diff --git a/daemons/controld/controld_te_callbacks.c b/daemons/controld/controld_te_callbacks.c index 368bc1177c..f326a66bf8 100644 --- a/daemons/controld/controld_te_callbacks.c +++ b/daemons/controld/controld_te_callbacks.c @@ -1,698 +1,693 @@ /* * Copyright 2004-2019 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include /* For ONLINESTATUS etc */ void te_update_confirm(const char *event, xmlNode * msg); extern char *te_uuid; gboolean shuttingdown = FALSE; crm_graph_t *transition_graph; crm_trigger_t *transition_trigger = NULL; /* #define RSC_OP_TEMPLATE "//"XML_TAG_DIFF_ADDED"//"XML_TAG_CIB"//"XML_CIB_TAG_STATE"[@uname='%s']"//"XML_LRM_TAG_RSC_OP"[@id='%s]" */ #define RSC_OP_TEMPLATE "//"XML_TAG_DIFF_ADDED"//"XML_TAG_CIB"//"XML_LRM_TAG_RSC_OP"[@id='%s']" static void te_update_diff_v1(const char *event, xmlNode *diff) { int lpc, max; xmlXPathObject *xpathObj = NULL; CRM_CHECK(diff != NULL, return); xml_log_patchset(LOG_TRACE, __FUNCTION__, diff); if (cib_config_changed(NULL, NULL, &diff)) { abort_transition(INFINITY, tg_restart, "Non-status change", diff); goto bail; /* configuration changed */ } /* Tickets Attributes - Added/Updated */ xpathObj = xpath_search(diff, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_CIB_TAG_TICKETS); if (numXpathResults(xpathObj) > 0) { xmlNode *aborted = getXpathResult(xpathObj, 0); abort_transition(INFINITY, tg_restart, "Ticket attribute: update", aborted); goto bail; } freeXpathObject(xpathObj); /* Tickets Attributes - Removed */ xpathObj = xpath_search(diff, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_CIB_TAG_TICKETS); if (numXpathResults(xpathObj) > 0) { xmlNode *aborted = getXpathResult(xpathObj, 0); abort_transition(INFINITY, tg_restart, "Ticket attribute: removal", aborted); goto bail; } freeXpathObject(xpathObj); /* Transient Attributes - Added/Updated */ xpathObj = xpath_search(diff, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_TAG_TRANSIENT_NODEATTRS "//" XML_CIB_TAG_NVPAIR); max = numXpathResults(xpathObj); for (lpc = 0; lpc < max; lpc++) { xmlNode *attr = getXpathResult(xpathObj, lpc); const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME); const char *value = NULL; if (safe_str_eq(CRM_OP_PROBED, name)) { value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE); } if (crm_is_true(value) == FALSE) { abort_transition(INFINITY, tg_restart, "Transient attribute: update", attr); crm_log_xml_trace(attr, "Abort"); goto bail; } } freeXpathObject(xpathObj); /* Transient Attributes - Removed */ xpathObj = xpath_search(diff, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_TAG_TRANSIENT_NODEATTRS); if (numXpathResults(xpathObj) > 0) { xmlNode *aborted = getXpathResult(xpathObj, 0); abort_transition(INFINITY, tg_restart, "Transient attribute: removal", aborted); goto bail; } freeXpathObject(xpathObj); /* * Updates by, or in response to, TE actions will never contain updates * for more than one resource at a time, so such updates indicate an * LRM refresh. * * In that case, start a new transition rather than check each result * individually, which can result in _huge_ speedups in large clusters. * * Unfortunately, we can only do so when there are no pending actions. * Otherwise, we could mistakenly throw away those results here, and * the cluster will stall waiting for them and time out the operation. */ if (transition_graph->pending == 0) { xpathObj = xpath_search(diff, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_LRM_TAG_RESOURCE); max = numXpathResults(xpathObj); if (max > 1) { crm_debug("Ignoring resource operation updates due to history refresh of %d resources", max); crm_log_xml_trace(diff, "lrm-refresh"); abort_transition(INFINITY, tg_restart, "History refresh", NULL); goto bail; } freeXpathObject(xpathObj); } /* Process operation updates */ xpathObj = xpath_search(diff, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_LRM_TAG_RSC_OP); max = numXpathResults(xpathObj); if (max > 0) { int lpc = 0; for (lpc = 0; lpc < max; lpc++) { xmlNode *rsc_op = getXpathResult(xpathObj, lpc); const char *node = get_node_id(rsc_op); process_graph_event(rsc_op, node); } } freeXpathObject(xpathObj); /* Detect deleted (as opposed to replaced or added) actions - eg. crm_resource -C */ xpathObj = xpath_search(diff, "//" XML_TAG_DIFF_REMOVED "//" XML_LRM_TAG_RSC_OP); max = numXpathResults(xpathObj); for (lpc = 0; lpc < max; lpc++) { int path_max = 0; const char *op_id = NULL; char *rsc_op_xpath = NULL; xmlXPathObject *op_match = NULL; xmlNode *match = getXpathResult(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if(match == NULL) { continue; }; op_id = ID(match); path_max = strlen(RSC_OP_TEMPLATE) + strlen(op_id) + 1; rsc_op_xpath = calloc(1, path_max); snprintf(rsc_op_xpath, path_max, RSC_OP_TEMPLATE, op_id); op_match = xpath_search(diff, rsc_op_xpath); if (numXpathResults(op_match) == 0) { /* Prevent false positives by matching cancelations too */ const char *node = get_node_id(match); crm_action_t *cancelled = get_cancel_action(op_id, node); if (cancelled == NULL) { crm_debug("No match for deleted action %s (%s on %s)", rsc_op_xpath, op_id, node); abort_transition(INFINITY, tg_restart, "Resource op removal", match); freeXpathObject(op_match); free(rsc_op_xpath); goto bail; } else { crm_debug("Deleted lrm_rsc_op %s on %s was for graph event %d", op_id, node, cancelled->id); } } freeXpathObject(op_match); free(rsc_op_xpath); } bail: freeXpathObject(xpathObj); } static void process_lrm_resource_diff(xmlNode *lrm_resource, const char *node) { for (xmlNode *rsc_op = __xml_first_child(lrm_resource); rsc_op != NULL; rsc_op = __xml_next(rsc_op)) { process_graph_event(rsc_op, node); } } static void process_resource_updates(const char *node, xmlNode *xml, xmlNode *change, const char *op, const char *xpath) { xmlNode *rsc = NULL; if (xml == NULL) { return; } else if (strcmp((const char*)xml->name, XML_CIB_TAG_LRM) == 0) { xml = first_named_child(xml, XML_LRM_TAG_RESOURCES); crm_trace("Got %p in %s", xml, XML_CIB_TAG_LRM); } CRM_ASSERT(strcmp((const char*)xml->name, XML_LRM_TAG_RESOURCES) == 0); /* * Updates by, or in response to, TE actions will never contain updates * for more than one resource at a time, so such updates indicate an * LRM refresh. * * In that case, start a new transition rather than check each result * individually, which can result in _huge_ speedups in large clusters. * * Unfortunately, we can only do so when there are no pending actions. * Otherwise, we could mistakenly throw away those results here, and * the cluster will stall waiting for them and time out the operation. */ if ((transition_graph->pending == 0) && xml->children && xml->children->next) { crm_log_xml_trace(change, "lrm-refresh"); abort_transition(INFINITY, tg_restart, "History refresh", NULL); return; } for (rsc = __xml_first_child(xml); rsc != NULL; rsc = __xml_next(rsc)) { crm_trace("Processing %s", ID(rsc)); process_lrm_resource_diff(rsc, node); } } #define NODE_PATT "/lrm[@id=" static char *get_node_from_xpath(const char *xpath) { char *nodeid = NULL; char *tmp = strstr(xpath, NODE_PATT); if(tmp) { tmp += strlen(NODE_PATT); tmp += 1; nodeid = strdup(tmp); tmp = strstr(nodeid, "\'"); CRM_ASSERT(tmp); tmp[0] = 0; } return nodeid; } static char *extract_node_uuid(const char *xpath) { char *mutable_path = strdup(xpath); char *node_uuid = NULL; char *search = NULL; char *match = NULL; match = strstr(mutable_path, "node_state[@id=\'"); if (match == NULL) { free(mutable_path); return NULL; } match += strlen("node_state[@id=\'"); search = strchr(match, '\''); if (search == NULL) { free(mutable_path); return NULL; } search[0] = 0; node_uuid = strdup(match); free(mutable_path); return node_uuid; } static void abort_unless_down(const char *xpath, const char *op, xmlNode *change, const char *reason) { char *node_uuid = NULL; crm_action_t *down = NULL; if(safe_str_neq(op, "delete")) { abort_transition(INFINITY, tg_restart, reason, change); return; } node_uuid = extract_node_uuid(xpath); if(node_uuid == NULL) { crm_err("Could not extract node ID from %s", xpath); abort_transition(INFINITY, tg_restart, reason, change); return; } down = match_down_event(node_uuid); if (down == NULL) { crm_trace("Not expecting %s to be down (%s)", node_uuid, xpath); abort_transition(INFINITY, tg_restart, reason, change); } else { crm_trace("Expecting changes to %s (%s)", node_uuid, xpath); } free(node_uuid); } static void process_op_deletion(const char *xpath, xmlNode *change) { char *mutable_key = strdup(xpath); char *key; char *node_uuid; - crm_action_t *cancel = NULL; // Extract the part of xpath between last pair of single quotes key = strrchr(mutable_key, '\''); if (key != NULL) { *key = '\0'; key = strrchr(mutable_key, '\''); } if (key == NULL) { crm_warn("Ignoring malformed CIB update (resource deletion of %s)", xpath); free(mutable_key); return; } ++key; node_uuid = extract_node_uuid(xpath); - cancel = get_cancel_action(key, node_uuid); - if (cancel) { - confirm_cancel_action(cancel); - - } else { + if (confirm_cancel_action(key, node_uuid) == FALSE) { abort_transition(INFINITY, tg_restart, "Resource operation removal", change); } free(mutable_key); free(node_uuid); } static void process_delete_diff(const char *xpath, const char *op, xmlNode *change) { if (strstr(xpath, "/" XML_LRM_TAG_RSC_OP "[")) { process_op_deletion(xpath, change); } else if (strstr(xpath, "/" XML_CIB_TAG_LRM "[")) { abort_unless_down(xpath, op, change, "Resource state removal"); } else if (strstr(xpath, "/" XML_CIB_TAG_STATE "[")) { abort_unless_down(xpath, op, change, "Node state removal"); } else { crm_trace("Ignoring delete of %s", xpath); } } static void process_node_state_diff(xmlNode *state, xmlNode *change, const char *op, const char *xpath) { xmlNode *lrm = first_named_child(state, XML_CIB_TAG_LRM); process_resource_updates(ID(state), lrm, change, op, xpath); } static void process_status_diff(xmlNode *status, xmlNode *change, const char *op, const char *xpath) { for (xmlNode *state = __xml_first_child(status); state != NULL; state = __xml_next(state)) { process_node_state_diff(state, change, op, xpath); } } static void process_cib_diff(xmlNode *cib, xmlNode *change, const char *op, const char *xpath) { xmlNode *status = first_named_child(cib, XML_CIB_TAG_STATUS); xmlNode *config = first_named_child(cib, XML_CIB_TAG_CONFIGURATION); if (status) { process_status_diff(status, change, op, xpath); } if (config) { abort_transition(INFINITY, tg_restart, "Non-status-only change", change); } } static void te_update_diff_v2(xmlNode *diff) { crm_log_xml_trace(diff, "Patch:Raw"); for (xmlNode *change = __xml_first_child(diff); change != NULL; change = __xml_next(change)) { xmlNode *match = NULL; const char *name = NULL; const char *xpath = crm_element_value(change, XML_DIFF_PATH); // Possible ops: create, modify, delete, move const char *op = crm_element_value(change, XML_DIFF_OP); // Ignore uninteresting updates if (op == NULL) { continue; } else if (xpath == NULL) { crm_trace("Ignoring %s change for version field", op); continue; } else if (strcmp(op, "move") == 0) { crm_trace("Ignoring move change at %s", xpath); continue; } // Find the result of create/modify ops if (strcmp(op, "create") == 0) { match = change->children; } else if (strcmp(op, "modify") == 0) { match = first_named_child(change, XML_DIFF_RESULT); if(match) { match = match->children; } } else if (strcmp(op, "delete") != 0) { crm_warn("Ignoring malformed CIB update (%s operation on %s is unrecognized)", op, xpath); continue; } if (match) { if (match->type == XML_COMMENT_NODE) { crm_trace("Ignoring %s operation for comment at %s", op, xpath); continue; } name = (const char *)match->name; } crm_trace("Handling %s operation for %s%s%s", op, (xpath? xpath : "CIB"), (name? " matched by " : ""), (name? name : "")); if (strstr(xpath, "/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION)) { abort_transition(INFINITY, tg_restart, "Configuration change", change); break; // Won't be packaged with operation results we may be waiting for } else if (strstr(xpath, "/" XML_CIB_TAG_TICKETS) || safe_str_eq(name, XML_CIB_TAG_TICKETS)) { abort_transition(INFINITY, tg_restart, "Ticket attribute change", change); break; // Won't be packaged with operation results we may be waiting for } else if (strstr(xpath, "/" XML_TAG_TRANSIENT_NODEATTRS "[") || safe_str_eq(name, XML_TAG_TRANSIENT_NODEATTRS)) { abort_unless_down(xpath, op, change, "Transient attribute change"); break; // Won't be packaged with operation results we may be waiting for } else if (strcmp(op, "delete") == 0) { process_delete_diff(xpath, op, change); } else if (name == NULL) { crm_warn("Ignoring malformed CIB update (%s at %s has no result)", op, xpath); } else if (strcmp(name, XML_TAG_CIB) == 0) { process_cib_diff(match, change, op, xpath); } else if (strcmp(name, XML_CIB_TAG_STATUS) == 0) { process_status_diff(match, change, op, xpath); } else if (strcmp(name, XML_CIB_TAG_STATE) == 0) { process_node_state_diff(match, change, op, xpath); } else if (strcmp(name, XML_CIB_TAG_LRM) == 