diff --git a/daemons/attrd/pacemaker-attrd.h b/daemons/attrd/pacemaker-attrd.h index f3e7e4ff33..be22629409 100644 --- a/daemons/attrd/pacemaker-attrd.h +++ b/daemons/attrd/pacemaker-attrd.h @@ -1,252 +1,246 @@ /* * Copyright 2013-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #ifndef PACEMAKER_ATTRD__H # define PACEMAKER_ATTRD__H #include #include #include #include #include #include #include /* * Legacy attrd (all pre-1.1.11 Pacemaker versions, plus all versions when used * with the no-longer-supported CMAN or corosync-plugin stacks) is unversioned. * * With atomic attrd, each attrd will send ATTRD_PROTOCOL_VERSION with every * peer request and reply. As of Pacemaker 2.0.0, at start-up each attrd will * also set a private attribute for itself with its version, so any attrd can * determine the minimum version supported by all peers. * * Protocol Pacemaker Significant changes * -------- --------- ------------------- * 1 1.1.11 PCMK__ATTRD_CMD_UPDATE (PCMK__XA_ATTR_NAME only), * PCMK__ATTRD_CMD_PEER_REMOVE, PCMK__ATTRD_CMD_REFRESH, * "flush", PCMK__ATTRD_CMD_SYNC_RESPONSE * 1 1.1.13 PCMK__ATTRD_CMD_UPDATE (with PCMK__XA_ATTR_REGEX), * PCMK__ATTRD_CMD_QUERY * 1 1.1.15 PCMK__ATTRD_CMD_UPDATE_BOTH, * PCMK__ATTRD_CMD_UPDATE_DELAY * 2 1.1.17 PCMK__ATTRD_CMD_CLEAR_FAILURE * 3 2.1.1 PCMK__ATTRD_CMD_SYNC_RESPONSE indicates remote nodes * 4 2.1.5 Multiple attributes can be updated in a single IPC * message * 5 2.1.5 Peers can request confirmation of a sent message * 6 2.1.7 PCMK__ATTRD_CMD_PEER_REMOVE supports PCMK__XA_REAP * 7 3.0.0 "flush" support dropped */ #define ATTRD_PROTOCOL_VERSION "7" #define ATTRD_SUPPORTS_MULTI_MESSAGE(x) ((x) >= 4) #define ATTRD_SUPPORTS_CONFIRMATION(x) ((x) >= 5) #define attrd_send_ack(client, id, flags) \ pcmk__ipc_send_ack((client), (id), (flags), PCMK__XE_ACK, \ ATTRD_PROTOCOL_VERSION, CRM_EX_INDETERMINATE) void attrd_init_mainloop(void); void attrd_run_mainloop(void); void attrd_set_requesting_shutdown(void); void attrd_clear_requesting_shutdown(void); void attrd_free_waitlist(void); bool attrd_shutting_down(bool if_requested); void attrd_shutdown(int nsig); void attrd_init_ipc(void); void attrd_ipc_fini(void); int attrd_cib_connect(int max_retry); void attrd_cib_disconnect(void); void attrd_cib_init(void); void attrd_cib_erase_transient_attrs(const char *node); bool attrd_value_needs_expansion(const char *value); int attrd_expand_value(const char *value, const char *old_value); /* regular expression to clear failures of all resources */ #define ATTRD_RE_CLEAR_ALL \ "^(" PCMK__FAIL_COUNT_PREFIX "|" PCMK__LAST_FAILURE_PREFIX ")-" /* regular expression to clear failure of all operations for one resource * (format takes resource name) - * - * @COMPAT attributes set < 1.1.17: - * also match older attributes that do not have the operation part */ -#define ATTRD_RE_CLEAR_ONE ATTRD_RE_CLEAR_ALL "%s(#.+_[0-9]+)?$" +#define ATTRD_RE_CLEAR_ONE ATTRD_RE_CLEAR_ALL "%s#.+_[0-9]+$" /* regular expression to clear failure of one operation for one resource * (format takes resource name, operation name, and interval) - * - * @COMPAT attributes set < 1.1.17: - * also match older attributes that do not have the operation part */ -#define ATTRD_RE_CLEAR_OP ATTRD_RE_CLEAR_ALL "%s(#%s_%u)?