diff --git a/daemons/controld/controld_based.c b/daemons/controld/controld_based.c index 4ea62fd344..e6a4612ed2 100644 --- a/daemons/controld/controld_based.c +++ b/daemons/controld/controld_based.c @@ -1,170 +1,170 @@ /* * 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 /* sleep */ #include #include #include #include #include int cib_retries = 0; static void do_cib_updated(const char *event, xmlNode * msg) { if (crm_patchset_contains_alert(msg, TRUE)) { mainloop_set_trigger(config_read); } } static void do_cib_replaced(const char *event, xmlNode * msg) { crm_debug("Updating the CIB after a replace: DC=%s", AM_I_DC ? "true" : "false"); if (AM_I_DC == FALSE) { return; } else if (fsa_state == S_FINALIZE_JOIN && is_set(fsa_input_register, R_CIB_ASKED)) { /* no need to restart the join - we asked for this replace op */ return; } /* start the join process again so we get everyone's LRM status */ populate_cib_nodes(node_update_quick|node_update_all, __FUNCTION__); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } /* A_CIB_STOP, A_CIB_START, O_CIB_RESTART */ void do_cib_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) { CRM_ASSERT(fsa_cib_conn != NULL); if (action & A_CIB_STOP) { if (fsa_cib_conn->state != cib_disconnected && last_resource_update != 0) { crm_info("Waiting for resource update %d to complete", last_resource_update); crmd_fsa_stall(FALSE); return; } crm_info("Disconnecting from the CIB manager"); clear_bit(fsa_input_register, R_CIB_CONNECTED); fsa_cib_conn->cmds->del_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, do_cib_updated); if (fsa_cib_conn->state != cib_disconnected) { fsa_cib_conn->cmds->set_slave(fsa_cib_conn, cib_scope_local); fsa_cib_conn->cmds->signoff(fsa_cib_conn); } crm_notice("Disconnected from the CIB manager"); } if (action & A_CIB_START) { int rc = pcmk_ok; if (cur_state == S_STOPPING) { crm_err("Ignoring request to connect to the CIB manager after shutdown"); return; } rc = fsa_cib_conn->cmds->signon(fsa_cib_conn, CRM_SYSTEM_CRMD, cib_command_nonblocking); if (rc != pcmk_ok) { /* a short wait that usually avoids stalling the FSA */ sleep(1); rc = fsa_cib_conn->cmds->signon(fsa_cib_conn, CRM_SYSTEM_CRMD, cib_command_nonblocking); } if (rc != pcmk_ok) { crm_info("Could not connect to the CIB manager: %s", pcmk_strerror(rc)); } else if (pcmk_ok != fsa_cib_conn->cmds->set_connection_dnotify(fsa_cib_conn, crmd_cib_connection_destroy)) { crm_err("Could not set dnotify callback"); } else if (pcmk_ok != fsa_cib_conn->cmds->add_notify_callback(fsa_cib_conn, T_CIB_REPLACE_NOTIFY, do_cib_replaced)) { crm_err("Could not set CIB notification callback (replace)"); } else if (pcmk_ok != fsa_cib_conn->cmds->add_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, do_cib_updated)) { crm_err("Could not set CIB notification callback (update)"); } else { set_bit(fsa_input_register, R_CIB_CONNECTED); cib_retries = 0; } if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) { cib_retries++; crm_warn("Couldn't complete CIB registration %d" " times... pause and retry", cib_retries); if (cib_retries < 30) { - crm_timer_start(wait_timer); + controld_start_timer(wait_timer); crmd_fsa_stall(FALSE); } else { crm_err("Could not complete CIB" " registration %d times..." " hard error", cib_retries); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } } } /*! * \internal * \brief Get CIB call options to use local scope if master unavailable * * \return CIB call options */ int crmd_cib_smart_opt() { int call_opt = cib_quorum_override; if (fsa_state == S_ELECTION || fsa_state == S_PENDING) { crm_info("Sending update to local CIB in state: %s", fsa_state2string(fsa_state)); call_opt |= cib_scope_local; } return call_opt; } /*! * \internal * \brief Check whether