__node_name(current)); } } if (assigned >= max_total) { pe_rsc_debug(collective, "Not assigning %s because maximum %d instances " "already assigned", instance->id, max_total); resource_location(instance, NULL, -INFINITY, "collective_limit_reached", collective->cluster); } else if (assign_instance(instance, NULL, max_per_node)) { assigned++; } } pe_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 pe_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->variant > pe_native) { for (iter = instance->children; (iter != NULL) && !pcmk_all_flags_set(*state, instance_all); iter = iter->next) { check_instance_state((const pe_resource_t *) iter->data, state); } return; } // If we get here, instance is a primitive if (instance->running_on != NULL) { instance_state |= instance_active; } // Check each of the instance's actions for runnable start or stop for (iter = instance->actions; (iter != NULL) && !pcmk_all_flags_set(instance_state, instance_starting |instance_stopping); iter = iter->next) { const pe_action_t *action = (const pe_action_t *) iter->data; const bool optional = pcmk_is_set(action->flags, pe_action_optional); if (pcmk__str_eq(RSC_START, action->task, pcmk__str_none)) { if (!optional && pcmk_is_set(action->flags, pe_action_runnable)) { pe_rsc_trace(instance, "Instance is starting due to %s", action->uuid); instance_state |= instance_starting; } else { pe_rsc_trace(instance, "%s doesn't affect %s state (%s)", action->uuid, instance->id, (optional? "optional" : "unrunnable")); } } else if (pcmk__str_eq(RSC_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, pe_action_pseudo|pe_action_runnable)) { pe_rsc_trace(instance, "Instance is stopping due to %s", action->uuid); instance_state |= instance_stopping; } else { pe_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(pe_resource_t *collective, GList *instances) { uint32_t state = 0; pe_action_t *stop = NULL; pe_action_t *stopped = NULL; pe_action_t *start = NULL; pe_action_t *started = NULL; pe_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) { pe_resource_t *instance = (pe_resource_t *) iter->data; instance->cmds->create_actions(instance); check_instance_state(instance, &state); } // Create pseudo-actions for rsc start and started start = pe__new_rsc_pseudo_action(collective, RSC_START, !pcmk_is_set(state, instance_starting), true); started = pe__new_rsc_pseudo_action(collective, RSC_STARTED, !pcmk_is_set(state, instance_starting), false); started->priority = INFINITY; if (pcmk_any_flags_set(state, instance_active|instance_starting)) { pe__set_action_flags(started, pe_action_runnable); } // Create pseudo-actions for rsc stop and stopped stop = pe__new_rsc_pseudo_action(collective, RSC_STOP, !pcmk_is_set(state, instance_stopping), true); stopped = pe__new_rsc_pseudo_action(collective, RSC_STOPPED, !pcmk_is_set(state, instance_stopping), true); stopped->priority = INFINITY; if (!pcmk_is_set(state, instance_restarting)) { pe__set_action_flags(stop, pe_action_migrate_runnable); } if (collective->variant == pe_clone) { 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 pe_resource_t *rsc) { if (rsc->variant == pe_container) { 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 pe_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 RSC_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 pe_resource_t *instance, const pe_node_t *node, enum rsc_role_e role, bool current) { pe_node_t *instance_node = NULL; CRM_CHECK((instance != NULL) && (node != NULL), return false); if ((role != RSC_ROLE_UNKNOWN) && (role != instance->fns->state(instance, current))) { pe_rsc_trace(instance, "%s is not a compatible instance (role is not %s)", instance->id, role2text(role)); return false; } if (!is_set_recursive(instance, pe_rsc_block, true)) { // We only want instances that haven't failed instance_node = instance->fns->location(instance, NULL, current); } if (instance_node == NULL) { pe_rsc_trace(instance, "%s is not a compatible instance (not assigned to a node)", instance->id); return false; } if (instance_node->details != node->details) { pe_rsc_trace(instance, "%s is not a compatible instance (assigned to %s not %s)", instance->id, pe__node_name(instance_node), pe__node_name(node)); return false; } return true; } /*! * \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 RSC_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 pe_resource_t * find_compatible_instance_on_node(const pe_resource_t *match_rsc, const pe_resource_t *rsc, const pe_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) { pe_resource_t *instance = (pe_resource_t *) iter->data; if (pcmk__instance_matches(instance, node, role, current)) { pe_rsc_trace(match_rsc, "Found %s %s instance %s compatible with %s on %s", role == RSC_ROLE_UNKNOWN? "matching" : role2text(role), rsc->id, instance->id, match_rsc->id, pe__node_name(node)); free_instance_list(rsc, instances); // Only frees list, not contents return instance; } } free_instance_list(rsc, instances); pe_rsc_trace(match_rsc, "No %s %s instance found compatible with %s on %s", ((role == RSC_ROLE_UNKNOWN)? "matching" : role2text(role)), rsc->id, match_rsc->id, pe__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 RSC_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 */ pe_resource_t * pcmk__find_compatible_instance(const pe_resource_t *match_rsc, const pe_resource_t *rsc, enum rsc_role_e role, bool current) { pe_resource_t *instance = NULL; GList *nodes = NULL; const pe_node_t *node = match_rsc->fns->location(match_rsc, NULL, current); // If match_rsc has a node, check only that node 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, (pe_node_t *) iter->data, role, current); } if (instance == NULL) { pe_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 pe_ordering flags to apply * \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 pe_action_t *first, const pe_action_t *then, pe_resource_t *then_instance, uint32_t type, bool current) { // Allow "then" instance to go down even without an interleave match if (current) { pe_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, pe_order_runnable_left |pe_order_implies_then)) { pe_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); } 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 pe_action_t * find_instance_action(const pe_action_t *action, const pe_resource_t *instance, const char *action_name, const pe_node_t *node, bool for_first) { const pe_resource_t *rsc = NULL; pe_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, CRMD_ACTION_STOP, CRMD_ACTION_STOPPED, NULL)) || (!for_first && pcmk__str_any_of(action->task, CRMD_ACTION_PROMOTE, CRMD_ACTION_PROMOTED, CRMD_ACTION_DEMOTE, CRMD_ACTION_DEMOTED, NULL))) { - rsc = pcmk__get_rsc_in_container(instance); + 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->actions, NULL, action_name, node); if (matching_action != NULL) { return matching_action; } if (pcmk_is_set(instance->flags, pe_rsc_orphan) || pcmk__str_any_of(action_name, RSC_STOP, RSC_DEMOTE, NULL)) { crm_trace("No %s action found for %s%s", action_name, pcmk_is_set(instance->flags, pe_rsc_orphan)? "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 pe_action_t *action) { const pe_resource_t *instance = action->rsc->children->data; // Any instance char *action_type = NULL; const char *action_name = action->task; enum action_tasks orig_task = no_action; if (pcmk__strcase_any_of(action->task, CRMD_ACTION_NOTIFY, CRMD_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 task2text(no_action)); action_name = strstr(action_type, "_notify_"); CRM_CHECK(action_name != NULL, return task2text(no_action)); action_name += strlen("_notify_"); } orig_task = get_complex_task(instance, action_name); free(action_type); return task2text(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 pe_action_optional to affect only mandatory * actions, and pe_action_runnable to affect only * runnable actions) * \param[in] type Group of enum pe_ordering flags to apply * * \return Group of enum pcmk__updated flags indicating what was updated */ static uint32_t update_interleaved_actions(pe_action_t *first, pe_action_t *then, const pe_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, "_" CRMD_ACTION_STOPPED "_0") || pcmk__ends_with(first->uuid, "_" CRMD_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) { pe_resource_t *first_instance = NULL; pe_resource_t *then_instance = iter->data; pe_action_t *first_action = NULL; pe_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, RSC_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->cmds->update_ordered_actions( first_action, then_action, node, first_instance->cmds->action_flags(first_action, node), filter, type, then->rsc->cluster); } 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 pe_action_t *first, const pe_action_t *then) { bool interleave = false; pe_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->variant < pe_clone) || (then->rsc->variant < pe_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, XML_RSC_ATTR_INTERLEAVE)); pe_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 pe_action_optional to affect only mandatory * actions, and pe_action_runnable to affect only * runnable actions) * \param[in] type Group of enum pe_ordering flags to apply * * \return Group of enum pcmk__updated flags indicating what was updated */ static uint32_t update_noninterleaved_actions(pe_resource_t *instance, pe_action_t *first, const pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type) { pe_action_t *instance_action = NULL; 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->actions, NULL, then->task, node); if (instance_action == NULL) { return changed; } // Check whether action is runnable instance_flags = instance->cmds->action_flags(instance_action, node); if (!pcmk_is_set(instance_flags, pe_action_runnable)) { return changed; } // If so, update actions for the instance changed = instance->cmds->update_ordered_actions(first, instance_action, node, flags, filter, type, instance->cluster); // 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) { pe_action_wrapper_t *after = after_iter->data; pcmk__update_action_for_orderings(after->action, instance->cluster); } } 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 pe_action_optional to affect only mandatory * actions, and pe_action_runnable to affect only * runnable actions) * \param[in] type Group of enum pe_ordering flags to apply * \param[in,out] data_set Cluster working set * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t pcmk__instance_update_ordered_actions(pe_action_t *first, pe_action_t *then, const pe_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pe_working_set_t *data_set) { 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, data_set); // Update the 'then' clone instances or bundle containers individually for (GList *iter = instances; iter != NULL; iter = iter->next) { pe_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 */ enum pe_action_flags pcmk__collective_action_flags(pe_action_t *action, const GList *instances, const pe_node_t *node) { bool any_runnable = false; enum pe_action_flags flags; const char *action_name = orig_action_name(action); // Set original assumptions (optional and runnable may be cleared below) flags = pe_action_optional|pe_action_runnable|pe_action_pseudo; for (const GList *iter = instances; iter != NULL; iter = iter->next) { const pe_resource_t *instance = iter->data; const pe_node_t *instance_node = NULL; pe_action_t *instance_action = NULL; enum pe_action_flags instance_flags; // Node is relevant only to primitive instances if (instance->variant == pe_native) { instance_node = node; } instance_action = find_first_action(instance->actions, NULL, action_name, instance_node); if (instance_action == NULL) { pe_rsc_trace(action->rsc, "%s has no %s action on %s", instance->id, action_name, pe__node_name(node)); continue; } pe_rsc_trace(action->rsc, "%s has %s for %s on %s", instance->id, instance_action->uuid, action_name, pe__node_name(node)); instance_flags = instance->cmds->action_flags(instance_action, node); // If any instance action is mandatory, so is the collective action if (pcmk_is_set(flags, pe_action_optional) && !pcmk_is_set(instance_flags, pe_action_optional)) { pe_rsc_trace(instance, "%s is mandatory because %s is", action->uuid, instance_action->uuid); pe__clear_action_summary_flags(flags, action, pe_action_optional); pe__clear_action_flags(action, pe_action_optional); } // If any instance action is runnable, so is the collective action if (pcmk_is_set(instance_flags, pe_action_runnable)) { any_runnable = true; } } if (!any_runnable) { pe_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, pe_action_runnable); if (node == NULL) { pe__clear_action_flags(action, pe_action_runnable); } } return flags; } /*! * \internal * \brief Add a collective resource's colocations to a list for an instance * * \param[in,out] list Colocation list to add to * \param[in] instance Clone or bundle instance or instance group member * \param[in] collective Clone or bundle resource with colocations to add * \param[in] with_this If true, add collective's "with this" colocations, * otherwise add its "this with" colocations */ void pcmk__add_collective_constraints(GList **list, const pe_resource_t *instance, const pe_resource_t *collective, bool with_this) { const GList *colocations = NULL; bool everywhere = false; CRM_CHECK((list != NULL) && (instance != NULL), return); if (collective == NULL) { return; } switch (collective->variant) { case pe_clone: case pe_container: break; default: return; } everywhere = can_run_everywhere(collective); if (with_this) { colocations = collective->rsc_cons_lhs; } else { colocations = collective->rsc_cons; } for (const GList *iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *colocation = iter->data; if (with_this && !pcmk__colocation_has_influence(colocation, instance)) { continue; } if (!everywhere || (colocation->score < 0) || (!with_this && (colocation->score == INFINITY))) { if (with_this) { pcmk__add_with_this(list, colocation); } else { pcmk__add_this_with(list, colocation); } } } } diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c index aefbf9aa14..024158db09 100644 --- a/lib/pacemaker/pcmk_sched_primitive.c +++ b/lib/pacemaker/pcmk_sched_primitive.c @@ -1,1573 +1,1573 @@ /* * Copyright 2004-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "libpacemaker_private.h" static void stop_resource(pe_resource_t *rsc, pe_node_t *node, bool optional); static void start_resource(pe_resource_t *rsc, pe_node_t *node, bool optional); static void demote_resource(pe_resource_t *rsc, pe_node_t *node, bool optional); static void promote_resource(pe_resource_t *rsc, pe_node_t *node, bool optional); static void assert_role_error(pe_resource_t *rsc, pe_node_t *node, bool optional); 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 */ { RSC_ROLE_UNKNOWN, /* Unknown */ RSC_ROLE_STOPPED, /* Stopped */ RSC_ROLE_STOPPED, /* Started */ RSC_ROLE_STOPPED, /* Unpromoted */ RSC_ROLE_STOPPED, /* Promoted */ }, /* Stopped */ { RSC_ROLE_STOPPED, /* Unknown */ RSC_ROLE_STOPPED, /* Stopped */ RSC_ROLE_STARTED, /* Started */ RSC_ROLE_UNPROMOTED, /* Unpromoted */ RSC_ROLE_UNPROMOTED, /* Promoted */ }, /* Started */ { RSC_ROLE_STOPPED, /* Unknown */ RSC_ROLE_STOPPED, /* Stopped */ RSC_ROLE_STARTED, /* Started */ RSC_ROLE_UNPROMOTED, /* Unpromoted */ RSC_ROLE_PROMOTED, /* Promoted */ }, /* Unpromoted */ { RSC_ROLE_STOPPED, /* Unknown */ RSC_ROLE_STOPPED, /* Stopped */ RSC_ROLE_STOPPED, /* Started */ RSC_ROLE_UNPROMOTED, /* Unpromoted */ RSC_ROLE_PROMOTED, /* Promoted */ }, /* Promoted */ { RSC_ROLE_STOPPED, /* Unknown */ RSC_ROLE_UNPROMOTED, /* Stopped */ RSC_ROLE_UNPROMOTED, /* Started */ RSC_ROLE_UNPROMOTED, /* Unpromoted */ RSC_ROLE_PROMOTED, /* Promoted */ }, }; /*! * \internal * \brief Function to schedule actions needed for a role change * * \param[in,out] rsc Resource whose role is changing * \param[in,out] node Node where resource will be in its next role * \param[in] optional Whether scheduled actions should be optional */ typedef void (*rsc_transition_fn)(pe_resource_t *rsc, pe_node_t *node, bool optional); static rsc_transition_fn rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = { /* This array lists the function needed to transition directly from one role * to another. NULL indicates that nothing is needed. * * Current role Transition function Next role * ------------ ------------------- ---------- */ /* Unknown */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ assert_role_error, /* Started */ assert_role_error, /* Unpromoted */ assert_role_error, /* Promoted */ }, /* Stopped */ { assert_role_error, /* Unknown */ NULL, /* Stopped */ start_resource, /* Started */ start_resource, /* Unpromoted */ assert_role_error, /* Promoted */ }, /* Started */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ NULL, /* Started */ NULL, /* Unpromoted */ promote_resource, /* Promoted */ }, /* Unpromoted */ { assert_role_error, /* Unknown */ stop_resource, /* Stopped */ stop_resource, /* Started */ NULL, /* Unpromoted */ promote_resource, /* Promoted */ }, /* Promoted */ { assert_role_error, /* Unknown */ demote_resource, /* Stopped */ demote_resource, /* Started */ demote_resource, /* Unpromoted */ NULL, /* Promoted */ }, }; /*! * \internal * \brief Get a list of a resource's allowed nodes sorted by node weight * * \param[in] rsc Resource to check * * \return List of allowed nodes sorted by node weight */ static GList * sorted_allowed_nodes(const pe_resource_t *rsc) { if (rsc->allowed_nodes != NULL) { GList *nodes = g_hash_table_get_values(rsc->allowed_nodes); if (nodes != NULL) { return pcmk__sort_nodes(nodes, pe__current_node(rsc)); } } return NULL; } /*! * \internal * \brief Assign a resource to its best allowed node, if possible * * \param[in,out] rsc Resource to choose a node for * \param[in] prefer If not NULL, prefer this node when all else equal * * \return true if \p rsc could be assigned to a node, otherwise false */ static bool assign_best_node(pe_resource_t *rsc, const pe_node_t *prefer) { GList *nodes = NULL; pe_node_t *chosen = NULL; pe_node_t *best = NULL; bool result = false; const pe_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc); if (prefer == NULL) { prefer = most_free_node; } if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { // We've already finished assignment of resources to nodes return rsc->allocated_to != NULL; } // Sort allowed nodes by weight nodes = sorted_allowed_nodes(rsc); if (nodes != NULL) { best = (pe_node_t *) nodes->data; // First node has best score } if ((prefer != NULL) && (nodes != NULL)) { // Get the allowed node version of prefer chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id); if (chosen == NULL) { pe_rsc_trace(rsc, "Preferred node %s for %s was unknown", pe__node_name(prefer), rsc->id); /* Favor the preferred node as long as its weight 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 weight is less than INFINITY. */ } else if (chosen->weight < best->weight) { pe_rsc_trace(rsc, "Preferred node %s for %s was unsuitable", pe__node_name(chosen), rsc->id); chosen = NULL; } else if (!pcmk__node_available(chosen, true, false)) { pe_rsc_trace(rsc, "Preferred node %s for %s was unavailable", pe__node_name(chosen), rsc->id); chosen = NULL; } else { pe_rsc_trace(rsc, "Chose preferred node %s for %s (ignoring %d candidates)", pe__node_name(chosen), rsc->id, g_list_length(nodes)); } } if ((chosen == NULL) && (best != NULL)) { /* Either there is no preferred node, or the preferred node is not * suitable, but another node is allowed to run the resource. */ chosen = best; if (!pe_rsc_is_unique_clone(rsc->parent) && (chosen->weight > 0) // Zero not acceptable && pcmk__node_available(chosen, false, false)) { /* If the resource is already running on a node, prefer that node if * it is just as good as the chosen node. * * We don't do this for unique clone instances, because * pcmk__assign_instances() has already assigned instances to their * running nodes when appropriate, and if we get here, we don't want * remaining unassigned instances to prefer a node that's already * running another instance. */ pe_node_t *running = pe__current_node(rsc); if (running == NULL) { // Nothing to do } else if (!pcmk__node_available(running, true, false)) { pe_rsc_trace(rsc, "Current node for %s (%s) can't run resources", rsc->id, pe__node_name(running)); } else { int nodes_with_best_score = 1; for (GList *iter = nodes->next; iter; iter = iter->next) { pe_node_t *allowed = (pe_node_t *) iter->data; if (allowed->weight != chosen->weight) { // The nodes are sorted by weight, so no more are equal break; } if (pe__same_node(allowed, running)) { // Scores are equal, so prefer the current node chosen = allowed; } nodes_with_best_score++; } if (nodes_with_best_score > 1) { do_crm_log(((chosen->weight >= INFINITY)? LOG_WARNING : LOG_INFO), "Chose %s for %s from %d nodes with score %s", pe__node_name(chosen), rsc->id, nodes_with_best_score, pcmk_readable_score(chosen->weight)); } } } pe_rsc_trace(rsc, "Chose %s for %s from %d candidates", pe__node_name(chosen), rsc->id, g_list_length(nodes)); } result = pcmk__finalize_assignment(rsc, chosen, false); g_list_free(nodes); return result; } /*! * \internal * \brief Apply a "this with" colocation to a node's allowed node scores * * \param[in,out] data Colocation to apply * \param[in,out] user_data Resource being assigned */ static void apply_this_with(gpointer data, gpointer user_data) { pcmk__colocation_t *colocation = (pcmk__colocation_t *) data; pe_resource_t *rsc = (pe_resource_t *) user_data; GHashTable *archive = NULL; pe_resource_t *other = colocation->primary; // In certain cases, we will need to revert the node scores if ((colocation->dependent_role >= RSC_ROLE_PROMOTED) || ((colocation->score < 0) && (colocation->score > -INFINITY))) { archive = pcmk__copy_node_table(rsc->allowed_nodes); } if (pcmk_is_set(other->flags, pe_rsc_provisional)) { pe_rsc_trace(rsc, "%s: Assigning colocation %s primary %s first" "(score=%d role=%s)", rsc->id, colocation->id, other->id, colocation->score, role2text(colocation->dependent_role)); other->cmds->assign(other, NULL); } // Apply the colocation score to this resource's allowed node scores rsc->cmds->apply_coloc_score(rsc, other, colocation, true); if ((archive != NULL) && !pcmk__any_node_available(rsc->allowed_nodes)) { pe_rsc_info(rsc, "%s: Reverting scores from colocation with %s " "because no nodes allowed", rsc->id, other->id); g_hash_table_destroy(rsc->allowed_nodes); rsc->allowed_nodes = archive; archive = NULL; } if (archive != NULL) { g_hash_table_destroy(archive); } } /*! * \internal * \brief Update a Pacemaker Remote node once its connection has been assigned * * \param[in] connection Connection resource that has been assigned */ static void remote_connection_assigned(const pe_resource_t *connection) { pe_node_t *remote_node = pe_find_node(connection->cluster->nodes, connection->id); CRM_CHECK(remote_node != NULL, return); if ((connection->allocated_to != NULL) && (connection->next_role != RSC_ROLE_STOPPED)) { crm_trace("Pacemaker Remote node %s will be online", remote_node->details->id); remote_node->details->online = TRUE; if (remote_node->details->unseen) { // Avoid unnecessary fence, since we will attempt connection remote_node->details->unclean = FALSE; } } else { crm_trace("Pacemaker Remote node %s will be shut down " "(%sassigned connection's next role is %s)", remote_node->details->id, ((connection->allocated_to == NULL)? "un" : ""), role2text(connection->next_role)); remote_node->details->shutdown = TRUE; } } /*! * \internal * \brief Assign a primitive resource to a node * * \param[in,out] rsc Resource to assign to a node * \param[in] prefer Node to prefer, if all else is equal * * \return Node that \p rsc is assigned to, if assigned entirely to one node */ pe_node_t * pcmk__primitive_assign(pe_resource_t *rsc, const pe_node_t *prefer) { GList *this_with_colocations = NULL; GList *with_this_colocations = NULL; GList *iter = NULL; pcmk__colocation_t *colocation = NULL; CRM_ASSERT(rsc != NULL); // Never assign a child without parent being assigned first if ((rsc->parent != NULL) && !pcmk_is_set(rsc->parent->flags, pe_rsc_allocating)) { pe_rsc_debug(rsc, "%s: Assigning parent %s first", rsc->id, rsc->parent->id); rsc->parent->cmds->assign(rsc->parent, prefer); } if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return rsc->allocated_to; // Assignment has already been done } // Ensure we detect assignment loops if (pcmk_is_set(rsc->flags, pe_rsc_allocating)) { pe_rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id); return NULL; } pe__set_resource_flags(rsc, pe_rsc_allocating); pe__show_node_weights(true, rsc, "Pre-assignment", rsc->allowed_nodes, rsc->cluster); this_with_colocations = pcmk__this_with_colocations(rsc); with_this_colocations = pcmk__with_this_colocations(rsc); // Apply mandatory colocations first, to satisfy as many as possible for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score <= -CRM_SCORE_INFINITY) || (colocation->score >= CRM_SCORE_INFINITY)) { apply_this_with(iter->data, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score <= -CRM_SCORE_INFINITY) || (colocation->score >= CRM_SCORE_INFINITY)) { pcmk__add_dependent_scores(iter->data, rsc); } } pe__show_node_weights(true, rsc, "Mandatory-colocations", rsc->allowed_nodes, rsc->cluster); // Then apply optional colocations for (iter = this_with_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score > -CRM_SCORE_INFINITY) && (colocation->score < CRM_SCORE_INFINITY)) { apply_this_with(iter->data, rsc); } } for (iter = with_this_colocations; iter != NULL; iter = iter->next) { colocation = iter->data; if ((colocation->score > -CRM_SCORE_INFINITY) && (colocation->score < CRM_SCORE_INFINITY)) { pcmk__add_dependent_scores(iter->data, rsc); } } g_list_free(this_with_colocations); g_list_free(with_this_colocations); if (rsc->next_role == RSC_ROLE_STOPPED) { pe_rsc_trace(rsc, "Banning %s from all nodes because it will be stopped", rsc->id); resource_location(rsc, NULL, -INFINITY, XML_RSC_ATTR_TARGET_ROLE, rsc->cluster); } else if ((rsc->next_role > rsc->role) && !pcmk_is_set(rsc->cluster->flags, pe_flag_have_quorum) && (rsc->cluster->no_quorum_policy == no_quorum_freeze)) { crm_notice("Resource %s cannot be elevated from %s to %s due to " "no-quorum-policy=freeze", rsc->id, role2text(rsc->role), role2text(rsc->next_role)); pe__set_next_role(rsc, rsc->role, "no-quorum-policy=freeze"); } pe__show_node_weights(!pcmk_is_set(rsc->cluster->flags, pe_flag_show_scores), rsc, __func__, rsc->allowed_nodes, rsc->cluster); // Unmanage resource if fencing is enabled but no device is configured if (pcmk_is_set(rsc->cluster->flags, pe_flag_stonith_enabled) && !pcmk_is_set(rsc->cluster->flags, pe_flag_have_stonith_resource)) { pe__clear_resource_flags(rsc, pe_rsc_managed); } if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { // Unmanaged resources stay on their current node const char *reason = NULL; pe_node_t *assign_to = NULL; pe__set_next_role(rsc, rsc->role, "unmanaged"); assign_to = pe__current_node(rsc); if (assign_to == NULL) { reason = "inactive"; } else if (rsc->role == RSC_ROLE_PROMOTED) { reason = "promoted"; } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { reason = "failed"; } else { reason = "active"; } pe_rsc_info(rsc, "Unmanaged resource %s assigned to %s: %s", rsc->id, (assign_to? assign_to->details->uname : "no node"), reason); pcmk__finalize_assignment(rsc, assign_to, true); } else if (pcmk_is_set(rsc->cluster->flags, pe_flag_stop_everything)) { pe_rsc_debug(rsc, "Forcing %s to stop: stop-all-resources", rsc->id); pcmk__finalize_assignment(rsc, NULL, true); } else if (pcmk_is_set(rsc->flags, pe_rsc_provisional) && assign_best_node(rsc, prefer)) { // Assignment successful } else if (rsc->allocated_to == NULL) { if (!pcmk_is_set(rsc->flags, pe_rsc_orphan)) { pe_rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id); } else if (rsc->running_on != NULL) { pe_rsc_info(rsc, "Stopping orphan resource %s", rsc->id); } } else { pe_rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, pe__node_name(rsc->allocated_to)); } pe__clear_resource_flags(rsc, pe_rsc_allocating); if (rsc->is_remote_node) { remote_connection_assigned(rsc); } return rsc->allocated_to; } /*! * \internal * \brief Schedule actions to bring resource down and back to current role * * \param[in,out] rsc Resource to restart * \param[in,out] current Node that resource should be brought down on * \param[in] need_stop Whether the resource must be stopped * \param[in] need_promote Whether the resource must be promoted * * \return Role that resource would have after scheduled actions are taken */ static void schedule_restart_actions(pe_resource_t *rsc, pe_node_t *current, bool need_stop, bool need_promote) { enum rsc_role_e role = rsc->role; enum rsc_role_e next_role; rsc_transition_fn fn = NULL; pe__set_resource_flags(rsc, pe_rsc_restarting); // Bring resource down to a stop on its current node while (role != RSC_ROLE_STOPPED) { next_role = rsc_state_matrix[role][RSC_ROLE_STOPPED]; pe_rsc_trace(rsc, "Creating %s action to take %s down from %s to %s", (need_stop? "required" : "optional"), rsc->id, role2text(role), role2text(next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, current, !need_stop); role = next_role; } // Bring resource up to its next role on its next node while ((rsc->role <= rsc->next_role) && (role != rsc->role) && !pcmk_is_set(rsc->flags, pe_rsc_block)) { bool required = need_stop; next_role = rsc_state_matrix[role][rsc->role]; if ((next_role == RSC_ROLE_PROMOTED) && need_promote) { required = true; } pe_rsc_trace(rsc, "Creating %s action to take %s up from %s to %s", (required? "required" : "optional"), rsc->id, role2text(role), role2text(next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, rsc->allocated_to, !required); role = next_role; } pe__clear_resource_flags(rsc, pe_rsc_restarting); } /*! * \internal * \brief If a resource's next role is not explicitly specified, set a default * * \param[in,out] rsc Resource to set next role for * * \return "explicit" if next role was explicitly set, otherwise "implicit" */ static const char * set_default_next_role(pe_resource_t *rsc) { if (rsc->next_role != RSC_ROLE_UNKNOWN) { return "explicit"; } if (rsc->allocated_to == NULL) { pe__set_next_role(rsc, RSC_ROLE_STOPPED, "assignment"); } else { pe__set_next_role(rsc, RSC_ROLE_STARTED, "assignment"); } return "implicit"; } /*! * \internal * \brief Create an action to represent an already pending start * * \param[in,out] rsc Resource to create start action for */ static void create_pending_start(pe_resource_t *rsc) { pe_action_t *start = NULL; pe_rsc_trace(rsc, "Creating action for %s to represent already pending start", rsc->id); start = start_action(rsc, rsc->allocated_to, TRUE); pe__set_action_flags(start, pe_action_print_always); } /*! * \internal * \brief Schedule actions needed to take a resource to its next role * * \param[in,out] rsc Resource to schedule actions for */ static void schedule_role_transition_actions(pe_resource_t *rsc) { enum rsc_role_e role = rsc->role; while (role != rsc->next_role) { enum rsc_role_e next_role = rsc_state_matrix[role][rsc->next_role]; rsc_transition_fn fn = NULL; pe_rsc_trace(rsc, "Creating action to take %s from %s to %s (ending at %s)", rsc->id, role2text(role), role2text(next_role), role2text(rsc->next_role)); fn = rsc_action_matrix[role][next_role]; if (fn == NULL) { break; } fn(rsc, rsc->allocated_to, false); role = next_role; } } /*! * \internal * \brief Create all actions needed for a given primitive resource * * \param[in,out] rsc Primitive resource to create actions for */ void pcmk__primitive_create_actions(pe_resource_t *rsc) { bool need_stop = false; bool need_promote = false; bool is_moving = false; bool allow_migrate = false; bool multiply_active = false; pe_node_t *current = NULL; unsigned int num_all_active = 0; unsigned int num_clean_active = 0; const char *next_role_source = NULL; CRM_ASSERT(rsc != NULL); next_role_source = set_default_next_role(rsc); pe_rsc_trace(rsc, "Creating all actions for %s transition from %s to %s " "(%s) on %s", rsc->id, role2text(rsc->role), role2text(rsc->next_role), next_role_source, pe__node_name(rsc->allocated_to)); current = rsc->fns->active_node(rsc, &num_all_active, &num_clean_active); g_list_foreach(rsc->dangling_migrations, pcmk__abort_dangling_migration, rsc); if ((current != NULL) && (rsc->allocated_to != NULL) && (current->details != rsc->allocated_to->details) && (rsc->next_role >= RSC_ROLE_STARTED)) { pe_rsc_trace(rsc, "Moving %s from %s to %s", rsc->id, pe__node_name(current), pe__node_name(rsc->allocated_to)); is_moving = true; allow_migrate = pcmk__rsc_can_migrate(rsc, current); // This is needed even if migrating (though I'm not sure why ...) need_stop = true; } // Check whether resource is partially migrated and/or multiply active if ((rsc->partial_migration_source != NULL) && (rsc->partial_migration_target != NULL) && allow_migrate && (num_all_active == 2) && pe__same_node(current, rsc->partial_migration_source) && pe__same_node(rsc->allocated_to, rsc->partial_migration_target)) { /* A partial migration is in progress, and the migration target remains * the same as when the migration began. */ pe_rsc_trace(rsc, "Partial migration of %s from %s to %s will continue", rsc->id, pe__node_name(rsc->partial_migration_source), pe__node_name(rsc->partial_migration_target)); } else if ((rsc->partial_migration_source != NULL) || (rsc->partial_migration_target != NULL)) { // A partial migration is in progress but can't be continued if (num_all_active > 2) { // The resource is migrating *and* multiply active! crm_notice("Forcing recovery of %s because it is migrating " "from %s to %s and possibly active elsewhere", rsc->id, pe__node_name(rsc->partial_migration_source), pe__node_name(rsc->partial_migration_target)); } else { // The migration source or target isn't available crm_notice("Forcing recovery of %s because it can no longer " "migrate from %s to %s", rsc->id, pe__node_name(rsc->partial_migration_source), pe__node_name(rsc->partial_migration_target)); } need_stop = true; rsc->partial_migration_source = rsc->partial_migration_target = NULL; allow_migrate = false; } else if (pcmk_is_set(rsc->flags, pe_rsc_needs_fencing)) { multiply_active = (num_all_active > 1); } else { /* If a resource has "requires" set to nothing or quorum, don't consider * it active on unclean nodes (similar to how all resources behave when * stonith-enabled is false). We can start such resources elsewhere * before fencing completes, and if we considered the resource active on * the failed node, we would attempt recovery for being active on * multiple nodes. */ multiply_active = (num_clean_active > 1); } if (multiply_active) { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); // Resource was (possibly) incorrectly multiply active pe_proc_err("%s resource %s might be active on %u nodes (%s)", pcmk__s(class, "Untyped"), rsc->id, num_all_active, recovery2text(rsc->recovery_type)); crm_notice("See https://wiki.clusterlabs.org/wiki/FAQ" "#Resource_is_Too_Active for more information"); switch (rsc->recovery_type) { case recovery_stop_start: need_stop = true; break; case recovery_stop_unexpected: need_stop = true; // stop_resource() will skip expected node pe__set_resource_flags(rsc, pe_rsc_stop_unexpected); break; default: break; } } else { pe__clear_resource_flags(rsc, pe_rsc_stop_unexpected); } if (pcmk_is_set(rsc->flags, pe_rsc_start_pending)) { create_pending_start(rsc); } if (is_moving) { // Remaining tests are only for resources staying where they are } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { if (pcmk_is_set(rsc->flags, pe_rsc_stop)) { need_stop = true; pe_rsc_trace(rsc, "Recovering %s", rsc->id); } else { pe_rsc_trace(rsc, "Recovering %s by demotion", rsc->id); if (rsc->next_role == RSC_ROLE_PROMOTED) { need_promote = true; } } } else if (pcmk_is_set(rsc->flags, pe_rsc_block)) { pe_rsc_trace(rsc, "Blocking further actions on %s", rsc->id); need_stop = true; } else if ((rsc->role > RSC_ROLE_STARTED) && (current != NULL) && (rsc->allocated_to != NULL)) { pe_action_t *start = NULL; pe_rsc_trace(rsc, "Creating start action for promoted resource %s", rsc->id); start = start_action(rsc, rsc->allocated_to, TRUE); if (!pcmk_is_set(start->flags, pe_action_optional)) { // Recovery of a promoted resource pe_rsc_trace(rsc, "%s restart is required for recovery", rsc->id); need_stop = true; } } // Create any actions needed to bring resource down and back up to same role schedule_restart_actions(rsc, current, need_stop, need_promote); // Create any actions needed to take resource from this role to the next schedule_role_transition_actions(rsc); pcmk__create_recurring_actions(rsc); if (allow_migrate) { pcmk__create_migration_actions(rsc, current); } } /*! * \internal * \brief Ban a resource from any allowed nodes that are Pacemaker Remote nodes * * \param[in] rsc Resource to check */ static void rsc_avoids_remote_nodes(const pe_resource_t *rsc) { GHashTableIter iter; pe_node_t *node = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (node->details->remote_rsc != NULL) { node->weight = -INFINITY; } } } /*! * \internal * \brief Return allowed nodes as (possibly sorted) list * * Convert a resource's hash table of allowed nodes to a list. If printing to * stdout, sort the list, to keep action ID numbers consistent for regression * test output (while avoiding the performance hit on a live cluster). * * \param[in] rsc Resource to check for allowed nodes * * \return List of resource's allowed nodes * \note Callers should take care not to rely on the list being sorted. */ static GList * allowed_nodes_as_list(const pe_resource_t *rsc) { GList *allowed_nodes = NULL; if (rsc->allowed_nodes) { allowed_nodes = g_hash_table_get_values(rsc->allowed_nodes); } if (!pcmk__is_daemon) { allowed_nodes = g_list_sort(allowed_nodes, pe__cmp_node_name); } return allowed_nodes; } /*! * \internal * \brief Create implicit constraints needed for a primitive resource * * \param[in,out] rsc Primitive resource to create implicit constraints for */ void pcmk__primitive_internal_constraints(pe_resource_t *rsc) { GList *allowed_nodes = NULL; bool check_unfencing = false; bool check_utilization = false; CRM_ASSERT(rsc != NULL); if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe_rsc_trace(rsc, "Skipping implicit constraints for unmanaged resource %s", rsc->id); return; } // Whether resource requires unfencing check_unfencing = !pcmk_is_set(rsc->flags, pe_rsc_fence_device) && pcmk_is_set(rsc->cluster->flags, pe_flag_enable_unfencing) && pcmk_is_set(rsc->flags, pe_rsc_needs_unfencing); // Whether a non-default placement strategy is used check_utilization = (g_hash_table_size(rsc->utilization) > 0) && !pcmk__str_eq(rsc->cluster->placement_strategy, "default", pcmk__str_casei); // Order stops before starts (i.e. restart) pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, pe_order_optional|pe_order_implies_then|pe_order_restart, rsc->cluster); // Promotable ordering: demote before stop, start before promote if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pe_rsc_promotable) || (rsc->role > RSC_ROLE_UNPROMOTED)) { pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, RSC_DEMOTE, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, pe_order_promoted_implies_first, rsc->cluster); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_PROMOTE, 0), NULL, pe_order_runnable_left, rsc->cluster); } // Don't clear resource history if probing on same node pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, CRM_OP_LRM_DELETE, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_STATUS, 0), NULL, pe_order_same_node|pe_order_then_cancels_first, rsc->cluster); // Certain checks need allowed nodes if (check_unfencing || check_utilization || (rsc->container != NULL)) { allowed_nodes = allowed_nodes_as_list(rsc); } if (check_unfencing) { g_list_foreach(allowed_nodes, pcmk__order_restart_vs_unfence, rsc); } if (check_utilization) { pcmk__create_utilization_constraints(rsc, allowed_nodes); } if (rsc->container != NULL) { pe_resource_t *remote_rsc = NULL; if (rsc->is_remote_node) { // rsc is the implicit remote connection for a guest or bundle node /* Guest resources are not allowed to run on Pacemaker Remote nodes, * to avoid nesting remotes. However, bundles are allowed. */ if (!pcmk_is_set(rsc->flags, pe_rsc_allow_remote_remotes)) { rsc_avoids_remote_nodes(rsc->container); } /* If someone cleans up a guest or bundle node's container, we will * likely schedule a (re-)probe of the container and recovery of the * connection. Order the connection stop after the container probe, * so that if we detect the container running, we will trigger a new * transition and avoid the unnecessary recovery. */ pcmk__order_resource_actions(rsc->container, RSC_STATUS, rsc, RSC_STOP, pe_order_optional); /* A user can specify that a resource must start on a Pacemaker Remote * node by explicitly configuring it with the container=NODENAME * meta-attribute. This is of questionable merit, since location * constraints can accomplish the same thing. But we support it, so here * we check whether a resource (that is not itself a remote connection) * has container set to a remote node or guest node resource. */ } else if (rsc->container->is_remote_node) { remote_rsc = rsc->container; } else { remote_rsc = pe__resource_contains_guest_node(rsc->cluster, rsc->container); } if (remote_rsc != NULL) { /* Force the resource on the Pacemaker Remote node instead of * colocating the resource with the container resource. */ for (GList *item = allowed_nodes; item; item = item->next) { pe_node_t *node = item->data; if (node->details->remote_rsc != remote_rsc) { node->weight = -INFINITY; } } } else { /* This resource is either a filler for a container that does NOT * represent a Pacemaker Remote node, or a Pacemaker Remote * connection resource for a guest node or bundle. */ int score; crm_trace("Order and colocate %s relative to its container %s", rsc->id, rsc->container->id); pcmk__new_ordering(rsc->container, pcmk__op_key(rsc->container->id, RSC_START, 0), NULL, rsc, pcmk__op_key(rsc->id, RSC_START, 0), NULL, pe_order_implies_then|pe_order_runnable_left, rsc->cluster); pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, RSC_STOP, 0), NULL, rsc->container, pcmk__op_key(rsc->container->id, RSC_STOP, 0), NULL, pe_order_implies_first, rsc->cluster); if (pcmk_is_set(rsc->flags, pe_rsc_allow_remote_remotes)) { score = 10000; /* Highly preferred but not essential */ } else { score = INFINITY; /* Force them to run on the same host */ } pcmk__new_colocation("resource-with-container", NULL, score, rsc, rsc->container, NULL, NULL, true, rsc->cluster); } } if (rsc->is_remote_node || pcmk_is_set(rsc->flags, pe_rsc_fence_device)) { /* Remote connections and fencing devices are not allowed to run on * Pacemaker Remote nodes */ rsc_avoids_remote_nodes(rsc); } g_list_free(allowed_nodes); } /*! * \internal * \brief Apply a colocation's score to node weights or resource priority * * Given a colocation constraint, apply its score to the dependent's * allowed node weights (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,out] primary Primary resource in colocation * \param[in] colocation Colocation constraint to apply - * \param[in] for_dependent true if called on behalf of dependent + * \param[in] for_dependent true if called on behalf of dependent */ void pcmk__primitive_apply_coloc_score(pe_resource_t *dependent, - const pe_resource_t *primary, + pe_resource_t *primary, const pcmk__colocation_t *colocation, bool for_dependent) { enum pcmk__coloc_affects filter_results; CRM_CHECK((colocation != NULL) && (dependent != NULL) && (primary != NULL), return); if (for_dependent) { // Always process on behalf of primary resource primary->cmds->apply_coloc_score(dependent, primary, colocation, false); return; } filter_results = pcmk__colocation_affects(dependent, primary, colocation, false); pe_rsc_trace(dependent, "%s %s with %s (%s, score=%d, filter=%d)", ((colocation->score > 0)? "Colocating" : "Anti-colocating"), dependent->id, primary->id, colocation->id, colocation->score, filter_results); switch (filter_results) { case pcmk__coloc_affects_role: pcmk__apply_coloc_to_priority(dependent, primary, colocation); break; case pcmk__coloc_affects_location: pcmk__apply_coloc_to_weights(dependent, primary, colocation); break; default: // pcmk__coloc_affects_nothing return; } } /* Primitive implementation of * resource_alloc_functions_t:with_this_colocations() */ void pcmk__with_primitive_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { // Primitives don't have children, so rsc should also be orig_rsc CRM_CHECK((rsc != NULL) && (rsc->variant == pe_native) && (rsc == orig_rsc) && (list != NULL), return); // Add primitive's own colocations plus any relevant ones from parent pcmk__add_with_this_list(list, rsc->rsc_cons_lhs); if (rsc->parent != NULL) { rsc->parent->cmds->with_this_colocations(rsc->parent, rsc, list); } } /* Primitive implementation of * resource_alloc_functions_t:this_with_colocations() */ void pcmk__primitive_with_colocations(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList **list) { // Primitives don't have children, so rsc should also be orig_rsc CRM_CHECK((rsc != NULL) && (rsc->variant == pe_native) && (rsc == orig_rsc) && (list != NULL), return); // Add primitive's own colocations plus any relevant ones from parent pcmk__add_this_with_list(list, rsc->rsc_cons); if (rsc->parent != NULL) { rsc->parent->cmds->this_with_colocations(rsc->parent, rsc, list); } } /*! * \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 */ enum pe_action_flags pcmk__primitive_action_flags(pe_action_t *action, const pe_node_t *node) { CRM_ASSERT(action != NULL); return action->flags; } /*! * \internal * \brief Check whether a node is a multiply active resource's expected node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p rsc is multiply