$" +#define ATTRD_RE_CLEAR_OP ATTRD_RE_CLEAR_ALL "%s#%s_%u$" int attrd_failure_regex(regex_t *regex, const char *rsc, const char *op, guint interval_ms); extern cib_t *the_cib; extern crm_exit_t attrd_exit_status; /* Alerts */ extern lrmd_t *the_lrmd; extern crm_trigger_t *attrd_config_read; void attrd_lrmd_disconnect(void); gboolean attrd_read_options(gpointer user_data); int attrd_send_attribute_alert(const char *node, int nodeid, const char *attr, const char *value); // Elections void attrd_election_init(void); void attrd_election_fini(void); void attrd_start_election_if_needed(void); bool attrd_election_won(void); void attrd_handle_election_op(const crm_node_t *peer, xmlNode *xml); bool attrd_check_for_new_writer(const crm_node_t *peer, const xmlNode *xml); void attrd_declare_winner(void); void attrd_remove_voter(const crm_node_t *peer); void attrd_xml_add_writer(xmlNode *xml); enum attrd_attr_flags { attrd_attr_none = 0U, attrd_attr_changed = (1U << 0), // Attribute value has changed since last write attrd_attr_uuid_missing = (1U << 1), // Whether we know we're missing a peer UUID attrd_attr_is_private = (1U << 2), // Whether to keep this attribute out of the CIB attrd_attr_force_write = (1U << 3), // Update attribute by ignoring delay }; typedef struct attribute_s { char *id; // Attribute name char *set_type; // PCMK_XE_INSTANCE_ATTRIBUTES or PCMK_XE_UTILIZATION char *set_id; // Set's XML ID to use when writing char *user; // ACL user to use for CIB writes int update; // Call ID of pending write int timeout_ms; // How long to wait for more changes before writing uint32_t flags; // Group of enum attrd_attr_flags GHashTable *values; // Key: node name, value: attribute_value_t mainloop_timer_t *timer; // Timer to use for timeout_ms } attribute_t; #define attrd_set_attr_flags(attr, flags_to_set) do { \ (attr)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Value for attribute", (attr)->id, \ (attr)->flags, (flags_to_set), #flags_to_set); \ } while (0) #define attrd_clear_attr_flags(attr, flags_to_clear) do { \ (attr)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Value for attribute", (attr)->id, \ (attr)->flags, (flags_to_clear), #flags_to_clear); \ } while (0) enum attrd_value_flags { attrd_value_none = 0U, attrd_value_remote = (1U << 0), // Value is for Pacemaker Remote node attrd_value_from_peer = (1U << 1), // Value is from peer sync response }; typedef struct attribute_value_s { char *nodename; // Node that this value is for char *current; // Attribute value char *requested; // Value specified in pending CIB write, if any uint32_t nodeid; // Cluster node ID of node that this value is for uint32_t flags; // Group of attrd_value_flags } attribute_value_t; #define attrd_set_value_flags(attr_value, flags_to_set) do { \ (attr_value)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Value for node", (attr_value)->nodename, \ (attr_value)->flags, (flags_to_set), #flags_to_set); \ } while (0) #define attrd_clear_value_flags(attr_value, flags_to_clear) do { \ (attr_value)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Value for node", (attr_value)->nodename, \ (attr_value)->flags, (flags_to_clear), #flags_to_clear); \ } while (0) extern pcmk_cluster_t *attrd_cluster; extern GHashTable *attributes; extern GHashTable *peer_protocol_vers; #define CIB_OP_TIMEOUT_S 120 int attrd_cluster_connect(void); void attrd_broadcast_value(const attribute_t *a, const attribute_value_t *v); void attrd_peer_update(const crm_node_t *peer, xmlNode *xml, const char *host, bool filter); void attrd_peer_sync(crm_node_t *peer); void attrd_peer_remove(const char *host, bool uncache, const char *source); void attrd_peer_clear_failure(pcmk__request_t *request); void attrd_peer_sync_response(const crm_node_t *peer, bool peer_won, xmlNode *xml); void attrd_broadcast_protocol(void); xmlNode *attrd_client_peer_remove(pcmk__request_t *request); xmlNode *attrd_client_clear_failure(pcmk__request_t *request); xmlNode *attrd_client_update(pcmk__request_t *request); xmlNode *attrd_client_refresh(pcmk__request_t *request); xmlNode *attrd_client_query(pcmk__request_t *request); gboolean attrd_send_message(crm_node_t *node, xmlNode *data, bool confirm); xmlNode *attrd_add_value_xml(xmlNode *parent, const attribute_t *a, const attribute_value_t *v, bool force_write); void attrd_clear_value_seen(void); void attrd_free_attribute(gpointer data); void attrd_free_attribute_value(gpointer data); attribute_t *attrd_populate_attribute(xmlNode *xml, const char *attr); char *attrd_set_id(const attribute_t *attr, const char *node_state_id); char *attrd_nvpair_id(const attribute_t *attr, const char *node_state_id); enum attrd_write_options { attrd_write_changed = 0, attrd_write_all = (1 << 0), attrd_write_no_delay = (1 << 1), }; void attrd_write_attributes(uint32_t options); void attrd_write_or_elect_attribute(attribute_t *a); extern int minimum_protocol_version; void attrd_remove_peer_protocol_ver(const char *host); void attrd_update_minimum_protocol_ver(const char *host, const char *value); mainloop_timer_t *attrd_add_timer(const char *id, int timeout_ms, attribute_t *attr); void attrd_unregister_handlers(void); void attrd_handle_request(pcmk__request_t *request); enum attrd_sync_point { attrd_sync_point_local, attrd_sync_point_cluster, }; typedef int (*attrd_confirmation_action_fn)(xmlNode *); void attrd_add_client_to_waitlist(pcmk__request_t *request); void attrd_ack_waitlist_clients(enum attrd_sync_point sync_point, const xmlNode *xml); int attrd_cluster_sync_point_update(xmlNode *xml); void attrd_do_not_expect_from_peer(const char *host); void attrd_do_not_wait_for_client(pcmk__client_t *client); void attrd_expect_confirmations(pcmk__request_t *request, attrd_confirmation_action_fn fn); void attrd_free_confirmations(void); void attrd_handle_confirmation(int callid, const char *host); void attrd_remove_client_from_waitlist(pcmk__client_t *client); const char *attrd_request_sync_point(xmlNode *xml); bool attrd_request_has_sync_point(xmlNode *xml); extern gboolean stand_alone; #endif /* PACEMAKER_ATTRD__H */ diff --git a/include/crm/crm.h b/include/crm/crm.h index 42cf30cb69..b53331362f 100644 --- a/include/crm/crm.h +++ b/include/crm/crm.h @@ -1,160 +1,159 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_CRM__H # define PCMK__CRM_CRM__H # include # include # include # include # include # include #include #ifdef __cplusplus extern "C" { #endif /** * \file * \brief A dumping ground * \ingroup core */ #ifndef PCMK_ALLOW_DEPRECATED /*! * \brief Allow use of deprecated Pacemaker APIs * * By default, external code using Pacemaker headers is allowed to use * deprecated Pacemaker APIs. If PCMK_ALLOW_DEPRECATED is defined to 0 before * including any Pacemaker headers, deprecated APIs will be unusable. It is * strongly recommended to leave this unchanged for production and release * builds, to avoid breakage when users upgrade to new Pacemaker releases that * deprecate more APIs. This should be defined to 0 only for development and * testing builds when desiring to check for usage of currently deprecated APIs. */ #define PCMK_ALLOW_DEPRECATED 1 #endif /*! * The CRM feature set assists with compatibility in mixed-version clusters. * The major version number increases when nodes with different versions * would not work (rolling upgrades are not allowed). The minor version * number increases when mixed-version clusters are allowed only during * rolling upgrades (a node with the oldest