diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c index a20f96aacc..c8129ee89b 100644 --- a/daemons/controld/controld_execd.c +++ b/daemons/controld/controld_execd.c @@ -1,2802 +1,2805 @@ /* * 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 // lrmd_event_data_t, lrmd_rsc_info_t, etc. #include #include #include +#include #include #include #include #include #include #include #include #define START_DELAY_THRESHOLD 5 * 60 * 1000 #define MAX_LRM_REG_FAILS 30 #define s_if_plural(i) (((i) == 1)? "" : "s") struct delete_event_s { int rc; const char *rsc; lrm_state_t *lrm_state; }; static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id); static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list); static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data); static int delete_rsc_status(lrm_state_t * lrm_state, const char *rsc_id, int call_options, const char *user_name); static lrmd_event_data_t *construct_op(lrm_state_t * lrm_state, xmlNode * rsc_op, const char *rsc_id, const char *operation); static void do_lrm_rsc_op(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *operation, xmlNode * msg, xmlNode * request); void send_direct_ack(const char *to_host, const char *to_sys, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op, const char *rsc_id); static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level); static int do_update_resource(const char *node_name, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op); static void lrm_connection_destroy(void) { if (is_set(fsa_input_register, R_LRM_CONNECTED)) { crm_crit("Connection to executor failed"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); clear_bit(fsa_input_register, R_LRM_CONNECTED); } else { crm_info("Disconnected from executor"); } } static char * make_stop_id(const char *rsc, int call_id) { return crm_strdup_printf("%s:%d", rsc, call_id); } static void copy_instance_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") == NULL) { g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value)); } } static void copy_meta_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") != NULL) { g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value)); } } /*! * \internal * \brief Remove a recurring operation from a resource's history * * \param[in,out] history Resource history to modify * \param[in] op Operation to remove * * \return TRUE if the operation was found and removed, FALSE otherwise */ static gboolean history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_event_data_t *existing = iter->data; if ((op->interval_ms == existing->interval_ms) && crm_str_eq(op->rsc_id, existing->rsc_id, TRUE) && safe_str_eq(op->op_type, existing->op_type)) { history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter); lrmd_free_event(existing); return TRUE; } } return FALSE; } /*! * \internal * \brief Free all recurring operations in resource history * * \param[in,out] history Resource history to modify */ static void history_free_recurring_ops(rsc_history_t *history) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_free_event(iter->data); } g_list_free(history->recurring_op_list); history->recurring_op_list = NULL; } /*! * \internal * \brief Free resource history * * \param[in,out] history Resource history to free */ void history_free(gpointer data) { rsc_history_t *history = (rsc_history_t*)data; if (history->stop_params) { g_hash_table_destroy(history->stop_params); } /* Don't need to free history->rsc.id because it's set to history->id */ free(history->rsc.type); free(history->rsc.standard); free(history->rsc.provider); lrmd_free_event(history->failed); lrmd_free_event(history->last); free(history->id); history_free_recurring_ops(history); free(history); } static void update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op) { int target_rc = 0; rsc_history_t *entry = NULL; if (op->rsc_deleted) { crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type); delete_rsc_status(lrm_state, op->rsc_id, cib_quorum_override, NULL); return; } if (safe_str_eq(op->op_type, RSC_NOTIFY)) { return; } crm_debug("Updating history for '%s' with %s op", op->rsc_id, op->op_type); entry = g_hash_table_lookup(lrm_state->resource_history, op->rsc_id); if (entry == NULL && rsc) { entry = calloc(1, sizeof(rsc_history_t)); entry->id = strdup(op->rsc_id); g_hash_table_insert(lrm_state->resource_history, entry->id, entry); entry->rsc.id = entry->id; entry->rsc.type = strdup(rsc->type); entry->rsc.standard = strdup(rsc->standard); if (rsc->provider) { entry->rsc.provider = strdup(rsc->provider); } else { entry->rsc.provider = NULL; } } else if (entry == NULL) { crm_info("Resource %s no longer exists, not updating cache", op->rsc_id); return; } entry->last_callid = op->call_id; target_rc = rsc_op_expected_rc(op); if (op->op_status == PCMK_LRM_OP_CANCELLED) { if (op->interval_ms > 0) { crm_trace("Removing cancelled recurring op: " CRM_OP_FMT, op->rsc_id, op->op_type, op->interval_ms); history_remove_recurring_op(entry, op); return; } else { crm_trace("Skipping " CRM_OP_FMT " rc=%d, status=%d", op->rsc_id, op->op_type, op->interval_ms, op->rc, op->op_status); } } else if (did_rsc_op_fail(op, target_rc)) { /* Store failed monitors here, otherwise the block below will cause them * to be forgotten when a stop happens. */ if (entry->failed) { lrmd_free_event(entry->failed); } entry->failed = lrmd_copy_event(op); } else if (op->interval_ms == 0) { if (entry->last) { lrmd_free_event(entry->last); } entry->last = lrmd_copy_event(op); if (op->params && (safe_str_eq(CRMD_ACTION_START, op->op_type) || safe_str_eq("reload", op->op_type) || safe_str_eq(CRMD_ACTION_STATUS, op->op_type))) { if (entry->stop_params) { g_hash_table_destroy(entry->stop_params); } entry->stop_params = crm_str_table_new(); g_hash_table_foreach(op->params, copy_instance_keys, entry->stop_params); } } if (op->interval_ms > 0) { /* Ensure there are no duplicates */ history_remove_recurring_op(entry, op); crm_trace("Adding recurring op: " CRM_OP_FMT, op->rsc_id, op->op_type, op->interval_ms); entry->recurring_op_list = g_list_prepend(entry->recurring_op_list, lrmd_copy_event(op)); } else if (entry->recurring_op_list && safe_str_eq(op->op_type, RSC_STATUS) == FALSE) { crm_trace("Dropping %d recurring ops because of: " CRM_OP_FMT, g_list_length(entry->recurring_op_list), op->rsc_id, op->op_type, op->interval_ms); history_free_recurring_ops(entry); } } /*! * \internal * \brief Send a direct OK ack for a resource task * * \param[in] lrm_state LRM connection * \param[in] input Input message being ack'ed * \param[in] rsc_id ID of affected resource * \param[in] rsc Affected resource (if available) * \param[in] task Operation task being ack'ed * \param[in] ack_host Name of host to send ack to * \param[in] ack_sys IPC system name to ack */ static void send_task_ok_ack(lrm_state_t *lrm_state, ha_msg_input_t *input, const char *rsc_id, lrmd_rsc_info_t *rsc, const char *task, const char *ack_host, const char *ack_sys) { lrmd_event_data_t *op = construct_op(lrm_state, input->xml, rsc_id, task); op->rc = PCMK_OCF_OK; op->op_status = PCMK_LRM_OP_DONE; send_direct_ack(ack_host, ack_sys, rsc, op, rsc_id); lrmd_free_event(op); } static inline const char * op_node_name(lrmd_event_data_t *op) { return op->remote_nodename? op->remote_nodename : fsa_our_uname; } void lrm_op_callback(lrmd_event_data_t * op) { CRM_CHECK(op != NULL, return); switch (op->type) { case lrmd_event_disconnect: if (op->remote_nodename == NULL) { /* If this is the local executor IPC connection, set the right * bits in the controller when the connection goes down. */ lrm_connection_destroy(); } break; case lrmd_event_exec_complete: { lrm_state_t *lrm_state = lrm_state_find(op_node_name(op)); CRM_ASSERT(lrm_state != NULL); process_lrm_event(lrm_state, op, NULL, NULL); } break; default: break; } } /* A_LRM_CONNECT */ void do_lrm_control(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { /* This only pertains to local executor connections. Remote connections are * handled as resources within the scheduler. Connecting and disconnecting * from remote executor instances is handled differently. */ lrm_state_t *lrm_state = NULL; if(fsa_our_uname == NULL) { return; /* Nothing to do */ } lrm_state = lrm_state_find_or_create(fsa_our_uname); if (lrm_state == NULL) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } if (action & A_LRM_DISCONNECT) { if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) { if (action == A_LRM_DISCONNECT) { crmd_fsa_stall(FALSE); return; } } clear_bit(fsa_input_register, R_LRM_CONNECTED); crm_info("Disconnecting from the executor"); lrm_state_disconnect(lrm_state); lrm_state_reset_tables(lrm_state, FALSE); crm_notice("Disconnected from the executor"); } if (action & A_LRM_CONNECT) { int ret = pcmk_ok; crm_debug("Connecting to the executor"); ret = lrm_state_ipc_connect(lrm_state); if (ret != pcmk_ok) { if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) { crm_warn("Failed to connect to the executor %d time%s (%d max)", lrm_state->num_lrm_register_fails, s_if_plural(lrm_state->num_lrm_register_fails), MAX_LRM_REG_FAILS); crm_timer_start(wait_timer); crmd_fsa_stall(FALSE); return; } } if (ret != pcmk_ok) { crm_err("Failed to connect to the executor the max allowed %d time%s", lrm_state->num_lrm_register_fails, s_if_plural(lrm_state->num_lrm_register_fails)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } set_bit(fsa_input_register, R_LRM_CONNECTED); crm_info("Connection to the executor established"); } if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __FUNCTION__); } } static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level) { int counter = 0; gboolean rc = TRUE; const char *when = "lrm disconnect"; GHashTableIter gIter; const char *key = NULL; rsc_history_t *entry = NULL; struct recurring_op_s *pending = NULL; crm_debug("Checking for active resources before exit"); if (cur_state == S_TERMINATE) { log_level = LOG_ERR; when = "shutdown"; } else if (is_set(fsa_input_register, R_SHUTDOWN)) { when = "shutdown... waiting"; } if (lrm_state->pending_ops && lrm_state_is_connected(lrm_state) == TRUE) { guint removed = g_hash_table_foreach_remove( lrm_state->pending_ops, stop_recurring_actions, lrm_state); guint nremaining = g_hash_table_size(lrm_state->pending_ops); if (removed || nremaining) { crm_notice("Stopped %u recurring operation%s at %s (%u remaining)", removed, s_if_plural(removed), when, nremaining); } } if (lrm_state->pending_ops) { g_hash_table_iter_init(&gIter, lrm_state->pending_ops); while (g_hash_table_iter_next(&gIter, NULL, (void **)&pending)) { /* Ignore recurring actions in the shutdown calculations */ if (pending->interval_ms == 0) { counter++; } } } if (counter > 0) { do_crm_log(log_level, "%d pending executor operation%s at %s", counter, s_if_plural(counter), when); if (cur_state == S_TERMINATE || !is_set(fsa_input_register, R_SENT_RSC_STOP)) { g_hash_table_iter_init(&gIter, lrm_state->pending_ops); while (g_hash_table_iter_next(&gIter, (gpointer*)&key, (gpointer*)&pending)) { do_crm_log(log_level, "Pending action: %s (%s)", key, pending->op_key); } } else { rc = FALSE; } return rc; } if (lrm_state->resource_history == NULL) { return rc; } if (is_set(fsa_input_register, R_SHUTDOWN)) { /* At this point we're not waiting, we're just shutting down */ when = "shutdown"; } counter = 0; g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (gpointer*)&entry)) { if (is_rsc_active(lrm_state, entry->id) == FALSE) { continue; } counter++; if (log_level == LOG_ERR) { crm_info("Found %s active at %s", entry->id, when); } else { crm_trace("Found %s active at %s", entry->id, when); } if (lrm_state->pending_ops) { GHashTableIter hIter; g_hash_table_iter_init(&hIter, lrm_state->pending_ops); while (g_hash_table_iter_next(&hIter, (gpointer*)&key, (gpointer*)&pending)) { if (crm_str_eq(entry->id, pending->rsc_id, TRUE)) { crm_notice("%sction %s (%s) incomplete at %s", pending->interval_ms == 0 ? "A" : "Recurring a", key, pending->op_key, when); } } } } if (counter) { crm_err("%d resource%s active at %s", counter, (counter == 1)? " was" : "s were", when); } return rc; } static char * build_parameter_list(const lrmd_event_data_t *op, const struct ra_metadata_s *metadata, xmlNode *result, enum ra_param_flags_e param_type, bool invert_for_xml) { int len = 0; int max = 0; char *list = NULL; GList *iter = NULL; /* Newer resource agents support the "private" parameter attribute to * indicate sensitive parameters. For backward compatibility with older * agents, this list is used if the agent doesn't specify any as "private". */ const char *secure_terms[] = { "password", "passwd", "user", }; if (is_not_set(metadata->ra_flags, ra_uses_private) && (param_type == ra_param_private)) { max = DIMOF(secure_terms); } for (iter = metadata->ra_params; iter != NULL; iter = iter->next) { struct ra_param_s *param = (struct ra_param_s *) iter->data; bool accept = FALSE; if (is_set(param->rap_flags, param_type)) { accept = TRUE; } else if (max) { for (int lpc = 0; lpc < max; lpc++) { if (safe_str_eq(secure_terms[lpc], param->rap_name)) { accept = TRUE; break; } } } if (accept) { int start = len; crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type)); len += strlen(param->rap_name) + 2; // include spaces around list = realloc_safe(list, len + 1); // include null terminator // spaces before and after make parsing simpler sprintf(list + start, " %s ", param->rap_name); } else { crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type)); } if (result && (invert_for_xml? !accept : accept)) { const char *v = g_hash_table_lookup(op->params, param->rap_name); if (v != NULL) { crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v); crm_xml_add(result, param->rap_name, v); } } } return list; } static void append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { char *list = NULL; char *digest = NULL; xmlNode *restart = NULL; CRM_LOG_ASSERT(op->params != NULL); if (op->interval_ms > 0) { /* monitors are not reloadable */ return; } if (is_set(metadata->ra_flags, ra_supports_reload)) { restart = create_xml_node(NULL, XML_TAG_PARAMS); /* Add any parameters with unique="1" to the "op-force-restart" list. * * (Currently, we abuse "unique=0" to indicate reloadability. This is * nonstandard and should eventually be replaced once the OCF standard * is updated with something better.) */ list = build_parameter_list(op, metadata, restart, ra_param_unique, FALSE); } else { /* Resource does not support reloads */ return; } digest = calculate_operation_digest(restart, version); /* Add "op-force-restart" and "op-restart-digest" to indicate the resource supports reload, * no matter if it actually supports any parameters with unique="1"). */ crm_xml_add(update, XML_LRM_ATTR_OP_RESTART, list? list: ""); crm_xml_add(update, XML_LRM_ATTR_RESTART_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, list); crm_log_xml_trace(restart, "restart digest source"); free_xml(restart); free(digest); free(list); } static void append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { char *list = NULL; char *digest = NULL; xmlNode *secure = NULL; CRM_LOG_ASSERT(op->params != NULL); /* * To keep XML_LRM_ATTR_OP_SECURE short, we want it to contain the * secure parameters but XML_LRM_ATTR_SECURE_DIGEST to be based on * the insecure ones */ secure = create_xml_node(NULL, XML_TAG_PARAMS); list = build_parameter_list(op, metadata, secure, ra_param_private, TRUE); if (list != NULL) { digest = calculate_operation_digest(secure, version); crm_xml_add(update, XML_LRM_ATTR_OP_SECURE, list); crm_xml_add(update, XML_LRM_ATTR_SECURE_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, list); crm_log_xml_trace(secure, "secure digest source"); } else { crm_trace("%s: no secure parameters", op->rsc_id); } free_xml(secure); free(digest); free(list); } static gboolean build_operation_update(xmlNode * parent, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op, const char *node_name, const char *src) { int target_rc = 0; xmlNode *xml_op = NULL; struct ra_metadata_s *metadata = NULL; const char *caller_version = NULL; lrm_state_t *lrm_state = NULL; if (op == NULL) { return FALSE; } target_rc = rsc_op_expected_rc(op); /* there is a small risk in formerly mixed clusters that it will * be sub-optimal. * * however with our upgrade policy, the update we send should * still be completely supported anyway */ caller_version = g_hash_table_lookup(op->params, XML_ATTR_CRM_VERSION); CRM_LOG_ASSERT(caller_version != NULL); if(caller_version == NULL) { caller_version = CRM_FEATURE_SET; } crm_trace("Building %s operation update with originator version: %s", op->rsc_id, caller_version); - xml_op = create_operation_update(parent, op, caller_version, target_rc, fsa_our_uname, src, LOG_DEBUG); + xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc, + fsa_our_uname, src, LOG_DEBUG); if (xml_op == NULL) { return TRUE; } if ((rsc == NULL) || (op->params == NULL) || !crm_op_needs_metadata(rsc->standard, op->op_type)) { crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)", op->op_type, op->rsc_id, op->params, rsc); return TRUE; } lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { crm_warn("Cannot calculate digests for operation " CRM_OP_FMT " because we have no connection to executor for %s", op->rsc_id, op->op_type, op->interval_ms, node_name); return TRUE; } metadata = metadata_cache_get(lrm_state->metadata_cache, rsc); if (metadata == NULL) { /* For now, we always collect resource agent meta-data via a local, * synchronous, direct execution of the agent. This has multiple issues: * the executor should execute agents, not the controller; meta-data for * Pacemaker Remote nodes should be collected on those nodes, not * locally; and the meta-data call shouldn't eat into the timeout of the * real action being performed. * * These issues are planned to be addressed by having the scheduler * schedule a meta-data cache check at the beginning of each transition. * Once that is working, this block will only be a fallback in case the * initial collection fails. */ char *metadata_str = NULL; int rc = lrm_state_get_metadata(lrm_state, rsc->standard, rsc->provider, rsc->type, &metadata_str, 0); if (rc != pcmk_ok) { crm_warn("Failed to get metadata for %s (%s:%s:%s)", rsc->id, rsc->standard, rsc->provider, rsc->type); return TRUE; } metadata = metadata_cache_update(lrm_state->metadata_cache, rsc, metadata_str); free(metadata_str); if (metadata == NULL) { crm_warn("Failed to update metadata for %s (%s:%s:%s)", rsc->id, rsc->standard, rsc->provider, rsc->type); return TRUE; } } #if ENABLE_VERSIONED_ATTRS crm_xml_add(xml_op, XML_ATTR_RA_VERSION, metadata->ra_version); #endif crm_trace("Including additional digests for %s::%s:%s", rsc->standard, rsc->provider, rsc->type); append_restart_list(op, metadata, xml_op, caller_version); append_secure_list(op, metadata, xml_op, caller_version); return TRUE; } static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id) { rsc_history_t *entry = NULL; entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->last == NULL) { return FALSE; } crm_trace("Processing %s: %s.