active with multiple-active set to * stop_unexpected, and \p node is the node where it will remain active * \note This assumes that the resource's next role cannot be changed to stopped * after this is called, which should be reasonable if status has already * been unpacked and resources have been assigned to nodes. */ static bool is_expected_node(const pe_resource_t *rsc, const pe_node_t *node) { return pcmk_all_flags_set(rsc->flags, pe_rsc_stop_unexpected|pe_rsc_restarting) && (rsc->next_role > RSC_ROLE_STOPPED) && pe__same_node(rsc->allocated_to, node); } /*! * \internal * \brief Schedule actions needed to stop a resource wherever it is active * * \param[in,out] rsc Resource being stopped * \param[in] node Node where resource is being stopped (ignored) * \param[in] optional Whether actions should be optional */ static void stop_resource(pe_resource_t *rsc, pe_node_t *node, bool optional) { for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pe_node_t *current = (pe_node_t *) iter->data; pe_action_t *stop = NULL; if (is_expected_node(rsc, current)) { /* We are scheduling restart actions for a multiply active resource * with multiple-active=stop_unexpected, and this is where it should * not be stopped. */ pe_rsc_trace(rsc, "Skipping stop of multiply active resource %s " "on expected node %s", rsc->id, pe__node_name(current)); continue; } if (rsc->partial_migration_target != NULL) { // Continue migration if node originally was and remains target if (pe__same_node(current, rsc->partial_migration_target) && pe__same_node(current, rsc->allocated_to)) { pe_rsc_trace(rsc, "Skipping stop of %s on %s " "because partial migration there will continue", rsc->id, pe__node_name(current)); continue; } else { pe_rsc_trace(rsc, "Forcing stop of %s on %s " "because migration target changed", rsc->id, pe__node_name(current)); optional = false; } } pe_rsc_trace(rsc, "Scheduling stop of %s on %s", rsc->id, pe__node_name(current)); stop = stop_action(rsc, current, optional); if (rsc->allocated_to == NULL) { pe_action_set_reason(stop, "node availability", true); } else if (pcmk_all_flags_set(rsc->flags, pe_rsc_restarting |pe_rsc_stop_unexpected)) { /* We are stopping a multiply active resource on a node that is * not its expected node, and we are still scheduling restart * actions, so the stop is for being multiply active. */ pe_action_set_reason(stop, "being multiply active", true); } if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { pe__clear_action_flags(stop, pe_action_runnable); } if (pcmk_is_set(rsc->cluster->flags, pe_flag_remove_after_stop)) { pcmk__schedule_cleanup(rsc, current, optional); } if (pcmk_is_set(rsc->flags, pe_rsc_needs_unfencing)) { pe_action_t *unfence = pe_fence_op(current, "on", true, NULL, false, rsc->cluster); order_actions(stop, unfence, pe_order_implies_first); if (!pcmk__node_unfenced(current)) { pe_proc_err("Stopping %s until %s can be unfenced", rsc->id, pe__node_name(current)); } } } } /*! * \internal * \brief Schedule actions needed to start a resource on a node * * \param[in,out] rsc Resource being started * \param[in,out] node Node where resource should be started * \param[in] optional Whether actions should be optional */ static void start_resource(pe_resource_t *rsc, pe_node_t *node, bool optional) { pe_action_t *start = NULL; CRM_ASSERT(node != NULL); pe_rsc_trace(rsc, "Scheduling %s start of %s on %s (score %d)", (optional? "optional" : "required"), rsc->id, pe__node_name(node), node->weight); start = start_action(rsc, node, TRUE); pcmk__order_vs_unfence(rsc, node, start, pe_order_implies_then); if (pcmk_is_set(start->flags, pe_action_runnable) && !optional) { pe__clear_action_flags(start, pe_action_optional); } if (is_expected_node(rsc, node)) { /* This could be a problem if the start becomes necessary for other * reasons later. */ pe_rsc_trace(rsc, "Start of multiply active resouce %s " "on expected node %s will be a pseudo-action", rsc->id, pe__node_name(node)); pe__set_action_flags(start, pe_action_pseudo); } } /*! * \internal * \brief Schedule actions needed to promote a resource on a node * * \param[in,out] rsc Resource being promoted * \param[in] node Node where resource should be promoted * \param[in] optional Whether actions should be optional */ static void promote_resource(pe_resource_t *rsc, pe_node_t *node, bool optional) { GList *iter = NULL; GList *action_list = NULL; bool runnable = true; CRM_ASSERT(node != NULL); // Any start must be runnable for promotion to be runnable action_list = pe__resource_actions(rsc, node, RSC_START, true); for (iter = action_list; iter != NULL; iter = iter->next) { pe_action_t *start = (pe_action_t *) iter->data; if (!pcmk_is_set(start->flags, pe_action_runnable)) { runnable = false; } } g_list_free(action_list); if (runnable) { pe_action_t *promote = promote_action(rsc, node, optional); pe_rsc_trace(rsc, "Scheduling %s promotion of %s on %s", (optional? "optional" : "required"), rsc->id, pe__node_name(node)); if (is_expected_node(rsc, node)) { /* This could be a problem if the promote becomes necessary for * other reasons later. */ pe_rsc_trace(rsc, "Promotion of multiply active resouce %s " "on expected node %s will be a pseudo-action", rsc->id, pe__node_name(node)); pe__set_action_flags(promote, pe_action_pseudo); } } else { pe_rsc_trace(rsc, "Not promoting %s on %s: start unrunnable", rsc->id, pe__node_name(node)); action_list = pe__resource_actions(rsc, node, RSC_PROMOTE, true); for (iter = action_list; iter != NULL; iter = iter->next) { pe_action_t *promote = (pe_action_t *) iter->data; pe__clear_action_flags(promote, pe_action_runnable); } g_list_free(action_list); } } /*! * \internal * \brief Schedule actions needed to demote a resource wherever it is active * * \param[in,out] rsc Resource being demoted * \param[in] node Node where resource should be demoted (ignored) * \param[in] optional Whether actions should be optional */ static void demote_resource(pe_resource_t *rsc, pe_node_t *node, bool optional) { /* Since this will only be called for a primitive (possibly as an instance * of a collective resource), the resource is multiply active if it is * running on more than one node, so we want to demote on all of them as * part of recovery, regardless of which one is the desired node. */ for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) { pe_node_t *current = (pe_node_t *) iter->data; if (is_expected_node(rsc, current)) { pe_rsc_trace(rsc, "Skipping demote of multiply active resource %s " "on expected node %s", rsc->id, pe__node_name(current)); } else { pe_rsc_trace(rsc, "Scheduling %s demotion of %s on %s", (optional? "optional" : "required"), rsc->id, pe__node_name(current)); demote_action(rsc, current, optional); } } } static void assert_role_error(pe_resource_t *rsc, pe_node_t *node, bool optional) { CRM_ASSERT(false); } /*! * \internal * \brief Schedule cleanup of a resource * * \param[in,out] rsc Resource to clean up * \param[in] node Node to clean up on * \param[in] optional Whether clean-up should be optional */ void pcmk__schedule_cleanup(pe_resource_t *rsc, const pe_node_t *node, bool optional) { /* If the cleanup is required, its orderings are optional, because they're * relevant only if both actions are required. Conversely, if the cleanup is * optional, the orderings make the then action required if the first action * becomes required. */ uint32_t flag = optional? pe_order_implies_then : pe_order_optional; CRM_CHECK((rsc != NULL) && (node != NULL), return); if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { pe_rsc_trace(rsc, "Skipping clean-up of %s on %s: resource failed", rsc->id, pe__node_name(node)); return; } if (node->details->unclean || !node->details->online) { pe_rsc_trace(rsc, "Skipping clean-up of %s on %s: node unavailable", rsc->id, pe__node_name(node)); return; } crm_notice("Scheduling clean-up of %s on %s", rsc->id, pe__node_name(node)); delete_action(rsc, node, optional); // stop -> clean-up -> start pcmk__order_resource_actions(rsc, RSC_STOP, rsc, RSC_DELETE, flag); pcmk__order_resource_actions(rsc, RSC_DELETE, rsc, RSC_START, flag); } /*! * \internal * \brief Add primitive meta-attributes relevant to graph actions to XML * * \param[in] rsc Primitive resource whose meta-attributes should be added * \param[in,out] xml Transition graph action attributes XML to add to */ void pcmk__primitive_add_graph_meta(const pe_resource_t *rsc, xmlNode *xml) { char *name = NULL; char *value = NULL; const pe_resource_t *parent = NULL; CRM_ASSERT((rsc != NULL) && (xml != NULL)); /* Clone instance numbers get set internally as meta-attributes, and are * needed in the transition graph (for example, to tell unique clone * instances apart). */ value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INCARNATION); if (value != NULL) { name = crm_meta_name(XML_RSC_ATTR_INCARNATION); crm_xml_add(xml, name, value); free(name); } // Not sure if this one is really needed ... value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_REMOTE_NODE); if (value != NULL) { name = crm_meta_name(XML_RSC_ATTR_REMOTE_NODE); crm_xml_add(xml, name, value); free(name); } /* The container meta-attribute can be set on the primitive itself or one of * its parents (for example, a group inside a container resource), so check * them all, and keep the highest one found. */ for (parent = rsc; parent != NULL; parent = parent->parent) { if (parent->container != NULL) { crm_xml_add(xml, CRM_META "_" XML_RSC_ATTR_CONTAINER, parent->container->id); } } /* Bundle replica children will get their external-ip set internally as a * meta-attribute. The graph action needs it, but under a different naming * convention than other meta-attributes. */ value = g_hash_table_lookup(rsc->meta, "external-ip"); if (value != NULL) { crm_xml_add(xml, "pcmk_external_ip", value); } } // Primitive implementation of resource_alloc_functions_t:add_utilization() void pcmk__primitive_add_utilization(const pe_resource_t *rsc, const pe_resource_t *orig_rsc, GList *all_rscs, GHashTable *utilization) { if (!pcmk_is_set(rsc->flags, pe_rsc_provisional)) { return; } pe_rsc_trace(orig_rsc, "%s: Adding primitive %s as colocated utilization", orig_rsc->id, rsc->id); pcmk__release_node_capacity(utilization, rsc); } /*! * \internal * \brief Get epoch time of node's shutdown attribute (or now if none) * * \param[in,out] node Node to check * * \return Epoch time corresponding to shutdown attribute if set or now if not */ static time_t shutdown_time(pe_node_t *node) { const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN); time_t result = 0; if (shutdown != NULL) { long long result_ll; if (pcmk__scan_ll(shutdown, &result_ll, 0LL) == pcmk_rc_ok) { result = (time_t) result_ll; } } return (result == 0)? get_effective_time(node->details->data_set) : result; } /*! * \internal * \brief Ban a resource from a node if it's not locked to the node * * \param[in] data Node to check * \param[in,out] user_data Resource to check */ static void ban_if_not_locked(gpointer data, gpointer user_data) { const pe_node_t *node = (const pe_node_t *) data; pe_resource_t *rsc = (pe_resource_t *) user_data; if (strcmp(node->details->uname, rsc->lock_node->details->uname) != 0) { resource_location(rsc, node, -CRM_SCORE_INFINITY, XML_CONFIG_ATTR_SHUTDOWN_LOCK, rsc->cluster); } } // Primitive implementation of resource_alloc_functions_t:shutdown_lock() void pcmk__primitive_shutdown_lock(pe_resource_t *rsc) { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); // Fence devices and remote connections can't be locked if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches) || pe__resource_is_remote_conn(rsc, rsc->cluster)) { return; } if (rsc->lock_node != NULL) { // The lock was obtained from resource history if (rsc->running_on != NULL) { /* The resource was started elsewhere even though it is now * considered locked. This shouldn't be possible, but as a * failsafe, we don't want to disturb the resource now. */ pe_rsc_info(rsc, "Cancelling shutdown lock because %s is already active", rsc->id); pe__clear_resource_history(rsc, rsc->lock_node, rsc->cluster); rsc->lock_node = NULL; rsc->lock_time = 0; } // Only a resource active on exactly one node can be locked } else if (pcmk__list_of_1(rsc->running_on)) { pe_node_t *node = rsc->running_on->data; if (node->details->shutdown) { if (node->details->unclean) { pe_rsc_debug(rsc, "Not locking %s to unclean %s for shutdown", rsc->id, pe__node_name(node)); } else { rsc->lock_node = node; rsc->lock_time = shutdown_time(node); } } } if (rsc->lock_node == NULL) { // No lock needed return; } if (rsc->cluster->shutdown_lock > 0) { time_t lock_expiration = rsc->lock_time + rsc->cluster->shutdown_lock; pe_rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)", rsc->id, pe__node_name(rsc->lock_node), (long long) lock_expiration); pe__update_recheck_time(++lock_expiration, rsc->cluster); } else { pe_rsc_info(rsc, "Locking %s to %s due to shutdown", rsc->id, pe__node_name(rsc->lock_node)); } // If resource is locked to one node, ban it from all other nodes g_list_foreach(rsc->cluster->nodes, ban_if_not_locked, rsc); } diff --git a/lib/pengine/Makefile.am b/lib/pengine/Makefile.am index c2a8c902db..805772b87a 100644 --- a/lib/pengine/Makefile.am +++ b/lib/pengine/Makefile.am @@ -1,81 +1,81 @@ # -# Copyright 2004-2022 the Pacemaker project contributors +# Copyright 2004-2023 the Pacemaker project contributors # # The version control history for this file may have further details. # # This source code is licensed under the GNU General Public License version 2 # or later (GPLv2+) WITHOUT ANY WARRANTY. # include $(top_srcdir)/mk/common.mk # Without "." here, check-recursive will run through the subdirectories first # and then run "make check" here. This will fail, because there's things in # the subdirectories that need check_LTLIBRARIES built first. Adding "." here # changes the order so the subdirectories are processed afterwards. SUBDIRS = . tests ## libraries lib_LTLIBRARIES = libpe_rules.la libpe_status.la check_LTLIBRARIES = libpe_rules_test.la libpe_status_test.la ## SOURCES -noinst_HEADERS = variant.h pe_status_private.h +noinst_HEADERS = pe_status_private.h libpe_rules_la_LDFLAGS = -version-info 30:0:4 libpe_rules_la_CFLAGS = $(CFLAGS_HARDENED_LIB) libpe_rules_la_LDFLAGS += $(LDFLAGS_HARDENED_LIB) libpe_rules_la_LIBADD = $(top_builddir)/lib/common/libcrmcommon.la libpe_rules_la_SOURCES = rules.c rules_alerts.c common.c libpe_status_la_LDFLAGS = -version-info 34:0:6 libpe_status_la_CFLAGS = $(CFLAGS_HARDENED_LIB) libpe_status_la_LDFLAGS += $(LDFLAGS_HARDENED_LIB) libpe_status_la_LIBADD = $(top_builddir)/lib/common/libcrmcommon.la # Use += rather than backlashed continuation lines for parsing by bumplibs libpe_status_la_SOURCES = libpe_status_la_SOURCES += bundle.c libpe_status_la_SOURCES += clone.c libpe_status_la_SOURCES += common.c libpe_status_la_SOURCES += complex.c libpe_status_la_SOURCES += failcounts.c libpe_status_la_SOURCES += group.c libpe_status_la_SOURCES += native.c libpe_status_la_SOURCES += pe_actions.c libpe_status_la_SOURCES += pe_health.c libpe_status_la_SOURCES += pe_digest.c libpe_status_la_SOURCES += pe_notif.c libpe_status_la_SOURCES += pe_output.c libpe_status_la_SOURCES += remote.c libpe_status_la_SOURCES += rules.c libpe_status_la_SOURCES += status.c libpe_status_la_SOURCES += tags.c libpe_status_la_SOURCES += unpack.c