feature set will be elected DC). The * minor-minor version number is ignored, but allows resource agents to detect * cluster support for various features. * * The feature set also affects the processing of old saved CIBs (such as for * many scheduler regression tests). * * Particular feature points currently tested by Pacemaker code: * * >2.1: Operation updates include timing data * >=3.0.5: XML v2 digests are created * >=3.0.8: Peers do not need acks for cancellations * >=3.0.9: DC will send its own shutdown request to all peers - * >=3.0.13: Fail counts include operation name and interval * >=3.2.0: DC supports PCMK_EXEC_INVALID and PCMK_EXEC_NOT_CONNECTED * >=3.19.0: DC supports PCMK__CIB_REQUEST_COMMIT_TRANSACT */ # define CRM_FEATURE_SET "3.19.7" /* Pacemaker's CPG protocols use fixed-width binary fields for the sender and * recipient of a CPG message. This imposes an arbitrary limit on cluster node * names. */ //! \brief Maximum length of a Corosync cluster node name (in bytes) #define MAX_NAME 256 # define CRM_META "CRM_meta" // NOTE: sbd (as of at least 1.5.2) uses this extern char *crm_system_name; /* Sub-systems */ # define CRM_SYSTEM_DC "dc" #define CRM_SYSTEM_DCIB "dcib" // Primary instance of CIB manager # define CRM_SYSTEM_CIB "cib" # define CRM_SYSTEM_CRMD "crmd" # define CRM_SYSTEM_LRMD "lrmd" # define CRM_SYSTEM_PENGINE "pengine" # define CRM_SYSTEM_TENGINE "tengine" # define CRM_SYSTEM_STONITHD "stonithd" # define CRM_SYSTEM_MCP "pacemakerd" // Names of internally generated node attributes // @TODO Replace these with PCMK_NODE_ATTR_* # define CRM_ATTR_UNAME "#uname" # define CRM_ATTR_ID "#id" # define CRM_ATTR_KIND "#kind" # define CRM_ATTR_ROLE "#role" # define CRM_ATTR_IS_DC "#is_dc" # define CRM_ATTR_CLUSTER_NAME "#cluster-name" # define CRM_ATTR_SITE_NAME "#site-name" # define CRM_ATTR_UNFENCED "#node-unfenced" # define CRM_ATTR_DIGESTS_ALL "#digests-all" # define CRM_ATTR_DIGESTS_SECURE "#digests-secure" # define CRM_ATTR_PROTOCOL "#attrd-protocol" # define CRM_ATTR_FEATURE_SET "#feature-set" /* Valid operations */ # define CRM_OP_NOOP "noop" # define CRM_OP_JOIN_ANNOUNCE "join_announce" # define CRM_OP_JOIN_OFFER "join_offer" # define CRM_OP_JOIN_REQUEST "join_request" # define CRM_OP_JOIN_ACKNAK "join_ack_nack" # define CRM_OP_JOIN_CONFIRM "join_confirm" # define CRM_OP_PING "ping" # define CRM_OP_NODE_INFO "node-info" # define CRM_OP_THROTTLE "throttle" # define CRM_OP_VOTE "vote" # define CRM_OP_NOVOTE "no-vote" # define CRM_OP_HELLO "hello" # define CRM_OP_PECALC "pe_calc" # define CRM_OP_QUIT "quit" # define CRM_OP_SHUTDOWN_REQ "req_shutdown" # define CRM_OP_SHUTDOWN PCMK_ACTION_DO_SHUTDOWN # define CRM_OP_REGISTER "register" # define CRM_OP_IPC_FWD "ipc_fwd" # define CRM_OP_INVOKE_LRM "lrm_invoke" # define CRM_OP_LRM_REFRESH "lrm_refresh" //!< Deprecated since 1.1.10 # define CRM_OP_LRM_DELETE PCMK_ACTION_LRM_DELETE # define CRM_OP_LRM_FAIL "lrm_fail" # define CRM_OP_PROBED "probe_complete" # define CRM_OP_REPROBE "probe_again" # define CRM_OP_CLEAR_FAILCOUNT PCMK_ACTION_CLEAR_FAILCOUNT # define CRM_OP_REMOTE_STATE "remote_state" # define CRM_OP_RM_NODE_CACHE "rm_node_cache" # define CRM_OP_MAINTENANCE_NODES PCMK_ACTION_MAINTENANCE_NODES /* Possible cluster membership states */ # define CRMD_JOINSTATE_DOWN "down" # define CRMD_JOINSTATE_PENDING "pending" # define CRMD_JOINSTATE_MEMBER "member" # define CRMD_JOINSTATE_NACK "banned" # include # include # include # include #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) #include #endif #ifdef __cplusplus } #endif #endif diff --git