an action type should be recorded in the CIB * * \param[in] action Action type * * \return TRUE if action should be recorded, FALSE otherwise */ bool controld_action_is_recordable(const char *action) { if (safe_str_eq(action, CRMD_ACTION_CANCEL) || safe_str_eq(action, CRMD_ACTION_DELETE) || safe_str_eq(action, CRMD_ACTION_NOTIFY) || safe_str_eq(action, CRMD_ACTION_METADATA)) { return FALSE; } return TRUE; } diff --git a/daemons/controld/controld_control.c b/daemons/controld/controld_control.c index ff2ace24c4..0681caf2bb 100644 --- a/daemons/controld/controld_control.c +++ b/daemons/controld/controld_control.c @@ -1,774 +1,774 @@ /* * Copyright 2004-2019 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include qb_ipcs_service_t *ipcs = NULL; #if SUPPORT_COROSYNC extern gboolean crm_connect_corosync(crm_cluster_t * cluster); #endif void crm_shutdown(int nsig); gboolean crm_read_options(gpointer user_data); gboolean fsa_has_quorum = FALSE; crm_trigger_t *fsa_source = NULL; crm_trigger_t *config_read = NULL; bool no_quorum_suicide_escalation = FALSE; /* A_HA_CONNECT */ void do_ha_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) { gboolean registered = FALSE; static crm_cluster_t *cluster = NULL; if (cluster == NULL) { cluster = calloc(1, sizeof(crm_cluster_t)); } if (action & A_HA_DISCONNECT) { crm_cluster_disconnect(cluster); crm_info("Disconnected from the cluster"); set_bit(fsa_input_register, R_HA_DISCONNECTED); } if (action & A_HA_CONNECT) { crm_set_status_callback(&peer_update_callback); crm_set_autoreap(FALSE); if (is_corosync_cluster()) { #if SUPPORT_COROSYNC registered = crm_connect_corosync(cluster); #endif } if (registered == TRUE) { controld_election_init(cluster->uname); fsa_our_uname = cluster->uname; fsa_our_uuid = cluster->uuid; if(cluster->uuid == NULL) { crm_err("Could not obtain local uuid"); registered = FALSE; } } if (registered == FALSE) { set_bit(fsa_input_register, R_HA_DISCONNECTED); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } populate_cib_nodes(node_update_none, __FUNCTION__); clear_bit(fsa_input_register, R_HA_DISCONNECTED); crm_info("Connected to the cluster"); } if (action & ~(A_HA_CONNECT | A_HA_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __FUNCTION__); } } /* A_SHUTDOWN */ void do_shutdown(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) { /* just in case */ set_bit(fsa_input_register, R_SHUTDOWN); controld_disconnect_fencer(FALSE); } /* A_SHUTDOWN_REQ */ void do_shutdown_req(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) { xmlNode *msg = NULL; set_bit(fsa_input_register, R_SHUTDOWN); crm_info("Sending shutdown request to all peers (DC is %s)", (fsa_our_dc? fsa_our_dc : "not set")); msg = create_request(CRM_OP_SHUTDOWN_REQ, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); /* set_bit(fsa_input_register, R_STAYDOWN); */ if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } free_xml(msg); } extern char *max_generation_from; extern xmlNode *max_generation_xml; extern GHashTable *resource_history; extern GHashTable *voted; void crmd_fast_exit(crm_exit_t exit_code) { if (is_set(fsa_input_register, R_STAYDOWN)) { crm_warn("Inhibiting respawn "CRM_XS" remapping exit code %d to %d", exit_code, CRM_EX_FATAL); exit_code = CRM_EX_FATAL; } else if ((exit_code == CRM_EX_OK) && is_set(fsa_input_register, R_IN_RECOVERY)) { crm_err("Could not recover from internal error"); exit_code = CRM_EX_ERROR; } crm_exit(exit_code); } crm_exit_t crmd_exit(crm_exit_t exit_code) { GListPtr gIter = NULL; GMainLoop *mloop = crmd_mainloop; static bool in_progress = FALSE; if (in_progress && (exit_code == CRM_EX_OK)) { crm_debug("Exit is already in progress"); return exit_code; } else if(in_progress) { crm_notice("Error during shutdown