libpe_status_la_SOURCES += utils.c # # libpe_rules_test and libpe_status_test are only used with unit tests, so we can # mock system calls. See lib/common/mock.c for details. # include $(top_srcdir)/mk/tap.mk libpe_rules_test_la_SOURCES = $(libpe_rules_la_SOURCES) libpe_rules_test_la_LDFLAGS = $(libpe_rules_la_LDFLAGS) -rpath $(libdir) $(LDFLAGS_WRAP) # See comments on libcrmcommon_test_la in lib/common/Makefile.am regarding these flags. libpe_rules_test_la_CFLAGS = $(libpe_rules_la_CFLAGS) -DPCMK__UNIT_TESTING \ -fno-builtin -fno-inline libpe_rules_test_la_LIBADD = $(top_builddir)/lib/common/libcrmcommon_test.la -lcmocka -lm libpe_status_test_la_SOURCES = $(libpe_status_la_SOURCES) libpe_status_test_la_LDFLAGS = $(libpe_status_la_LDFLAGS) -rpath $(libdir) $(LDFLAGS_WRAP) # See comments on libcrmcommon_test_la in lib/common/Makefile.am regarding these flags. libpe_status_test_la_CFLAGS = $(libpe_status_la_CFLAGS) -DPCMK__UNIT_TESTING \ -fno-builtin -fno-inline libpe_status_test_la_LIBADD = $(top_builddir)/lib/common/libcrmcommon_test.la -lcmocka -lm clean-generic: rm -f *.log *.debug *~ diff --git a/lib/pengine/bundle.c b/lib/pengine/bundle.c index 388a702d4e..bc888d42fe 100644 --- a/lib/pengine/bundle.c +++ b/lib/pengine/bundle.c @@ -1,2005 +1,2185 @@ /* * Copyright 2004-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include -#define PE__VARIANT_BUNDLE 1 -#include "./variant.h" +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; + + pe_resource_t *child; + + GList *replicas; // pe__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) \ + CRM_ASSERT(rsc != NULL); \ + CRM_ASSERT(rsc->variant == pe_container); \ + CRM_ASSERT(rsc->variant_opaque != NULL); \ + data = (pe__bundle_variant_data_t *) rsc->variant_opaque; /*! * \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 pe_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 + */ +pe_resource_t * +pe__bundled_resource(const pe_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 pe_resource_t * +pe__get_rsc_in_container(const pe_resource_t *instance) +{ + const pe__bundle_variant_data_t *data = NULL; + const pe_resource_t *top = pe__const_top_resource(instance, true); + + if ((top == NULL) || (top->variant != pe_container)) { + return NULL; + } + get_bundle_variant_data(data, top); + + for (const GList *iter = data->replicas; iter != NULL; iter = iter->next) { + const pe__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 pe_resource_t *bundle, const pe_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) { + pe__bundle_replica_t *replica = iter->data; + + if (pe__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 + */ +pe_resource_t * +pe__first_container(const pe_resource_t *bundle) +{ + const pe__bundle_variant_data_t *bundle_data = NULL; + const pe__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(pe_resource_t *bundle, + bool (*fn)(pe__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((pe__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, pe__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 = create_xml_node(NULL, XML_CIB_TAG_RESOURCE); crm_xml_add(rsc, XML_ATTR_ID, name); crm_xml_add(rsc, XML_AGENT_ATTR_CLASS, PCMK_RESOURCE_CLASS_OCF); crm_xml_add(rsc, XML_AGENT_ATTR_PROVIDER, provider); crm_xml_add(rsc, XML_ATTR_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) { pe_err("Specifying the 'control-port' for %s requires 'replicas-per-host=1'", data->prefix); data->nreplicas_per_host = 1; // @TODO to be sure: pe__clear_resource_flags(rsc, pe_rsc_unique); } return TRUE; } return FALSE; } static int create_ip_resource(pe_resource_t *parent, pe__bundle_variant_data_t *data, pe__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); crm_xml_sanitize_id(id); xml_ip = create_resource(id, "heartbeat", "IPaddr2"); free(id); xml_obj = create_xml_node(xml_ip, XML_TAG_ATTR_SETS); crm_xml_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 = create_xml_node(xml_ip, "operations"); crm_create_op_xml(xml_obj, ID(xml_ip), "monitor", "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (pe__unpack_resource(xml_ip, &replica->ip, parent, parent->cluster) != 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(pe_resource_t *parent, const pe__bundle_variant_data_t *data, pe__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); crm_xml_sanitize_id(id); xml_container = create_resource(id, "heartbeat", agent_str); free(id); xml_obj = create_xml_node(xml_container, XML_TAG_ATTR_SETS); crm_xml_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", XML_BOOLEAN_TRUE); crm_create_nvpair_xml(xml_obj, NULL, "force_kill", XML_BOOLEAN_FALSE); crm_create_nvpair_xml(xml_obj, NULL, "reuse", XML_BOOLEAN_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_remote_port=", data->control_port, NULL); } else { g_string_append_printf(buffer, " %sPCMK_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, "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 = create_xml_node(xml_container, "operations"); crm_create_op_xml(xml_obj, ID(xml_container), "monitor", "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (pe__unpack_resource(xml_container, &replica->container, parent, parent->cluster) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } pe__set_resource_flags(replica->container, pe_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(pe_resource_t *rsc, const char *uname) { gpointer match = g_hash_table_lookup(rsc->allowed_nodes, uname); if (match) { ((pe_node_t *) match)->weight = -INFINITY; ((pe_node_t *) match)->rsc_discover_mode = pe_discover_never; } if (rsc->children) { g_list_foreach(rsc->children, (GFunc) disallow_node, (gpointer) uname); } } static int create_remote_resource(pe_resource_t *parent, pe__bundle_variant_data_t *data, pe__bundle_replica_t *replica) { if (replica->child && valid_network(data)) { GHashTableIter gIter; pe_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->cluster->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->cluster->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 data set cleanup to use as * the node ID and uname. */ free(id); id = NULL; uname = 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 = pe_find_node(parent->cluster->nodes, uname); if (node == NULL) { node = pe_create_node(uname, uname, "remote", "-INFINITY", parent->cluster); } else { node->weight = -INFINITY; } node->rsc_discover_mode = pe_discover_never; /* unpack_remote_nodes() ensures that each remote node and guest node * has a pe_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->cluster->resources, (GFunc) disallow_node, (gpointer) uname); replica->node = pe__copy_node(node); replica->node->weight = 500; replica->node->rsc_discover_mode = pe_discover_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)); { pe_node_t *copy = pe__copy_node(replica->node); copy->weight = -INFINITY; g_hash_table_insert(replica->child->parent->allowed_nodes, (gpointer) replica->node->details->id, copy); } if (pe__unpack_resource(xml_remote, &replica->remote, parent, parent->cluster) != 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 (pe__is_guest_or_remote_node(node)) { /* Remote resources can only run on 'normal' cluster node */ node->weight = -INFINITY; } } replica->node->details->remote_rsc = replica->remote; // Ensure pe__is_guest_node() functions correctly immediately replica->remote->container = replica->container; /* A bundle's #kind is closer to "container" (guest node) than the * "remote" set by pe_create_node(). */ g_hash_table_insert(replica->node->details->attrs, strdup(CRM_ATTR_KIND), strdup("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(pe_resource_t *parent, pe__bundle_variant_data_t *data, pe__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)) { add_hash_param(replica->child->meta, "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. */ pe__set_resource_flags(replica->remote, pe_rsc_allow_remote_remotes); } 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 = calloc(1, sizeof(pe__bundle_mount_t)); CRM_ASSERT(mount != NULL); mount->source = strdup(source); mount->target = strdup(target); pcmk__str_update(&mount->options, 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 pe__bundle_replica_t * replica_for_remote(pe_resource_t *remote) { pe_resource_t *top = remote; pe__bundle_variant_data_t *bundle_data = NULL; if (top == NULL) { return NULL; } while (top->parent != NULL) { top = top->parent; } get_bundle_variant_data(bundle_data, top); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; if (replica->remote == remote) { return replica; } } CRM_LOG_ASSERT(FALSE); return NULL; } bool pe__bundle_needs_remote_name(pe_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->cluster); value = g_hash_table_lookup(params, XML_RSC_ATTR_REMOTE_RA_ADDR); return pcmk__str_eq(value, "#uname", pcmk__str_casei) && xml_contains_remote_node(rsc->xml); } const char * pe__add_bundle_remote_name(pe_resource_t *rsc, pe_working_set_t *data_set, xmlNode *xml, const char *field) { // REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside pe_node_t *node = NULL; pe__bundle_replica_t *replica = NULL; if (!pe__bundle_needs_remote_name(rsc)) { return NULL; } replica = replica_for_remote(rsc); if (replica == NULL) { return NULL; } node = replica->container->allocated_to; if (node == NULL) { /* If it won't be running anywhere after the * transition, go with where it's running now. */ node = pe__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, pe__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", ID(mount_xml), flags, \ (flags_to_set), #flags_to_set); \ } while (0) gboolean pe__unpack_bundle(pe_resource_t *rsc, pe_working_set_t *data_set) { const char *value = NULL; xmlNode *xml_obj = NULL; xmlNode *xml_resource = NULL; pe__bundle_variant_data_t *bundle_data = NULL; bool need_log_mount = TRUE; CRM_ASSERT(rsc != NULL); pe_rsc_trace(rsc, "Processing resource %s...", rsc->id); bundle_data = calloc(1, sizeof(pe__bundle_variant_data_t)); rsc->variant_opaque = bundle_data; bundle_data->prefix = strdup(rsc->id); xml_obj = first_named_child(rsc->xml, PE__CONTAINER_AGENT_DOCKER_S); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_DOCKER; } else { xml_obj = first_named_child(rsc->xml, PE__CONTAINER_AGENT_RKT_S); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_RKT; } else { xml_obj = first_named_child(rsc->xml, PE__CONTAINER_AGENT_PODMAN_S); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_PODMAN; } else { return FALSE; } } } // Use 0 for default, minimum, and invalid promoted-max value = crm_element_value(xml_obj, XML_RSC_ATTR_PROMOTED_MAX); if (value == NULL) { // @COMPAT deprecated since 2.0.0 value = crm_element_value(xml_obj, "masters"); } pcmk__scan_min_int(value, &bundle_data->promoted_max, 0); // Default replicas to promoted-max if it was specified and 1 otherwise value = crm_element_value(xml_obj, "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, "replicas-per-host"); pcmk__scan_min_int(value, &bundle_data->nreplicas_per_host, 1); if (bundle_data->nreplicas_per_host == 1) { pe__clear_resource_flags(rsc, pe_rsc_unique); } bundle_data->container_command = crm_element_value_copy(xml_obj, "run-command"); bundle_data->launcher_options = crm_element_value_copy(xml_obj, "options"); bundle_data->image = crm_element_value_copy(xml_obj, "image"); bundle_data->container_network = crm_element_value_copy(xml_obj, "network"); xml_obj = first_named_child(rsc->xml, "network"); if(xml_obj) { bundle_data->ip_range_start = crm_element_value_copy(xml_obj, "ip-range-start"); bundle_data->host_netmask = crm_element_value_copy(xml_obj, "host-netmask"); bundle_data->host_network = crm_element_value_copy(xml_obj, "host-interface"); bundle_data->control_port = crm_element_value_copy(xml_obj, "control-port"); value = crm_element_value(xml_obj, "add-host"); if (crm_str_to_boolean(value, &bundle_data->add_host) != 1) { bundle_data->add_host = TRUE; } for (xmlNode *xml_child = pcmk__xe_first_child(xml_obj); xml_child != NULL; xml_child = pcmk__xe_next(xml_child)) { pe__bundle_port_t *port = calloc(1, sizeof(pe__bundle_port_t)); port->source = crm_element_value_copy(xml_child, "port"); if(port->source == NULL) { port->source = crm_element_value_copy(xml_child, "range"); } else { port->target = crm_element_value_copy(xml_child, "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 { pe_err("Invalid port directive %s", ID(xml_child)); port_free(port); } } } xml_obj = first_named_child(rsc->xml, "storage"); for (xmlNode *xml_child = pcmk__xe_first_child(xml_obj); xml_child != NULL; xml_child = pcmk__xe_next(xml_child)) { const char *source = crm_element_value(xml_child, "source-dir"); const char *target = crm_element_value(xml_child, "target-dir"); const char *options = crm_element_value(xml_child, "options"); int flags = pe__bundle_mount_none; if (source == NULL) { source = crm_element_value(xml_child, "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 { pe_err("Invalid mount directive %s", ID(xml_child)); } } xml_obj = first_named_child(rsc->xml, "primitive"); if (xml_obj && valid_network(bundle_data)) { char *value = NULL; xmlNode *xml_set = NULL; xml_resource = create_xml_node(NULL, XML_CIB_TAG_INCARNATION); /* @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.) */ crm_xml_set_id(xml_resource, "%s-%s", bundle_data->prefix, (bundle_data->promoted_max? "master" : (const char *)xml_resource->name)); xml_set = create_xml_node(xml_resource, XML_TAG_META_SETS); crm_xml_set_id(xml_set, "%s-%s-meta", bundle_data->prefix, xml_resource->name); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_ORDERED, XML_BOOLEAN_TRUE); value = pcmk__itoa(bundle_data->nreplicas); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_INCARNATION_MAX, value); free(value); value = pcmk__itoa(bundle_data->nreplicas_per_host); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_INCARNATION_NODEMAX, value); free(value); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_UNIQUE, pcmk__btoa(bundle_data->nreplicas_per_host > 1)); if (bundle_data->promoted_max) { crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_PROMOTABLE, XML_BOOLEAN_TRUE); value = pcmk__itoa(bundle_data->promoted_max); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_PROMOTED_MAX, value); free(value); } //crm_xml_add(xml_obj, XML_ATTR_ID, bundle_data->prefix); add_node_copy(xml_resource, xml_obj); } else if(xml_obj) { pe_err("Cannot control %s inside %s without either ip-range-start or control-port", rsc->id, 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, data_set) != 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 = calloc(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 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) { pe__bundle_replica_t *replica = calloc(1, sizeof(pe__bundle_replica_t)); replica->child = childIter->data; replica->child->exclusive_discover = TRUE; replica->offset = lpc++; // Ensure the child's notify gets set based on the underlying primitive's value if (pcmk_is_set(replica->child->flags, pe_rsc_notify)) { pe__set_resource_flags(bundle_data->child, pe_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, XML_RSC_ATTR_TARGET); } bundle_data->container_host_options = g_string_free(buffer, FALSE); if (bundle_data->attribute_target) { g_hash_table_replace(rsc->meta, strdup(XML_RSC_ATTR_TARGET), strdup(bundle_data->attribute_target)); g_hash_table_replace(bundle_data->child->meta, strdup(XML_RSC_ATTR_TARGET), strdup(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++) { pe__bundle_replica_t *replica = calloc(1, sizeof(pe__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) { pe__bundle_replica_t *replica = gIter->data; if (create_replica_resources(rsc, bundle_data, replica) != pcmk_rc_ok) { pe_err("Failed unpacking resource %s", rsc->id); rsc->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(pe_resource_t *rsc, gboolean all) { if (rsc) { gboolean child_active = rsc->fns->active(rsc, all); if (child_active && !all) { return TRUE; } else if (!child_active && all) { return FALSE; } } return -1; } gboolean pe__bundle_active(pe_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) { pe__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 */ pe_resource_t * pe__find_bundle_replica(const pe_resource_t *bundle, const pe_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) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica && replica->node); if (replica->node->details == node->details) { return replica->child; } } return NULL; } /*! * \internal * \deprecated This function will be removed in a future release */ static void print_rsc_in_list(pe_resource_t *rsc, const char *pre_text, long options, void *print_data) { if (rsc != NULL) { if (options & pe_print_html) { status_print("
  • "); } rsc->fns->print(rsc, pre_text, options, print_data); if (options & pe_print_html) { status_print("