%d=%d", rsc_id, entry->last->op_type, entry->last->interval_ms, entry->last->rc); if (entry->last->rc == PCMK_OCF_OK && safe_str_eq(entry->last->op_type, CRMD_ACTION_STOP)) { return FALSE; } else if (entry->last->rc == PCMK_OCF_OK && safe_str_eq(entry->last->op_type, CRMD_ACTION_MIGRATE)) { /* a stricter check is too complex... * leave that to the PE */ return FALSE; } else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) { return FALSE; } else if ((entry->last->interval_ms == 0) && (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) { /* Badly configured resources can't be reliably stopped */ return FALSE; } return TRUE; } static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list) { GHashTableIter iter; rsc_history_t *entry = NULL; g_hash_table_iter_init(&iter, lrm_state->resource_history); while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) { GList *gIter = NULL; xmlNode *xml_rsc = create_xml_node(rsc_list, XML_LRM_TAG_RESOURCE); crm_xml_add(xml_rsc, XML_ATTR_ID, entry->id); crm_xml_add(xml_rsc, XML_ATTR_TYPE, entry->rsc.type); crm_xml_add(xml_rsc, XML_AGENT_ATTR_CLASS, entry->rsc.standard); crm_xml_add(xml_rsc, XML_AGENT_ATTR_PROVIDER, entry->rsc.provider); if (entry->last && entry->last->params) { const char *container = g_hash_table_lookup(entry->last->params, CRM_META"_"XML_RSC_ATTR_CONTAINER); if (container) { crm_trace("Resource %s is a part of container resource %s", entry->id, container); crm_xml_add(xml_rsc, XML_RSC_ATTR_CONTAINER, container); } } build_operation_update(xml_rsc, &(entry->rsc), entry->failed, lrm_state->node_name, __FUNCTION__); build_operation_update(xml_rsc, &(entry->rsc), entry->last, lrm_state->node_name, __FUNCTION__); for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) { build_operation_update(xml_rsc, &(entry->rsc), gIter->data, lrm_state->node_name, __FUNCTION__); } } return FALSE; } static xmlNode * do_lrm_query_internal(lrm_state_t *lrm_state, int update_flags) { xmlNode *xml_state = NULL; xmlNode *xml_data = NULL; xmlNode *rsc_list = NULL; crm_node_t *peer = NULL; peer = crm_get_peer_full(0, lrm_state->node_name, CRM_GET_PEER_ANY); CRM_CHECK(peer != NULL, return NULL); xml_state = create_node_state_update(peer, update_flags, NULL, __FUNCTION__); if (xml_state == NULL) { return NULL; } xml_data = create_xml_node(xml_state, XML_CIB_TAG_LRM); crm_xml_add(xml_data, XML_ATTR_ID, peer->uuid); rsc_list = create_xml_node(xml_data, XML_LRM_TAG_RESOURCES); /* Build a list of active (not always running) resources */ build_active_RAs(lrm_state, rsc_list); crm_log_xml_trace(xml_state, "Current executor state"); return xml_state; } xmlNode * do_lrm_query(gboolean is_replace, const char *node_name) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (!lrm_state) { crm_err("Could not find executor state for node %s", node_name); return NULL; } return do_lrm_query_internal(lrm_state, node_update_cluster|node_update_peer); } static void notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc) { lrmd_event_data_t *op = NULL; const char *from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM); const char *from_host = crm_element_value(input->msg, F_CRM_HOST_FROM); crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "localhost"), rsc_id, ((rc == pcmk_ok)? "" : " not")); op = construct_op(lrm_state, input->xml, rsc_id, CRMD_ACTION_DELETE); if (rc == pcmk_ok) { op->op_status = PCMK_LRM_OP_DONE; op->rc = PCMK_OCF_OK; } else { op->op_status = PCMK_LRM_OP_ERROR; op->rc = PCMK_OCF_UNKNOWN_ERROR; } send_direct_ack(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); if (safe_str_neq(from_sys, CRM_SYSTEM_TENGINE)) { /* this isn't expected - trigger a new transition */ time_t now = time(NULL); char *now_s = crm_itoa(now); crm_debug("Triggering a refresh after %s deleted %s from the executor", from_sys, rsc_id); update_attr_delegate(fsa_cib_conn, cib_none, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, "last-lrm-refresh", now_s, FALSE, NULL, NULL); free(now_s); } } static gboolean lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data) { struct delete_event_s *event = user_data; struct pending_deletion_op_s *op = value; if (crm_str_eq(event->rsc, op->rsc, TRUE)) { notify_deleted(event->lrm_state, op->input, event->rsc, event->rc); return TRUE; } return FALSE; } static gboolean lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data) { const char *rsc = user_data; struct recurring_op_s *pending = value; if (crm_str_eq(rsc, pending->rsc_id, TRUE)) { crm_info("Removing op %s:%d for deleted resource %s", pending->op_key, pending->call_id, rsc); return TRUE; } return FALSE; } /* * Remove the rsc from the CIB * * Avoids refreshing the entire LRM section of this host */ #define RSC_TEMPLATE "//"XML_CIB_TAG_STATE"[@uname='%s']//"XML_LRM_TAG_RESOURCE"[@id='%s']" static int delete_rsc_status(lrm_state_t * lrm_state, const char *rsc_id, int call_options, const char *user_name) { char *rsc_xpath = NULL; int rc = pcmk_ok; CRM_CHECK(rsc_id != NULL, return -ENXIO); rsc_xpath = crm_strdup_printf(RSC_TEMPLATE, lrm_state->node_name, rsc_id); rc = cib_internal_op(fsa_cib_conn, CIB_OP_DELETE, NULL, rsc_xpath, NULL, NULL, call_options | cib_xpath, user_name); free(rsc_xpath); return rc; } static void delete_rsc_entry(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, GHashTableIter * rsc_gIter, int rc, const char *user_name) { struct delete_event_s event; CRM_CHECK(rsc_id != NULL, return); if (rc == pcmk_ok) { char *rsc_id_copy = strdup(rsc_id); if (rsc_gIter) g_hash_table_iter_remove(rsc_gIter); else g_hash_table_remove(lrm_state->resource_history, rsc_id_copy); crm_debug("sync: Sending delete op for %s", rsc_id_copy); delete_rsc_status(lrm_state, rsc_id_copy, cib_quorum_override, user_name); g_hash_table_foreach_remove(lrm_state->pending_ops, lrm_remove_deleted_op, rsc_id_copy); free(rsc_id_copy); } if (input) { notify_deleted(lrm_state, input, rsc_id, rc); } event.rc = rc; event.rsc = rsc_id; event.lrm_state = lrm_state; g_hash_table_foreach_remove(lrm_state->deletion_ops, lrm_remove_deleted_rsc, &event); } /*! * \internal * \brief Erase an LRM history entry from the CIB, given the operation data * * \param[in] lrm_state LRM state of the desired node * \param[in] op Operation whose history should be deleted */ static void erase_lrm_history_by_op(lrm_state_t *lrm_state, lrmd_event_data_t *op) { xmlNode *xml_top = NULL; CRM_CHECK(op != NULL, return); xml_top = create_xml_node(NULL, XML_LRM_TAG_RSC_OP); crm_xml_add_int(xml_top, XML_LRM_ATTR_CALLID, op->call_id); crm_xml_add(xml_top, XML_ATTR_TRANSITION_KEY, op->user_data); if (op->interval_ms > 0) { char *op_id = generate_op_key(op->rsc_id, op->op_type, op->interval_ms); /* Avoid deleting last_failure too (if it was a result of this recurring op failing) */ crm_xml_add(xml_top, XML_ATTR_ID, op_id); free(op_id); } crm_debug("Erasing resource operation history for " CRM_OP_FMT " (call=%d)", op->rsc_id, op->op_type, op->interval_ms, op->call_id); fsa_cib_conn->cmds->remove(fsa_cib_conn, XML_CIB_TAG_STATUS, xml_top, cib_quorum_override); crm_log_xml_trace(xml_top, "op:cancel"); free_xml(xml_top); } /* Define xpath to find LRM resource history entry by node and resource */ #define XPATH_HISTORY \ "/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \ "/" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='%s']" \ "/" XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES \ "/" XML_LRM_TAG_RESOURCE "[@" XML_ATTR_ID "='%s']" \ "/" XML_LRM_TAG_RSC_OP /* ... and also by operation key */ #define XPATH_HISTORY_ID XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s']" /* ... and also by operation key and operation call ID */ #define XPATH_HISTORY_CALL XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_CALLID "='%d']" /* ... and also by operation key and original operation key */ #define XPATH_HISTORY_ORIG XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_TASK_KEY "='%s']" /*! * \internal * \brief Erase an LRM history entry from the CIB, given operation identifiers * * \param[in] lrm_state LRM state of the node to clear history for * \param[in] rsc_id Name of resource to clear history for * \param[in] key Operation key of operation to clear history for * \param[in] orig_op If specified, delete only if it has this original op * \param[in] call_id If specified, delete entry only if it has this call ID */ static void erase_lrm_history_by_id(lrm_state_t *lrm_state, const char *rsc_id, const char *key, const char *orig_op, int call_id) { char *op_xpath = NULL; CRM_CHECK((rsc_id != NULL) && (key != NULL), return); if (call_id > 0) { op_xpath = crm_strdup_printf(XPATH_HISTORY_CALL, lrm_state->node_name, rsc_id, key, call_id); } else if (orig_op) { op_xpath = crm_strdup_printf(XPATH_HISTORY_ORIG, lrm_state->node_name, rsc_id, key, orig_op); } else { op_xpath = crm_strdup_printf(XPATH_HISTORY_ID, lrm_state->node_name, rsc_id, key); } crm_debug("Erasing resource operation history for %s on %s (call=%d)", key, rsc_id, call_id); fsa_cib_conn->cmds->remove(fsa_cib_conn, op_xpath, NULL, cib_quorum_override | cib_xpath); free(op_xpath); } static inline gboolean last_failed_matches_op(rsc_history_t *entry, const char *op, guint interval_ms) { if (entry == NULL) { return FALSE; } if (op == NULL) { return TRUE; } return (safe_str_eq(op, entry->failed->op_type) && (interval_ms == entry->failed->interval_ms)); } /*! * \internal * \brief Clear a resource's last failure * * Erase a resource's last failure on a particular node from both the * LRM resource history in the CIB, and the resource history remembered * for the LRM state. * * \param[in] rsc_id Resource name * \param[in] node_name Node name * \param[in] operation If specified, only clear if matching this operation * \param[in] interval_ms If operation is specified, it has this interval */ void lrm_clear_last_failure(const char *rsc_id, const char *node_name, const char *operation, guint interval_ms) { char *op_key = NULL; char *orig_op_key = NULL; lrm_state_t *lrm_state = NULL; lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { return; } /* Erase from CIB */ op_key = generate_op_key(rsc_id, "last_failure", 0); if (operation) { orig_op_key = generate_op_key(rsc_id, operation, interval_ms); } erase_lrm_history_by_id(lrm_state, rsc_id, op_key, orig_op_key, 0); free(op_key); free(orig_op_key); /* Remove from memory */ if (lrm_state->resource_history) { rsc_history_t *entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (last_failed_matches_op(entry, operation, interval_ms)) { lrmd_free_event(entry->failed); entry->failed = NULL; } } } /* Returns: gboolean - cancellation is in progress */ static gboolean cancel_op(lrm_state_t * lrm_state, const char *rsc_id, const char *key, int op, gboolean remove) { int rc = pcmk_ok; char *local_key = NULL; struct recurring_op_s *pending = NULL; CRM_CHECK(op != 0, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); if (key == NULL) { local_key = make_stop_id(rsc_id, op); key = local_key; } pending = g_hash_table_lookup(lrm_state->pending_ops, key); if (pending) { if (remove && pending->remove == FALSE) { pending->remove = TRUE; crm_debug("Scheduling %s for removal", key); } if (pending->cancelled) { crm_debug("Operation %s already cancelled", key); free(local_key); return FALSE; } pending->cancelled = TRUE; } else { crm_info("No pending op found for %s", key); free(local_key); return FALSE; } crm_debug("Cancelling op %d for %s (%s)", op, rsc_id, key); rc = lrm_state_cancel(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms); if (rc == pcmk_ok) { crm_debug("Op %d for %s (%s): cancelled", op, rsc_id, key); free(local_key); return TRUE; } crm_debug("Op %d for %s (%s): Nothing to cancel", op, rsc_id, key); /* The caller needs to make sure the entry is * removed from the pending_ops list * * Usually by returning TRUE inside the worker function * supplied to g_hash_table_foreach_remove() * * Not removing the entry from pending_ops will block * the node from shutting down */ free(local_key); return FALSE; } struct cancel_data { gboolean done; gboolean remove; const char *key; lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean cancel_action_by_key(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct cancel_data *data = user_data; struct recurring_op_s *op = (struct recurring_op_s *)value; if (crm_str_eq(op->op_key, data->key, TRUE)) { data->done = TRUE; remove = !cancel_op(data->lrm_state, data->rsc->id, key, op->call_id, data->remove); } return remove; } static gboolean cancel_op_key(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *key, gboolean remove) { guint removed = 0; struct cancel_data data; CRM_CHECK(rsc != NULL, return FALSE); CRM_CHECK(key != NULL, return FALSE); data.key = key; data.rsc = rsc; data.done = FALSE; data.remove = remove; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove(lrm_state->pending_ops, cancel_action_by_key, &data); crm_trace("Removed %u op cache entries, new size: %u", removed, g_hash_table_size(lrm_state->pending_ops)); return data.done; } /*! * \internal * \brief Retrieve resource information from LRM * * \param[in] lrm_state LRM connection to use * \param[in] rsc_xml XML containing resource configuration * \param[in] do_create If true, register resource with LRM if not already * \param[out] rsc_info Where to store resource information obtained from LRM * * \retval pcmk_ok Success (and rsc_info holds newly allocated result) * \retval -EINVAL Required information is missing from arguments * \retval -ENOTCONN No active connection to LRM * \retval -ENODEV Resource not found * \retval -errno Error communicating with executor when registering resource * * \note Caller is responsible for freeing result on success. */ static int get_lrm_resource(lrm_state_t *lrm_state, xmlNode *rsc_xml, gboolean do_create, lrmd_rsc_info_t **rsc_info) { const char *id = ID(rsc_xml); CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL); CRM_CHECK(id, return -EINVAL); if (lrm_state_is_connected(lrm_state) == FALSE) { return -ENOTCONN; } crm_trace("Retrieving resource information for %s from the executor", id); *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); // If resource isn't known by ID, try clone name, if provided if (!*rsc_info) { const char *long_id = crm_element_value(rsc_xml, XML_ATTR_ID_LONG); if (long_id) { *rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0); } } if ((*rsc_info == NULL) && do_create) { const char *class = crm_element_value(rsc_xml, XML_AGENT_ATTR_CLASS); const char *provider = crm_element_value(rsc_xml, XML_AGENT_ATTR_PROVIDER); const char *type = crm_element_value(rsc_xml, XML_ATTR_TYPE); int rc; crm_trace("Registering resource %s with the executor", id); rc = lrm_state_register_rsc(lrm_state, id, class, provider, type, lrmd_opt_drop_recurring); if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Could not register resource %s with the executor on %s: %s " CRM_XS " rc=%d", id, lrm_state->node_name, pcmk_strerror(rc), rc); /* Register this as an internal error if this involves the local * executor. Otherwise, we're likely dealing with an unresponsive * remote node, which is not an FSA failure. */ if (lrm_state_is_local(lrm_state) == TRUE) { register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } return rc; } *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); } return *rsc_info? pcmk_ok : -ENODEV; } static void delete_resource(lrm_state_t * lrm_state, const char *id, lrmd_rsc_info_t * rsc, GHashTableIter * gIter, const char *sys, const char *host, const char *user, ha_msg_input_t * request, gboolean unregister) { int rc = pcmk_ok; crm_info("Removing resource %s for %s (%s) on %s", id, sys, user ? user : "internal", host); if (rsc && unregister) { rc = lrm_state_unregister_rsc(lrm_state, id, 0); } if (rc == pcmk_ok) { crm_trace("Resource '%s' deleted", id); } else if (rc == -EINPROGRESS) { crm_info("Deletion of resource '%s' pending", id); if (request) { struct pending_deletion_op_s *op = NULL; char *ref = crm_element_value_copy(request->msg, XML_ATTR_REFERENCE); op = calloc(1, sizeof(struct pending_deletion_op_s)); op->rsc = strdup(rsc->id); op->input = copy_ha_msg_input(request); g_hash_table_insert(lrm_state->deletion_ops, ref, op); } return; } else { crm_warn("Deletion of resource '%s' for %s (%s) on %s failed: %d", id, sys, user ? user : "internal", host, rc); } delete_rsc_entry(lrm_state, request, id, gIter, rc, user); } static int get_fake_call_id(lrm_state_t *lrm_state, const char *rsc_id) { int call_id = 999999999; rsc_history_t *entry = NULL; if(lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* Make sure the call id is greater than the last successful operation, * otherwise the failure will not result in a possible recovery of the resource * as it could appear the failure occurred before the successful start */ if (entry) { call_id = entry->last_callid + 1; } if (call_id < 0) { call_id = 1; } return call_id; } static void fake_op_status(lrm_state_t *lrm_state, lrmd_event_data_t *op, int op_status, enum ocf_exitcode op_exitcode) { op->call_id = get_fake_call_id(lrm_state, op->rsc_id); op->t_run = time(NULL); op->t_rcchange = op->t_run; op->op_status = op_status; op->rc = op_exitcode; } static void force_reprobe(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node) { GHashTableIter gIter; rsc_history_t *entry = NULL; crm_info("Clearing resource history on node %s", lrm_state->node_name); g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { /* only unregister the resource during a reprobe if it is not a remote connection * resource. otherwise unregistering the connection will terminate remote-node * membership */ gboolean unregister = TRUE; if (is_remote_lrmd_ra(NULL, NULL, entry->id)) { lrm_state_t *remote_lrm_state = lrm_state_find(entry->id); if (remote_lrm_state) { /* when forcing a reprobe, make sure to clear remote node before * clearing the remote node's connection resource */ force_reprobe(remote_lrm_state, from_sys, from_host, user_name, TRUE); } unregister = FALSE; } delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys, from_host, user_name, NULL, unregister); } /* Now delete the copy in the CIB */ erase_status_tag(lrm_state->node_name, XML_CIB_TAG_LRM, cib_scope_local); /* Finally, _delete_ the value in pacemaker-attrd -- setting it to FALSE * would result in the scheduler sending us back here again */ update_attrd(lrm_state->node_name, CRM_OP_PROBED, NULL, user_name, is_remote_node); } /*! * \internal * \brief Fail a requested action without actually executing it * * For an action that can't be executed, process it similarly to an actual * execution result, with specified error status (except for notify actions, * which will always be treated as successful). * * \param[in] lrm_state Executor connection that action is for * \param[in] action Action XML from request * \param[in] rc Desired return code to use * \param[in] op_status Desired operation status to use */ static void synthesize_lrmd_failure(lrm_state_t *lrm_state, xmlNode *action, int op_status, enum ocf_exitcode rc) { lrmd_event_data_t *op = NULL; const char *operation = crm_element_value(action, XML_LRM_ATTR_TASK); const char *target_node = crm_element_value(action, XML_LRM_ATTR_TARGET); xmlNode *xml_rsc = find_xml_node(action, XML_CIB_TAG_RESOURCE, TRUE); if ((xml_rsc == NULL) || (ID(xml_rsc) == NULL)) { /* @TODO Should we do something else, like direct ack? */ crm_info("Can't fake %s failure (%d) on %s without resource configuration", crm_element_value(action, XML_LRM_ATTR_TASK_KEY), rc, target_node); return; } else if(operation == NULL) { /* This probably came from crm_resource -C, nothing to do */ crm_info("Can't fake %s failure (%d) on %s without operation", ID(xml_rsc), rc, target_node); return; } op = construct_op(lrm_state, action, ID(xml_rsc), operation); if (safe_str_eq(operation, RSC_NOTIFY)) { // Notifications can't fail fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_OK); } else { fake_op_status(lrm_state, op, op_status, rc); } crm_info("Faking " CRM_OP_FMT " result (%d) on %s", op->rsc_id, op->op_type, op->interval_ms, op->rc, target_node); // Process the result as if it came from the LRM process_lrm_event(lrm_state, op, NULL, action); lrmd_free_event(op); } /*! * \internal * \brief Get target of an LRM operation * * \param[in] xml LRM operation data XML * * \return LRM operation target node name (local node or Pacemaker Remote node) */ static const char * lrm_op_target(xmlNode *xml) { const char *target = NULL; if (xml) { target = crm_element_value(xml, XML_LRM_ATTR_TARGET); } if (target == NULL) { target = fsa_our_uname; } return target; } static void fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { lrmd_event_data_t *op = NULL; lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = find_xml_node(xml, XML_CIB_TAG_RESOURCE, TRUE); CRM_CHECK(xml_rsc != NULL, return); /* The executor simply executes operations and reports the results, without * any concept of success or failure, so to fail a resource, we must fake * what a failure looks like. * * To do this, we create a fake executor operation event for the resource, * and pass that event to the executor client callback so it will be * processed as if it came from the executor. */ op = construct_op(lrm_state, xml, ID(xml_rsc), "asyncmon"); fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_UNKNOWN_ERROR); free((char*) op->user_data); op->user_data = NULL; op->interval_ms = 0; #if ENABLE_ACL if (user_name && is_privileged(user_name) == FALSE) { crm_err("%s does not have permission to fail %s", user_name, ID(xml_rsc)); send_direct_ack(from_host, from_sys, NULL, op, ID(xml_rsc)); lrmd_free_event(op); return; } #endif if (get_lrm_resource(lrm_state, xml_rsc, TRUE, &rsc) == pcmk_ok) { crm_info("Failing resource %s...", rsc->id); op->exit_reason = strdup("Simulated failure"); process_lrm_event(lrm_state, op, NULL, xml); op->op_status = PCMK_LRM_OP_DONE; op->rc = PCMK_OCF_OK; lrmd_free_rsc_info(rsc); } else { crm_info("Cannot find/create resource in order to fail it..."); crm_log_xml_warn(xml, "bad input"); } send_direct_ack(from_host, from_sys, NULL, op, ID(xml_rsc)); lrmd_free_event(op); } static void handle_refresh_op(lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { int rc = pcmk_ok; xmlNode *fragment = do_lrm_query_internal(lrm_state, node_update_all); fsa_cib_update(XML_CIB_TAG_STATUS, fragment, cib_quorum_override, rc, user_name); crm_info("Forced a local resource history refresh: call=%d", rc); if (safe_str_neq(CRM_SYSTEM_CRMD, from_sys)) { xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, fragment, from_host, from_sys, CRM_SYSTEM_LRMD, fsa_our_uuid); crm_debug("ACK'ing refresh from %s (%s)", from_sys, from_host); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(reply); } free_xml(fragment); } static void handle_query_op(xmlNode *msg, lrm_state_t *lrm_state) { xmlNode *data = do_lrm_query_internal(lrm_state, node_update_all); xmlNode *reply = create_reply(msg, data); if (relay_message(reply, TRUE) == FALSE) { crm_err("Unable to route reply"); crm_log_xml_err(reply, "reply"); } free_xml(reply); free_xml(data); } static void handle_reprobe_op(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node) { crm_notice("Forcing the status of all resources to be redetected"); force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node); if (safe_str_neq(CRM_SYSTEM_PENGINE, from_sys) && safe_str_neq(CRM_SYSTEM_TENGINE, from_sys)) { xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, NULL, from_host, from_sys, CRM_SYSTEM_LRMD, fsa_our_uuid); crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(reply); } } static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys) { char *op_key = NULL; char *meta_key = NULL; int call = 0; const char *call_id = NULL; const char *op_task = NULL; const char *interval_ms_s = NULL; gboolean in_progress = FALSE; xmlNode *params = find_xml_node(input->xml, XML_TAG_ATTRS, TRUE); CRM_CHECK(params != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_INTERVAL_MS); interval_ms_s = crm_element_value(params, meta_key); free(meta_key); CRM_CHECK(interval_ms_s != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_TASK); op_task = crm_element_value(params, meta_key); free(meta_key); CRM_CHECK(op_task != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_CALLID); call_id = crm_element_value(params, meta_key); free(meta_key); op_key = generate_op_key(rsc->id, op_task, crm_parse_ms(interval_ms_s)); crm_debug("Scheduler requested op %s (call=%s) be cancelled", op_key, (call_id? call_id : "NA")); call = crm_parse_int(call_id, "0"); if (call == 0) { // Normal case when the scheduler cancels a recurring op in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE); } else { // Normal case when the scheduler cancels an orphan op in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE); } // Acknowledge cancellation operation if for a remote connection resource if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) { char *op_id = make_stop_id(rsc->id, call); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) { crm_info("Nothing known about operation %d for %s", call, op_key); } erase_lrm_history_by_id(lrm_state, rsc->id, op_key, NULL, call); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); /* needed at least for cancellation of a remote operation */ g_hash_table_remove(lrm_state->pending_ops, op_id); free(op_id); } else { /* No ack is needed since abcdaa8, but peers with older versions * in a rolling upgrade need one. We didn't bump the feature set * at that commit, so we can only compare against the previous * CRM version (3.0.8). If any peers have feature set 3.0.9 but * not abcdaa8, they will time out waiting for the ack (no * released versions of Pacemaker are affected). */ const char *peer_version = crm_element_value(params, XML_ATTR_CRM_VERSION); if (compare_version(peer_version, "3.0.8") <= 0) { crm_info("Sending compatibility ack for %s cancellation to %s (CRM version %s)", op_key, from_host, peer_version); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); } } free(op_key); return TRUE; } static void do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host, bool crm_rsc_delete, const char *user_name) { gboolean unregister = TRUE; #if ENABLE_ACL int cib_rc = delete_rsc_status(lrm_state, rsc->id, cib_dryrun|cib_sync_call, user_name); if (cib_rc != pcmk_ok) { lrmd_event_data_t *op = NULL; crm_err("Could not delete resource status of %s for %s (user %s) on %s: %s" CRM_XS " rc=%d", rsc->id, from_sys, (user_name? user_name : "unknown"), from_host, pcmk_strerror(cib_rc), cib_rc); op = construct_op(lrm_state, input->xml, rsc->id, CRMD_ACTION_DELETE); op->op_status = PCMK_LRM_OP_ERROR; if (cib_rc == -EACCES) { op->rc = PCMK_OCF_INSUFFICIENT_PRIV; } else { op->rc = PCMK_OCF_UNKNOWN_ERROR; } send_direct_ack(from_host, from_sys, NULL, op, rsc->id); lrmd_free_event(op); return; } #endif if (crm_rsc_delete && is_remote_lrmd_ra(NULL, NULL, rsc->id)) { unregister = FALSE; } delete_resource(lrm_state, rsc->id, rsc, NULL, from_sys, from_host, user_name, input, unregister); } /* A_LRM_INVOKE */ void do_lrm_invoke(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { lrm_state_t *lrm_state = NULL; const char *crm_op = NULL; const char *from_sys = NULL; const char *from_host = NULL; const char *operation = NULL; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *user_name = NULL; const char *target_node = NULL; gboolean is_remote_node = FALSE; bool crm_rsc_delete = FALSE; target_node = lrm_op_target(input->xml); is_remote_node = safe_str_neq(target_node, fsa_our_uname); lrm_state = lrm_state_find(target_node); if ((lrm_state == NULL) && is_remote_node) { crm_err("Failing action because local node has never had connection to remote node %s", target_node); synthesize_lrmd_failure(NULL, input->xml, PCMK_LRM_OP_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR); return; } CRM_ASSERT(lrm_state != NULL); #if ENABLE_ACL user_name = crm_acl_get_set_user(input->msg, F_CRM_USER, NULL); crm_trace("Executor command from user '%s'", user_name); #endif crm_op = crm_element_value(input->msg, F_CRM_TASK); from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM); if (safe_str_neq(from_sys, CRM_SYSTEM_TENGINE)) { from_host = crm_element_value(input->msg, F_CRM_HOST_FROM); } crm_trace("Executor %s command from %s", crm_op, from_sys); if (safe_str_eq(crm_op, CRM_OP_LRM_DELETE)) { crm_rsc_delete = TRUE; // Only crm_resource uses this op operation = CRMD_ACTION_DELETE; } else if (safe_str_eq(crm_op, CRM_OP_LRM_FAIL)) { fail_lrm_resource(input->xml, lrm_state, user_name, from_host, from_sys); return; } else if (input->xml != NULL) { operation = crm_element_value(input->xml, XML_LRM_ATTR_TASK); } if (safe_str_eq(crm_op, CRM_OP_LRM_REFRESH)) { handle_refresh_op(lrm_state, user_name, from_host, from_sys); } else if (safe_str_eq(crm_op, CRM_OP_LRM_QUERY)) { handle_query_op(input->msg, lrm_state); } else if (safe_str_eq(operation, CRM_OP_PROBED)) { update_attrd(lrm_state->node_name, CRM_OP_PROBED, XML_BOOLEAN_TRUE, user_name, is_remote_node); } else if (safe_str_eq(operation, CRM_OP_REPROBE) || safe_str_eq(crm_op, CRM_OP_REPROBE)) { handle_reprobe_op(lrm_state, from_sys, from_host, user_name, is_remote_node); } else if (operation != NULL) { lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = find_xml_node(input->xml, XML_CIB_TAG_RESOURCE, TRUE); gboolean create_rsc = safe_str_neq(operation, CRMD_ACTION_DELETE); int rc; // We can't return anything meaningful without a resource ID CRM_CHECK(xml_rsc && ID(xml_rsc), return); rc = get_lrm_resource(lrm_state, xml_rsc, create_rsc, &rsc); if (rc == -ENOTCONN) { synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR); return; } else if ((rc < 0) && !create_rsc) { /* Delete of malformed or nonexistent resource * (deleting something that does not exist is a success) */ crm_notice("Not registering resource '%s' for a %s event " CRM_XS " get-rc=%d (%s) transition-key=%s", ID(xml_rsc), operation, rc, pcmk_strerror(rc), ID(input->xml)); delete_rsc_entry(lrm_state, input, ID(xml_rsc), NULL, pcmk_ok, user_name); send_task_ok_ack(lrm_state, input, ID(xml_rsc), NULL, operation, from_host, from_sys); return; } else if (rc == -EINVAL) { // Resource operation on malformed resource crm_err("Invalid resource definition for %s", ID(xml_rsc)); crm_log_xml_warn(input->msg, "invalid resource"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_ERROR, PCMK_OCF_NOT_CONFIGURED); // fatal error return; } else if (rc < 0) { // Error communicating with the executor crm_err("Could not register resource '%s' with executor: %s " CRM_XS " rc=%d", ID(xml_rsc), pcmk_strerror(rc), rc); crm_log_xml_warn(input->msg, "failed registration"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_ERROR, PCMK_OCF_INVALID_PARAM); // hard error return; } if (safe_str_eq(operation, CRMD_ACTION_CANCEL)) { if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) { crm_log_xml_warn(input->xml, "Bad command"); } } else if (safe_str_eq(operation, CRMD_ACTION_DELETE)) { do_lrm_delete(input, lrm_state, rsc, from_sys, from_host, crm_rsc_delete, user_name); } else { do_lrm_rsc_op(lrm_state, rsc, operation, input->xml, input->msg); } lrmd_free_rsc_info(rsc); } else { crm_err("Cannot perform operation %s of unknown type", crm_str(crm_op)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } static lrmd_event_data_t * construct_op(lrm_state_t * lrm_state, xmlNode * rsc_op, const char *rsc_id, const char *operation) { lrmd_event_data_t *op = NULL; const char *op_delay = NULL; const char *op_timeout = NULL; const char *interval_ms_s = NULL; GHashTable *params = NULL; const char *transition = NULL; CRM_ASSERT(rsc_id && operation); op = calloc(1, sizeof(lrmd_event_data_t)); CRM_ASSERT(op != NULL); op->type = lrmd_event_exec_complete; op->op_type = strdup(operation); op->op_status = PCMK_LRM_OP_PENDING; op->rc = -1; op->rsc_id = strdup(rsc_id); op->interval_ms = 0; op->timeout = 0; op->start_delay = 0; if (rsc_op == NULL) { CRM_LOG_ASSERT(safe_str_eq(CRMD_ACTION_STOP, operation)); op->user_data = NULL; /* the stop_all_resources() case * by definition there is no DC (or they'd be shutting * us down). * So we should put our version here. */ op->params = crm_str_table_new(); g_hash_table_insert(op->params, strdup(XML_ATTR_CRM_VERSION), strdup(CRM_FEATURE_SET)); crm_trace("Constructed %s op for %s", operation, rsc_id); return op; } params = xml2list(rsc_op); g_hash_table_remove(params, CRM_META "_op_target_rc"); op_delay = crm_meta_value(params, XML_OP_ATTR_START_DELAY); op_timeout = crm_meta_value(params, XML_ATTR_TIMEOUT); interval_ms_s = crm_meta_value(params, XML_LRM_ATTR_INTERVAL_MS); op->interval_ms = crm_parse_ms(interval_ms_s); op->timeout = crm_parse_int(op_timeout, "0"); op->start_delay = crm_parse_int(op_delay, "0"); #if ENABLE_VERSIONED_ATTRS // Resolve any versioned parameters if (lrm_state && safe_str_neq(op->op_type, RSC_METADATA) && safe_str_neq(op->op_type, CRMD_ACTION_DELETE) && !is_remote_lrmd_ra(NULL, NULL, rsc_id)) { // Resource info *should* already be cached, so we don't get executor call lrmd_rsc_info_t *rsc = lrm_state_get_rsc_info(lrm_state, rsc_id, 0); struct ra_metadata_s *metadata; metadata = metadata_cache_get(lrm_state->metadata_cache, rsc); if (metadata) { xmlNode *versioned_attrs = NULL; GHashTable *hash = NULL; char *key = NULL; char *value = NULL; GHashTableIter iter; versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_ATTRS); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_iter_steal(&iter); g_hash_table_replace(params, key, value); } g_hash_table_destroy(hash); versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_META); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_replace(params, crm_meta_name(key), strdup(value)); if (safe_str_eq(key, XML_ATTR_TIMEOUT)) { op->timeout = crm_parse_int(value, "0"); } else if (safe_str_eq(key, XML_OP_ATTR_START_DELAY)) { op->start_delay = crm_parse_int(value, "0"); } } g_hash_table_destroy(hash); versioned_attrs = first_named_child(rsc_op, XML_TAG_RSC_VER_ATTRS); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_iter_steal(&iter); g_hash_table_replace(params, key, value); } g_hash_table_destroy(hash); } lrmd_free_rsc_info(rsc); } #endif if (safe_str_neq(operation, RSC_STOP)) { op->params = params; } else { rsc_history_t *entry = NULL; if (lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* If we do not have stop parameters cached, use * whatever we are given */ if (!entry || !entry->stop_params) { op->params = params; } else { /* Copy the cached parameter list so that we stop the resource * with the old attributes, not the new ones */ op->params = crm_str_table_new(); g_hash_table_foreach(params, copy_meta_keys, op->params); g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params); g_hash_table_destroy(params); params = NULL; } } /* sanity */ if (op->timeout <= 0) { op->timeout = op->interval_ms; } if (op->start_delay < 0) { op->start_delay = 0; } transition = crm_element_value(rsc_op, XML_ATTR_TRANSITION_KEY); CRM_CHECK(transition != NULL, return op); op->user_data = strdup(transition); if (op->interval_ms != 0) { if (safe_str_eq(operation, CRMD_ACTION_START) || safe_str_eq(operation, CRMD_ACTION_STOP)) { crm_err("Start and Stop actions cannot have an interval: %u", op->interval_ms); op->interval_ms = 0; } } crm_trace("Constructed %s op for %s: interval=%u", operation, rsc_id, op->interval_ms); return op; } void send_direct_ack(const char *to_host, const char *to_sys, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op, const char *rsc_id) { xmlNode *reply = NULL; xmlNode *update, *iter; crm_node_t *peer = NULL; CRM_CHECK(op != NULL, return); if (op->rsc_id == NULL) { CRM_ASSERT(rsc_id != NULL); op->rsc_id = strdup(rsc_id); } if (to_sys == NULL) { to_sys = CRM_SYSTEM_TENGINE; } peer = crm_get_peer(0, fsa_our_uname); update = create_node_state_update(peer, node_update_none, NULL, __FUNCTION__); iter = create_xml_node(update, XML_CIB_TAG_LRM); crm_xml_add(iter, XML_ATTR_ID, fsa_our_uuid); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE); crm_xml_add(iter, XML_ATTR_ID, op->rsc_id); build_operation_update(iter, rsc, op, fsa_our_uname, __FUNCTION__); reply = create_request(CRM_OP_INVOKE_LRM, update, to_host, to_sys, CRM_SYSTEM_LRMD, NULL); crm_log_xml_trace(update, "ACK Update"); crm_debug("ACK'ing resource op " CRM_OP_FMT " from %s: %s", op->rsc_id, op->op_type, op->interval_ms, op->user_data, crm_element_value(reply, XML_ATTR_REFERENCE)); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(update); free_xml(reply); } gboolean verify_stopped(enum crmd_fsa_state cur_state, int log_level) { gboolean res = TRUE; GList *lrm_state_list = lrm_state_get_list(); GList *state_entry; for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) { lrm_state_t *lrm_state = state_entry->data; if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) { /* keep iterating through all even when false is returned */ res = FALSE; } } set_bit(fsa_input_register, R_SENT_RSC_STOP); g_list_free(lrm_state_list); lrm_state_list = NULL; return res; } struct stop_recurring_action_s { lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct stop_recurring_action_s *event = user_data; struct recurring_op_s *op = (struct recurring_op_s *)value; if ((op->interval_ms != 0) && crm_str_eq(op->rsc_id, event->rsc->id, TRUE)) { crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key); remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE); } return remove; } static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; lrm_state_t *lrm_state = user_data; struct recurring_op_s *op = (struct recurring_op_s *)value; if (op->interval_ms != 0) { crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (const char *) key); remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE); } return remove; } static void record_pending_op(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op) { const char *record_pending = NULL; CRM_CHECK(node_name != NULL, return); CRM_CHECK(rsc != NULL, return); CRM_CHECK(op != NULL, return); // Never record certain operation types as pending if ((op->op_type == NULL) || (op->params == NULL) || !controld_action_is_recordable(op->op_type)) { return; } // defaults to true record_pending = crm_meta_value(op->params, XML_OP_ATTR_PENDING); if (record_pending && !crm_is_true(record_pending)) { return; } op->call_id = -1; op->op_status = PCMK_LRM_OP_PENDING; op->rc = PCMK_OCF_UNKNOWN; op->t_run = time(NULL); op->t_rcchange = op->t_run; /* write a "pending" entry to the CIB, inhibit notification */ crm_debug("Recording pending op " CRM_OP_FMT " on %s in the CIB", op->rsc_id, op->op_type, op->interval_ms, node_name); do_update_resource(node_name, rsc, op); } static void do_lrm_rsc_op(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *operation, xmlNode * msg, xmlNode * request) { int call_id = 0; char *op_id = NULL; lrmd_event_data_t *op = NULL; lrmd_key_value_t *params = NULL; fsa_data_t *msg_data = NULL; const char *transition = NULL; gboolean stop_recurring = FALSE; bool send_nack = FALSE; CRM_CHECK(rsc != NULL, return); CRM_CHECK(operation != NULL, return); if (msg != NULL) { transition = crm_element_value(msg, XML_ATTR_TRANSITION_KEY); if (transition == NULL) { crm_log_xml_err(msg, "Missing transition number"); } } op = construct_op(lrm_state, msg, rsc->id, operation); CRM_CHECK(op != NULL, return); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) && (op->interval_ms == 0) && strcmp(operation, CRMD_ACTION_MIGRATE) == 0) { /* pcmk remote connections are a special use case. * We never ever want to stop monitoring a connection resource until * the entire migration has completed. If the connection is unexpectedly * severed, even during a migration, this is an event we must detect.