process, exiting now with status %d (%s)", exit_code, crm_exit_str(exit_code)); crm_write_blackbox(SIGTRAP, NULL); crmd_fast_exit(exit_code); } in_progress = TRUE; crm_trace("Preparing to exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); /* Suppress secondary errors resulting from us disconnecting everything */ set_bit(fsa_input_register, R_HA_DISCONNECTED); /* Close all IPC servers and clients to ensure any and all shared memory files are cleaned up */ if(ipcs) { crm_trace("Closing IPC server"); mainloop_del_ipc_server(ipcs); ipcs = NULL; } controld_close_attrd_ipc(); pe_subsystem_free(); controld_disconnect_fencer(TRUE); if ((exit_code == CRM_EX_OK) && (crmd_mainloop == NULL)) { crm_debug("No mainloop detected"); exit_code = CRM_EX_ERROR; } /* On an error, just get out. * * Otherwise, make the effort to have mainloop exit gracefully so * that it (mostly) cleans up after itself and valgrind has less * to report on - allowing real errors stand out */ if (exit_code != CRM_EX_OK) { crm_notice("Forcing immediate exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); crm_write_blackbox(SIGTRAP, NULL); crmd_fast_exit(exit_code); } /* Clean up as much memory as possible for valgrind */ for (gIter = fsa_message_queue; gIter != NULL; gIter = gIter->next) { fsa_data_t *fsa_data = gIter->data; crm_info("Dropping %s: [ state=%s cause=%s origin=%s ]", fsa_input2string(fsa_data->fsa_input), fsa_state2string(fsa_state), fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin); delete_fsa_input(fsa_data); } clear_bit(fsa_input_register, R_MEMBERSHIP); g_list_free(fsa_message_queue); fsa_message_queue = NULL; metadata_cache_fini(); controld_election_fini(); /* Tear down the CIB manager connection, but don't free it yet -- it could * be used when we drain the mainloop later. */ cib_free_callbacks(fsa_cib_conn); fsa_cib_conn->cmds->signoff(fsa_cib_conn); verify_stopped(fsa_state, LOG_WARNING); clear_bit(fsa_input_register, R_LRM_CONNECTED); lrm_state_destroy_all(); /* This basically will not work, since mainloop has a reference to it */ mainloop_destroy_trigger(fsa_source); fsa_source = NULL; mainloop_destroy_trigger(config_read); config_read = NULL; mainloop_destroy_trigger(transition_trigger); transition_trigger = NULL; crm_client_cleanup(); crm_peer_destroy(); controld_free_fsa_timers(); te_cleanup_stonith_history_sync(NULL, TRUE); controld_free_sched_timer(); free(fsa_our_dc_version); fsa_our_dc_version = NULL; free(fsa_our_uname); fsa_our_uname = NULL; free(fsa_our_uuid); fsa_our_uuid = NULL; free(fsa_our_dc); fsa_our_dc = NULL; free(fsa_cluster_name); fsa_cluster_name = NULL; free(te_uuid); te_uuid = NULL; free(failed_stop_offset); failed_stop_offset = NULL; free(failed_start_offset); failed_start_offset = NULL; free(max_generation_from); max_generation_from = NULL; free_xml(max_generation_xml); max_generation_xml = NULL; mainloop_destroy_signal(SIGPIPE); mainloop_destroy_signal(SIGUSR1); mainloop_destroy_signal(SIGTERM); mainloop_destroy_signal(SIGTRAP); /* leave SIGCHLD engaged as we might still want to drain some service-actions */ if (mloop) { GMainContext *ctx = g_main_loop_get_context(crmd_mainloop); /* Don't re-enter this block */ crmd_mainloop = NULL; /* no signals on final draining anymore */ mainloop_destroy_signal(SIGCHLD); crm_trace("Draining mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx)); { int lpc = 0; while((g_main_context_pending(ctx) && lpc < 10)) { lpc++; crm_trace("Iteration %d", lpc); g_main_context_dispatch(ctx); } } crm_trace("Closing mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx)); g_main_loop_quit(mloop); /* Won't do anything yet, since we're inside it now */ g_main_loop_unref(mloop); } else { mainloop_destroy_signal(SIGCHLD); } cib_delete(fsa_cib_conn); fsa_cib_conn = NULL; throttle_fini(); /* Graceful */ crm_trace("Done preparing for exit with status %d (%s)", exit_code, crm_exit_str(exit_code)); return exit_code; } /* A_EXIT_0, A_EXIT_1 */ void do_exit(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) { crm_exit_t exit_code = CRM_EX_OK; int log_level = LOG_INFO; const char *exit_type = "gracefully"; if (action & A_EXIT_1) { log_level = LOG_ERR; exit_type = "forcefully"; exit_code = CRM_EX_ERROR; } verify_stopped(cur_state, LOG_ERR); do_crm_log(log_level, "Performing %s - %s exiting the controller", fsa_action2string(action), exit_type); crm_info("[%s] stopped (%d)", crm_system_name, exit_code); crmd_exit(exit_code); } static void sigpipe_ignore(int nsig) { return; } /* A_STARTUP */ void do_startup(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) { crm_debug("Registering Signal Handlers"); mainloop_add_signal(SIGTERM, crm_shutdown); mainloop_add_signal(SIGPIPE, sigpipe_ignore); fsa_source = mainloop_add_trigger(G_PRIORITY_HIGH, crm_fsa_trigger, NULL); config_read = mainloop_add_trigger(G_PRIORITY_HIGH, crm_read_options, NULL); transition_trigger = mainloop_add_trigger(G_PRIORITY_LOW, te_graph_trigger, NULL); crm_debug("Creating CIB manager and executor objects"); fsa_cib_conn = cib_new(); lrm_state_init_local(); if (controld_init_fsa_timers() == FALSE) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } static int32_t crmd_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid) { crm_trace("Connection %p", c); if (crm_client_new(c, uid, gid) == NULL) { return -EIO; } return 0; } static void crmd_ipc_created(qb_ipcs_connection_t * c) { crm_trace("Connection %p", c); } static int32_t crmd_ipc_dispatch(qb_ipcs_connection_t * c, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; crm_client_t *client = crm_client_get(c); xmlNode *msg = crm_ipcs_recv(client, data, size, &id, &flags); crm_trace("Invoked: %s", crm_client_name(client)); crm_ipcs_send_ack(client, id, flags, "ack", __FUNCTION__, __LINE__); if (msg == NULL) { return 0; } #if ENABLE_ACL CRM_ASSERT(client->user != NULL); crm_acl_get_set_user(msg, F_CRM_USER, client->user); #endif crm_trace("Processing msg from %s", crm_client_name(client)); crm_log_xml_trace(msg, "controller[inbound]"); crm_xml_add(msg, F_CRM_SYS_FROM, client->id); if (crmd_authorize_message(msg, client, NULL)) { route_message(C_IPC_MESSAGE, msg); } trigger_fsa(fsa_source); free_xml(msg); return 0; } static int32_t crmd_ipc_closed(qb_ipcs_connection_t * c) { crm_client_t *client = crm_client_get(c); if (client) { crm_trace("Disconnecting %sregistered client %s (%p/%p)", (client->userdata? "" : "un"), crm_client_name(client), c, client); free(client->userdata); crm_client_destroy(client); trigger_fsa(fsa_source); } return 0; } static void crmd_ipc_destroy(qb_ipcs_connection_t * c) { crm_trace("Connection %p", c); crmd_ipc_closed(c); } /* A_STOP */ void do_stop(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) { crm_trace("Closing IPC server"); mainloop_del_ipc_server(ipcs); ipcs = NULL; register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL); } /* A_STARTED */ void do_started(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) { static struct qb_ipcs_service_handlers crmd_callbacks = { .connection_accept = crmd_ipc_accept, .connection_created = crmd_ipc_created, .msg_process = crmd_ipc_dispatch, .connection_closed = crmd_ipc_closed, .connection_destroyed = crmd_ipc_destroy }; if (cur_state != S_STARTING) { crm_err("Start cancelled... %s", fsa_state2string(cur_state)); return; } else if (is_set(fsa_input_register, R_MEMBERSHIP) == FALSE) { crm_info("Delaying start, no membership data (%.16llx)", R_MEMBERSHIP); crmd_fsa_stall(TRUE); return; } else if (is_set(fsa_input_register, R_LRM_CONNECTED) == FALSE) { crm_info("Delaying start, not connected to executor (%.16llx)", R_LRM_CONNECTED); crmd_fsa_stall(TRUE); return; } else if (is_set(fsa_input_register, R_CIB_CONNECTED) == FALSE) { crm_info("Delaying start, CIB not connected (%.16llx)", R_CIB_CONNECTED); crmd_fsa_stall(TRUE); return; } else if (is_set(fsa_input_register, R_READ_CONFIG) == FALSE) { crm_info("Delaying start, Config not read (%.16llx)", R_READ_CONFIG); crmd_fsa_stall(TRUE); return; } else if (is_set(fsa_input_register, R_PEER_DATA) == FALSE) { crm_info("Delaying start, No peer data (%.16llx)", R_PEER_DATA); crmd_fsa_stall(TRUE); return; } crm_debug("Init server comms"); ipcs = crmd_ipc_server_init(&crmd_callbacks); if (ipcs == NULL) { crm_err("Failed to create IPC server: shutting down and inhibiting respawn"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } else { crm_notice("Pacemaker controller successfully started and accepting connections"); } controld_trigger_fencer_connect(); clear_bit(fsa_input_register, R_STARTING); register_fsa_input(msg_data->fsa_cause, I_PENDING, NULL); } /* A_RECOVER */ void do_recover(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) { set_bit(fsa_input_register, R_IN_RECOVERY); crm_warn("Fast-tracking shutdown in response to errors"); register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL); } /* *INDENT-OFF* */ static pe_cluster_option crmd_opts[] = { /* name, old-name, validate, values, default, short description, long description */ { "dc-version", NULL, "string", NULL, "none", NULL, "Version of Pacemaker on the cluster's DC.", "Includes the hash which identifies the exact changeset it was built from. Used for diagnostic purposes." }, { "cluster-infrastructure", NULL, "string", NULL, "corosync", NULL, "The messaging stack on which Pacemaker is currently running.", "Used for informational and diagnostic purposes." }, { "cluster-name", NULL, "string", NULL, NULL, NULL, "An arbitrary name for the cluster", "This optional value is mostly for users' convenience as desired " "in administration, but may also be used in Pacemaker configuration " "rules via the #cluster-name node attribute, and by higher-level tools " "and resource agents." }, { XML_CONFIG_ATTR_DC_DEADTIME, NULL, "time", NULL, "20s", &check_time, "How long to wait for a response from other nodes during startup.", "The \"correct\" value will depend on the speed/load of your network and the type of switches used." }, { XML_CONFIG_ATTR_RECHECK, NULL, "time", "Zero disables polling. Positive values are an interval in seconds (unless other SI units are specified. eg. 5min)", "15min", &check_timer, "Polling interval for time based changes to options, resource parameters and constraints.", "The Cluster is primarily event driven, however the configuration can have elements that change based on time." " To ensure these changes take effect, we can optionally poll the cluster's status for changes." }, { "load-threshold", NULL, "percentage", NULL, "80%", &check_utilization, "The maximum amount of system resources that should be used by nodes in the cluster", "The cluster will slow down its recovery process when the amount of system resources used" " (currently CPU) approaches this limit", }, { "node-action-limit", NULL, "integer", NULL, "0", &check_number, "The maximum number of jobs that can be scheduled per node. Defaults to 2x cores"}, { XML_CONFIG_ATTR_FENCE_REACTION, NULL, "string", NULL, "stop", NULL, "How a cluster node should react if notified of its own fencing", "A cluster node may receive notification of its own fencing if fencing " "is misconfigured, or if fabric fencing is in use that doesn't cut " "cluster communication. Allowed values are \"stop\" to