0) { process_resource_updates(ID(match), match, change, op, xpath); } else if (strcmp(name, XML_LRM_TAG_RESOURCES) == 0) { char *local_node = get_node_from_xpath(xpath); process_resource_updates(local_node, match, change, op, xpath); free(local_node); } else if (strcmp(name, XML_LRM_TAG_RESOURCE) == 0) { char *local_node = get_node_from_xpath(xpath); process_lrm_resource_diff(match, local_node); free(local_node); } else if (strcmp(name, XML_LRM_TAG_RSC_OP) == 0) { char *local_node = get_node_from_xpath(xpath); process_graph_event(match, local_node); free(local_node); } else { crm_warn("Ignoring malformed CIB update (%s at %s has unrecognized result %s)", op, xpath, name); } } } void te_update_diff(const char *event, xmlNode * msg) { xmlNode *diff = NULL; const char *op = NULL; int rc = -EINVAL; int format = 1; int p_add[] = { 0, 0, 0 }; int p_del[] = { 0, 0, 0 }; CRM_CHECK(msg != NULL, return); crm_element_value_int(msg, F_CIB_RC, &rc); if (transition_graph == NULL) { crm_trace("No graph"); return; } else if (rc < pcmk_ok) { crm_trace("Filter rc=%d (%s)", rc, pcmk_strerror(rc)); return; } else if (transition_graph->complete && fsa_state != S_IDLE && fsa_state != S_TRANSITION_ENGINE && fsa_state != S_POLICY_ENGINE) { crm_trace("Filter state=%s, complete=%d", fsa_state2string(fsa_state), transition_graph->complete); return; } op = crm_element_value(msg, F_CIB_OPERATION); diff = get_message_xml(msg, F_CIB_UPDATE_RESULT); xml_patch_versions(diff, p_add, p_del); crm_debug("Processing (%s) diff: %d.%d.%d -> %d.%d.%d (%s)", op, p_del[0], p_del[1], p_del[2], p_add[0], p_add[1], p_add[2], fsa_state2string(fsa_state)); crm_element_value_int(diff, "format", &format); switch (format) { case 1: te_update_diff_v1(event, diff); break; case 2: te_update_diff_v2(diff); break; default: crm_warn("Ignoring malformed CIB update (unknown patch format %d)", format); } } gboolean process_te_message(xmlNode * msg, xmlNode * xml_data) { const char *from = crm_element_value(msg, F_ORIG); const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO); const char *sys_from = crm_element_value(msg, F_CRM_SYS_FROM); const char *ref = crm_element_value(msg, F_CRM_REFERENCE); const char *op = crm_element_value(msg, F_CRM_TASK); const char *type = crm_element_value(msg, F_CRM_MSG_TYPE); crm_trace("Processing %s (%s) message", op, ref); crm_log_xml_trace(msg, "ipc"); if (op == NULL) { /* error */ } else if (sys_to == NULL || strcasecmp(sys_to, CRM_SYSTEM_TENGINE) != 0) { crm_trace("Bad sys-to %s", crm_str(sys_to)); return FALSE; } else if (safe_str_eq(op, CRM_OP_INVOKE_LRM) && safe_str_eq(sys_from, CRM_SYSTEM_LRMD) /* && safe_str_eq(type, XML_ATTR_RESPONSE) */ ) { xmlXPathObject *xpathObj = NULL; crm_log_xml_trace(msg, "Processing (N)ACK"); crm_debug("Processing (N)ACK %s from %s", crm_element_value(msg, F_CRM_REFERENCE), from); xpathObj = xpath_search(xml_data, "//" XML_LRM_TAG_RSC_OP); if (numXpathResults(xpathObj)) { int lpc = 0, max = numXpathResults(xpathObj); for (lpc = 0; lpc < max; lpc++) { xmlNode *rsc_op = getXpathResult(xpathObj, lpc); const char *node = get_node_id(rsc_op); process_graph_event(rsc_op, node); } freeXpathObject(xpathObj); } else { crm_log_xml_err(msg, "Invalid (N)ACK"); freeXpathObject(xpathObj); return FALSE; } } else { crm_err("Unknown command: %s::%s from %s", type, op, sys_from); } crm_trace("finished processing message"); return TRUE; } void cib_action_updated(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { if (rc < pcmk_ok) { crm_err("Update %d FAILED: %s", call_id, pcmk_strerror(rc)); } } /*! * \brief Handle a timeout in node-to-node communication * * \param[in] data Pointer to action timer * * \return FALSE (indicating that source should be not be re-added) */ gboolean action_timer_callback(gpointer data) { crm_action_timer_t *timer = NULL; const char *task = NULL; const char *on_node = NULL; const char *via_node = NULL; CRM_CHECK(data != NULL, return FALSE); timer = (crm_action_timer_t *) data; stop_te_timer(timer); CRM_CHECK(timer->action != NULL, return FALSE); task = crm_element_value(timer->action->xml, XML_LRM_ATTR_TASK); on_node = crm_element_value(timer->action->xml, XML_LRM_ATTR_TARGET); via_node = crm_element_value(timer->action->xml, XML_LRM_ATTR_ROUTER_NODE); if (transition_graph->complete) { crm_notice("Node %s did not send %s result (via %s) within %dms " "(ignoring because transition