a/lib/pengine/failcounts.c b/lib/pengine/failcounts.c index eef38d58db..1d25e8e499 100644 --- a/lib/pengine/failcounts.c +++ b/lib/pengine/failcounts.c @@ -1,480 +1,469 @@ /* * Copyright 2008-2024 the Pacemaker project contributors * * 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 static gboolean is_matched_failure(const char *rsc_id, const xmlNode *conf_op_xml, const xmlNode *lrm_op_xml) { gboolean matched = FALSE; const char *conf_op_name = NULL; const char *lrm_op_task = NULL; const char *conf_op_interval_spec = NULL; guint conf_op_interval_ms = 0; guint lrm_op_interval_ms = 0; const char *lrm_op_id = NULL; char *last_failure_key = NULL; if (rsc_id == NULL || conf_op_xml == NULL || lrm_op_xml == NULL) { return FALSE; } // Get name and interval from configured op conf_op_name = crm_element_value(conf_op_xml, PCMK_XA_NAME); conf_op_interval_spec = crm_element_value(conf_op_xml, PCMK_META_INTERVAL); pcmk_parse_interval_spec(conf_op_interval_spec, &conf_op_interval_ms); // Get name and interval from op history entry lrm_op_task = crm_element_value(lrm_op_xml, PCMK_XA_OPERATION); crm_element_value_ms(lrm_op_xml, PCMK_META_INTERVAL, &lrm_op_interval_ms); if ((conf_op_interval_ms != lrm_op_interval_ms) || !pcmk__str_eq(conf_op_name, lrm_op_task, pcmk__str_casei)) { return FALSE; } lrm_op_id = pcmk__xe_id(lrm_op_xml); last_failure_key = pcmk__op_key(rsc_id, "last_failure", 0); if (pcmk__str_eq(last_failure_key, lrm_op_id, pcmk__str_casei)) { matched = TRUE; } else { char *expected_op_key = pcmk__op_key(rsc_id, conf_op_name, conf_op_interval_ms); if (pcmk__str_eq(expected_op_key, lrm_op_id, pcmk__str_casei)) { int rc = 0; int target_rc = pe__target_rc_from_xml(lrm_op_xml); crm_element_value_int(lrm_op_xml, PCMK__XA_RC_CODE, &rc); if (rc != target_rc) { matched = TRUE; } } free(expected_op_key); } free(last_failure_key); return matched; } static gboolean block_failure(const pcmk_node_t *node, pcmk_resource_t *rsc, const xmlNode *xml_op) { char *xml_name = clone_strip(rsc->id); /* @TODO This xpath search occurs after template expansion, but it is unable * to properly detect on-fail in id-ref, operation meta-attributes, or * op_defaults, or evaluate rules. * * Also, PCMK_META_ON_FAIL defaults to PCMK_VALUE_BLOCK (in * unpack_operation()) for stop actions when stonith is disabled. * * Ideally, we'd unpack the operation before this point, and pass in a * meta-attributes table that takes all that into consideration. */ char *xpath = crm_strdup_printf("//" PCMK_XE_PRIMITIVE "[@" PCMK_XA_ID "='%s']" "//" PCMK_XE_OP "[@" PCMK_META_ON_FAIL "='" PCMK_VALUE_BLOCK "']", xml_name); xmlXPathObject *xpathObj = xpath_search(rsc->private->xml, xpath); gboolean should_block = FALSE; free(xpath); if (xpathObj) { int max = numXpathResults(xpathObj); int lpc = 0; for (lpc = 0; lpc < max; lpc++) { xmlNode *pref = getXpathResult(xpathObj, lpc); if (xml_op) { should_block = is_matched_failure(xml_name, pref, xml_op); if (should_block) { break; } } else { const char *conf_op_name = NULL; const char *conf_op_interval_spec = NULL; guint conf_op_interval_ms = 0; char *lrm_op_xpath = NULL; xmlXPathObject *lrm_op_xpathObj = NULL; // Get name and interval from configured op conf_op_name = crm_element_value(pref, PCMK_XA_NAME); conf_op_interval_spec = crm_element_value(pref, PCMK_META_INTERVAL); pcmk_parse_interval_spec(conf_op_interval_spec, &conf_op_interval_ms); #define XPATH_FMT "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" \ "//" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" \ "/" PCMK__XE_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='%s']" \ "[@" PCMK_META_INTERVAL "='%u']" lrm_op_xpath = crm_strdup_printf(XPATH_FMT, node->private->name, xml_name, conf_op_name, conf_op_interval_ms); lrm_op_xpathObj = xpath_search(rsc->private->scheduler->input, lrm_op_xpath); free(lrm_op_xpath); if (lrm_op_xpathObj) { int max2 = numXpathResults(lrm_op_xpathObj); int lpc2 = 0; for (lpc2 = 0; lpc2 < max2; lpc2++) { xmlNode *lrm_op_xml = getXpathResult(lrm_op_xpathObj, lpc2); should_block = is_matched_failure(xml_name, pref, lrm_op_xml); if (should_block) { break; } } } freeXpathObject(lrm_op_xpathObj); if (should_block) { break; } } } } free(xml_name); freeXpathObject(xpathObj); return should_block; } /*! * \internal * \brief Get resource name as used in failure-related node attributes * * \param[in] rsc Resource to check * * \return Newly allocated string containing resource's fail name * \note The caller is responsible for freeing the result. */ static inline char * rsc_fail_name(const pcmk_resource_t *rsc) { const char *name = pcmk__s(rsc->private->history_id, rsc->id); return pcmk_is_set(rsc->flags, pcmk__rsc_unique)? strdup(name) : clone_strip(name); } /*! * \internal * \brief Compile regular expression to match a failure-related node attribute * * \param[in] prefix Attribute prefix to match * \param[in] rsc_name Resource name to match as used in failure attributes - * \param[in] is_legacy Whether DC uses per-resource fail counts * \param[in] is_unique Whether the resource is a globally unique clone * \param[out] re Where to store resulting regular expression * * \return Standard Pacemaker return code * \note Fail attributes are named like PREFIX-RESOURCE#OP_INTERVAL. * The caller is responsible for freeing re with regfree(). */ static int -generate_fail_regex(const char *prefix, const char *rsc_name, - gboolean is_legacy, gboolean is_unique, regex_t *re) +generate_fail_regex(const char *prefix, const char *rsc_name, bool is_unique, + regex_t *re) { - char *pattern; - - /* @COMPAT DC < 1.1.17: Fail counts used to be per-resource rather than - * per-operation. - */ - const char *op_pattern = (is_legacy? "" : "#.+_[0-9]+"); + char *pattern = NULL; + const char *op_pattern = "#.+_[0-9]+"; /* Ignore instance numbers for anything other than globally unique clones. * Anonymous clone fail counts could contain an instance number if the * clone was initially unique, failed, then was converted to anonymous. * @COMPAT Also, before 1.1.8, anonymous clone fail counts always contained * clone instance numbers. */ const char *instance_pattern = (is_unique? "" : "(:[0-9]+)?"); pattern = crm_strdup_printf("^%s-%s%s%s$", prefix, rsc_name, instance_pattern, op_pattern); if (regcomp(re, pattern, REG_EXTENDED|REG_NOSUB) != 0) { free(pattern); return EINVAL; } free(pattern); return pcmk_rc_ok; } /*! * \internal * \brief Compile regular expressions to match failure-related node attributes * * \param[in] rsc Resource being checked for failures * \param[out] failcount_re Storage for regular expression for fail count * \param[out] lastfailure_re Storage for regular expression for last failure * * \return Standard Pacemaker return code * \note On success, the caller is responsible for freeing the expressions with * regfree(). */ static int -generate_fail_regexes(const pcmk_resource_t *rsc, - regex_t *failcount_re, regex_t *lastfailure_re) +generate_fail_regexes(const pcmk_resource_t *rsc, regex_t *failcount_re, + regex_t *lastfailure_re) { int rc = pcmk_rc_ok; char *rsc_name = rsc_fail_name(rsc); - const char *version = crm_element_value(rsc->private->scheduler->input, - PCMK_XA_CRM_FEATURE_SET); - - // @COMPAT Pacemaker <= 1.1.16 used a single fail count per resource - gboolean is_legacy = (compare_version(version, "3.0.13") < 0); - if (generate_fail_regex(PCMK__FAIL_COUNT_PREFIX, rsc_name, is_legacy, + if (generate_fail_regex(PCMK__FAIL_COUNT_PREFIX, rsc_name, pcmk_is_set(rsc->flags, pcmk__rsc_unique), failcount_re) != pcmk_rc_ok) { rc = EINVAL; } else if (generate_fail_regex(PCMK__LAST_FAILURE_PREFIX, rsc_name, - is_legacy, pcmk_is_set(rsc->flags, pcmk__rsc_unique), lastfailure_re) != pcmk_rc_ok) { rc = EINVAL; regfree(failcount_re); } free(rsc_name); return rc; } // Data for fail-count-related iterators struct failcount_data { const pcmk_node_t *node;// Node to check for fail count pcmk_resource_t *rsc; // Resource to check for fail count uint32_t flags; // Fail count flags const xmlNode *xml_op; // History entry for expiration purposes (or NULL) regex_t failcount_re; // Fail count regular expression to match regex_t lastfailure_re; // Last failure regular expression to match int failcount; // Fail count so far time_t last_failure; // Time of most recent failure so far }; /*! * \internal * \brief Update fail count and last failure appropriately for a node attribute * * \param[in] key Node attribute name * \param[in] value Node attribute value * \param[in] user_data Fail count data to update */ static void update_failcount_for_attr(gpointer key, gpointer value, gpointer user_data) { struct failcount_data *fc_data = user_data; // If this is a matching fail count attribute, update fail count if (regexec(&(fc_data->failcount_re), (const char *) key, 0, NULL, 0) == 0) { fc_data->failcount = pcmk__add_scores(fc_data->failcount, char2score(value)); pcmk__rsc_trace(fc_data->rsc, "Added %s (%s) to %s fail count (now %s)", (const char *) key, (const char *) value, fc_data->rsc->id, pcmk_readable_score(fc_data->failcount)); return; } // If this is a matching last failure attribute, update last failure if (regexec(&(fc_data->lastfailure_re), (const char *) key, 0, NULL, 0) == 0) { long long last_ll; if (pcmk__scan_ll(value, &last_ll, 0LL) == pcmk_rc_ok) { fc_data->last_failure = (time_t) QB_MAX(fc_data->last_failure, last_ll); } } } /*! * \internal * \brief Update fail count and last failure appropriately for launched resource * * \param[in] data Launched resource * \param[in] user_data Fail count data to update */ static void update_launched_failcount(gpointer data, gpointer user_data) { pcmk_resource_t *launched = data; struct failcount_data *fc_data = user_data; time_t launched_last_failure = 0; fc_data->failcount += pe_get_failcount(fc_data->node, launched, &launched_last_failure, fc_data->flags, fc_data->xml_op); fc_data->last_failure = QB_MAX(fc_data->last_failure, launched_last_failure); } #define readable_expiration(rsc) \ pcmk__readable_interval((rsc)->private->failure_expiration_ms) /*! * \internal * \brief Get a resource's fail count on a node * * \param[in] node Node to check * \param[in,out] rsc Resource to check * \param[out] last_failure If not NULL, where to set time of most recent * failure of \p rsc on \p node * \param[in] flags Group of enum pcmk__fc_flags * \param[in] xml_op If not NULL, consider only the action in this * history entry when determining whether on-fail * is configured as "blocked", otherwise consider * all actions configured for \p rsc * * \return Fail count for \p rsc on \p node according to \p flags */ int pe_get_failcount(const pcmk_node_t *node, pcmk_resource_t *rsc, time_t *last_failure, uint32_t flags, const xmlNode *xml_op) { struct