    • \n"); } } } /*! * \internal * \deprecated This function will be removed in a future release */ static void bundle_print_xml(pe_resource_t *rsc, const char *pre_text, long options, void *print_data) { pe__bundle_variant_data_t *bundle_data = NULL; char *child_text = NULL; CRM_CHECK(rsc != NULL, return); if (pre_text == NULL) { pre_text = ""; } child_text = crm_strdup_printf("%s ", pre_text); get_bundle_variant_data(bundle_data, rsc); status_print("%sid); status_print("type=\"%s\" ", container_agent_str(bundle_data->agent_type)); status_print("image=\"%s\" ", bundle_data->image); status_print("unique=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_unique)); status_print("managed=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_managed)); status_print("failed=\"%s\" ", pe__rsc_bool_str(rsc, pe_rsc_failed)); status_print(">\n"); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); status_print("%s \n", pre_text, replica->offset); print_rsc_in_list(replica->ip, child_text, options, print_data); print_rsc_in_list(replica->child, child_text, options, print_data); print_rsc_in_list(replica->container, child_text, options, print_data); print_rsc_in_list(replica->remote, child_text, options, print_data); status_print("%s \n", pre_text); } status_print("%s\n", pre_text); free(child_text); } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pe_resource_t *", "GList *", "GList *") int pe__bundle_xml(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pe_resource_t *rsc = va_arg(args, pe_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->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) { pe__bundle_replica_t *replica = gIter->data; char *id = NULL; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->remote, only_rsc, print_everything); if (!print_everything && !print_ip && !print_child && !print_ctnr && !print_remote) { continue; } if (!printed_header) { printed_header = TRUE; desc = pe__resource_description(rsc, show_opts); rc = pe__name_and_nvpairs_xml(out, true, "bundle", 8, "id", rsc->id, "type", container_agent_str(bundle_data->agent_type), "image", bundle_data->image, "unique", pe__rsc_bool_str(rsc, pe_rsc_unique), "maintenance", pe__rsc_bool_str(rsc, pe_rsc_maintenance), "managed", pe__rsc_bool_str(rsc, pe_rsc_managed), "failed", pe__rsc_bool_str(rsc, pe_rsc_failed), "description", desc); CRM_ASSERT(rc == pcmk_rc_ok); } id = pcmk__itoa(replica->offset); rc = pe__name_and_nvpairs_xml(out, true, "replica", 1, "id", id); free(id); CRM_ASSERT(rc == pcmk_rc_ok); if (print_ip) { out->message(out, crm_map_element_name(replica->ip->xml), show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), show_opts, replica->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, pe__bundle_replica_t *replica, pe_node_t *node, uint32_t show_opts) { pe_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 pe_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pe_rsc_maintenance)) { return " (maintenance)"; } if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { return " (unmanaged)"; } return ""; } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pe_resource_t *", "GList *", "GList *") int pe__bundle_html(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pe_resource_t *rsc = va_arg(args, pe_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->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) { pe__bundle_replica_t *replica = gIter->data; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->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, pe_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, crm_map_element_name(replica->ip->xml), new_show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), new_show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), new_show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), new_show_opts, replica->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, pe_rsc_unique) ? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); pe__bundle_replica_output_html(out, replica, pe__current_node(replica->container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } static void pe__bundle_replica_output_text(pcmk__output_t *out, pe__bundle_replica_t *replica, pe_node_t *node, uint32_t show_opts) { const pe_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", "pe_resource_t *", "GList *", "GList *") int pe__bundle_text(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pe_resource_t *rsc = va_arg(args, pe_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); get_bundle_variant_data(bundle_data, rsc); CRM_ASSERT(rsc != NULL); if (rsc->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) { pe__bundle_replica_t *replica = gIter->data; gboolean print_ip, print_child, print_ctnr, print_remote; CRM_ASSERT(replica); if (pcmk__rsc_filtered_by_node(replica->container, only_node)) { continue; } print_ip = replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, print_everything); print_child = replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, print_everything); print_ctnr = !replica->container->fns->is_filtered(replica->container, only_rsc, print_everything); print_remote = replica->remote != NULL && !replica->remote->fns->is_filtered(replica->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, pe_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, crm_map_element_name(replica->ip->xml), new_show_opts, replica->ip, only_node, only_rsc); } if (print_child) { out->message(out, crm_map_element_name(replica->child->xml), new_show_opts, replica->child, only_node, only_rsc); } if (print_ctnr) { out->message(out, crm_map_element_name(replica->container->xml), new_show_opts, replica->container, only_node, only_rsc); } if (print_remote) { out->message(out, crm_map_element_name(replica->remote->xml), new_show_opts, replica->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, pe_rsc_unique) ? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); pe__bundle_replica_output_text(out, replica, pe__current_node(replica->container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } /*! * \internal * \deprecated This function will be removed in a future release */ static void print_bundle_replica(pe__bundle_replica_t *replica, const char *pre_text, long options, void *print_data) { pe_node_t *node = NULL; pe_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); } node = pe__current_node(replica->container); common_print(rsc, pre_text, buffer, node, options, print_data); } /*! * \internal * \deprecated This function will be removed in a future release */ void pe__print_bundle(pe_resource_t *rsc, const char *pre_text, long options, void *print_data) { pe__bundle_variant_data_t *bundle_data = NULL; char *child_text = NULL; CRM_CHECK(rsc != NULL, return); if (options & pe_print_xml) { bundle_print_xml(rsc, pre_text, options, print_data); return; } get_bundle_variant_data(bundle_data, rsc); if (pre_text == NULL) { pre_text = " "; } status_print("%sContainer bundle%s: %s [%s]%s%s\n", pre_text, ((bundle_data->nreplicas > 1)? " set" : ""), rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pe_rsc_unique) ? " (unique)" : "", pcmk_is_set(rsc->flags, pe_rsc_managed) ? "" : " (unmanaged)"); if (options & pe_print_html) { status_print("
    \n
      \n"); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pe__bundle_replica_t *replica = gIter->data; CRM_ASSERT(replica); if (options & pe_print_html) { status_print("
    • "); } if (pcmk_is_set(options, pe_print_implicit)) { child_text = crm_strdup_printf(" %s", pre_text); if (pcmk__list_of_multiple(bundle_data->replicas)) { status_print(" %sReplica[%d]\n", pre_text, replica->offset); } if (options & pe_print_html) { status_print("
      \n
        \n"); } print_rsc_in_list(replica->ip, child_text, options, print_data); print_rsc_in_list(replica->container, child_text, options, print_data); print_rsc_in_list(replica->remote, child_text, options, print_data); print_rsc_in_list(replica->child, child_text, options, print_data); if (options & pe_print_html) { status_print("
      \n"); } } else { child_text = crm_strdup_printf("%s ", pre_text); print_bundle_replica(replica, child_text, options, print_data); } free(child_text); if (options & pe_print_html) { status_print("
      • \n"); } } if (options & pe_print_html) { status_print("
    \n"); } } static void free_bundle_replica(pe__bundle_replica_t *replica) { if (replica == NULL) { return; } if (replica->node) { free(replica->node); replica->node = NULL; } if (replica->ip) { free_xml(replica->ip->xml); replica->ip->xml = NULL; replica->ip->fns->free(replica->ip); replica->ip = NULL; } if (replica->container) { free_xml(replica->container->xml); replica->container->xml = NULL; replica->container->fns->free(replica->container); replica->container = NULL; } if (replica->remote) { free_xml(replica->remote->xml); replica->remote->xml = NULL; replica->remote->fns->free(replica->remote); replica->remote = NULL; } free(replica->ipaddr); free(replica); } void pe__free_bundle(pe_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); pe_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) { free_xml(bundle_data->child->xml); bundle_data->child->xml = NULL; bundle_data->child->fns->free(bundle_data->child); } common_free(rsc); } enum rsc_role_e pe__bundle_resource_state(const pe_resource_t *rsc, gboolean current) { enum rsc_role_e container_role = RSC_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 pe_resource_t *rsc) { if ((rsc == NULL) || (rsc->variant != pe_container)) { return 0; } else { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); return bundle_data->nreplicas; } } void pe__count_bundle(pe_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) { pe__bundle_replica_t *replica = item->data; if (replica->ip) { replica->ip->fns->count(replica->ip); } if (replica->child) { replica->child->fns->count(replica->child); } if (replica->container) { replica->container->fns->count(replica->container); } if (replica->remote) { replica->remote->fns->count(replica->remote); } } } gboolean pe__bundle_is_filtered(const pe_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) { pe__bundle_replica_t *replica = gIter->data; if (replica->ip != NULL && !replica->ip->fns->is_filtered(replica->ip, only_rsc, FALSE)) { passes = TRUE; break; } else if (replica->child != NULL && !replica->child->fns->is_filtered(replica->child, only_rsc, FALSE)) { passes = TRUE; break; } else if (!replica->container->fns->is_filtered(replica->container, only_rsc, FALSE)) { passes = TRUE; break; } else if (replica->remote != NULL && !replica->remote->fns->is_filtered(replica->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 pe_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) { pe__bundle_replica_t *replica = iter->data; containers = g_list_append(containers, replica->container); } return containers; } // Bundle implementation of resource_object_functions_t:active_node() pe_node_t * pe__bundle_active_node(const pe_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean) { pe_node_t *active = NULL; pe_node_t *node = NULL; pe_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) { pe__bundle_replica_t *replica = iter->data; if (replica->container->running_on != NULL) { containers = g_list_append(containers, replica->container); } } if (containers == NULL) { return NULL; } /* If the bundle has only a single active container, just use that * container's method. If live migration is ever supported for bundle * containers, this will allow us to prefer the migration source when there * is only one container and it is migrating. For now, this just lets us * avoid creating the nodes table. */ if (pcmk__list_of_1(containers)) { container = containers->data; node = container->fns->active_node(container, count_all, count_clean); g_list_free(containers); return node; } // Add all containers' active nodes to a hash table (for uniqueness) nodes = g_hash_table_new(NULL, NULL); for (iter = containers; iter != NULL; iter = iter->next) { container = iter->data; for (GList *node_iter = container->running_on; node_iter != NULL; node_iter = node_iter->next) { node = node_iter->data; // If insert returns true, we haven't counted this node yet if (g_hash_table_insert(nodes, (gpointer) node->details, (gpointer) node) && !pe__count_active_node(rsc, node, &active, count_all, count_clean)) { goto done; } } } done: g_list_free(containers); g_hash_table_destroy(nodes); return active; } /*! * \internal * \brief Get maximum bundle resource instances per node * * \param[in] rsc Bundle resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int pe__bundle_max_per_node(const pe_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/tests/native/native_find_rsc_test.c b/lib/pengine/tests/native/native_find_rsc_test.c index 22aaf41b35..03f20281d0 100644 --- a/lib/pengine/tests/native/native_find_rsc_test.c +++ b/lib/pengine/tests/native/native_find_rsc_test.c @@ -1,677 +1,670 @@ /* - * Copyright 2022 the Pacemaker project contributors + * Copyright 2022-2023 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include -/* Needed to access replicas inside a bundle. */ -#define PE__VARIANT_BUNDLE 1 -#include - xmlNode *input = NULL; pe_working_set_t *data_set = NULL; pe_node_t *cluster01, *cluster02, *httpd_bundle_0; pe_resource_t *exim_group, *inactive_group, *promotable_clone, *inactive_clone; pe_resource_t *httpd_bundle, *mysql_clone_group; static int setup(void **state) { char *path = NULL; crm_xml_init(); path = crm_strdup_printf("%s/crm_mon.xml", getenv("PCMK_CTS_CLI_DIR")); input = filename2xml(path); free(path); if (input == NULL) { return 1; } data_set = pe_new_working_set(); if (data_set == NULL) { return 1; } pe__set_working_set_flags(data_set, pe_flag_no_counts|pe_flag_no_compat); data_set->input = input; cluster_status(data_set); /* Get references to the cluster nodes so we don't have to find them repeatedly. */ cluster01 = pe_find_node(data_set->nodes, "cluster01"); cluster02 = pe_find_node(data_set->nodes, "cluster02"); httpd_bundle_0 = pe_find_node(data_set->nodes, "httpd-bundle-0"); /* Get references to several resources we use frequently. */ for (GList *iter = data_set->resources; iter != NULL; iter = iter->next) { pe_resource_t *rsc = (pe_resource_t *) iter->data; if (strcmp(rsc->id, "exim-group") == 0) { exim_group = rsc; } else if (strcmp(rsc->id, "httpd-bundle") == 0) { httpd_bundle = rsc; } else if (strcmp(rsc->id, "inactive-clone") == 0) { inactive_clone = rsc; } else if (strcmp(rsc->id, "inactive-group") == 0) { inactive_group = rsc; } else if (strcmp(rsc->id, "mysql-clone-group") == 0) { mysql_clone_group = rsc; } else if (strcmp(rsc->id, "promotable-clone") == 0) { promotable_clone = rsc; } } return 0; } static int teardown(void **state) { pe_free_working_set(data_set); return 0; } static void bad_args(void **state) { pe_resource_t *rsc = (pe_resource_t *) g_list_first(data_set->resources)->data; char *id = rsc->id; char *name = NULL; assert_non_null(rsc); assert_null(native_find_rsc(NULL, "dummy", NULL, 0)); assert_null(native_find_rsc(rsc, NULL, NULL, 0)); /* No resources exist with these names. */ name = crm_strdup_printf("%sX", rsc->id); assert_null(native_find_rsc(rsc, name, NULL, 0)); free(name); name = crm_strdup_printf("x%s", rsc->id); assert_null(native_find_rsc(rsc, name, NULL, 0)); free(name); name = g_ascii_strup(rsc->id, -1); assert_null(native_find_rsc(rsc, name, NULL, 0)); g_free(name); /* Fails because resource ID is NULL. */ rsc->id = NULL; assert_null(native_find_rsc(rsc, id, NULL, 0)); rsc->id = id; } static void primitive_rsc(void **state) { pe_resource_t *dummy = NULL; /* Find the "dummy" resource, which is the only one with that ID in the set. */ for (GList *iter = data_set->resources; iter != NULL; iter = iter->next) { pe_resource_t *rsc = (pe_resource_t *) iter->data; if (strcmp(rsc->id, "dummy") == 0) { dummy = rsc; break; } } assert_non_null(dummy); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(dummy, native_find_rsc(dummy, "dummy", NULL, 0)); assert_ptr_equal(dummy, native_find_rsc(dummy, "dummy", NULL, pe_find_current)); /* Fails because resource is not a clone (nor cloned). */ assert_null(native_find_rsc(dummy, "dummy", NULL, pe_find_clone)); assert_null(native_find_rsc(dummy, "dummy", cluster02, pe_find_clone)); /* Fails because dummy is not running on cluster01, even with the right flags. */ assert_null(native_find_rsc(dummy, "dummy", cluster01, pe_find_current)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(dummy, "dummy", cluster02, 0)); /* Passes because dummy is running on cluster02. */ assert_ptr_equal(dummy, native_find_rsc(dummy, "dummy", cluster02, pe_find_current)); } static void group_rsc(void **state) { assert_non_null(exim_group); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(exim_group, native_find_rsc(exim_group, "exim-group", NULL, 0)); assert_ptr_equal(exim_group, native_find_rsc(exim_group, "exim-group", NULL, pe_find_current)); /* Fails because resource is not a clone (nor cloned). */ assert_null(native_find_rsc(exim_group, "exim-group", NULL, pe_find_clone)); assert_null(native_find_rsc(exim_group, "exim-group", cluster01, pe_find_clone)); /* Fails because none of exim-group's children are running on cluster01, even with the right flags. */ assert_null(native_find_rsc(exim_group, "exim-group", cluster01, pe_find_current)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(exim_group, "exim-group", cluster01, 0)); /* Passes because one of exim-group's children is running on cluster02. */ assert_ptr_equal(exim_group, native_find_rsc(exim_group, "exim-group", cluster02, pe_find_current)); } static void inactive_group_rsc(void **state) { assert_non_null(inactive_group); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(inactive_group, native_find_rsc(inactive_group, "inactive-group", NULL, 0)); assert_ptr_equal(inactive_group, native_find_rsc(inactive_group, "inactive-group", NULL, pe_find_current)); assert_ptr_equal(inactive_group, native_find_rsc(inactive_group, "inactive-group", NULL, pe_find_inactive)); /* Fails because resource is not a clone (nor cloned). */ assert_null(native_find_rsc(inactive_group, "inactive-group", NULL, pe_find_clone)); assert_null(native_find_rsc(inactive_group, "inactive-group", cluster01, pe_find_clone)); /* Fails because none of inactive-group's children are running. */ assert_null(native_find_rsc(inactive_group, "inactive-group", cluster01, pe_find_current)); assert_null(native_find_rsc(inactive_group, "inactive-group", cluster02, pe_find_current)); /* Passes because of flags. */ assert_ptr_equal(inactive_group, native_find_rsc(inactive_group, "inactive-group", cluster01, pe_find_inactive)); /* Passes because of flags. */ assert_ptr_equal(inactive_group, native_find_rsc(inactive_group, "inactive-group", cluster02, pe_find_inactive)); } static void group_member_rsc(void **state) { pe_resource_t *public_ip = NULL; /* Find the "Public-IP" resource, a member of "exim-group". */ for (GList *iter = exim_group->children; iter != NULL; iter = iter->next) { pe_resource_t *rsc = (pe_resource_t *) iter->data; if (strcmp(rsc->id, "Public-IP") == 0) { public_ip = rsc; break; } } assert_non_null(public_ip); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(public_ip, native_find_rsc(public_ip, "Public-IP", NULL, 0)); assert_ptr_equal(public_ip, native_find_rsc(public_ip, "Public-IP", NULL, pe_find_current)); /* Fails because resource is not a clone (nor cloned). */ assert_null(native_find_rsc(public_ip, "Public-IP", NULL, pe_find_clone)); assert_null(native_find_rsc(public_ip, "Public-IP", cluster02, pe_find_clone)); /* Fails because Public-IP is not running on cluster01, even with the right flags. */ assert_null(native_find_rsc(public_ip, "Public-IP", cluster01, pe_find_current)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(public_ip, "Public-IP", cluster02, 0)); /* Passes because Public-IP is running on cluster02. */ assert_ptr_equal(public_ip, native_find_rsc(public_ip, "Public-IP", cluster02, pe_find_current)); } static void inactive_group_member_rsc(void **state) { pe_resource_t *inactive_dummy_1 = NULL; /* Find the "inactive-dummy-1" resource, a member of "inactive-group". */ for (GList *iter = inactive_group->children; iter != NULL; iter = iter->next) { pe_resource_t *rsc = (pe_resource_t *) iter->data; if (strcmp(rsc->id, "inactive-dummy-1") == 0) { inactive_dummy_1 = rsc; break; } } assert_non_null(inactive_dummy_1); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(inactive_dummy_1, native_find_rsc(inactive_dummy_1, "inactive-dummy-1", NULL, 0)); assert_ptr_equal(inactive_dummy_1, native_find_rsc(inactive_dummy_1, "inactive-dummy-1", NULL, pe_find_current)); /* Fails because resource is not a clone (nor cloned). */ assert_null(native_find_rsc(inactive_dummy_1, "inactive-dummy-1", NULL, pe_find_clone)); assert_null(native_find_rsc(inactive_dummy_1, "inactive-dummy-1", cluster01, pe_find_clone)); /* Fails because inactive-dummy-1 is not running. */ assert_null(native_find_rsc(inactive_dummy_1, "inactive-dummy-1", cluster01, pe_find_current)); assert_null(native_find_rsc(inactive_dummy_1, "inactive-dummy-1", cluster02, pe_find_current)); /* Passes because of flags. */ assert_ptr_equal(inactive_dummy_1, native_find_rsc(inactive_dummy_1, "inactive-dummy-1", cluster01, pe_find_inactive)); /* Passes because of flags. */ assert_ptr_equal(inactive_dummy_1, native_find_rsc(inactive_dummy_1, "inactive-dummy-1", cluster02, pe_find_inactive)); } static void clone_rsc(void **state) { assert_non_null(promotable_clone); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(promotable_clone, native_find_rsc(promotable_clone, "promotable-clone", NULL, 0)); assert_ptr_equal(promotable_clone, native_find_rsc(promotable_clone, "promotable-clone", NULL, pe_find_current)); assert_ptr_equal(promotable_clone, native_find_rsc(promotable_clone, "promotable-clone", NULL, pe_find_clone)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(promotable_clone, "promotable-clone", cluster01, 0)); /* Passes because one of ping-clone's children is running on cluster01. */ assert_ptr_equal(promotable_clone, native_find_rsc(promotable_clone, "promotable-clone", cluster01, pe_find_current)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(promotable_clone, "promotable-clone", cluster02, 0)); /* Passes because one of ping_clone's children is running on cluster02. */ assert_ptr_equal(promotable_clone, native_find_rsc(promotable_clone, "promotable-clone", cluster02, pe_find_current)); /* Passes for previous reasons, plus includes pe_find_clone check. */ assert_ptr_equal(promotable_clone, native_find_rsc(promotable_clone, "promotable-clone", cluster01, pe_find_clone|pe_find_current)); assert_ptr_equal(promotable_clone, native_find_rsc(promotable_clone, "promotable-clone", cluster02, pe_find_clone|pe_find_current)); } static void inactive_clone_rsc(void **state) { assert_non_null(inactive_clone); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(inactive_clone, native_find_rsc(inactive_clone, "inactive-clone", NULL, 0)); assert_ptr_equal(inactive_clone, native_find_rsc(inactive_clone, "inactive-clone", NULL, pe_find_current)); assert_ptr_equal(inactive_clone, native_find_rsc(inactive_clone, "inactive-clone", NULL, pe_find_clone)); assert_ptr_equal(inactive_clone, native_find_rsc(inactive_clone, "inactive-clone", NULL, pe_find_inactive)); /* Fails because none of inactive-clone's children are running. */ assert_null(native_find_rsc(inactive_clone, "inactive-clone", cluster01, pe_find_current|pe_find_clone)); assert_null(native_find_rsc(inactive_clone, "inactive-clone", cluster02, pe_find_current|pe_find_clone)); /* Passes because of flags. */ assert_ptr_equal(inactive_clone, native_find_rsc(inactive_clone, "inactive-clone", cluster01, pe_find_inactive)); /* Passes because of flags. */ assert_ptr_equal(inactive_clone, native_find_rsc(inactive_clone, "inactive-clone", cluster02, pe_find_inactive)); } static void clone_instance_rsc(void **state) { pe_resource_t *promotable_0 = NULL; pe_resource_t *promotable_1 = NULL; /* Find the "promotable-rsc:0" and "promotable-rsc:1" resources, members of "promotable-clone". */ for (GList *iter = promotable_clone->children; iter != NULL; iter = iter->next) { pe_resource_t *rsc = (pe_resource_t *) iter->data; if (strcmp(rsc->id, "promotable-rsc:0") == 0) { promotable_0 = rsc; } else if (strcmp(rsc->id, "promotable-rsc:1") == 0) { promotable_1 = rsc; } } assert_non_null(promotable_0); assert_non_null(promotable_1); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(promotable_0, native_find_rsc(promotable_0, "promotable-rsc:0", NULL, 0)); assert_ptr_equal(promotable_0, native_find_rsc(promotable_0, "promotable-rsc:0", NULL, pe_find_current)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_1, "promotable-rsc:1", NULL, 0)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_1, "promotable-rsc:1", NULL, pe_find_current)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(promotable_0, "promotable-rsc:0", cluster02, 0)); assert_null(native_find_rsc(promotable_1, "promotable-rsc:1", cluster01, 0)); /* Check that the resource is running on the node we expect. */ assert_ptr_equal(promotable_0, native_find_rsc(promotable_0, "promotable-rsc:0", cluster02, pe_find_current)); assert_null(native_find_rsc(promotable_0, "promotable-rsc:0", cluster01, pe_find_current)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_1, "promotable-rsc:1", cluster01, pe_find_current)); assert_null(native_find_rsc(promotable_1, "promotable-rsc:1", cluster02, pe_find_current)); /* Passes because NULL was passed for node and primitive name was given, with correct flags. */ assert_ptr_equal(promotable_0, native_find_rsc(promotable_0, "promotable-rsc", NULL, pe_find_clone)); /* Passes because pe_find_any matches any instance's base name. */ assert_ptr_equal(promotable_0, native_find_rsc(promotable_0, "promotable-rsc", NULL, pe_find_any)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_1, "promotable-rsc", NULL, pe_find_any)); /* Passes because pe_find_anon matches. */ assert_ptr_equal(promotable_0, native_find_rsc(promotable_0, "promotable-rsc", NULL, pe_find_anon)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_1, "promotable-rsc", NULL, pe_find_anon)); /* Check that the resource is running on the node we expect. */ assert_ptr_equal(promotable_0, native_find_rsc(promotable_0, "promotable-rsc", cluster02, pe_find_any|pe_find_current)); assert_ptr_equal(promotable_0, native_find_rsc(promotable_0, "promotable-rsc", cluster02, pe_find_anon|pe_find_current)); assert_null(native_find_rsc(promotable_0, "promotable-rsc", cluster01, pe_find_any|pe_find_current)); assert_null(native_find_rsc(promotable_0, "promotable-rsc", cluster01, pe_find_anon|pe_find_current)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_1, "promotable-rsc", cluster01, pe_find_any|pe_find_current)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_1, "promotable-rsc", cluster01, pe_find_anon|pe_find_current)); assert_null(native_find_rsc(promotable_1, "promotable-rsc", cluster02, pe_find_any|pe_find_current)); assert_null(native_find_rsc(promotable_1, "promotable-rsc", cluster02, pe_find_anon|pe_find_current)); /* Fails because incorrect flags were given along with primitive name. */ assert_null(native_find_rsc(promotable_0, "promotable-rsc", NULL, pe_find_current)); assert_null(native_find_rsc(promotable_1, "promotable-rsc", NULL, pe_find_current)); /* And then we check failure possibilities again, except passing promotable_clone * instead of promotable_X as the first argument to native_find_rsc. */ /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(promotable_clone, "promotable-rsc:0", cluster02, 0)); assert_null(native_find_rsc(promotable_clone, "promotable-rsc:1", cluster01, 0)); /* Check that the resource is running on the node we expect. */ assert_ptr_equal(promotable_0, native_find_rsc(promotable_clone, "promotable-rsc:0", cluster02, pe_find_current)); assert_ptr_equal(promotable_0, native_find_rsc(promotable_clone, "promotable-rsc", cluster02, pe_find_any|pe_find_current)); assert_ptr_equal(promotable_0, native_find_rsc(promotable_clone, "promotable-rsc", cluster02, pe_find_anon|pe_find_current)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_clone, "promotable-rsc:1", cluster01, pe_find_current)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_clone, "promotable-rsc", cluster01, pe_find_any|pe_find_current)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_clone, "promotable-rsc", cluster01, pe_find_anon|pe_find_current)); } static void renamed_rsc(void **state) { pe_resource_t *promotable_0 = NULL; pe_resource_t *promotable_1 = NULL; /* Find the "promotable-rsc:0" and "promotable-rsc:1" resources, members of "promotable-clone". */ for (GList *iter = promotable_clone->children; iter != NULL; iter = iter->next) { pe_resource_t *rsc = (pe_resource_t *) iter->data; if (strcmp(rsc->id, "promotable-rsc:0") == 0) { promotable_0 = rsc; } else if (strcmp(rsc->id, "promotable-rsc:1") == 0) { promotable_1 = rsc; } } assert_non_null(promotable_0); assert_non_null(promotable_1); /* Passes because pe_find_renamed means the base name matches clone_name. */ assert_ptr_equal(promotable_0, native_find_rsc(promotable_0, "promotable-rsc", NULL, pe_find_renamed)); assert_ptr_equal(promotable_1, native_find_rsc(promotable_1, "promotable-rsc", NULL, pe_find_renamed)); } static