not in progress)", (on_node? on_node : ""), (task? task : "unknown action"), (via_node? via_node : "controller"), timer->timeout); } else { /* fail the action */ crm_err("Node %s did not send %s result (via %s) within %dms " "(action timeout plus cluster-delay)", (on_node? on_node : ""), (task? task : "unknown action"), (via_node? via_node : "controller"), timer->timeout); print_action(LOG_ERR, "Aborting transition, action lost: ", timer->action); timer->action->failed = TRUE; te_action_confirmed(timer->action, transition_graph); abort_transition(INFINITY, tg_restart, "Action lost", NULL); // Record timeout in the CIB if appropriate if ((timer->action->type == action_type_rsc) && controld_action_is_recordable(task)) { controld_record_action_timeout(timer->action); } } return FALSE; } diff --git a/daemons/controld/controld_te_events.c b/daemons/controld/controld_te_events.c index 8fb9139a65..34e4daa6a2 100644 --- a/daemons/controld/controld_te_events.c +++ b/daemons/controld/controld_te_events.c @@ -1,536 +1,534 @@ /* * Copyright 2004-2019 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include char *failed_stop_offset = NULL; char *failed_start_offset = NULL; gboolean fail_incompletable_actions(crm_graph_t * graph, const char *down_node) { const char *target_uuid = NULL; const char *router = NULL; const char *router_uuid = NULL; xmlNode *last_action = NULL; GListPtr gIter = NULL; GListPtr gIter2 = NULL; if (graph == NULL || graph->complete) { return FALSE; } gIter = graph->synapses; for (; gIter != NULL; gIter = gIter->next) { synapse_t *synapse = (synapse_t *) gIter->data; if (synapse->confirmed || synapse->failed) { /* We've already been here */ continue; } gIter2 = synapse->actions; for (; gIter2 != NULL; gIter2 = gIter2->next) { crm_action_t *action = (crm_action_t *) gIter2->data; if (action->type == action_type_pseudo || action->confirmed) { continue; } else if (action->type == action_type_crm) { const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); if (safe_str_eq(task, CRM_OP_FENCE)) { continue; } } target_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID); router = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); if (router) { crm_node_t *node = crm_get_peer(0, router); if (node) { router_uuid = node->uuid; } } if (safe_str_eq(target_uuid, down_node) || safe_str_eq(router_uuid, down_node)) { action->failed = TRUE; synapse->failed = TRUE; last_action = action->xml; stop_te_timer(action->timer); update_graph(graph, action); if (synapse->executed) { crm_notice("Action %d (%s) was pending on %s (offline)", action->id, crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY), down_node); } else { crm_info("Action %d (%s) is scheduled for %s (offline)", action->id, crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY), down_node); } } } } if (last_action != NULL) { crm_info("Node %s shutdown resulted in un-runnable actions", down_node); abort_transition(INFINITY, tg_restart, "Node failure", last_action); return TRUE; } return FALSE; } /*! * \internal * \brief Update failure-related node attributes if warranted * * \param[in] event XML describing operation that (maybe) failed * \param[in] event_node_uuid Node that event occurred on * \param[in] rc Actual operation return code * \param[in] target_rc Expected operation return code * \param[in] do_update If TRUE, do update regardless of operation type * \param[in] ignore_failures If TRUE, update last failure but not fail count * * \return TRUE if this was not a direct nack, success or lrm status refresh */ static gboolean update_failcount(xmlNode * event, const char *event_node_uuid, int rc, int target_rc, gboolean do_update, gboolean ignore_failures) { guint interval_ms = 0; char *task = NULL; char *rsc_id = NULL; const char *value = NULL; const char *id = crm_element_value(event, XML_LRM_ATTR_TASK_KEY); const char *on_uname = crm_peer_uname(event_node_uuid); const char *origin = crm_element_value(event, XML_ATTR_ORIGIN); // Nothing needs to be done for success or status refresh if (rc == target_rc) { return FALSE; } else if (safe_str_eq(origin, "build_active_RAs")) { crm_debug("No update for %s (rc=%d) on %s: Old failure from lrm status refresh", id, rc, on_uname); return FALSE; } /* Sanity check */ CRM_CHECK(on_uname != NULL, return TRUE); CRM_CHECK(parse_op_key(id, &rsc_id, &task, &interval_ms), crm_err("Couldn't