*/ stop_recurring = FALSE; } else if ((op->interval_ms == 0) && strcmp(operation, CRMD_ACTION_STATUS) != 0 && strcmp(operation, CRMD_ACTION_NOTIFY) != 0) { /* stop any previous monitor operations before changing the resource state */ stop_recurring = TRUE; } if (stop_recurring == TRUE) { guint removed = 0; struct stop_recurring_action_s data; data.rsc = rsc; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove( lrm_state->pending_ops, stop_recurring_action_by_rsc, &data); if (removed) { crm_debug("Stopped %u recurring operation%s in preparation for " CRM_OP_FMT, removed, s_if_plural(removed), rsc->id, operation, op->interval_ms); } } /* now do the op */ crm_info("Performing key=%s op=" CRM_OP_FMT, transition, rsc->id, operation, op->interval_ms); if (is_set(fsa_input_register, R_SHUTDOWN) && safe_str_eq(operation, RSC_START)) { register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); send_nack = TRUE; } else if (fsa_state != S_NOT_DC && fsa_state != S_POLICY_ENGINE /* Recalculating */ && fsa_state != S_TRANSITION_ENGINE && safe_str_neq(operation, CRMD_ACTION_STOP)) { send_nack = TRUE; } if(send_nack) { crm_notice("Discarding attempt to perform action %s on %s in state %s (shutdown=%s)", operation, rsc->id, fsa_state2string(fsa_state), is_set(fsa_input_register, R_SHUTDOWN)?"true":"false"); op->rc = PCMK_OCF_UNKNOWN_ERROR; op->op_status = PCMK_LRM_OP_INVALID; send_direct_ack(NULL, NULL, rsc, op, rsc->id); lrmd_free_event(op); free(op_id); return; } record_pending_op(lrm_state->node_name, rsc, op); op_id = generate_op_key(rsc->id, op->op_type, op->interval_ms); if (op->interval_ms > 0) { /* cancel it so we can then restart it without conflict */ cancel_op_key(lrm_state, rsc, op_id, FALSE); } if (op->params) { char *key = NULL; char *value = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, op->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { params = lrmd_key_value_add(params, key, value); } } call_id = lrm_state_exec(lrm_state, rsc->id, op->op_type, op->user_data, op->interval_ms, op->timeout, op->start_delay, params); if (call_id <= 0 && lrm_state_is_local(lrm_state)) { crm_err("Operation %s on %s failed: %d", operation, rsc->id, call_id); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } else if (call_id <= 0) { crm_err("Operation %s on resource %s failed to execute on remote node %s: %d", operation, rsc->id, lrm_state->node_name, call_id); fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_UNKNOWN_ERROR); process_lrm_event(lrm_state, op, NULL, NULL); } else { /* record all operations so we can wait * for them to complete during shutdown */ char *call_id_s = make_stop_id(rsc->id, call_id); struct recurring_op_s *pending = NULL; pending = calloc(1, sizeof(struct recurring_op_s)); crm_trace("Recording pending op: %d - %s %s", call_id, op_id, call_id_s); pending->call_id = call_id; pending->interval_ms = op->interval_ms; pending->op_type = strdup(operation); pending->op_key = strdup(op_id); pending->rsc_id = strdup(rsc->id); pending->start_time = time(NULL); pending->user_data = op->user_data? strdup(op->user_data) : NULL; g_hash_table_replace(lrm_state->pending_ops, call_id_s, pending); if ((op->interval_ms > 0) && (op->start_delay > START_DELAY_THRESHOLD)) { int target_rc = 0; crm_info("Faking confirmation of %s: execution postponed for over 5 minutes", op_id); decode_transition_key(op->user_data, NULL, NULL, NULL, &target_rc); op->rc = target_rc; op->op_status = PCMK_LRM_OP_DONE; send_direct_ack(NULL, NULL, rsc, op, rsc->id); } pending->params = op->params; op->params = NULL; } free(op_id); lrmd_free_event(op); return; } int last_resource_update = 0; static void cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { switch (rc) { case pcmk_ok: case -pcmk_err_diff_failed: case -pcmk_err_diff_resync: crm_trace("Resource update %d complete: rc=%d", call_id, rc); break; default: crm_warn("Resource update %d failed: (rc=%d) %s", call_id, rc, pcmk_strerror(rc)); } if (call_id == last_resource_update) { last_resource_update = 0; trigger_fsa(fsa_source); } } static int do_update_resource(const char *node_name, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op) { /* */ int rc = pcmk_ok; xmlNode *update, *iter = NULL; int call_opt = crmd_cib_smart_opt(); const char *uuid = NULL; CRM_CHECK(op != NULL, return 0); iter = create_xml_node(iter, XML_CIB_TAG_STATUS); update = iter; iter = create_xml_node(iter, XML_CIB_TAG_STATE); if (safe_str_eq(node_name, fsa_our_uname)) { uuid = fsa_our_uuid; } else { /* remote nodes uuid and uname are equal */ uuid = node_name; crm_xml_add(iter, XML_NODE_IS_REMOTE, "true"); } CRM_LOG_ASSERT(uuid != NULL); if(uuid == NULL) { rc = -EINVAL; goto done; } crm_xml_add(iter, XML_ATTR_UUID, uuid); crm_xml_add(iter, XML_ATTR_UNAME, node_name); crm_xml_add(iter, XML_ATTR_ORIGIN, __FUNCTION__); iter = create_xml_node(iter, XML_CIB_TAG_LRM); crm_xml_add(iter, XML_ATTR_ID, uuid); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE); crm_xml_add(iter, XML_ATTR_ID, op->rsc_id); build_operation_update(iter, rsc, op, node_name, __FUNCTION__); if (rsc) { const char *container = NULL; crm_xml_add(iter, XML_ATTR_TYPE, rsc->type); crm_xml_add(iter, XML_AGENT_ATTR_CLASS, rsc->standard); crm_xml_add(iter, XML_AGENT_ATTR_PROVIDER, rsc->provider); if (op->params) { container = g_hash_table_lookup(op->params, CRM_META"_"XML_RSC_ATTR_CONTAINER); } if (container) { crm_trace("Resource %s is a part of container resource %s", op->rsc_id, container); crm_xml_add(iter, XML_RSC_ATTR_CONTAINER, container); } } else { crm_warn("Resource %s no longer exists in the executor", op->rsc_id); send_direct_ack(NULL, NULL, rsc, op, op->rsc_id); goto cleanup; } crm_log_xml_trace(update, __FUNCTION__); /* make it an asynchronous call and be done with it * * Best case: * the resource state will be discovered during * the next signup or election. * * Bad case: * we are shutting down and there is no DC at the time, * but then why were we shutting down then anyway? * (probably because of an internal error) * * Worst case: * we get shot for having resources "running" that really weren't * * the alternative however means blocking here for too long, which * isn't acceptable */ fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, rc, NULL); if (rc > 0) { last_resource_update = rc; } done: /* the return code is a call number, not an error code */ crm_trace("Sent resource state update message: %d for %s=%u on %s", rc, op->op_type, op->interval_ms, op->rsc_id); fsa_register_cib_callback(rc, FALSE, NULL, cib_rsc_callback); cleanup: free_xml(update); return rc; } void do_lrm_event(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input cur_input, fsa_data_t * msg_data) { CRM_CHECK(FALSE, return); } static char * unescape_newlines(const char *string) { char *pch = NULL; char *ret = NULL; static const char *escaped_newline = "\\n"; if (!string) { return NULL; } ret = strdup(string); pch = strstr(ret, escaped_newline); while (pch != NULL) { /* 2 chars for 2 chars, null-termination irrelevant */ memcpy(pch, "\n ", 2 * sizeof(char)); pch = strstr(pch, escaped_newline); } return ret; } static bool did_lrm_rsc_op_fail(lrm_state_t *lrm_state, const char * rsc_id, const char * op_type, guint interval_ms) { rsc_history_t *entry = NULL; CRM_CHECK(lrm_state != NULL, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); CRM_CHECK(op_type != NULL, return FALSE); entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->failed == NULL) { return FALSE; } if (crm_str_eq(entry->failed->rsc_id, rsc_id, TRUE) && safe_str_eq(entry->failed->op_type, op_type) && entry->failed->interval_ms == interval_ms) { return TRUE; } return FALSE; } void process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op, struct recurring_op_s *pending, xmlNode *action_xml) { char *op_id = NULL; char *op_key = NULL; int update_id = 0; gboolean remove = FALSE; gboolean removed = FALSE; bool need_direct_ack = FALSE; lrmd_rsc_info_t *rsc = NULL; const char *node_name = NULL; CRM_CHECK(op != NULL, return); CRM_CHECK(op->rsc_id != NULL, return); // Remap new status codes for older DCs if (compare_version(fsa_our_dc_version, "3.2.0") < 0) { switch (op->op_status) { case PCMK_LRM_OP_NOT_CONNECTED: op->op_status = PCMK_LRM_OP_ERROR; op->rc = PCMK_OCF_CONNECTION_DIED; break; case PCMK_LRM_OP_INVALID: op->op_status = PCMK_LRM_OP_ERROR; op->rc = CRM_DIRECT_NACK_RC; break; default: break; } } op_id = make_stop_id(op->rsc_id, op->call_id); op_key = generate_op_key(op->rsc_id, op->op_type, op->interval_ms); // Get resource info if available (from executor state or action XML) if (lrm_state) { rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0); } if ((rsc == NULL) && action_xml) { xmlNode *xml = find_xml_node(action_xml, XML_CIB_TAG_RESOURCE, TRUE); const char *standard = crm_element_value(xml, XML_AGENT_ATTR_CLASS); const char *provider = crm_element_value(xml, XML_AGENT_ATTR_PROVIDER); const char *type = crm_element_value(xml, XML_ATTR_TYPE); if (standard && type) { crm_info("%s agent information not cached, using %s%s%s:%s from action XML", op->rsc_id, standard, (provider? ":" : ""), (provider? provider : ""), type); rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type); } else { crm_err("Can't process %s result because %s agent information not cached or in XML", op_key, op->rsc_id); } } // Get node name if available (from executor state or action XML) if (lrm_state) { node_name = lrm_state->node_name; } else if (action_xml) { node_name = crm_element_value(action_xml, XML_LRM_ATTR_TARGET); } if(pending == NULL) { remove = TRUE; if (lrm_state) { pending = g_hash_table_lookup(lrm_state->pending_ops, op_id); } } if (op->op_status == PCMK_LRM_OP_ERROR) { switch(op->rc) { case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_MASTER: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_MASTER: // Leave it to the TE/scheduler to decide if this is an error op->op_status = PCMK_LRM_OP_DONE; break; default: /* Nothing to do */ break; } } if (op->op_status != PCMK_LRM_OP_CANCELLED) { /* We might not record the result, so directly acknowledge it to the * originator instead, so it doesn't time out waiting for the result * (especially important if part of a transition). */ need_direct_ack = TRUE; if (controld_action_is_recordable(op->op_type)) { if (node_name && rsc) { // We should record the result, and happily, we can update_id = do_update_resource(node_name, rsc, op); need_direct_ack = FALSE; } else if (op->rsc_deleted) { /* We shouldn't record the result (likely the resource was * refreshed, cleaned, or removed while this operation was * in flight). */ crm_notice("Not recording %s result in CIB because " "resource information was removed since it was initiated", op_key); } else { /* This shouldn't be possible; the executor didn't consider the * resource deleted, but we couldn't find resource or node * information. */ crm_err("Unable to record %s result in CIB: %s", op_key, (node_name? "No resource information" : "No node name")); } } } else if (op->interval_ms == 0) { /* A non-recurring operation was cancelled. Most likely, the * never-initiated action was removed from the executor's pending * operations list upon resource removal. */ need_direct_ack = TRUE; } else if (pending == NULL) { /* This recurring operation was cancelled, but was not pending. No * transition actions are waiting on it, nothing needs to be done. */ } else if (op->user_data == NULL) { /* This recurring operation was cancelled and pending, but we don't * have a transition key. This should never happen. */ crm_err("Recurring operation %s was cancelled without transition information", op_key); } else if (pending->remove) { /* This recurring operation was cancelled (by us) and pending, and we * have been waiting for it to finish. */ if (lrm_state) { erase_lrm_history_by_op(lrm_state, op); } /* If the recurring operation had failed, the lrm_rsc_op is recorded as * "last_failure" which won't get erased from the cib given the logic on * purpose in erase_lrm_history_by_op(). So that the cancel action won't * have a chance to get confirmed by DC with process_op_deletion(). * Cluster transition would get stuck waiting for the remaining action * timer to time out. * * Directly acknowledge the cancel operation in this case. */ if (did_lrm_rsc_op_fail(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms)) { need_direct_ack = TRUE; } } else if (op->rsc_deleted) { /* This recurring operation was cancelled (but not by us, and the * executor does not have resource information, likely due to resource * cleanup, refresh, or removal) and pending. */ crm_debug("Recurring op %s was cancelled due to resource deletion", op_key); need_direct_ack = TRUE; } else { /* This recurring operation was cancelled (but not by us, likely by the * executor before stopping the resource) and pending. We don't need to * do anything special. */ } if (need_direct_ack) { send_direct_ack(NULL, NULL, NULL, op, op->rsc_id); } if(remove == FALSE) { /* The caller will do this afterwards, but keep the logging consistent */ removed = TRUE; } else if (lrm_state && ((op->interval_ms == 0) || (op->op_status == PCMK_LRM_OP_CANCELLED))) { gboolean found = g_hash_table_remove(lrm_state->pending_ops, op_id); if (op->interval_ms != 0) { removed = TRUE; } else if (found) { removed = TRUE; crm_trace("Op %s (call=%d, stop-id=%s, remaining=%u): Confirmed", op_key, op->call_id, op_id, g_hash_table_size(lrm_state->pending_ops)); } } if (node_name == NULL) { node_name = "unknown node"; // for logging } switch (op->op_status) { case PCMK_LRM_OP_CANCELLED: crm_info("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s confirmed=%s", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_lrm_status_str(op->op_status), op->call_id, op_key, (removed? "true" : "false")); break; case PCMK_LRM_OP_DONE: crm_notice("Result of %s operation for %s on %s: %d (%s) " CRM_XS " call=%d key=%s confirmed=%s cib-update=%d", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, op->rc, services_ocf_exitcode_str(op->rc), op->call_id, op_key, (removed? "true" : "false"), update_id); break; case PCMK_LRM_OP_TIMEOUT: crm_err("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s timeout=%dms", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_lrm_status_str(op->op_status), op->call_id, op_key, op->timeout); break; default: crm_err("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s confirmed=%s status=%d cib-update=%d", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_lrm_status_str(op->op_status), op->call_id, op_key, (removed? "true" : "false"), op->op_status, update_id); } if (op->output) { char *prefix = crm_strdup_printf("%s-" CRM_OP_FMT ":%d", node_name, op->rsc_id, op->op_type, op->interval_ms, op->call_id); if (op->rc) { crm_log_output(LOG_NOTICE, prefix, op->output); } else { crm_log_output(LOG_DEBUG, prefix, op->output); } free(prefix); } if (lrm_state) { if (safe_str_neq(op->op_type, RSC_METADATA)) { crmd_alert_resource_op(lrm_state->node_name, op); } else if (rsc && (op->rc == PCMK_OCF_OK)) { char *metadata = unescape_newlines(op->output); metadata_cache_update(lrm_state->metadata_cache, rsc, metadata); free(metadata); } } if (op->rsc_deleted) { crm_info("Deletion of resource '%s' complete after %s", op->rsc_id, op_key); if (lrm_state) { delete_rsc_entry(lrm_state, NULL, op->rsc_id, NULL, pcmk_ok, NULL); } } /* If a shutdown was escalated while operations were pending, * then the FSA will be stalled right now... allow it to continue */ mainloop_set_trigger(fsa_source); if (lrm_state && rsc) { update_history_cache(lrm_state, rsc, op); } lrmd_free_rsc_info(rsc); free(op_key); free(op_id); } diff --git a/daemons/controld/controld_te_actions.c b/daemons/controld/controld_te_actions.c index 8f3b9d3cfd..5c9769a6e7 100644 --- a/daemons/controld/controld_te_actions.c +++ b/daemons/controld/controld_te_actions.c @@ -1,636 +1,639 @@ /* * Copyright 2004-2019 the Pacemaker project contributors * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include +#include // lrmd_event_data_t, lrmd_free_event() #include #include +#include #include #include #include #include #include #include #include char *te_uuid = NULL; GHashTable *te_targets = NULL; void send_rsc_command(crm_action_t * action); static void te_update_job_count(crm_action_t * action, int offset); static void te_start_action_timer(crm_graph_t * graph, crm_action_t * action) { action->timer = calloc(1, sizeof(crm_action_timer_t)); action->timer->timeout = action->timeout; action->timer->action = action; action->timer->source_id = g_timeout_add(action->timer->timeout + graph->network_delay, action_timer_callback, (void *)action->timer); CRM_ASSERT(action->timer->source_id != 0); } static gboolean te_pseudo_action(crm_graph_t * graph, crm_action_t * pseudo) { const char *task = crm_element_value(pseudo->xml, XML_LRM_ATTR_TASK); /* send to peers as well? */ if (safe_str_eq(task, CRM_OP_MAINTENANCE_NODES)) { GHashTableIter iter; crm_node_t *node = NULL; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *cmd = NULL; if (safe_str_eq(fsa_our_uname, node->uname)) { continue; } cmd = create_request(task, pseudo->xml, node->uname, CRM_SYSTEM_CRMD, CRM_SYSTEM_TENGINE, NULL); send_cluster_message(node, crm_msg_crmd, cmd, FALSE); free_xml(cmd); } remote_ra_process_maintenance_nodes(pseudo->xml); } else { /* Check action for Pacemaker Remote node side effects */ remote_ra_process_pseudo(pseudo->xml); } crm_debug("Pseudo-action %d (%s) fired and confirmed", pseudo->id, crm_element_value(pseudo->xml, XML_LRM_ATTR_TASK_KEY)); te_action_confirmed(pseudo, graph); return TRUE; } static int get_target_rc(crm_action_t * action) { const char *target_rc_s = crm_meta_value(action->params, XML_ATTR_TE_TARGET_RC); if (target_rc_s != NULL) { return crm_parse_int(target_rc_s, "0"); } return 0; } static gboolean te_crm_command(crm_graph_t * graph, crm_action_t * action) { char *counter = NULL; xmlNode *cmd = NULL; gboolean is_local = FALSE; const char *id = NULL; const char *task = NULL; const char *value = NULL; const char *on_node = NULL; const char *router_node = NULL; gboolean rc = TRUE; gboolean no_wait = FALSE; id = ID(action->xml); task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); router_node = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); if (!router_node) { router_node = on_node; } CRM_CHECK(on_node != NULL && strlen(on_node) != 0, crm_err("Corrupted command (id=%s) %s: no node", crm_str(id), crm_str(task)); return FALSE); if (safe_str_eq(router_node, fsa_our_uname)) { is_local = TRUE; } value = crm_meta_value(action->params, XML_ATTR_TE_NOWAIT); if (crm_is_true(value)) { no_wait = TRUE; } crm_info("Executing crm-event (%s)%s%s: %s on %s", crm_str(id), (is_local? " locally" : ""), (no_wait? " without waiting" : ""), crm_str(task), on_node); if (is_local && safe_str_eq(task, CRM_OP_SHUTDOWN)) { /* defer until everything else completes */ crm_info("crm-event (%s) is a local shutdown", crm_str(id)); graph->completion_action = tg_shutdown; graph->abort_reason = "local shutdown"; te_action_confirmed(action, graph); return TRUE; } else if (safe_str_eq(task, CRM_OP_SHUTDOWN)) { crm_node_t *peer = crm_get_peer(0, router_node); crm_update_peer_expected(__FUNCTION__, peer, CRMD_JOINSTATE_DOWN); } cmd = create_request(task, action->xml, router_node, CRM_SYSTEM_CRMD, CRM_SYSTEM_TENGINE, NULL); counter = generate_transition_key(transition_graph->id, action->id, get_target_rc(action), te_uuid); crm_xml_add(cmd, XML_ATTR_TRANSITION_KEY, counter); rc = send_cluster_message(crm_get_peer(0, router_node), crm_msg_crmd, cmd, TRUE); free(counter); free_xml(cmd); if (rc == FALSE) { crm_err("Action %d failed: send", action->id); return FALSE; } else if (no_wait) { te_action_confirmed(action, graph); } else { if (action->timeout <= 0) { crm_err("Action %d: %s on %s had an invalid timeout (%dms). Using %dms instead", action->id, task, on_node, action->timeout, graph->network_delay); action->timeout = graph->network_delay; } te_start_action_timer(graph, action); } return TRUE; } void controld_record_action_timeout(crm_action_t *action) { lrmd_event_data_t *op = NULL; xmlNode *state = NULL; xmlNode *rsc = NULL; xmlNode *xml_op = NULL; xmlNode *action_rsc = NULL; int rc = pcmk_ok; const char *rsc_id = NULL; const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); const