void bundle_rsc(void **state) { assert_non_null(httpd_bundle); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(httpd_bundle, native_find_rsc(httpd_bundle, "httpd-bundle", NULL, 0)); assert_ptr_equal(httpd_bundle, native_find_rsc(httpd_bundle, "httpd-bundle", NULL, pe_find_current)); /* Fails because resource is not a clone (nor cloned). */ assert_null(native_find_rsc(httpd_bundle, "httpd-bundle", NULL, pe_find_clone)); assert_null(native_find_rsc(httpd_bundle, "httpd-bundle", cluster01, pe_find_clone)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(httpd_bundle, "httpd-bundle", cluster01, 0)); /* Passes because one of httpd_bundle's children is running on cluster01. */ assert_ptr_equal(httpd_bundle, native_find_rsc(httpd_bundle, "httpd-bundle", cluster01, pe_find_current)); } -static void -bundle_replica_rsc(void **state) { - pe__bundle_variant_data_t *bundle_data = NULL; - pe__bundle_replica_t *replica_0 = NULL; - - pe_resource_t *ip_0 = NULL; - pe_resource_t *child_0 = NULL; - pe_resource_t *container_0 = NULL; - pe_resource_t *remote_0 = NULL; - - get_bundle_variant_data(bundle_data, httpd_bundle); - replica_0 = (pe__bundle_replica_t *) bundle_data->replicas->data; - - ip_0 = replica_0->ip; - child_0 = replica_0->child; - container_0 = replica_0->container; - remote_0 = replica_0->remote; +static bool +bundle_first_replica(pe__bundle_replica_t *replica, void *user_data) +{ + pe_resource_t *ip_0 = replica->ip; + pe_resource_t *child_0 = replica->child; + pe_resource_t *container_0 = replica->container; + pe_resource_t *remote_0 = replica->remote; assert_non_null(ip_0); assert_non_null(child_0); assert_non_null(container_0); assert_non_null(remote_0); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(ip_0, native_find_rsc(ip_0, "httpd-bundle-ip-192.168.122.131", NULL, 0)); assert_ptr_equal(child_0, native_find_rsc(child_0, "httpd:0", NULL, 0)); assert_ptr_equal(container_0, native_find_rsc(container_0, "httpd-bundle-docker-0", NULL, 0)); assert_ptr_equal(remote_0, native_find_rsc(remote_0, "httpd-bundle-0", NULL, 0)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(ip_0, "httpd-bundle-ip-192.168.122.131", cluster01, 0)); assert_null(native_find_rsc(child_0, "httpd:0", httpd_bundle_0, 0)); assert_null(native_find_rsc(container_0, "httpd-bundle-docker-0", cluster01, 0)); assert_null(native_find_rsc(remote_0, "httpd-bundle-0", cluster01, 0)); /* Check that the resource is running on the node we expect. */ assert_ptr_equal(ip_0, native_find_rsc(ip_0, "httpd-bundle-ip-192.168.122.131", cluster01, pe_find_current)); assert_null(native_find_rsc(ip_0, "httpd-bundle-ip-192.168.122.131", cluster02, pe_find_current)); assert_null(native_find_rsc(ip_0, "httpd-bundle-ip-192.168.122.131", httpd_bundle_0, pe_find_current)); assert_ptr_equal(child_0, native_find_rsc(child_0, "httpd:0", httpd_bundle_0, pe_find_current)); assert_null(native_find_rsc(child_0, "httpd:0", cluster01, pe_find_current)); assert_null(native_find_rsc(child_0, "httpd:0", cluster02, pe_find_current)); assert_ptr_equal(container_0, native_find_rsc(container_0, "httpd-bundle-docker-0", cluster01, pe_find_current)); assert_null(native_find_rsc(container_0, "httpd-bundle-docker-0", cluster02, pe_find_current)); assert_null(native_find_rsc(container_0, "httpd-bundle-docker-0", httpd_bundle_0, pe_find_current)); assert_ptr_equal(remote_0, native_find_rsc(remote_0, "httpd-bundle-0", cluster01, pe_find_current)); assert_null(native_find_rsc(remote_0, "httpd-bundle-0", cluster02, pe_find_current)); assert_null(native_find_rsc(remote_0, "httpd-bundle-0", httpd_bundle_0, pe_find_current)); /* Passes because pe_find_any matches any replica's base name. */ assert_ptr_equal(child_0, native_find_rsc(child_0, "httpd", NULL, pe_find_any)); /* Passes because pe_find_anon matches. */ assert_ptr_equal(child_0, native_find_rsc(child_0, "httpd", NULL, pe_find_anon)); /* Check that the resource is running on the node we expect. */ assert_ptr_equal(child_0, native_find_rsc(child_0, "httpd", httpd_bundle_0, pe_find_any|pe_find_current)); assert_ptr_equal(child_0, native_find_rsc(child_0, "httpd", httpd_bundle_0, pe_find_anon|pe_find_current)); assert_null(native_find_rsc(child_0, "httpd", cluster01, pe_find_any|pe_find_current)); assert_null(native_find_rsc(child_0, "httpd", cluster01, pe_find_anon|pe_find_current)); assert_null(native_find_rsc(child_0, "httpd", cluster02, pe_find_any|pe_find_current)); assert_null(native_find_rsc(child_0, "httpd", cluster02, pe_find_anon|pe_find_current)); /* Fails because incorrect flags were given along with base name. */ assert_null(native_find_rsc(child_0, "httpd", NULL, pe_find_current)); /* And then we check failure possibilities again, except passing httpd-bundle * instead of X_0 as the first argument to native_find_rsc. */ /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(httpd_bundle, "httpd-bundle-ip-192.168.122.131", cluster01, 0)); assert_null(native_find_rsc(httpd_bundle, "httpd:0", httpd_bundle_0, 0)); assert_null(native_find_rsc(httpd_bundle, "httpd-bundle-docker-0", cluster01, 0)); assert_null(native_find_rsc(httpd_bundle, "httpd-bundle-0", cluster01, 0)); /* Check that the resource is running on the node we expect. */ assert_ptr_equal(ip_0, native_find_rsc(httpd_bundle, "httpd-bundle-ip-192.168.122.131", cluster01, pe_find_current)); assert_ptr_equal(child_0, native_find_rsc(httpd_bundle, "httpd:0", httpd_bundle_0, pe_find_current)); assert_ptr_equal(container_0, native_find_rsc(httpd_bundle, "httpd-bundle-docker-0", cluster01, pe_find_current)); assert_ptr_equal(remote_0, native_find_rsc(httpd_bundle, "httpd-bundle-0", cluster01, pe_find_current)); + return false; // Do not iterate through any further replicas +} + +static void +bundle_replica_rsc(void **state) +{ + pe__foreach_bundle_replica(httpd_bundle, bundle_first_replica, NULL); } static void clone_group_rsc(void **rsc) { assert_non_null(mysql_clone_group); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(mysql_clone_group, native_find_rsc(mysql_clone_group, "mysql-clone-group", NULL, 0)); assert_ptr_equal(mysql_clone_group, native_find_rsc(mysql_clone_group, "mysql-clone-group", NULL, pe_find_current)); assert_ptr_equal(mysql_clone_group, native_find_rsc(mysql_clone_group, "mysql-clone-group", NULL, pe_find_clone)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(mysql_clone_group, "mysql-clone-group", cluster01, 0)); /* Passes because one of mysql-clone-group's children is running on cluster01. */ assert_ptr_equal(mysql_clone_group, native_find_rsc(mysql_clone_group, "mysql-clone-group", cluster01, pe_find_current)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(mysql_clone_group, "mysql-clone-group", cluster02, 0)); /* Passes because one of mysql-clone-group's children is running on cluster02. */ assert_ptr_equal(mysql_clone_group, native_find_rsc(mysql_clone_group, "mysql-clone-group", cluster02, pe_find_current)); /* Passes for previous reasons, plus includes pe_find_clone check. */ assert_ptr_equal(mysql_clone_group, native_find_rsc(mysql_clone_group, "mysql-clone-group", cluster01, pe_find_clone|pe_find_current)); assert_ptr_equal(mysql_clone_group, native_find_rsc(mysql_clone_group, "mysql-clone-group", cluster02, pe_find_clone|pe_find_current)); } static void clone_group_instance_rsc(void **rsc) { pe_resource_t *mysql_group_0 = NULL; pe_resource_t *mysql_group_1 = NULL; /* Find the "mysql-group:0" and "mysql-group:1" resources, members of "mysql-clone-group". */ for (GList *iter = mysql_clone_group->children; iter != NULL; iter = iter->next) { pe_resource_t *rsc = (pe_resource_t *) iter->data; if (strcmp(rsc->id, "mysql-group:0") == 0) { mysql_group_0 = rsc; } else if (strcmp(rsc->id, "mysql-group:1") == 0) { mysql_group_1 = rsc; } } assert_non_null(mysql_group_0); assert_non_null(mysql_group_1); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_group_0, "mysql-group:0", NULL, 0)); assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_group_0, "mysql-group:0", NULL, pe_find_current)); assert_ptr_equal(mysql_group_1, native_find_rsc(mysql_group_1, "mysql-group:1", NULL, 0)); assert_ptr_equal(mysql_group_1, native_find_rsc(mysql_group_1, "mysql-group:1", NULL, pe_find_current)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(mysql_group_0, "mysql-group:0", cluster02, 0)); assert_null(native_find_rsc(mysql_group_1, "mysql-group:1", cluster01, 0)); /* Check that the resource is running on the node we expect. */ assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_group_0, "mysql-group:0", cluster02, pe_find_current)); assert_null(native_find_rsc(mysql_group_0, "mysql-group:0", cluster01, pe_find_current)); assert_ptr_equal(mysql_group_1, native_find_rsc(mysql_group_1, "mysql-group:1", cluster01, pe_find_current)); assert_null(native_find_rsc(mysql_group_1, "mysql-group:1", cluster02, pe_find_current)); /* Passes because NULL was passed for node and base name was given, with correct flags. */ assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_group_0, "mysql-group" , NULL, pe_find_clone)); /* Passes because pe_find_any matches any base name. */ assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_group_0, "mysql-group" , NULL, pe_find_any)); assert_ptr_equal(mysql_group_1, native_find_rsc(mysql_group_1, "mysql-group" , NULL, pe_find_any)); /* Passes because pe_find_anon matches. */ assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_group_0, "mysql-group" , NULL, pe_find_anon)); assert_ptr_equal(mysql_group_1, native_find_rsc(mysql_group_1, "mysql-group" , NULL, pe_find_anon)); /* Check that the resource is running on the node we expect. */ assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_group_0, "mysql-group", cluster02, pe_find_any|pe_find_current)); assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_group_0, "mysql-group", cluster02, pe_find_anon|pe_find_current)); assert_null(native_find_rsc(mysql_group_0, "mysql-group", cluster01, pe_find_any|pe_find_current)); assert_null(native_find_rsc(mysql_group_0, "mysql-group", cluster01, pe_find_anon|pe_find_current)); assert_ptr_equal(mysql_group_1, native_find_rsc(mysql_group_1, "mysql-group", cluster01, pe_find_any|pe_find_current)); assert_ptr_equal(mysql_group_1, native_find_rsc(mysql_group_1, "mysql-group", cluster01, pe_find_anon|pe_find_current)); assert_null(native_find_rsc(mysql_group_1, "mysql-group", cluster02, pe_find_any|pe_find_current)); assert_null(native_find_rsc(mysql_group_1, "mysql-group", cluster02, pe_find_anon|pe_find_current)); /* Fails because incorrect flags were given along with base name. */ assert_null(native_find_rsc(mysql_group_0, "mysql-group", NULL, pe_find_current)); assert_null(native_find_rsc(mysql_group_1, "mysql-group", NULL, pe_find_current)); /* And then we check failure possibilities again, except passing mysql_clone_group * instead of mysql_group_X as the first argument to native_find_rsc. */ /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(mysql_clone_group, "mysql-group:0", cluster02, 0)); assert_null(native_find_rsc(mysql_clone_group, "mysql-group:1", cluster01, 0)); /* Check that the resource is running on the node we expect. */ assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_clone_group, "mysql-group:0", cluster02, pe_find_current)); assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_clone_group, "mysql-group", cluster02, pe_find_any|pe_find_current)); assert_ptr_equal(mysql_group_0, native_find_rsc(mysql_clone_group, "mysql-group", cluster02, pe_find_anon|pe_find_current)); assert_ptr_equal(mysql_group_1, native_find_rsc(mysql_clone_group, "mysql-group:1", cluster01, pe_find_current)); assert_ptr_equal(mysql_group_1, native_find_rsc(mysql_clone_group, "mysql-group", cluster01, pe_find_any|pe_find_current)); assert_ptr_equal(mysql_group_1, native_find_rsc(mysql_clone_group, "mysql-group", cluster01, pe_find_anon|pe_find_current)); } static void clone_group_member_rsc(void **state) { pe_resource_t *mysql_proxy = NULL; /* Find the "mysql-proxy" resource, a member of "mysql-group". */ for (GList *iter = mysql_clone_group->children; iter != NULL; iter = iter->next) { pe_resource_t *rsc = (pe_resource_t *) iter->data; if (strcmp(rsc->id, "mysql-group:0") == 0) { for (GList *iter2 = rsc->children; iter2 != NULL; iter2 = iter2->next) { pe_resource_t *child = (pe_resource_t *) iter2->data; if (strcmp(child->id, "mysql-proxy:0") == 0) { mysql_proxy = child; break; } } break; } } assert_non_null(mysql_proxy); /* Passes because NULL was passed for node, regardless of flags. */ assert_ptr_equal(mysql_proxy, native_find_rsc(mysql_proxy, "mysql-proxy:0", NULL, 0)); assert_ptr_equal(mysql_proxy, native_find_rsc(mysql_proxy, "mysql-proxy:0", NULL, pe_find_current)); /* Passes because resource's parent is a clone. */ assert_ptr_equal(mysql_proxy, native_find_rsc(mysql_proxy, "mysql-proxy:0", NULL, pe_find_clone)); assert_ptr_equal(mysql_proxy, native_find_rsc(mysql_proxy, "mysql-proxy:0", cluster02, pe_find_clone|pe_find_current)); /* Fails because mysql-proxy:0 is not running on cluster01, even with the right flags. */ assert_null(native_find_rsc(mysql_proxy, "mysql-proxy:0", cluster01, pe_find_current)); /* Fails because pe_find_current is required if a node is given. */ assert_null(native_find_rsc(mysql_proxy, "mysql-proxy:0", cluster02, 0)); /* Passes because mysql-proxy:0 is running on cluster02. */ assert_ptr_equal(mysql_proxy, native_find_rsc(mysql_proxy, "mysql-proxy:0", cluster02, pe_find_current)); } /* TODO: Add tests for finding on allocated node (passing a node without * pe_find_current, after scheduling, for a resource that is starting/stopping/moving. */ PCMK__UNIT_TEST(setup, teardown, cmocka_unit_test(bad_args), cmocka_unit_test(primitive_rsc), cmocka_unit_test(group_rsc), cmocka_unit_test(inactive_group_rsc), cmocka_unit_test(group_member_rsc), cmocka_unit_test(inactive_group_member_rsc), cmocka_unit_test(clone_rsc), cmocka_unit_test(inactive_clone_rsc), cmocka_unit_test(clone_instance_rsc), cmocka_unit_test(renamed_rsc), cmocka_unit_test(bundle_rsc), cmocka_unit_test(bundle_replica_rsc), cmocka_unit_test(clone_group_rsc), cmocka_unit_test(clone_group_instance_rsc), cmocka_unit_test(clone_group_member_rsc)) diff --git a/lib/pengine/variant.h b/lib/pengine/variant.h deleted file mode 100644 index daa3781420..0000000000 --- a/lib/pengine/variant.h +++ /dev/null @@ -1,91 +0,0 @@ -/* - * Copyright 2004-2022 the Pacemaker project contributors - * - * The version control history for this file may have further details. - * - * This source code is licensed under the GNU Lesser General Public License - * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. - */ - -#ifndef PE_VARIANT__H -# define PE_VARIANT__H - -# if PE__VARIANT_BUNDLE - -typedef struct { - int offset; - char *ipaddr; - pe_node_t *node; - pe_resource_t *ip; - pe_resource_t *child; - pe_resource_t *container; - pe_resource_t *remote; -} pe__bundle_replica_t; - -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; - - pe_resource_t *child; - - GList *replicas; // pe__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) \ - CRM_ASSERT(rsc != NULL); \ - CRM_ASSERT(rsc->variant == pe_container); \ - CRM_ASSERT(rsc->variant_opaque != NULL); \ - data = (pe__bundle_variant_data_t *)rsc->variant_opaque; \ - -# endif - -#endif