parse: %s", ID(event)); goto bail); /* Decide whether update is necessary and what value to use */ if ((interval_ms > 0) || safe_str_eq(task, CRMD_ACTION_PROMOTE) || safe_str_eq(task, CRMD_ACTION_DEMOTE)) { do_update = TRUE; } else if (safe_str_eq(task, CRMD_ACTION_START)) { do_update = TRUE; if (failed_start_offset == NULL) { failed_start_offset = strdup(CRM_INFINITY_S); } value = failed_start_offset; } else if (safe_str_eq(task, CRMD_ACTION_STOP)) { do_update = TRUE; if (failed_stop_offset == NULL) { failed_stop_offset = strdup(CRM_INFINITY_S); } value = failed_stop_offset; } /* Fail count will be either incremented or set to infinity */ if (value == NULL || safe_str_neq(value, CRM_INFINITY_S)) { value = XML_NVPAIR_ATTR_VALUE "++"; } if (do_update) { char *now = crm_itoa(time(NULL)); char *attr_name = NULL; gboolean is_remote_node = FALSE; if (g_hash_table_lookup(crm_remote_peer_cache, event_node_uuid)) { is_remote_node = TRUE; } crm_info("Updating %s for %s on %s after failed %s: rc=%d (update=%s, time=%s)", (ignore_failures? "last failure" : "failcount"), rsc_id, on_uname, task, rc, value, now); /* Update the fail count, if we're not ignoring failures */ if (!ignore_failures) { attr_name = crm_failcount_name(rsc_id, task, interval_ms); update_attrd(on_uname, attr_name, value, NULL, is_remote_node); free(attr_name); } /* Update the last failure time (even if we're ignoring failures, * so that failure can still be detected and shown, e.g. by crm_mon) */ attr_name = crm_lastfailure_name(rsc_id, task, interval_ms); update_attrd(on_uname, attr_name, now, NULL, is_remote_node); free(attr_name); free(now); } bail: free(rsc_id); free(task); return TRUE; } /*! * \internal * \brief Return simplified operation status based on operation return code * * \param[in] action CRM action instance of operation * \param[in] orig_status Original reported operation status * \param[in] rc Actual operation return code * \param[in] target_rc Expected operation return code * * \return PCMK_LRM_OP_DONE if rc equals target_rc, PCMK_LRM_OP_ERROR otherwise * * \note This assumes that PCMK_LRM_OP_PENDING operations have already been * filtered (otherwise they will get simplified as well). */ static int status_from_rc(crm_action_t * action, int orig_status, int rc, int target_rc) { if (target_rc == rc) { crm_trace("Target rc: == %d", rc); if (orig_status != PCMK_LRM_OP_DONE) { crm_trace("Re-mapping op status to PCMK_LRM_OP_DONE for rc=%d", rc); } return PCMK_LRM_OP_DONE; } if (orig_status != PCMK_LRM_OP_INVALID) { const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); const char *uname = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); crm_warn("Action %d (%s) on %s failed (target: %d vs. rc: %d): %s", action->id, task, uname, target_rc, rc, services_lrm_status_str(PCMK_LRM_OP_ERROR)); } return PCMK_LRM_OP_ERROR; } /*! * \internal * \brief Confirm action and update transition graph, aborting transition on failures * * \param[in,out] action CRM action instance of this operation * \param[in] event Event instance of this operation * \param[in] orig_status Original reported operation status * \param[in] op_rc Actual operation return code * \param[in] target_rc Expected operation return code * \param[in] ignore_failures Whether to ignore operation failures * * \note This assumes that PCMK_LRM_OP_PENDING operations have already been * filtered (otherwise they may be treated as failures). */ static void match_graph_event(crm_action_t *action, xmlNode *event, int op_status, int op_rc, int target_rc, gboolean ignore_failures) { const char *target = NULL; const char *this_event = NULL; const char *ignore_s = ""; // Remap operation status to DONE or ERROR based on return code op_status = status_from_rc(action, op_status, op_rc, target_rc); // Mark action as failed if not ignoring failures if (op_status == PCMK_LRM_OP_ERROR) { if (ignore_failures) { ignore_s = ", ignoring failure"; } else { action->failed = TRUE; } } stop_te_timer(action->timer); te_action_confirmed(action, transition_graph); if (action->failed) { abort_transition(action->synapse->priority + 1, tg_restart, "Event failed", event); } this_event = crm_element_value(event, XML_LRM_ATTR_TASK_KEY); target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); crm_info("Action %s (%d) confirmed on %s (rc=%d%s)", crm_str(this_event), action->id, crm_str(target), op_rc, ignore_s); } crm_action_t * controld_get_action(int id) { for (GList *item = transition_graph->synapses; item; item = item->next) { synapse_t *synapse = (synapse_t *) item->data; for (GList *item2 = synapse->actions; item2; item2 = item2->next) { crm_action_t *action = (crm_action_t *) item2->data; if (action->id == id) { return action; } } } return NULL; } crm_action_t * get_cancel_action(const char *id, const char *node) { GListPtr gIter = NULL; GListPtr gIter2 = NULL; gIter = transition_graph->synapses; for (; gIter != NULL; gIter = gIter->next) { synapse_t *synapse = (synapse_t *) gIter->data; gIter2 = synapse->actions; for (; gIter2 != NULL; gIter2 = gIter2->next) { const char *task = NULL; const char *target = NULL; crm_action_t *action = (crm_action_t *) gIter2->data; task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); if (safe_str_neq(CRMD_ACTION_CANCEL, task)) { continue; } task = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); if (safe_str_neq(task, id)) { crm_trace("Wrong key %s for %s on %s", task, id, node); continue; } target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID); if (node && safe_str_neq(target, node)) { crm_trace("Wrong node %s for %s on %s", target, id, node); continue; } crm_trace("Found %s on %s", id, node); return action; } } return NULL; } -void -confirm_cancel_action(crm_action_t *cancel) +bool +confirm_cancel_action(const char *id, const char *node_id) { const char *op_key = NULL; const char *node_name = NULL; + crm_action_t *cancel = get_cancel_action(id, node_id); - CRM_ASSERT(cancel != NULL); - + if (cancel == NULL) { + return FALSE; + } op_key = crm_element_value(cancel->xml, XML_LRM_ATTR_TASK_KEY); node_name = crm_element_value(cancel->xml, XML_LRM_ATTR_TARGET); stop_te_timer(cancel->timer); te_action_confirmed(cancel, transition_graph); crm_info("Cancellation of %s on %s confirmed (action %d)", op_key, node_name, cancel->id); + return TRUE; } /* downed nodes are listed like: ... */ #define XPATH_DOWNED "//" XML_GRAPH_TAG_DOWNED \ "/" XML_CIB_TAG_NODE "[@" XML_ATTR_UUID "='%s']" /*! * \brief Find a transition event that would have made a specified node down * * \param[in] target UUID of node to match * * \return Matching event if found, NULL otherwise */ crm_action_t * match_down_event(const char *target) { crm_action_t *match = NULL; xmlXPathObjectPtr xpath_ret = NULL; GListPtr gIter, gIter2; char *xpath = crm_strdup_printf(XPATH_DOWNED, target); for (gIter = transition_graph->synapses; gIter != NULL && match == NULL; gIter = gIter->next) { for (gIter2 = ((synapse_t*)gIter->data)->actions; gIter2 != NULL && match == NULL; gIter2 = gIter2->next) { match = (crm_action_t*)gIter2->data; if (match->executed) { xpath_ret = xpath_search(match->xml, xpath); if (numXpathResults(xpath_ret) < 1) { match = NULL; } freeXpathObject(xpath_ret); } else { // Only actions that were actually started can match match = NULL; } } } free(xpath); if (match != NULL) { crm_debug("Shutdown action %d (%s) found for node %s", match->id, crm_element_value(match->xml, XML_LRM_ATTR_TASK_KEY), target); } else { crm_debug("No reason to expect node %s to be down", target); } return match; } void process_graph_event(xmlNode *event, const char *event_node) { int rc = -1; int status = -1; int callid = -1; int action_num = -1; crm_action_t *action = NULL; int target_rc = -1; int transition_num = -1; char *update_te_uuid = NULL; gboolean ignore_failures = FALSE; const char *id = NULL; const char *desc = NULL; const char *magic = NULL; CRM_ASSERT(event != NULL); /* */ id = crm_element_value(event, XML_LRM_ATTR_TASK_KEY); crm_element_value_int(event, XML_LRM_ATTR_RC, &rc); crm_element_value_int(event, XML_LRM_ATTR_OPSTATUS, &status); crm_element_value_int(event, XML_LRM_ATTR_CALLID, &callid); magic = crm_element_value(event, XML_ATTR_TRANSITION_KEY); if (magic == NULL) { /* non-change */ return; } if (decode_transition_key(magic, &update_te_uuid, &transition_num, &action_num, &target_rc) == FALSE) { crm_err("Invalid event %s.