char *target_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID); int call_options = cib_quorum_override | cib_scope_local; int target_rc = get_target_rc(action); crm_warn("%s %d: %s on %s timed out", crm_element_name(action->xml), action->id, task_uuid, target); action_rsc = find_xml_node(action->xml, XML_CIB_TAG_RESOURCE, TRUE); if (action_rsc == NULL) { return; } rsc_id = ID(action_rsc); CRM_CHECK(rsc_id != NULL, crm_log_xml_err(action->xml, "Bad:action"); return); /* update the CIB */ state = create_xml_node(NULL, XML_CIB_TAG_STATE); crm_xml_add(state, XML_ATTR_UUID, target_uuid); crm_xml_add(state, XML_ATTR_UNAME, target); rsc = create_xml_node(state, XML_CIB_TAG_LRM); crm_xml_add(rsc, XML_ATTR_ID, target_uuid); rsc = create_xml_node(rsc, XML_LRM_TAG_RESOURCES); rsc = create_xml_node(rsc, XML_LRM_TAG_RESOURCE); crm_xml_add(rsc, XML_ATTR_ID, rsc_id); crm_copy_xml_element(action_rsc, rsc, XML_ATTR_TYPE); crm_copy_xml_element(action_rsc, rsc, XML_AGENT_ATTR_CLASS); crm_copy_xml_element(action_rsc, rsc, XML_AGENT_ATTR_PROVIDER); /* If the executor gets a timeout while waiting for the action to complete, * that will be reported via the usual callback. This timeout means that we * didn't hear from the executor or the controller that relayed the action * to the executor. * * @TODO Using PCMK_OCF_UNKNOWN_ERROR instead of PCMK_OCF_TIMEOUT is one way * to distinguish those situations, but perhaps PCMK_OCF_TIMEOUT would be * preferable anyway. */ op = convert_graph_action(NULL, action, PCMK_LRM_OP_TIMEOUT, PCMK_OCF_UNKNOWN_ERROR); op->call_id = -1; op->user_data = generate_transition_key(transition_graph->id, action->id, target_rc, te_uuid); - xml_op = create_operation_update(rsc, op, CRM_FEATURE_SET, target_rc, target, __FUNCTION__, LOG_INFO); + xml_op = pcmk__create_history_xml(rsc, op, CRM_FEATURE_SET, target_rc, + target, __FUNCTION__, LOG_INFO); lrmd_free_event(op); crm_log_xml_trace(xml_op, "Action timeout"); rc = fsa_cib_conn->cmds->update(fsa_cib_conn, XML_CIB_TAG_STATUS, state, call_options); fsa_register_cib_callback(rc, FALSE, NULL, cib_action_updated); free_xml(state); crm_trace("Sent CIB update (call ID %d) for timeout of action %d (%s on %s)", rc, action->id, task_uuid, target); action->sent_update = TRUE; } static gboolean te_rsc_command(crm_graph_t * graph, crm_action_t * action) { /* never overwrite stop actions in the CIB with * anything other than completed results * * Writing pending stops makes it look like the * resource is running again */ xmlNode *cmd = NULL; xmlNode *rsc_op = NULL; gboolean rc = TRUE; gboolean no_wait = FALSE; gboolean is_local = FALSE; char *counter = NULL; const char *task = NULL; const char *value = NULL; const char *on_node = NULL; const char *router_node = NULL; const char *task_uuid = NULL; CRM_ASSERT(action != NULL); CRM_ASSERT(action->xml != NULL); action->executed = FALSE; on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); CRM_CHECK(on_node != NULL && strlen(on_node) != 0, crm_err("Corrupted command(id=%s) %s: no node", ID(action->xml), crm_str(task)); return FALSE); rsc_op = action->xml; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); task_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); router_node = crm_element_value(rsc_op, XML_LRM_ATTR_ROUTER_NODE); if (!router_node) { router_node = on_node; } counter = generate_transition_key(transition_graph->id, action->id, get_target_rc(action), te_uuid); crm_xml_add(rsc_op, XML_ATTR_TRANSITION_KEY, counter); if (safe_str_eq(router_node, fsa_our_uname)) { is_local = TRUE; } value = crm_meta_value(action->params, XML_ATTR_TE_NOWAIT); if (crm_is_true(value)) { no_wait = TRUE; } crm_notice("Initiating %s operation %s%s on %s%s "CRM_XS" action %d", task, task_uuid, (is_local? " locally" : ""), on_node, (no_wait? " without waiting" : ""), action->id); cmd = create_request(CRM_OP_INVOKE_LRM, rsc_op, router_node, CRM_SYSTEM_LRMD, CRM_SYSTEM_TENGINE, NULL); if (is_local) { /* shortcut local resource commands */ ha_msg_input_t data = { .msg = cmd, .xml = rsc_op, }; fsa_data_t msg = { .id = 0, .data = &data, .data_type = fsa_dt_ha_msg, .fsa_input = I_NULL, .fsa_cause = C_FSA_INTERNAL, .actions = A_LRM_INVOKE, .origin = __FUNCTION__, }; do_lrm_invoke(A_LRM_INVOKE, C_FSA_INTERNAL, fsa_state, I_NULL, &msg); } else { rc = send_cluster_message(crm_get_peer(0, router_node), crm_msg_lrmd, cmd, TRUE); } free(counter); free_xml(cmd); action->executed = TRUE; if (rc == FALSE) { crm_err("Action %d failed: send", action->id); return FALSE; } else if (no_wait) { crm_info("Action %d confirmed - no wait", action->id); action->confirmed = TRUE; /* Just mark confirmed. * Don't bump the job count only to immediately decrement it */ update_graph(transition_graph, action); trigger_graph(); } else if (action->confirmed == TRUE) { crm_debug("Action %d: %s %s on %s(timeout %dms) was already confirmed.", action->id, task, task_uuid, on_node, action->timeout); } else { if (action->timeout <= 0) { crm_err("Action %d: %s %s on %s had an invalid timeout (%dms). Using %dms instead", action->id, task, task_uuid, on_node, action->timeout, graph->network_delay); action->timeout = graph->network_delay; } te_update_job_count(action, 1); te_start_action_timer(graph, action); } return TRUE; } struct te_peer_s { char *name; int jobs; int migrate_jobs; }; static void te_peer_free(gpointer p) { struct te_peer_s *peer = p; free(peer->name); free(peer); } void te_reset_job_counts(void) { GHashTableIter iter; struct te_peer_s *peer = NULL; if(te_targets == NULL) { te_targets = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, te_peer_free); } g_hash_table_iter_init(&iter, te_targets); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & peer)) { peer->jobs = 0; peer->migrate_jobs = 0; } } static void te_update_job_count_on(const char *target, int offset, bool migrate) { struct te_peer_s *r = NULL; if(target == NULL || te_targets == NULL) { return; } r = g_hash_table_lookup(te_targets, target); if(r == NULL) { r = calloc(1, sizeof(struct te_peer_s)); r->name = strdup(target); g_hash_table_insert(te_targets, r->name, r); } r->jobs += offset; if(migrate) { r->migrate_jobs += offset; } crm_trace("jobs[%s] = %d", target, r->jobs); } static void te_update_job_count(crm_action_t * action, int offset) { const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); if (action->type != action_type_rsc || target == NULL) { /* No limit on these */ return; } /* if we have a router node, this means the action is performing * on a remote node. For now, we count all actions occurring on a * remote node against the job list on the cluster node hosting * the connection resources */ target = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); if ((target == NULL) && (safe_str_eq(task, CRMD_ACTION_MIGRATE) || safe_str_eq(task, CRMD_ACTION_MIGRATED))) { const char *t1 = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_SOURCE); const char *t2 = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_TARGET); te_update_job_count_on(t1, offset, TRUE); te_update_job_count_on(t2, offset, TRUE); return; } else if (target == NULL) { target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); } te_update_job_count_on(target, offset, FALSE); } static gboolean te_should_perform_action_on(crm_graph_t * graph, crm_action_t * action, const char *target) { int limit = 0; struct te_peer_s *r = NULL; const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); const char *id = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); if(target == NULL) { /* No limit on these */ return TRUE; } else if(te_targets == NULL) { return FALSE; } r = g_hash_table_lookup(te_targets, target); limit = throttle_get_job_limit(target); if(r == NULL) { r = calloc(1, sizeof(struct te_peer_s)); r->name = strdup(target); g_hash_table_insert(te_targets, r->name, r); } if(limit <= r->jobs) { crm_trace("Peer %s is over their job limit of %d (%d): deferring %s", target, limit, r->jobs, id); return FALSE; } else if(graph->migration_limit > 0 && r->migrate_jobs >= graph->migration_limit) { if (safe_str_eq(task, CRMD_ACTION_MIGRATE) || safe_str_eq(task, CRMD_ACTION_MIGRATED)) { crm_trace("Peer %s is over their migration job limit of %d (%d): deferring %s", target, graph->migration_limit, r->migrate_jobs, id); return FALSE; } } crm_trace("Peer %s has not hit their limit yet. current jobs = %d limit= %d limit", target, r->jobs, limit); return TRUE; } static gboolean te_should_perform_action(crm_graph_t * graph, crm_action_t * action) { const char *target = NULL; const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); if (action->type != action_type_rsc) { /* No limit on these */ return TRUE; } /* if we have a router node, this means the action is performing * on a remote node. For now, we count all actions occurring on a * remote node against the job list on the cluster node hosting * the connection resources */ target = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); if ((target == NULL) && (safe_str_eq(task, CRMD_ACTION_MIGRATE) || safe_str_eq(task, CRMD_ACTION_MIGRATED))) { target = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_SOURCE); if(te_should_perform_action_on(graph, action, target) == FALSE) { return FALSE; } target = crm_meta_value(action->params, XML_LRM_ATTR_MIGRATE_TARGET); } else if (target == NULL) { target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); } return te_should_perform_action_on(graph, action, target); } /*! * \brief Confirm a graph action (and optionally update graph) * * \param[in] action Action to confirm * \param[in] graph Update and trigger this graph (if non-NULL) */ void te_action_confirmed(crm_action_t *action, crm_graph_t *graph) { if (action->confirmed == FALSE) { if ((action->type == action_type_rsc) && (crm_element_value(action->xml, XML_LRM_ATTR_TARGET) != NULL)) { te_update_job_count(action, -1); } action->confirmed = TRUE; } if (graph) { update_graph(graph, action); trigger_graph(); } } crm_graph_functions_t te_graph_fns = { te_pseudo_action, te_rsc_command, te_crm_command, te_fence_node, te_should_perform_action, }; void notify_crmd(crm_graph_t * graph) { const char *type = "unknown"; enum crmd_fsa_input event = I_NULL; crm_debug("Processing transition completion in state %s", fsa_state2string(fsa_state)); if (graph->complete == FALSE) { CRM_CHECK(graph->complete,); graph->complete = TRUE; } switch (graph->completion_action) { case tg_stop: type = "stop"; if (fsa_state == S_TRANSITION_ENGINE) { event = I_TE_SUCCESS; } break; case tg_done: type = "done"; if (fsa_state == S_TRANSITION_ENGINE) { event = I_TE_SUCCESS; } break; case tg_restart: type = "restart"; if (fsa_state == S_TRANSITION_ENGINE) { if (transition_timer->period_ms > 0) { crm_timer_stop(transition_timer); crm_timer_start(transition_timer); } else { event = I_PE_CALC; } } else if (fsa_state == S_POLICY_ENGINE) { register_fsa_action(A_PE_INVOKE); } break; case tg_shutdown: type = "shutdown"; if (is_set(fsa_input_register, R_SHUTDOWN)) { event = I_STOP; } else { crm_err("We didn't ask to be shut down, yet the scheduler is telling us to"); event = I_TERMINATE; } } crm_debug("Transition %d status: %s - %s", graph->id, type, crm_str(graph->abort_reason)); graph->abort_reason = NULL; graph->completion_action = tg_done; clear_bit(fsa_input_register, R_IN_TRANSITION); if (event != I_NULL) { register_fsa_input(C_FSA_INTERNAL, event, NULL); } else if (fsa_source) { mainloop_set_trigger(fsa_source); } } diff --git a/include/crm/common/internal.h b/include/crm/common/internal.h index 6775d14632..b2eec000a6 100644 --- a/include/crm/common/internal.h +++ b/include/crm/common/internal.h @@ -1,172 +1,168 @@ /* * Copyright 2015-2019 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 CRM_COMMON_INTERNAL__H #define CRM_COMMON_INTERNAL__H #include /* for gboolean */ #include /* for struct dirent */ #include /* for getpid() */ #include /* for uid_t and gid_t */ #include /* internal I/O utilities (from io.c) */ char *generate_series_filename(const char *directory, const char *series, int sequence, gboolean bzip); int get_last_sequence(const char *directory, const char *series); void write_last_sequence(const char *directory, const char *series, int sequence, int max); int crm_chown_last_sequence(const char *directory, const char *series, uid_t uid, gid_t gid); bool pcmk__daemon_can_write(const char *dir, const char *file); void crm_sync_directory(const char *name); char *crm_read_contents(const char *filename); int crm_write_sync(int fd, const char *contents); int crm_set_nonblocking(int fd); const char *crm_get_tmpdir(void); /* internal procfs utilities (from procfs.c) */ int crm_procfs_process_info(struct dirent *entry, char *name, int *pid); int crm_procfs_pid_of(const char *name); unsigned int crm_procfs_num_cores(void); /* internal XML schema functions (from xml.c) */ void crm_schema_init(void); void crm_schema_cleanup(void); /* internal functions related to process IDs (from pid.c) */ /*! * \internal * \brief Detect if process per PID and optionally exe path (component) exists * * \param[in] pid PID of process assumed alive, disproving of which to try * \param[in] daemon exe path (component) to possibly match with procfs entry * * \return -1 on invalid PID specification, -2 when the calling process has no * (is refused an) ability to (dis)prove the predicate, * 0 if the negation of the predicate is confirmed (check-through-kill * indicates so, or the subsequent check-through-procfs-match on * \p daemon when provided and procfs available at the standard path), * 1 if it cannot be disproved (reliably [modulo race conditions] * when \p daemon provided, procfs available at the standard path * and the calling process has permissions to access the respective * procfs location, less so otherwise, since mere check-through-kill * is exercised without powers to exclude PID recycled in the interim). * * \note This function cannot be used to verify \e authenticity of the process. */ int crm_pid_active(long pid, const char *daemon); long crm_pidfile_inuse(const char *filename, long mypid, const char *daemon); long crm_read_pidfile(const char *filename); int crm_lock_pidfile(const char *filename, const char *name); /* interal functions related to resource operations (from operations.c) */ char *generate_op_key(const char *rsc_id, const char *op_type, guint interval_ms); char *generate_notify_key(const char *rsc_id, const char *notify_type, const char *op_type); char *generate_transition_key(int action, int transition_id, int target_rc, const char *node); void filter_action_parameters(xmlNode *param_set, const char *version); -xmlNode *create_operation_update(xmlNode *parent, lrmd_event_data_t *event, - const char *caller_version, int target_rc, - const char *node, const char *origin, - int level); // miscellaneous utilities (from utils.c) const char *pcmk_message_name(const char *name); /* internal generic string functions (from strings.c) */ long long crm_int_helper(const char *text, char **end_text); guint crm_parse_ms(const char *text); bool crm_starts_with(const char *str, const char *prefix); gboolean crm_ends_with(const char *s, const char *match); gboolean crm_ends_with_ext(const char *s, const char *match); char *add_list_element(char *list, const char *value); bool crm_compress_string(const char *data, int length, int max, char **result, unsigned int *result_len); gint crm_alpha_sort(gconstpointer a, gconstpointer b); static inline char * crm_concat(const char *prefix, const char *suffix, char join) { CRM_ASSERT(prefix && suffix); return crm_strdup_printf("%s%c%s", prefix, join, suffix); } static inline int crm_strlen_zero(const char *s) { return !s || *s == '\0'; } static inline char * crm_getpid_s() { return crm_strdup_printf("%lu", (unsigned long) getpid()); } /* convenience functions for failure-related node attributes */ #define CRM_FAIL_COUNT_PREFIX "fail-count" #define CRM_LAST_FAILURE_PREFIX "last-failure" /*! * \internal * \brief Generate a failure-related node attribute name for a resource * * \param[in] prefix Start of attribute name * \param[in] rsc_id Resource name * \param[in] op Operation name * \param[in] interval_ms Operation interval * * \return Newly allocated string with attribute name * * \note Failure attributes are named like PREFIX-RSC#OP_INTERVAL (for example, * "fail-count-myrsc#monitor_30000"). The '#' is used because it is not * a valid character in a resource ID, to reliably distinguish where the * operation name begins. The '_' is used simply to be more comparable to * action labels like "myrsc_monitor_30000". */ static inline char * crm_fail_attr_name(const char *prefix, const char *rsc_id, const char *op, guint interval_ms) { CRM_CHECK(prefix && rsc_id && op, return NULL); return crm_strdup_printf("%s-%s#%s_%u", prefix, rsc_id, op, interval_ms); } static inline char * crm_failcount_name(const char *rsc_id, const char *op, guint interval_ms) { return crm_fail_attr_name(CRM_FAIL_COUNT_PREFIX, rsc_id, op, interval_ms); } static inline char * crm_lastfailure_name(const char *rsc_id, const char *op, guint interval_ms) { return crm_fail_attr_name(CRM_LAST_FAILURE_PREFIX, rsc_id, op, interval_ms); } #endif /* CRM_COMMON_INTERNAL__H */ diff --git a/include/pcmki/pcmki_sched_utils.h b/include/pcmki/pcmki_sched_utils.h index 43612353ba..73e7284f7a 100644 --- a/include/pcmki/pcmki_sched_utils.h +++ b/include/pcmki/pcmki_sched_utils.h @@ -1,94 +1,100 @@ /* * 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 Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PENGINE_AUTILS__H # define PENGINE_AUTILS__H #include // bool #include // GList, GHashTable, gboolean, guint #include // GListPtr +#include // lrmd_event_data_t #include // cib_t #include #include #include /* Constraint helper functions */ extern rsc_colocation_t *invert_constraint(rsc_colocation_t * constraint); pe__location_t *copy_constraint(pe__location_t *constraint); pe__location_t *rsc2node_new(const char *id, pe_resource_t *rsc, int weight, const char *discovery_mode, pe_node_t *node, pe_working_set_t *data_set); extern gboolean rsc_colocation_new(const char *id, const char *node_attr, int score, resource_t * rsc_lh, resource_t * rsc_rh, const char *state_lh, const char *state_rh, pe_working_set_t * data_set); extern gboolean rsc_ticket_new(const char *id, resource_t * rsc_lh, ticket_t * ticket, const char *state_lh, const char *loss_policy, pe_working_set_t * data_set); GList *sort_nodes_by_weight(GList *nodes, pe_node_t *active_node, pe_working_set_t *data_set); extern gboolean can_run_resources(const node_t * node); extern gboolean native_assign_node(resource_t * rsc, GListPtr candidates, node_t * chosen, gboolean force); void native_deallocate(resource_t * rsc); extern void log_action(unsigned int log_level, const char *pre_text, action_t * action, gboolean details); gboolean can_run_any(GHashTable * nodes); pe_resource_t *find_compatible_child(pe_resource_t *local_child, pe_resource_t *rsc, enum rsc_role_e filter, gboolean current, pe_working_set_t *data_set); resource_t *find_compatible_child_by_node(resource_t * local_child, node_t * local_node, resource_t * rsc, enum rsc_role_e filter, gboolean current); gboolean is_child_compatible(resource_t *child_rsc, node_t * local_node, enum rsc_role_e filter, gboolean current); bool assign_node(resource_t * rsc, node_t * node, gboolean force); enum pe_action_flags summary_action_flags(action_t * action, GListPtr children, node_t * node); enum action_tasks clone_child_action(action_t * action); int copies_per_node(resource_t * rsc); enum filter_colocation_res { influence_nothing = 0, influence_rsc_location, influence_rsc_priority, }; extern enum filter_colocation_res filter_colocation_constraint(resource_t * rsc_lh, resource_t * rsc_rh, rsc_colocation_t * constraint, gboolean