failcount_data fc_data = { .node = node, .rsc = rsc, .flags = flags, .xml_op = xml_op, .failcount = 0, .last_failure = (time_t) 0, }; // Calculate resource failcount as sum of all matching operation failcounts CRM_CHECK(generate_fail_regexes(rsc, &fc_data.failcount_re, &fc_data.lastfailure_re) == pcmk_rc_ok, return 0); g_hash_table_foreach(node->private->attrs, update_failcount_for_attr, &fc_data); regfree(&(fc_data.failcount_re)); regfree(&(fc_data.lastfailure_re)); // If failure blocks the resource, disregard any failure timeout if ((fc_data.failcount > 0) && (rsc->private->failure_expiration_ms > 0) && block_failure(node, rsc, xml_op)) { pcmk__config_warn("Ignoring failure timeout (%s) for %s " "because it conflicts with " PCMK_META_ON_FAIL "=" PCMK_VALUE_BLOCK, readable_expiration(rsc), rsc->id); rsc->private->failure_expiration_ms = 0; } // If all failures have expired, ignore fail count if (pcmk_is_set(flags, pcmk__fc_effective) && (fc_data.failcount > 0) && (fc_data.last_failure > 0) && (rsc->private->failure_expiration_ms > 0)) { time_t now = get_effective_time(rsc->private->scheduler); const guint expiration = rsc->private->failure_expiration_ms / 1000; if (now > (fc_data.last_failure + expiration)) { pcmk__rsc_debug(rsc, "Failcount for %s on %s expired after %s", rsc->id, pcmk__node_name(node), readable_expiration(rsc)); fc_data.failcount = 0; } } /* Add the fail count of any launched resources, except that we never want * the fail counts of a bundle container's launched resources to count * towards the container's fail count. * * Most importantly, a Pacemaker Remote connection to a bundle container * is launched by the container, but can reside on a different node than the * container itself. Counting its fail count on its node towards the * container's fail count on that node could lead to attempting to stop the * container on the wrong node. */ if (pcmk_is_set(flags, pcmk__fc_launched) && (rsc->private->launched != NULL) && !pcmk__is_bundled(rsc)) { g_list_foreach(rsc->private->launched, update_launched_failcount, &fc_data); if (fc_data.failcount > 0) { pcmk__rsc_info(rsc, "Container %s and the resources within it " "have failed %s time%s on %s", rsc->id, pcmk_readable_score(fc_data.failcount), pcmk__plural_s(fc_data.failcount), pcmk__node_name(node)); } } else if (fc_data.failcount > 0) { pcmk__rsc_info(rsc, "%s has failed %s time%s on %s", rsc->id, pcmk_readable_score(fc_data.failcount), pcmk__plural_s(fc_data.failcount), pcmk__node_name(node)); } if (last_failure != NULL) { if ((fc_data.failcount > 0) && (fc_data.last_failure > 0)) { *last_failure = fc_data.last_failure; } else { *last_failure = 0; } } return fc_data.failcount; } /*! * \brief Schedule a controller operation to clear a fail count * * \param[in,out] rsc Resource with failure * \param[in] node Node failure occurred on * \param[in] reason Readable description why needed (for logging) * \param[in,out] scheduler Scheduler data cluster * * \return Scheduled action */ pcmk_action_t * pe__clear_failcount(pcmk_resource_t *rsc, const pcmk_node_t *node, const char *reason, pcmk_scheduler_t *scheduler) { char *key = NULL; pcmk_action_t *clear = NULL; CRM_CHECK(rsc && node && reason && scheduler, return NULL); key = pcmk__op_key(rsc->id, PCMK_ACTION_CLEAR_FAILCOUNT, 0); clear = custom_action(rsc, key, PCMK_ACTION_CLEAR_FAILCOUNT, node, FALSE, scheduler); pcmk__insert_meta(clear, PCMK__META_OP_NO_WAIT, PCMK_VALUE_TRUE); crm_notice("Clearing failure of %s on %s because %s " QB_XS " %s", rsc->id, pcmk__node_name(node), reason, clear->uuid); return clear; }