%d detected: %s", id, callid, magic); abort_transition(INFINITY, tg_restart, "Bad event", event); return; } if (status == PCMK_LRM_OP_PENDING) { goto bail; } if (transition_num == -1) { desc = "initiated outside of the cluster"; abort_transition(INFINITY, tg_restart, "Unexpected event", event); } else if ((action_num < 0) || (crm_str_eq(update_te_uuid, te_uuid, TRUE) == FALSE)) { desc = "initiated by a different node"; abort_transition(INFINITY, tg_restart, "Foreign event", event); } else if (transition_graph->id != transition_num) { guint interval_ms = 0; if (parse_op_key(id, NULL, NULL, &interval_ms) && (interval_ms != 0)) { /* Recurring actions have the transition number they were first * scheduled in. */ if (status == PCMK_LRM_OP_CANCELLED) { - const char *node_id = get_node_id(event); - - action = get_cancel_action(id, node_id); - if (action) { - confirm_cancel_action(action); - } + confirm_cancel_action(id, get_node_id(event)); goto bail; } desc = "arrived after initial scheduling"; abort_transition(INFINITY, tg_restart, "Change in recurring result", event); } else { desc = "arrived really late"; abort_transition(INFINITY, tg_restart, "Old event", event); } } else if (transition_graph->complete) { desc = "arrived late"; abort_transition(INFINITY, tg_restart, "Inactive graph", event); } else { action = controld_get_action(action_num); if (action == NULL) { desc = "unknown"; abort_transition(INFINITY, tg_restart, "Unknown event", event); } else { /* Actions already confirmed skip matching. */ /* ex. Ignoring xxx_last_0 or xxx_last_failure_0 generated by create_operation_update() in order to prevent duplicate fail-count from increasing. */ if (action->confirmed == TRUE) { crm_log_xml_debug(event, "No update by already confirmed events :"); goto bail; } ignore_failures = safe_str_eq( crm_meta_value(action->params, XML_OP_ATTR_ON_FAIL), "ignore"); match_graph_event(action, event, status, rc, target_rc, ignore_failures); } } if (action && (rc == target_rc)) { crm_trace("Processed update to %s: %s", id, magic); } else { if ((status != PCMK_LRM_OP_INVALID) && update_failcount(event, event_node, rc, target_rc, (transition_num == -1), ignore_failures)) { desc = "failed"; } crm_info("Detected action (%d.%d) %s.%d=%s: %s", transition_num, action_num, id, callid, services_ocf_exitcode_str(rc), desc); } bail: free(update_te_uuid); } diff --git a/daemons/controld/controld_transition.h b/daemons/controld/controld_transition.h index 3a520d9827..4e69293c2c 100644 --- a/daemons/controld/controld_transition.h +++ b/daemons/controld/controld_transition.h @@ -1,65 +1,65 @@ /* * Copyright 2004-2019 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. */ #ifndef TENGINE__H # define TENGINE__H # include # include # include # include /* tengine */ extern crm_action_t *match_down_event(const char *target); extern crm_action_t *get_cancel_action(const char *id, const char *node); -void confirm_cancel_action(crm_action_t *cancel); +bool confirm_cancel_action(const char *id, const char *node_id); void controld_record_action_timeout(crm_action_t *action); extern gboolean fail_incompletable_actions(crm_graph_t * graph, const char *down_node); void process_graph_event(xmlNode *event, const char *event_node); /* utils */ crm_action_t *controld_get_action(int id); extern gboolean stop_te_timer(crm_action_timer_t * timer); extern const char *get_rsc_state(const char *task, enum op_status status); /* unpack */ extern gboolean process_te_message(xmlNode * msg, xmlNode * xml_data); extern crm_graph_t *transition_graph; extern crm_trigger_t *transition_trigger; extern char *te_uuid; extern void notify_crmd(crm_graph_t * graph); void cib_action_updated(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data); gboolean action_timer_callback(gpointer data); gboolean te_graph_trigger(gpointer user_data); void te_update_diff(const char *event, xmlNode *msg); extern void trigger_graph_processing(const char *fn, int line); void abort_after_delay(int abort_priority, enum transition_action abort_action, const char *abort_text, guint delay_ms); extern void abort_transition_graph(int abort_priority, enum transition_action abort_action, const char *abort_text, xmlNode * reason, const char *fn, int line); # define trigger_graph() trigger_graph_processing(__FUNCTION__, __LINE__) # define abort_transition(pri, action, text, reason) \ abort_transition_graph(pri, action, text, reason,__FUNCTION__,__LINE__); extern crm_trigger_t *transition_trigger; extern char *failed_stop_offset; extern char *failed_start_offset; void te_action_confirmed(crm_action_t *action, crm_graph_t *graph); void te_reset_job_counts(void); #endif