preview); extern int compare_capacity(const node_t * node1, const node_t * node2); extern void calculate_utilization(GHashTable * current_utilization, GHashTable * utilization, gboolean plus); extern void process_utilization(resource_t * rsc, node_t ** prefer, pe_working_set_t * data_set); pe_action_t *create_pseudo_resource_op(resource_t * rsc, const char *task, bool optional, bool runnable, pe_working_set_t *data_set); pe_action_t *pe_cancel_op(pe_resource_t *rsc, const char *name, guint interval_ms, pe_node_t *node, pe_working_set_t *data_set); pe_action_t *sched_shutdown_op(pe_node_t *node, pe_working_set_t *data_set); +xmlNode *pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *event, + const char *caller_version, int target_rc, + const char *node, const char *origin, + int level); + # define LOAD_STOPPED "load_stopped" void modify_configuration( pe_working_set_t * data_set, cib_t *cib, const char *quorum, const char *watchdog, GListPtr node_up, GListPtr node_down, GListPtr node_fail, GListPtr op_inject, GListPtr ticket_grant, GListPtr ticket_revoke, GListPtr ticket_standby, GListPtr ticket_activate); int run_simulation(pe_working_set_t * data_set, cib_t *cib, GListPtr op_fail_list, bool quiet); #endif diff --git a/lib/common/operations.c b/lib/common/operations.c index 480bddc912..a1ff8ade6a 100644 --- a/lib/common/operations.c +++ b/lib/common/operations.c @@ -1,638 +1,436 @@ /* * 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 Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #ifndef _GNU_SOURCE # define _GNU_SOURCE #endif #include #include #include #include #include #include #include #include #include /*! * \brief Generate an operation key * * \param[in] rsc_id ID of resource being operated on * \param[in] op_type Operation name * \param[in] interval_ms Operation interval * * \return Newly allocated memory containing operation key as string * * \note It is the caller's responsibility to free() the result. */ char * generate_op_key(const char *rsc_id, const char *op_type, guint interval_ms) { CRM_ASSERT(rsc_id != NULL); CRM_ASSERT(op_type != NULL); return crm_strdup_printf(CRM_OP_FMT, rsc_id, op_type, interval_ms); } gboolean parse_op_key(const char *key, char **rsc_id, char **op_type, guint *interval_ms) { char *notify = NULL; char *mutable_key = NULL; char *mutable_key_ptr = NULL; size_t len = 0, offset = 0; unsigned long long ch = 0; guint local_interval_ms = 0; // Initialize output variables in case of early return if (rsc_id) { *rsc_id = NULL; } if (op_type) { *op_type = NULL; } if (interval_ms) { *interval_ms = 0; } CRM_CHECK(key && *key, return FALSE); // Parse interval at end of string len = strlen(key); offset = len - 1; while ((offset > 0) && isdigit(key[offset])) { ch = key[offset] - '0'; for (int digits = len - offset; digits > 1; --digits) { ch = ch * 10; } local_interval_ms += ch; offset--; } crm_trace("Operation key '%s' has interval %ums", key, local_interval_ms); if (interval_ms) { *interval_ms = local_interval_ms; } CRM_CHECK((offset != (len - 1)) && (key[offset] == '_'), return FALSE); mutable_key = strndup(key, offset); offset--; while (offset > 0 && key[offset] != '_') { offset--; } CRM_CHECK(key[offset] == '_', free(mutable_key); return FALSE); mutable_key_ptr = mutable_key + offset + 1; crm_trace(" Action: %s", mutable_key_ptr); if (op_type) { *op_type = strdup(mutable_key_ptr); } mutable_key[offset] = 0; offset--; notify = strstr(mutable_key, "_post_notify"); if (notify && safe_str_eq(notify, "_post_notify")) { notify[0] = 0; } notify = strstr(mutable_key, "_pre_notify"); if (notify && safe_str_eq(notify, "_pre_notify")) { notify[0] = 0; } crm_trace(" Resource: %s", mutable_key); if (rsc_id) { *rsc_id = mutable_key; } else { free(mutable_key); } return TRUE; } char * generate_notify_key(const char *rsc_id, const char *notify_type, const char *op_type) { CRM_CHECK(rsc_id != NULL, return NULL); CRM_CHECK(op_type != NULL, return NULL); CRM_CHECK(notify_type != NULL, return NULL); return crm_strdup_printf("%s_%s_notify_%s_0", rsc_id, notify_type, op_type); } -static char * -generate_transition_magic(const char *transition_key, int op_status, int op_rc) -{ - CRM_CHECK(transition_key != NULL, return NULL); - return crm_strdup_printf("%d:%d;%s", op_status, op_rc, transition_key); -} - /*! * \brief Parse a transition magic string into its constituent parts * * \param[in] magic Magic string to parse (must be non-NULL) * \param[out] uuid If non-NULL, where to store copy of parsed UUID * \param[out] transition_id If non-NULL, where to store parsed transition ID * \param[out] action_id If non-NULL, where to store parsed action ID * \param[out] op_status If non-NULL, where to store parsed result status * \param[out] op_rc If non-NULL, where to store parsed actual rc * \param[out] target_rc If non-NULL, where to stored parsed target rc * * \return TRUE if key was valid, FALSE otherwise * \note If uuid is supplied and this returns TRUE, the caller is responsible * for freeing the memory for *uuid using free(). */ gboolean decode_transition_magic(const char *magic, char **uuid, int *transition_id, int *action_id, int *op_status, int *op_rc, int *target_rc) { int res = 0; char *key = NULL; gboolean result = TRUE; int local_op_status = -1; int local_op_rc = -1; CRM_CHECK(magic != NULL, return FALSE); #ifdef SSCANF_HAS_M res = sscanf(magic, "%d:%d;%ms", &local_op_status, &local_op_rc, &key); #else key = calloc(1, strlen(magic) - 3); // magic must have >=4 other characters CRM_ASSERT(key); res = sscanf(magic, "%d:%d;%s", &local_op_status, &local_op_rc, key); #endif if (res == EOF) { crm_err("Could not decode transition information '%s': %s", magic, pcmk_strerror(errno)); result = FALSE; } else if (res < 3) { crm_warn("Transition information '%s' incomplete (%d of 3 expected items)", magic, res); result = FALSE; } else { if (op_status) { *op_status = local_op_status; } if (op_rc) { *op_rc = local_op_rc; } result = decode_transition_key(key, uuid, transition_id, action_id, target_rc); } free(key); return result; } char * generate_transition_key(int transition_id, int action_id, int target_rc, const char *node) { CRM_CHECK(node != NULL, return NULL); return crm_strdup_printf("%d:%d:%d:%-*s", action_id, transition_id, target_rc, 36, node); } /*! * \brief Parse a transition key into its constituent parts * * \param[in] key Transition key to parse (must be non-NULL) * \param[out] uuid If non-NULL, where to store copy of parsed UUID * \param[out] transition_id If non-NULL, where to store parsed transition ID * \param[out] action_id If non-NULL, where to store parsed action ID * \param[out] target_rc If non-NULL, where to stored parsed target rc * * \return TRUE if key was valid, FALSE otherwise * \note If uuid is supplied and this returns TRUE, the caller is responsible * for freeing the memory for *uuid using free(). */ gboolean decode_transition_key(const char *key, char **uuid, int *transition_id, int *action_id, int *target_rc) { int local_transition_id = -1; int local_action_id = -1; int local_target_rc = -1; char local_uuid[37] = { '\0' }; // Initialize any supplied output arguments if (uuid) { *uuid = NULL; } if (transition_id) { *transition_id = -1; } if (action_id) { *action_id = -1; } if (target_rc) { *target_rc = -1; } CRM_CHECK(key != NULL, return FALSE); if (sscanf(key, "%d:%d:%d:%36s", &local_action_id, &local_transition_id, &local_target_rc, local_uuid) != 4) { crm_err("Invalid transition key '%s'", key); return FALSE; } if (strlen(local_uuid) != 36) { crm_warn("Invalid UUID '%s' in transition key '%s'", local_uuid, key); } if (uuid) { *uuid = strdup(local_uuid); CRM_ASSERT(*uuid); } if (transition_id) { *transition_id = local_transition_id; } if (action_id) { *action_id = local_action_id; } if (target_rc) { *target_rc = local_target_rc; } return TRUE; } void filter_action_parameters(xmlNode * param_set, const char *version) { char *key = NULL; char *timeout = NULL; char *interval_ms_s = NULL; const char *attr_filter[] = { XML_ATTR_ID, XML_ATTR_CRM_VERSION, XML_LRM_ATTR_OP_DIGEST, XML_LRM_ATTR_TARGET, XML_LRM_ATTR_TARGET_UUID, "pcmk_external_ip" }; gboolean do_delete = FALSE; int lpc = 0; static int meta_len = 0; if (meta_len == 0) { meta_len = strlen(CRM_META); } if (param_set == NULL) { return; } for (lpc = 0; lpc < DIMOF(attr_filter); lpc++) { xml_remove_prop(param_set, attr_filter[lpc]); } key = crm_meta_name(XML_LRM_ATTR_INTERVAL_MS); interval_ms_s = crm_element_value_copy(param_set, key); free(key); key = crm_meta_name(XML_ATTR_TIMEOUT); timeout = crm_element_value_copy(param_set, key); if (param_set) { xmlAttrPtr xIter = param_set->properties; while (xIter) { const char *prop_name = (const char *)xIter->name; xIter = xIter->next; do_delete = FALSE; if (strncasecmp(prop_name, CRM_META, meta_len) == 0) { do_delete = TRUE; } if (do_delete) { xml_remove_prop(param_set, prop_name); } } } if (interval_ms_s && strcmp(interval_ms_s, "0")) { /* Re-instate the operation's timeout value */ if (timeout != NULL) { crm_xml_add(param_set, key, timeout); } } free(interval_ms_s); free(timeout); free(key); } -#define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" -static void -append_digest(lrmd_event_data_t * op, xmlNode * update, const char *version, const char *magic, - int level) -{ - /* this will enable us to later determine that the - * resource's parameters have changed and we should force - * a restart - */ - char *digest = NULL; - xmlNode *args_xml = NULL; - - if (op->params == NULL) { - return; - } - - args_xml = create_xml_node(NULL, XML_TAG_PARAMS); - g_hash_table_foreach(op->params, hash2field, args_xml); - filter_action_parameters(args_xml, version); - digest = calculate_operation_digest(args_xml, version); - -#if 0 - if (level < get_crm_log_level() - && op->interval_ms == 0 && crm_str_eq(op->op_type, CRMD_ACTION_START, TRUE)) { - char *digest_source = dump_xml_unformatted(args_xml); - - do_crm_log(level, "Calculated digest %s for %s (%s). Source: %s\n", - digest, ID(update), magic, digest_source); - free(digest_source); - } -#endif - crm_xml_add(update, XML_LRM_ATTR_OP_DIGEST, digest); - - free_xml(args_xml); - free(digest); -} - int rsc_op_expected_rc(lrmd_event_data_t * op) { int rc = 0; if (op && op->user_data) { decode_transition_key(op->user_data, NULL, NULL, NULL, &rc); } return rc; } gboolean did_rsc_op_fail(lrmd_event_data_t * op, int target_rc) { switch (op->op_status) { case PCMK_LRM_OP_CANCELLED: case PCMK_LRM_OP_PENDING: return FALSE; break; case PCMK_LRM_OP_NOTSUPPORTED: case PCMK_LRM_OP_TIMEOUT: case PCMK_LRM_OP_ERROR: case PCMK_LRM_OP_NOT_CONNECTED: case PCMK_LRM_OP_INVALID: return TRUE; break; default: if (target_rc != op->rc) { return TRUE; } } return FALSE; } /*! * \brief Create a CIB XML element for an operation * * \param[in] parent If not NULL, make new XML node a child of this one * \param[in] prefix Generate an ID using this prefix * \param[in] task Operation task to set * \param[in] interval_spec Operation interval to set * \param[in] timeout If not NULL, operation timeout to set * * \return New XML object on success, NULL otherwise */ xmlNode * crm_create_op_xml(xmlNode *parent, const char *prefix, const char *task, const char *interval_spec, const char *timeout) { xmlNode *xml_op; CRM_CHECK(prefix && task && interval_spec, return NULL); xml_op = create_xml_node(parent, XML_ATTR_OP); crm_xml_set_id(xml_op, "%s-%s-%s", prefix, task, interval_spec); crm_xml_add(xml_op, XML_LRM_ATTR_INTERVAL, interval_spec); crm_xml_add(xml_op, "name", task); if (timeout) { crm_xml_add(xml_op, XML_ATTR_TIMEOUT, timeout); } return xml_op; } -xmlNode * -create_operation_update(xmlNode * parent, lrmd_event_data_t * op, const char * caller_version, - int target_rc, const char * node, const char * origin, int level) -{ - char *key = NULL; - char *magic = NULL; - char *op_id = NULL; - char *op_id_additional = NULL; - char *local_user_data = NULL; - const char *exit_reason = NULL; - - xmlNode *xml_op = NULL; - const char *task = NULL; - - CRM_CHECK(op != NULL, return NULL); - do_crm_log(level, "%s: Updating resource %s after %s op %s (interval=%u)", - origin, op->rsc_id, op->op_type, services_lrm_status_str(op->op_status), - op->interval_ms); - - crm_trace("DC version: %s", caller_version); - - task = op->op_type; - - /* Record a successful reload as a start, and a failed reload as a monitor, - * to make life easier for the scheduler when determining the current state. - */ - if (crm_str_eq(task, "reload", TRUE)) { - if (op->op_status == PCMK_LRM_OP_DONE) { - task = CRMD_ACTION_START; - } else { - task = CRMD_ACTION_STATUS; - } - } - - key = generate_op_key(op->rsc_id, task, op->interval_ms); - if (crm_str_eq(task, CRMD_ACTION_NOTIFY, TRUE)) { - const char *n_type = crm_meta_value(op->params, "notify_type"); - const char *n_task = crm_meta_value(op->params, "notify_operation"); - - CRM_LOG_ASSERT(n_type != NULL); - CRM_LOG_ASSERT(n_task != NULL); - op_id = generate_notify_key(op->rsc_id, n_type, n_task); - - if (op->op_status != PCMK_LRM_OP_PENDING) { - /* Ignore notify errors. - * - * @TODO It might be better to keep the correct result here, and - * ignore it in process_graph_event(). - */ - op->op_status = PCMK_LRM_OP_DONE; - op->rc = 0; - } - - } else if (did_rsc_op_fail(op, target_rc)) { - op_id = generate_op_key(op->rsc_id, "last_failure", 0); - if (op->interval_ms == 0) { - // Ensure 'last' gets updated, in case record-pending is true - op_id_additional = generate_op_key(op->rsc_id, "last", 0); - } - exit_reason = op->exit_reason; - - } else if (op->interval_ms > 0) { - op_id = strdup(key); - - } else { - op_id = generate_op_key(op->rsc_id, "last", 0); - } - - again: - xml_op = find_entity(parent, XML_LRM_TAG_RSC_OP, op_id); - if (xml_op == NULL) { - xml_op = create_xml_node(parent, XML_LRM_TAG_RSC_OP); - } - - if (op->user_data == NULL) { - crm_debug("Generating fake transition key for: " CRM_OP_FMT " %d from %s", - op->rsc_id, op->op_type, op->interval_ms, - op->call_id, origin); - local_user_data = generate_transition_key(-1, op->call_id, target_rc, FAKE_TE_ID); - op->user_data = local_user_data; - } - - if(magic == NULL) { - magic = generate_transition_magic(op->user_data, op->op_status, op->rc); - } - - crm_xml_add(xml_op, XML_ATTR_ID, op_id); - crm_xml_add(xml_op, XML_LRM_ATTR_TASK_KEY, key); - crm_xml_add(xml_op, XML_LRM_ATTR_TASK, task); - crm_xml_add(xml_op, XML_ATTR_ORIGIN, origin); - crm_xml_add(xml_op, XML_ATTR_CRM_VERSION, caller_version); - crm_xml_add(xml_op, XML_ATTR_TRANSITION_KEY, op->user_data); - crm_xml_add(xml_op, XML_ATTR_TRANSITION_MAGIC, magic); - crm_xml_add(xml_op, XML_LRM_ATTR_EXIT_REASON, exit_reason == NULL ? "" : exit_reason); - crm_xml_add(xml_op, XML_LRM_ATTR_TARGET, node); /* For context during triage */ - - crm_xml_add_int(xml_op, XML_LRM_ATTR_CALLID, op->call_id); - crm_xml_add_int(xml_op, XML_LRM_ATTR_RC, op->rc); - crm_xml_add_int(xml_op, XML_LRM_ATTR_OPSTATUS, op->op_status); - crm_xml_add_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, op->interval_ms); - - if (compare_version("2.1", caller_version) <= 0) { - if (op->t_run || op->t_rcchange || op->exec_time || op->queue_time) { - crm_trace("Timing data (" CRM_OP_FMT "): last=%u change=%u exec=%u queue=%u", - op->rsc_id, op->op_type, op->interval_ms, - op->t_run, op->t_rcchange, op->exec_time, op->queue_time); - - if (op->interval_ms == 0) { - /* The values are the same for non-recurring ops */ - crm_xml_add_int(xml_op, XML_RSC_OP_LAST_RUN, op->t_run); - crm_xml_add_int(xml_op, XML_RSC_OP_LAST_CHANGE, op->t_run); - - } else if(op->t_rcchange) { - /* last-run is not accurate for recurring ops */ - crm_xml_add_int(xml_op, XML_RSC_OP_LAST_CHANGE, op->t_rcchange); - - } else { - /* ...but is better than nothing otherwise */ - crm_xml_add_int(xml_op, XML_RSC_OP_LAST_CHANGE, op->t_run); - } - - crm_xml_add_int(xml_op, XML_RSC_OP_T_EXEC, op->exec_time); - crm_xml_add_int(xml_op, XML_RSC_OP_T_QUEUE, op->queue_time); - } - } - - if (crm_str_eq(op->op_type, CRMD_ACTION_MIGRATE, TRUE) - || crm_str_eq(op->op_type, CRMD_ACTION_MIGRATED, TRUE)) { - /* - * Record migrate_source and migrate_target always for migrate ops. - */ - const char *name = XML_LRM_ATTR_MIGRATE_SOURCE; - - crm_xml_add(xml_op, name, crm_meta_value(op->params, name)); - - name = XML_LRM_ATTR_MIGRATE_TARGET; - crm_xml_add(xml_op, name, crm_meta_value(op->params, name)); - } - - append_digest(op, xml_op, caller_version, magic, LOG_DEBUG); - - if (op_id_additional) { - free(op_id); - op_id = op_id_additional; - op_id_additional = NULL; - goto again; - } - - if (local_user_data) { - free(local_user_data); - op->user_data = NULL; - } - free(magic); - free(op_id); - free(key); - return xml_op; -} - /*! * \brief Check whether an operation requires resource agent meta-data * * \param[in] rsc_class Resource agent class (or NULL to skip class check) * \param[in] op Operation action (or NULL to skip op check) * * \return TRUE if operation needs meta-data, FALSE otherwise * \note At least one of rsc_class and op must be specified. */ bool crm_op_needs_metadata(const char *rsc_class, const char *op) { /* Agent meta-data is used to determine whether a reload is possible, and to * evaluate versioned parameters -- so if this op is not relevant to those * features, we don't need the meta-data. */ CRM_CHECK(rsc_class || op, return FALSE); if (rsc_class && is_not_set(pcmk_get_ra_caps(rsc_class), pcmk_ra_cap_params)) { /* Meta-data is only needed for resource classes that use parameters */ return FALSE; } /* Meta-data is only needed for these actions */ if (op && strcmp(op, CRMD_ACTION_START) && strcmp(op, CRMD_ACTION_STATUS) && strcmp(op, CRMD_ACTION_PROMOTE) && strcmp(op, CRMD_ACTION_DEMOTE) && strcmp(op, CRMD_ACTION_RELOAD) && strcmp(op, CRMD_ACTION_MIGRATE) && strcmp(op, CRMD_ACTION_MIGRATED) && strcmp(op, CRMD_ACTION_NOTIFY)) { return FALSE; } return TRUE; } diff --git a/lib/pacemaker/pcmk_sched_transition.c b/lib/pacemaker/pcmk_sched_transition.c index 056dfff65f..1c644722d7 100644 --- a/lib/pacemaker/pcmk_sched_transition.c +++ b/lib/pacemaker/pcmk_sched_transition.c @@ -1,861 +1,863 @@ /* * Copyright 2009-2019 the Pacemaker project contributors * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include +#include // lrmd_event_data_t, lrmd_free_event() #include #include #include #include #include static bool fake_quiet = FALSE; static cib_t *fake_cib = NULL; static GListPtr fake_resource_list = NULL; static GListPtr fake_op_fail_list = NULL; gboolean bringing_nodes_online = FALSE; #define STATUS_PATH_MAX 512 #define quiet_log(fmt, args...) do { \ if(fake_quiet) { \ crm_trace(fmt, ##args); \ } else { \ printf(fmt , ##args); \ } \ } while(0) #define NEW_NODE_TEMPLATE "//"XML_CIB_TAG_NODE"[@uname='%s']" #define NODE_TEMPLATE "//"XML_CIB_TAG_STATE"[@uname='%s']" #define RSC_TEMPLATE "//"XML_CIB_TAG_STATE"[@uname='%s']//"XML_LRM_TAG_RESOURCE"[@id='%s']" static void inject_transient_attr(xmlNode * cib_node, const char *name, const char *value) { xmlNode *attrs = NULL; xmlNode *instance_attrs = NULL; xmlChar *node_path; const char *node_uuid = ID(cib_node); node_path = xmlGetNodePath(cib_node); quiet_log(" + Injecting attribute %s=%s into %s '%s'\n", name, value, node_path, ID(cib_node)); free(node_path); attrs = first_named_child(cib_node, XML_TAG_TRANSIENT_NODEATTRS); if (attrs == NULL) { attrs = create_xml_node(cib_node, XML_TAG_TRANSIENT_NODEATTRS); crm_xml_add(attrs, XML_ATTR_ID, node_uuid); } instance_attrs = first_named_child(attrs, XML_TAG_ATTR_SETS); if (instance_attrs == NULL) { instance_attrs = create_xml_node(attrs, XML_TAG_ATTR_SETS); crm_xml_add(instance_attrs, XML_ATTR_ID, node_uuid); } crm_create_nvpair_xml(instance_attrs, NULL, name, value); } static void update_failcounts(xmlNode * cib_node, const char *resource, const char *task, guint interval_ms, int rc) { if (rc == 0) { return; } else if ((rc == 7) && (interval_ms == 0)) { return; } else { char *name = NULL; char *now = crm_itoa(time(NULL)); name = crm_failcount_name(resource, task, interval_ms); inject_transient_attr(cib_node, name, "value++"); free(name); name = crm_lastfailure_name(resource, task, interval_ms); inject_transient_attr(cib_node, name, now); free(name); free(now); } } static void create_node_entry(cib_t * cib_conn, const char *node) { int rc = pcmk_ok; char *xpath = crm_strdup_printf(NEW_NODE_TEMPLATE, node); rc = cib_conn->cmds->query(cib_conn, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); if (rc == -ENXIO) { xmlNode *cib_object = create_xml_node(NULL, XML_CIB_TAG_NODE); crm_xml_add(cib_object, XML_ATTR_ID, node); // Use node name as ID crm_xml_add(cib_object, XML_ATTR_UNAME, node); cib_conn->cmds->create(cib_conn, XML_CIB_TAG_NODES, cib_object, cib_sync_call | cib_scope_local); /* Not bothering with subsequent query to see if it exists, we'll bomb out later in the call to query_node_uuid()... */ free_xml(cib_object); } free(xpath); } static lrmd_event_data_t * create_op(xmlNode *cib_resource, const char *task, guint interval_ms, int outcome) { lrmd_event_data_t *op = NULL; xmlNode *xop = NULL; op = calloc(1, sizeof(lrmd_event_data_t)); op->rsc_id = strdup(ID(cib_resource)); op->interval_ms = interval_ms; op->op_type = strdup(task); op->rc = outcome; op->op_status = 0; op->params = NULL; /* TODO: Fill me in */ op->t_run = time(NULL); op->t_rcchange = op->t_run; op->call_id = 0; for (xop = __xml_first_child(cib_resource); xop != NULL; xop = __xml_next(xop)) { int tmp = 0; crm_element_value_int(xop, XML_LRM_ATTR_CALLID, &tmp); if (tmp > op->call_id) { op->call_id = tmp; } } op->call_id++; return op; } static xmlNode * inject_op(xmlNode * cib_resource, lrmd_event_data_t * op, int target_rc) { - return create_operation_update(cib_resource, op, CRM_FEATURE_SET, target_rc, - NULL, crm_system_name, LOG_TRACE); + return pcmk__create_history_xml(cib_resource, op, CRM_FEATURE_SET, + target_rc, NULL, crm_system_name, + LOG_TRACE); } static xmlNode * inject_node_state(cib_t * cib_conn, const char *node, const char *uuid) { int rc = pcmk_ok; xmlNode *cib_object = NULL; char *xpath = crm_strdup_printf(NODE_TEMPLATE, node); if (bringing_nodes_online) { create_node_entry(cib_conn, node); } rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object, cib_xpath | cib_sync_call | cib_scope_local); if (cib_object && ID(cib_object) == NULL) { crm_err("Detected multiple node_state entries for xpath=%s, bailing", xpath); crm_log_xml_warn(cib_object, "Duplicates"); free(xpath); crm_exit(CRM_EX_SOFTWARE); return NULL; // not reached, but makes static analysis happy } if (rc == -ENXIO) { char *found_uuid = NULL; if (uuid == NULL) { query_node_uuid(cib_conn, node, &found_uuid, NULL); } else { found_uuid = strdup(uuid); } cib_object = create_xml_node(NULL, XML_CIB_TAG_STATE); crm_xml_add(cib_object, XML_ATTR_UUID, found_uuid); crm_xml_add(cib_object, XML_ATTR_UNAME, node); cib_conn->cmds->create(cib_conn, XML_CIB_TAG_STATUS, cib_object, cib_sync_call | cib_scope_local); free_xml(cib_object); free(found_uuid); rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object, cib_xpath | cib_sync_call | cib_scope_local); crm_trace("injecting node state for %s. rc is %d", node, rc); } free(xpath); CRM_ASSERT(rc == pcmk_ok); return cib_object; } static xmlNode * modify_node(cib_t * cib_conn, char *node, gboolean up) { xmlNode *cib_node = inject_node_state(cib_conn, node, NULL); if (up) { crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_YES); crm_xml_add(cib_node, XML_NODE_IS_PEER, ONLINESTATUS); crm_xml_add(cib_node, XML_NODE_JOIN_STATE, CRMD_JOINSTATE_MEMBER); crm_xml_add(cib_node, XML_NODE_EXPECTED, CRMD_JOINSTATE_MEMBER); } else { crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_NO); crm_xml_add(cib_node, XML_NODE_IS_PEER, OFFLINESTATUS); crm_xml_add(cib_node, XML_NODE_JOIN_STATE, CRMD_JOINSTATE_DOWN); crm_xml_add(cib_node, XML_NODE_EXPECTED, CRMD_JOINSTATE_DOWN); } crm_xml_add(cib_node, XML_ATTR_ORIGIN, crm_system_name); return cib_node; } static xmlNode * find_resource_xml(xmlNode * cib_node, const char *resource) { xmlNode *match = NULL; const char *node = crm_element_value(cib_node, XML_ATTR_UNAME); char *xpath = crm_strdup_printf(RSC_TEMPLATE, node, resource); match = get_xpath_object(xpath, cib_node, LOG_TRACE); free(xpath); return match; } static xmlNode * inject_resource(xmlNode * cib_node, const char *resource, const char *lrm_name, const char *rclass, const char *rtype, const char *rprovider) { xmlNode *lrm = NULL; xmlNode *container = NULL; xmlNode *cib_resource = NULL; char *xpath = NULL; cib_resource = find_resource_xml(cib_node, resource); if (cib_resource != NULL) { /* If an existing LRM history entry uses the resource name, * continue using it, even if lrm_name is different. */ return cib_resource; } // Check for history entry under preferred name if (strcmp(resource, lrm_name)) { cib_resource = find_resource_xml(cib_node, lrm_name); if (cib_resource != NULL) { return cib_resource; } } /* One day, add query for class, provider, type */ if (rclass == NULL || rtype == NULL) { fprintf(stderr, "Resource %s not found in the status section of %s." " Please supply the class and type to continue\n", resource, ID(cib_node)); return NULL; } else if (safe_str_neq(rclass, PCMK_RESOURCE_CLASS_OCF) && safe_str_neq(rclass, PCMK_RESOURCE_CLASS_STONITH) && safe_str_neq(rclass, PCMK_RESOURCE_CLASS_SERVICE) && safe_str_neq(rclass, PCMK_RESOURCE_CLASS_UPSTART) && safe_str_neq(rclass, PCMK_RESOURCE_CLASS_SYSTEMD) && safe_str_neq(rclass, PCMK_RESOURCE_CLASS_LSB)) { fprintf(stderr, "Invalid class for %s: %s\n", resource, rclass); return NULL; } else if (is_set(pcmk_get_ra_caps(rclass), pcmk_ra_cap_provider) && (rprovider == NULL)) { fprintf(stderr, "Please specify the provider for resource %s\n", resource); return NULL; } xpath = (char *)xmlGetNodePath(cib_node); crm_info("Injecting new resource %s into %s '%s'", lrm_name, xpath, ID(cib_node)); free(xpath); lrm = first_named_child(cib_node, XML_CIB_TAG_LRM); if (lrm == NULL) { const char *node_uuid = ID(cib_node); lrm = create_xml_node(cib_node, XML_CIB_TAG_LRM); crm_xml_add(lrm, XML_ATTR_ID, node_uuid); } container = first_named_child(lrm, XML_LRM_TAG_RESOURCES); if (container == NULL) { container = create_xml_node(lrm, XML_LRM_TAG_RESOURCES); } cib_resource = create_xml_node(container, XML_LRM_TAG_RESOURCE); // If we're creating a new entry, use the preferred name crm_xml_add(cib_resource, XML_ATTR_ID, lrm_name); crm_xml_add(cib_resource, XML_AGENT_ATTR_CLASS, rclass); crm_xml_add(cib_resource, XML_AGENT_ATTR_PROVIDER, rprovider); crm_xml_add(cib_resource, XML_ATTR_TYPE, rtype); return cib_resource; } #define XPATH_MAX 1024 static int find_ticket_state(cib_t * the_cib, const char *ticket_id, xmlNode ** ticket_state_xml) { int offset = 0; int rc = pcmk_ok; xmlNode *xml_search = NULL; char *xpath_string = NULL; CRM_ASSERT(ticket_state_xml != NULL); *ticket_state_xml = NULL; xpath_string = calloc(1, XPATH_MAX); offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "%s", "/cib/status/tickets"); if (ticket_id) { offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "/%s[@id=\"%s\"]", XML_CIB_TAG_TICKET_STATE, ticket_id); } CRM_LOG_ASSERT(offset > 0); rc = the_cib->cmds->query(the_cib, xpath_string, &xml_search, cib_sync_call | cib_scope_local | cib_xpath); if (rc != pcmk_ok) { goto bail; } crm_log_xml_debug(xml_search, "Match"); if (xml_has_children(xml_search)) { if (ticket_id) { fprintf(stdout, "Multiple ticket_states match ticket_id=%s\n", ticket_id); } *ticket_state_xml = xml_search; } else { *ticket_state_xml = xml_search; } bail: free(xpath_string); return rc; } static int set_ticket_state_attr(const char *ticket_id, const char *attr_name, const char *attr_value, cib_t * cib, int cib_options) { int rc = pcmk_ok; xmlNode *xml_top = NULL; xmlNode *ticket_state_xml = NULL; rc = find_ticket_state(cib, ticket_id, &ticket_state_xml); if (rc == pcmk_ok) { crm_debug("Found a match state for ticket: id=%s", ticket_id); xml_top = ticket_state_xml; } else if (rc != -ENXIO) { return rc; } else { xmlNode *xml_obj = NULL; xml_top = create_xml_node(NULL, XML_CIB_TAG_STATUS); xml_obj = create_xml_node(xml_top, XML_CIB_TAG_TICKETS); ticket_state_xml = create_xml_node(xml_obj, XML_CIB_TAG_TICKET_STATE); crm_xml_add(ticket_state_xml, XML_ATTR_ID, ticket_id); } crm_xml_add(ticket_state_xml, attr_name, attr_value); crm_log_xml_debug(xml_top, "Update"); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, xml_top, cib_options); free_xml(xml_top); return rc; } void modify_configuration(pe_working_set_t * data_set, cib_t *cib, const char *quorum, const char *watchdog, GListPtr node_up, GListPtr node_down, GListPtr node_fail, GListPtr op_inject, GListPtr ticket_grant, GListPtr ticket_revoke, GListPtr ticket_standby, GListPtr ticket_activate) { int rc = pcmk_ok; GListPtr gIter = NULL; xmlNode *cib_op = NULL; xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; lrmd_event_data_t *op = NULL; if (quorum) { xmlNode *top = create_xml_node(NULL, XML_TAG_CIB); quiet_log(" + Setting quorum: %s\n", quorum); /* crm_xml_add(top, XML_ATTR_DC_UUID, dc_uuid); */ crm_xml_add(top, XML_ATTR_HAVE_QUORUM, quorum); rc = cib->cmds->modify(cib, NULL, top, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } if (watchdog) { quiet_log(" + Setting watchdog: %s\n", watchdog); rc = update_attr_delegate(cib, cib_sync_call | cib_scope_local, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, XML_ATTR_HAVE_WATCHDOG, watchdog, FALSE, NULL, NULL); CRM_ASSERT(rc == pcmk_ok); } for (gIter = node_up; gIter != NULL; gIter = gIter->next) { char *node = (char *)gIter->data; quiet_log(" + Bringing node %s online\n", node); cib_node = modify_node(cib, node, TRUE); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); } for (gIter = node_down; gIter != NULL; gIter = gIter->next) { char xpath[STATUS_PATH_MAX]; char *node = (char *)gIter->data; quiet_log(" + Taking node %s offline\n", node); cib_node = modify_node(cib, node, FALSE); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", node, XML_CIB_TAG_LRM); cib->cmds->remove(cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", node, XML_TAG_TRANSIENT_NODEATTRS); cib->cmds->remove(cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); } for (gIter = node_fail; gIter != NULL; gIter = gIter->next) { char *node = (char *)gIter->data; quiet_log(" + Failing node %s\n", node); cib_node = modify_node(cib, node, TRUE); crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_NO); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); } for (gIter = ticket_grant; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; quiet_log(" + Granting ticket %s\n", ticket_id); rc = set_ticket_state_attr(ticket_id, "granted", "true", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = ticket_revoke; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; quiet_log(" + Revoking ticket %s\n", ticket_id); rc = set_ticket_state_attr(ticket_id, "granted", "false", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = ticket_standby; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; quiet_log(" + Making ticket %s standby\n", ticket_id); rc = set_ticket_state_attr(ticket_id, "standby", "true", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = ticket_activate; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; quiet_log(" + Activating ticket %s\n", ticket_id); rc = set_ticket_state_attr(ticket_id, "standby", "false", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = op_inject; gIter != NULL; gIter = gIter->next) { char *spec = (char *)gIter->data; int rc = 0; int outcome = 0; guint interval_ms = 0; char *key = NULL; char *node = NULL; char *task = NULL; char *resource = NULL; const char *rtype = NULL; const char *rclass = NULL; const char *rprovider = NULL; resource_t *rsc = NULL; quiet_log(" + Injecting %s into the configuration\n", spec); key = calloc(1, strlen(spec) + 1); node = calloc(1, strlen(spec) + 1); rc = sscanf(spec, "%[^@]@%[^=]=%d", key, node, &outcome); if (rc != 3) { fprintf(stderr, "Invalid operation spec: %s. Only found %d fields\n", spec, rc); free(key); free(node); continue; } parse_op_key(key, &resource, &task, &interval_ms); rsc = pe_find_resource(data_set->resources, resource); if (rsc == NULL) { fprintf(stderr, " - Invalid resource name: %s\n", resource); } else { rclass = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); rtype = crm_element_value(rsc->xml, XML_ATTR_TYPE); rprovider = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER); cib_node = inject_node_state(cib, node, NULL); CRM_ASSERT(cib_node != NULL); update_failcounts(cib_node, resource, task, interval_ms, outcome); cib_resource = inject_resource(cib_node, resource, resource, rclass, rtype, rprovider); CRM_ASSERT(cib_resource != NULL); op = create_op(cib_resource, task, interval_ms, outcome); CRM_ASSERT(op != NULL); cib_op = inject_op(cib_resource, op, 0); CRM_ASSERT(cib_op != NULL); lrmd_free_event(op); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } free(task); free(node); free(key); } } static gboolean exec_pseudo_action(crm_graph_t * graph, crm_action_t * action) { const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); action->confirmed = TRUE; quiet_log(" * Pseudo action: %s%s%s\n", task, node ? " on " : "", node ? node : ""); update_graph(graph, action); return TRUE; } static gboolean exec_rsc_action(crm_graph_t * graph, crm_action_t * action) { int rc = 0; GListPtr gIter = NULL; lrmd_event_data_t *op = NULL; int target_outcome = 0; const char *rtype = NULL; const char *rclass = NULL; const char *resource = NULL; const char *rprovider = NULL; const char *lrm_name = NULL; const char *operation = crm_element_value(action->xml, "operation"); const char *target_rc_s = crm_meta_value(action->params, XML_ATTR_TE_TARGET_RC); xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; xmlNode *action_rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE); char *node = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET); char *uuid = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET_UUID); const char *router_node = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); if (safe_str_eq(operation, CRM_OP_PROBED) || safe_str_eq(operation, CRM_OP_REPROBE)) { crm_info("Skipping %s op for %s", operation, node); goto done; } if (action_rsc == NULL) { crm_log_xml_err(action->xml, "Bad"); free(node); free(uuid); return FALSE; } /* Look for the preferred name * If not found, try the expected 'local' name * If not found use the preferred name anyway */ resource = crm_element_value(action_rsc, XML_ATTR_ID); CRM_ASSERT(resource != NULL); // makes static analysis happy lrm_name = resource; // Preferred name when writing history if (pe_find_resource(fake_resource_list, resource) == NULL) { const char *longname = crm_element_value(action_rsc, XML_ATTR_ID_LONG); if (longname && pe_find_resource(fake_resource_list, longname)) { resource = longname; } } if (safe_str_eq(operation, "delete") || safe_str_eq(operation, RSC_METADATA)) { quiet_log(" * Resource action: %-15s %s on %s\n", resource, operation, node); goto done; } rclass = crm_element_value(action_rsc, XML_AGENT_ATTR_CLASS); rtype = crm_element_value(action_rsc, XML_ATTR_TYPE); rprovider = crm_element_value(action_rsc, XML_AGENT_ATTR_PROVIDER); if (target_rc_s != NULL) { target_outcome = crm_parse_int(target_rc_s, "0"); } CRM_ASSERT(fake_cib->cmds->query(fake_cib, NULL, NULL, cib_sync_call | cib_scope_local) == pcmk_ok); cib_node = inject_node_state(fake_cib, node, (router_node? node : uuid)); CRM_ASSERT(cib_node != NULL); cib_resource = inject_resource(cib_node, resource, lrm_name, rclass, rtype, rprovider); if (cib_resource == NULL) { crm_err("invalid resource in transition"); free(node); free(uuid); free_xml(cib_node); return FALSE; } op = convert_graph_action(cib_resource, action, 0, target_outcome); if (op->interval_ms) { quiet_log(" * Resource action: %-15s %s=%u on %s\n", resource, op->op_type, op->interval_ms, node); } else { quiet_log(" * Resource action: %-15s %s on %s\n", resource, op->op_type, node); } for (gIter = fake_op_fail_list; gIter != NULL; gIter = gIter->next) { char *spec = (char *)gIter->data; char *key = NULL; const char *match_name = NULL; // Allow user to specify anonymous clone with or without instance number key = crm_strdup_printf(CRM_OP_FMT "@%s=", resource, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = resource; } free(key); if ((match_name == NULL) && strcmp(resource, lrm_name)) { key = crm_strdup_printf(CRM_OP_FMT "@%s=", lrm_name, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = lrm_name; } free(key); } if (match_name != NULL) { rc = sscanf(spec, "%*[^=]=%d", (int *) &op->rc); // ${match_name}_${task}_${interval_in_ms}@${node}=${rc} if (rc != 1) { fprintf(stderr, "Invalid failed operation spec: %s. Result code must be integer\n", spec); continue; } action->failed = TRUE; graph->abort_priority = INFINITY; printf("\tPretending action %d failed with rc=%d\n", action->id, op->rc); update_failcounts(cib_node, match_name, op->op_type, op->interval_ms, op->rc); break; } } inject_op(cib_resource, op, target_outcome); lrmd_free_event(op); rc = fake_cib->cmds->modify(fake_cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); done: free(node); free(uuid); free_xml(cib_node); action->confirmed = TRUE; update_graph(graph, action); return TRUE; } static gboolean exec_crmd_action(crm_graph_t * graph, crm_action_t * action) { const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); xmlNode *rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE); action->confirmed = TRUE; if(rsc) { quiet_log(" * Cluster action: %s for %s on %s\n", task, ID(rsc), node); } else { quiet_log(" * Cluster action: %s on %s\n", task, node); } update_graph(graph, action); return TRUE; } static gboolean exec_stonith_action(crm_graph_t * graph, crm_action_t * action) { const char *op = crm_meta_value(action->params, "stonith_action"); char *target = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET); quiet_log(" * Fencing %s (%s)\n", target, op); if(safe_str_neq(op, "on")) { int rc = 0; char xpath[STATUS_PATH_MAX]; xmlNode *cib_node = modify_node(fake_cib, target, FALSE); crm_xml_add(cib_node, XML_ATTR_ORIGIN, __FUNCTION__); CRM_ASSERT(cib_node != NULL); rc = fake_cib->cmds->replace(fake_cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", target, XML_CIB_TAG_LRM); fake_cib->cmds->remove(fake_cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", target, XML_TAG_TRANSIENT_NODEATTRS); fake_cib->cmds->remove(fake_cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); free_xml(cib_node); } action->confirmed = TRUE; update_graph(graph, action); free(target); return TRUE; } int run_simulation(pe_working_set_t * data_set, cib_t *cib, GListPtr op_fail_list, bool quiet) { crm_graph_t *transition = NULL; enum transition_status graph_rc = -1; crm_graph_functions_t exec_fns = { exec_pseudo_action, exec_rsc_action, exec_crmd_action, exec_stonith_action, }; fake_cib = cib; fake_quiet = quiet; fake_op_fail_list = op_fail_list; quiet_log("\nExecuting cluster transition:\n"); set_graph_functions(&exec_fns); transition = unpack_graph(data_set->graph, crm_system_name); print_graph(LOG_DEBUG, transition); fake_resource_list = data_set->resources; do { graph_rc = run_graph(transition); } while (graph_rc == transition_active); fake_resource_list = NULL; if (graph_rc != transition_complete) { fprintf(stdout, "Transition failed: %s\n", transition_status(graph_rc)); print_graph(LOG_ERR, transition); } destroy_graph(transition); if (graph_rc != transition_complete) { fprintf(stdout, "An invalid transition was produced\n"); } if (quiet == FALSE) { xmlNode *cib_object = NULL; int rc = fake_cib->cmds->query(fake_cib, NULL, &cib_object, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); pe_reset_working_set(data_set); data_set->input = cib_object; } if (graph_rc != transition_complete) { return graph_rc; } return 0; } diff --git a/lib/pacemaker/pcmk_sched_utils.c b/lib/pacemaker/pcmk_sched_utils.c index 7b5cb7df5a..68a8bba02f 100644 --- a/lib/pacemaker/pcmk_sched_utils.c +++ b/lib/pacemaker/pcmk_sched_utils.c @@ -1,467 +1,686 @@ /* * Copyright 2004-2019 Andrew Beekhof * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include +#include // lrmd_event_data_t #include pe__location_t * rsc2node_new(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) { pe__location_t *new_con = NULL; if (rsc == NULL || id == NULL) { pe_err("Invalid constraint %s for rsc=%p", crm_str(id), rsc); return NULL; } else if (foo_node == NULL) { CRM_CHECK(node_weight == 0, return NULL); } new_con = calloc(1, sizeof(pe__location_t)); if (new_con != NULL) { new_con->id = strdup(id); new_con->rsc_lh = rsc; new_con->node_list_rh = NULL; new_con->role_filter = RSC_ROLE_UNKNOWN; if (discover_mode == NULL || safe_str_eq(discover_mode, "always")) { new_con->discover_mode = pe_discover_always; } else if (safe_str_eq(discover_mode, "never")) { new_con->discover_mode = pe_discover_never; } else if (safe_str_eq(discover_mode, "exclusive")) { new_con->discover_mode = pe_discover_exclusive; rsc->exclusive_discover = TRUE; } else { pe_err("Invalid %s value %s in location constraint", XML_LOCATION_ATTR_DISCOVERY, discover_mode); } if (foo_node != NULL) { node_t *copy = node_copy(foo_node); copy->weight = node_weight; new_con->node_list_rh = g_list_prepend(NULL, copy); } data_set->placement_constraints = g_list_prepend(data_set->placement_constraints, new_con); rsc->rsc_location = g_list_prepend(rsc->rsc_location, new_con); } return new_con; } gboolean can_run_resources(const node_t * node) { if (node == NULL) { return FALSE; } #if 0 if (node->weight < 0) { return FALSE; } #endif if (node->details->online == FALSE || node->details->shutdown || node->details->unclean || node->details->standby || node->details->maintenance) { crm_trace("%s: online=%d, unclean=%d, standby=%d, maintenance=%d", node->details->uname, node->details->online, node->details->unclean, node->details->standby, node->details->maintenance); return FALSE; } return TRUE; } struct node_weight_s { pe_node_t *active; pe_working_set_t *data_set; }; /* return -1 if 'a' is more preferred * return 1 if 'b' is more preferred */ static gint sort_node_weight(gconstpointer a, gconstpointer b, gpointer data) { const node_t *node1 = (const node_t *)a; const node_t *node2 = (const node_t *)b; struct node_weight_s *nw = data; int node1_weight = 0; int node2_weight = 0; int result = 0; if (a == NULL) { return 1; } if (b == NULL) { return -1; } node1_weight = node1->weight; node2_weight = node2->weight; if (can_run_resources(node1) == FALSE) { node1_weight = -INFINITY; } if (can_run_resources(node2) == FALSE) { node2_weight = -INFINITY; } if (node1_weight > node2_weight) { crm_trace("%s (%d) > %s (%d) : weight", node1->details->uname, node1_weight, node2->details->uname, node2_weight); return -1; } if (node1_weight < node2_weight) { crm_trace("%s (%d) < %s (%d) : weight", node1->details->uname, node1_weight, node2->details->uname, node2_weight); return 1; } crm_trace("%s (%d) == %s (%d) : weight", node1->details->uname, node1_weight, node2->details->uname, node2_weight); if (safe_str_eq(nw->data_set->placement_strategy, "minimal")) { goto equal; } if (safe_str_eq(nw->data_set->placement_strategy, "balanced")) { result = compare_capacity(node1, node2); if (result < 0) { crm_trace("%s > %s : capacity (%d)", node1->details->uname, node2->details->uname, result); return -1; } else if (result > 0) { crm_trace("%s < %s : capacity (%d)", node1->details->uname, node2->details->uname, result); return 1; } } /* now try to balance resources across the cluster */ if (node1->details->num_resources < node2->details->num_resources) { crm_trace("%s (%d) > %s (%d) : resources", node1->details->uname, node1->details->num_resources, node2->details->uname, node2->details->num_resources); return -1; } else if (node1->details->num_resources > node2->details->num_resources) { crm_trace("%s (%d) < %s (%d) : resources", node1->details->uname, node1->details->num_resources, node2->details->uname, node2->details->num_resources); return 1; } if (nw->active && nw->active->details == node1->details) { crm_trace("%s (%d) > %s (%d) : active", node1->details->uname, node1->details->num_resources, node2->details->uname, node2->details->num_resources); return -1; } else if (nw->active && nw->active->details == node2->details) { crm_trace("%s (%d) < %s (%d) : active", node1->details->uname, node1->details->num_resources, node2->details->uname, node2->details->num_resources); return 1; } equal: crm_trace("%s = %s", node1->details->uname, node2->details->uname); return strcmp(node1->details->uname, node2->details->uname); } GList * sort_nodes_by_weight(GList *nodes, pe_node_t *active_node, pe_working_set_t *data_set) { struct node_weight_s nw = { active_node, data_set }; return g_list_sort_with_data(nodes, sort_node_weight, &nw); } void native_deallocate(resource_t * rsc) { if (rsc->allocated_to) { node_t *old = rsc->allocated_to; crm_info("Deallocating %s from %s", rsc->id, old->details->uname); set_bit(rsc->flags, pe_rsc_provisional); rsc->allocated_to = NULL; old->details->allocated_rsc = g_list_remove(old->details->allocated_rsc, rsc); old->details->num_resources--; /* old->count--; */ calculate_utilization(old->details->utilization, rsc->utilization, TRUE); free(old); } } gboolean native_assign_node(resource_t * rsc, GListPtr nodes, node_t * chosen, gboolean force) { CRM_ASSERT(rsc->variant == pe_native); if (force == FALSE && chosen != NULL) { bool unset = FALSE; if(chosen->weight < 0) { unset = TRUE; // Allow the graph to assume that the remote resource will come up } else if (!can_run_resources(chosen) && !pe__is_guest_node(chosen)) { unset = TRUE; } if(unset) { crm_debug("All nodes for resource %s are unavailable" ", unclean or shutting down (%s: %d, %d)", rsc->id, chosen->details->uname, can_run_resources(chosen), chosen->weight); rsc->next_role = RSC_ROLE_STOPPED; chosen = NULL; } } /* todo: update the old node for each resource to reflect its * new resource count */ native_deallocate(rsc); clear_bit(rsc->flags, pe_rsc_provisional); if (chosen == NULL) { GListPtr gIter = NULL; char *rc_inactive = crm_itoa(PCMK_OCF_NOT_RUNNING); crm_debug("Could not allocate a node for %s", rsc->id); rsc->next_role = RSC_ROLE_STOPPED; for (gIter = rsc->actions; gIter != NULL; gIter = gIter->next) { action_t *op = (action_t *) gIter->data; const char *interval_ms_s = g_hash_table_lookup(op->meta, XML_LRM_ATTR_INTERVAL_MS); crm_debug("Processing %s", op->uuid); if(safe_str_eq(RSC_STOP, op->task)) { update_action_flags(op, pe_action_optional | pe_action_clear, __FUNCTION__, __LINE__); } else if(safe_str_eq(RSC_START, op->task)) { update_action_flags(op, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__); /* set_bit(rsc->flags, pe_rsc_block); */ } else if (interval_ms_s && safe_str_neq(interval_ms_s, "0")) { if(safe_str_eq(rc_inactive, g_hash_table_lookup(op->meta, XML_ATTR_TE_TARGET_RC))) { /* This is a recurring monitor for the stopped state, leave it alone */ } else { /* Normal monitor operation, cancel it */ update_action_flags(op, pe_action_runnable | pe_action_clear, __FUNCTION__, __LINE__); } } } free(rc_inactive); return FALSE; } crm_debug("Assigning %s to %s", chosen->details->uname, rsc->id); rsc->allocated_to = node_copy(chosen); chosen->details->allocated_rsc = g_list_prepend(chosen->details->allocated_rsc, rsc); chosen->details->num_resources++; chosen->count++; calculate_utilization(chosen->details->utilization, rsc->utilization, FALSE); dump_rsc_utilization(show_utilization ? 0 : utilization_log_level, __FUNCTION__, rsc, chosen); return TRUE; } void log_action(unsigned int log_level, const char *pre_text, action_t * action, gboolean details) { const char *node_uname = NULL; const char *node_uuid = NULL; if (action == NULL) { crm_trace("%s%s: ", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": "); return; } if (is_set(action->flags, pe_action_pseudo)) { node_uname = NULL; node_uuid = NULL; } else if (action->node != NULL) { node_uname = action->node->details->uname; node_uuid = action->node->details->id; } else { node_uname = ""; node_uuid = NULL; } switch (text2task(action->task)) { case stonith_node: case shutdown_crm: crm_trace("%s%s%sAction %d: %s%s%s%s%s%s", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": ", is_set(action->flags, pe_action_pseudo) ? "Pseudo " : is_set(action->flags, pe_action_optional) ? "Optional " : is_set(action->flags, pe_action_runnable) ? is_set(action->flags, pe_action_processed) ? "" : "(Provisional) " : "!!Non-Startable!! ", action->id, action->uuid, node_uname ? "\ton " : "", node_uname ? node_uname : "", node_uuid ? "\t\t(" : "", node_uuid ? node_uuid : "", node_uuid ? ")" : ""); break; default: crm_trace("%s%s%sAction %d: %s %s%s%s%s%s%s", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": ", is_set(action->flags, pe_action_optional) ? "Optional " : is_set(action->flags, pe_action_pseudo) ? "Pseudo " : is_set(action->flags, pe_action_runnable) ? is_set(action->flags, pe_action_processed) ? "" : "(Provisional) " : "!!Non-Startable!! ", action->id, action->uuid, action->rsc ? action->rsc->id : "", node_uname ? "\ton " : "", node_uname ? node_uname : "", node_uuid ? "\t\t(" : "", node_uuid ? node_uuid : "", node_uuid ? ")" : ""); break; } if (details) { GListPtr gIter = NULL; crm_trace("\t\t====== Preceding Actions"); gIter = action->actions_before; for (; gIter != NULL; gIter = gIter->next) { action_wrapper_t *other = (action_wrapper_t *) gIter->data; log_action(log_level + 1, "\t\t", other->action, FALSE); } crm_trace("\t\t====== Subsequent Actions"); gIter = action->actions_after; for (; gIter != NULL; gIter = gIter->next) { action_wrapper_t *other = (action_wrapper_t *) gIter->data; log_action(log_level + 1, "\t\t", other->action, FALSE); } crm_trace("\t\t====== End"); } else { crm_trace("\t\t(before=%d, after=%d)", g_list_length(action->actions_before), g_list_length(action->actions_after)); } } gboolean can_run_any(GHashTable * nodes) { GHashTableIter iter; node_t *node = NULL; if (nodes == NULL) { return FALSE; } g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if (can_run_resources(node) && node->weight >= 0) { return TRUE; } } return FALSE; } pe_action_t * create_pseudo_resource_op(resource_t * rsc, const char *task, bool optional, bool runnable, pe_working_set_t *data_set) { pe_action_t *action = custom_action(rsc, generate_op_key(rsc->id, task, 0), task, NULL, optional, TRUE, data_set); update_action_flags(action, pe_action_pseudo, __FUNCTION__, __LINE__); update_action_flags(action, pe_action_runnable, __FUNCTION__, __LINE__); if(runnable) { update_action_flags(action, pe_action_runnable, __FUNCTION__, __LINE__); } return action; } /*! * \internal * \brief Create an executor cancel op * * \param[in] rsc Resource of action to cancel * \param[in] task Name of action to cancel * \param[in] interval_ms Interval of action to cancel * \param[in] node Node of action to cancel * \param[in] data_set Working set of cluster * * \return Created op */ pe_action_t * pe_cancel_op(pe_resource_t *rsc, const char *task, guint interval_ms, pe_node_t *node, pe_working_set_t *data_set) { pe_action_t *cancel_op; char *interval_ms_s = crm_strdup_printf("%u", interval_ms); // @TODO dangerous if possible to schedule another action with this key char *key = generate_op_key(rsc->id, task, interval_ms); cancel_op = custom_action(rsc, key, RSC_CANCEL, node, FALSE, TRUE, data_set); free(cancel_op->task); cancel_op->task = strdup(RSC_CANCEL); free(cancel_op->cancel_task); cancel_op->cancel_task = strdup(task); add_hash_param(cancel_op->meta, XML_LRM_ATTR_TASK, task); add_hash_param(cancel_op->meta, XML_LRM_ATTR_INTERVAL_MS, interval_ms_s); free(interval_ms_s); return cancel_op; } /*! * \internal * \brief Create a shutdown op for a scheduler transition * * \param[in] node Node being shut down * \param[in] data_set Working set of cluster * * \return Created op */ pe_action_t * sched_shutdown_op(pe_node_t *node, pe_working_set_t *data_set) { char *shutdown_id = crm_strdup_printf("%s-%s", CRM_OP_SHUTDOWN, node->details->uname); pe_action_t *shutdown_op = custom_action(NULL, shutdown_id, CRM_OP_SHUTDOWN, node, FALSE, TRUE, data_set); crm_notice("Scheduling shutdown of node %s", node->details->uname); shutdown_constraints(node, shutdown_op, data_set); add_hash_param(shutdown_op->meta, XML_ATTR_TE_NOWAIT, XML_BOOLEAN_TRUE); return shutdown_op; } + +static char * +generate_transition_magic(const char *transition_key, int op_status, int op_rc) +{ + CRM_CHECK(transition_key != NULL, return NULL); + return crm_strdup_printf("%d:%d;%s", op_status, op_rc, transition_key); +} + +static void +append_digest(lrmd_event_data_t *op, xmlNode *update, const char *version, + const char *magic, int level) +{ + /* this will enable us to later determine that the + * resource's parameters have changed and we should force + * a restart + */ + char *digest = NULL; + xmlNode *args_xml = NULL; + + if (op->params == NULL) { + return; + } + + args_xml = create_xml_node(NULL, XML_TAG_PARAMS); + g_hash_table_foreach(op->params, hash2field, args_xml); + filter_action_parameters(args_xml, version); + digest = calculate_operation_digest(args_xml, version); + +#if 0 + if (level < get_crm_log_level() + && op->interval_ms == 0 && crm_str_eq(op->op_type, CRMD_ACTION_START, TRUE)) { + char *digest_source = dump_xml_unformatted(args_xml); + + do_crm_log(level, "Calculated digest %s for %s (%s). Source: %s\n", + digest, ID(update), magic, digest_source); + free(digest_source); + } +#endif + crm_xml_add(update, XML_LRM_ATTR_OP_DIGEST, digest); + + free_xml(args_xml); + free(digest); +} + +#define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" + +/*! + * \internal + * \brief Create XML for resource operation history update + * + * \param[in,out] parent Parent XML node to add to + * \param[in,out] op Operation event data + * \param[in] caller_version DC feature set + * \param[in] target_rc Expected result of operation + * \param[in] node Name of node on which operation was performed + * \param[in] origin Arbitrary description of update source + * \param[in] level A log message will be logged at this level + * + * \return Newly created XML node for history update + */ +xmlNode * +pcmk__create_history_xml(xmlNode *parent, lrmd_event_data_t *op, + const char *caller_version, int target_rc, + const char *node, const char *origin, int level) +{ + char *key = NULL; + char *magic = NULL; + char *op_id = NULL; + char *op_id_additional = NULL; + char *local_user_data = NULL; + const char *exit_reason = NULL; + + xmlNode *xml_op = NULL; + const char *task = NULL; + + CRM_CHECK(op != NULL, return NULL); + do_crm_log(level, "%s: Updating resource %s after %s op %s (interval=%u)", + origin, op->rsc_id, op->op_type, services_lrm_status_str(op->op_status), + op->interval_ms); + + crm_trace("DC version: %s", caller_version); + + task = op->op_type; + + /* Record a successful reload as a start, and a failed reload as a monitor, + * to make life easier for the scheduler when determining the current state. + */ + if (crm_str_eq(task, "reload", TRUE)) { + if (op->op_status == PCMK_LRM_OP_DONE) { + task = CRMD_ACTION_START; + } else { + task = CRMD_ACTION_STATUS; + } + } + + key = generate_op_key(op->rsc_id, task, op->interval_ms); + if (crm_str_eq(task, CRMD_ACTION_NOTIFY, TRUE)) { + const char *n_type = crm_meta_value(op->params, "notify_type"); + const char *n_task = crm_meta_value(op->params, "notify_operation"); + + CRM_LOG_ASSERT(n_type != NULL); + CRM_LOG_ASSERT(n_task != NULL); + op_id = generate_notify_key(op->rsc_id, n_type, n_task); + + if (op->op_status != PCMK_LRM_OP_PENDING) { + /* Ignore notify errors. + * + * @TODO It might be better to keep the correct result here, and + * ignore it in process_graph_event(). + */ + op->op_status = PCMK_LRM_OP_DONE; + op->rc = 0; + } + + } else if (did_rsc_op_fail(op, target_rc)) { + op_id = generate_op_key(op->rsc_id, "last_failure", 0); + if (op->interval_ms == 0) { + // Ensure 'last' gets updated, in case record-pending is true + op_id_additional = generate_op_key(op->rsc_id, "last", 0); + } + exit_reason = op->exit_reason; + + } else if (op->interval_ms > 0) { + op_id = strdup(key); + + } else { + op_id = generate_op_key(op->rsc_id, "last", 0); + } + + again: + xml_op = find_entity(parent, XML_LRM_TAG_RSC_OP, op_id); + if (xml_op == NULL) { + xml_op = create_xml_node(parent, XML_LRM_TAG_RSC_OP); + } + + if (op->user_data == NULL) { + crm_debug("Generating fake transition key for: " CRM_OP_FMT " %d from %s", + op->rsc_id, op->op_type, op->interval_ms, + op->call_id, origin); + local_user_data = generate_transition_key(-1, op->call_id, target_rc, FAKE_TE_ID); + op->user_data = local_user_data; + } + + if(magic == NULL) { + magic = generate_transition_magic(op->user_data, op->op_status, op->rc); + } + + crm_xml_add(xml_op, XML_ATTR_ID, op_id); + crm_xml_add(xml_op, XML_LRM_ATTR_TASK_KEY, key); + crm_xml_add(xml_op, XML_LRM_ATTR_TASK, task); + crm_xml_add(xml_op, XML_ATTR_ORIGIN, origin); + crm_xml_add(xml_op, XML_ATTR_CRM_VERSION, caller_version); + crm_xml_add(xml_op, XML_ATTR_TRANSITION_KEY, op->user_data); + crm_xml_add(xml_op, XML_ATTR_TRANSITION_MAGIC, magic); + crm_xml_add(xml_op, XML_LRM_ATTR_EXIT_REASON, exit_reason == NULL ? "" : exit_reason); + crm_xml_add(xml_op, XML_LRM_ATTR_TARGET, node); /* For context during triage */ + + crm_xml_add_int(xml_op, XML_LRM_ATTR_CALLID, op->call_id); + crm_xml_add_int(xml_op, XML_LRM_ATTR_RC, op->rc); + crm_xml_add_int(xml_op, XML_LRM_ATTR_OPSTATUS, op->op_status); + crm_xml_add_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, op->interval_ms); + + if (compare_version("2.1", caller_version) <= 0) { + if (op->t_run || op->t_rcchange || op->exec_time || op->queue_time) { + crm_trace("Timing data (" CRM_OP_FMT "): last=%u change=%u exec=%u queue=%u", + op->rsc_id, op->op_type, op->interval_ms, + op->t_run, op->t_rcchange, op->exec_time, op->queue_time); + + if (op->interval_ms == 0) { + /* The values are the same for non-recurring ops */ + crm_xml_add_int(xml_op, XML_RSC_OP_LAST_RUN, op->t_run); + crm_xml_add_int(xml_op, XML_RSC_OP_LAST_CHANGE, op->t_run); + + } else if(op->t_rcchange) { + /* last-run is not accurate for recurring ops */ + crm_xml_add_int(xml_op, XML_RSC_OP_LAST_CHANGE, op->t_rcchange); + + } else { + /* ...but is better than nothing otherwise */ + crm_xml_add_int(xml_op, XML_RSC_OP_LAST_CHANGE, op->t_run); + } + + crm_xml_add_int(xml_op, XML_RSC_OP_T_EXEC, op->exec_time); + crm_xml_add_int(xml_op, XML_RSC_OP_T_QUEUE, op->queue_time); + } + } + + if (crm_str_eq(op->op_type, CRMD_ACTION_MIGRATE, TRUE) + || crm_str_eq(op->op_type, CRMD_ACTION_MIGRATED, TRUE)) { + /* + * Record migrate_source and migrate_target always for migrate ops. + */ + const char *name = XML_LRM_ATTR_MIGRATE_SOURCE; + + crm_xml_add(xml_op, name, crm_meta_value(op->params, name)); + + name = XML_LRM_ATTR_MIGRATE_TARGET; + crm_xml_add(xml_op, name, crm_meta_value(op->params, name)); + } + + append_digest(op, xml_op, caller_version, magic, LOG_DEBUG); + + if (op_id_additional) { + free(op_id); + op_id = op_id_additional; + op_id_additional = NULL; + goto again; + } + + if (local_user_data) { + free(local_user_data); + op->user_data = NULL; + } + free(magic); + free(op_id); + free(key); + return xml_op; +}