attempt to " "immediately stop pacemaker and stay stopped, or \"panic\" to attempt " "to immediately reboot the local node, falling back to stop on failure." }, { XML_CONFIG_ATTR_ELECTION_FAIL, NULL, "time", NULL, "2min", &check_timer, "*** Advanced Use Only ***.", "If need to adjust this value, it probably indicates the presence of a bug." }, { XML_CONFIG_ATTR_FORCE_QUIT, NULL, "time", NULL, "20min", &check_timer, "*** Advanced Use Only ***.", "If need to adjust this value, it probably indicates the presence of a bug." }, { "join-integration-timeout", "crmd-integration-timeout", "time", NULL, "3min", &check_timer, "*** Advanced Use Only ***", "If need to adjust this value, it probably indicates the presence of a bug" }, { "join-finalization-timeout", "crmd-finalization-timeout", "time", NULL, "30min", &check_timer, "*** Advanced Use Only ***", "If you need to adjust this value, it probably indicates the presence of a bug" }, { "transition-delay", "crmd-transition-delay", "time", NULL, "0s", &check_timer, "*** Advanced Use Only *** Enabling this option will slow down cluster recovery under all conditions", "Delay cluster recovery for the configured interval to allow for additional/related events to occur.\n" "Useful if your configuration is sensitive to the order in which ping updates arrive." }, { "stonith-watchdog-timeout", NULL, "time", NULL, NULL, &check_sbd_timeout, "How long to wait before we can assume nodes are safely down", NULL }, { "stonith-max-attempts",NULL,"integer",NULL,"10",&check_positive_number, "How many times stonith can fail before it will no longer be attempted on a target" }, { "no-quorum-policy", NULL, "enum", "stop, freeze, ignore, suicide", "stop", &check_quorum, NULL, NULL }, }; /* *INDENT-ON* */ void crmd_metadata(void) { config_metadata("pacemaker-controld", "1.0", "controller properties", "Cluster properties used by Pacemaker's controller," " formerly known as crmd", crmd_opts, DIMOF(crmd_opts)); } static void verify_crmd_options(GHashTable * options) { verify_all_options(options, crmd_opts, DIMOF(crmd_opts)); } static const char * crmd_pref(GHashTable * options, const char *name) { return get_cluster_pref(options, crmd_opts, DIMOF(crmd_opts), name); } static void config_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { const char *value = NULL; GHashTable *config_hash = NULL; crm_time_t *now = crm_time_new(NULL); xmlNode *crmconfig = NULL; xmlNode *alerts = NULL; if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Local CIB query resulted in an error: %s", pcmk_strerror(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); if (rc == -EACCES || rc == -pcmk_err_schema_validation) { crm_err("The cluster is mis-configured - shutting down and staying down"); set_bit(fsa_input_register, R_STAYDOWN); } goto bail; } crmconfig = output; if ((crmconfig) && (crm_element_name(crmconfig)) && (strcmp(crm_element_name(crmconfig), XML_CIB_TAG_CRMCONFIG) != 0)) { crmconfig = first_named_child(crmconfig, XML_CIB_TAG_CRMCONFIG); } if (!crmconfig) { fsa_data_t *msg_data = NULL; crm_err("Local CIB query for " XML_CIB_TAG_CRMCONFIG " section failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); goto bail; } crm_debug("Call %d : Parsing CIB options", call_id); config_hash = crm_str_table_new(); pe_unpack_nvpairs(crmconfig, crmconfig, XML_CIB_TAG_PROPSET, NULL, config_hash, CIB_OPTIONS_FIRST, FALSE, now, NULL); verify_crmd_options(config_hash); value = crmd_pref(config_hash, XML_CONFIG_ATTR_DC_DEADTIME); election_trigger->period_ms = crm_parse_interval_spec(value); value = crmd_pref(config_hash, "node-action-limit"); /* Also checks migration-limit */ throttle_update_job_max(value); value = crmd_pref(config_hash, "load-threshold"); if(value) { throttle_set_load_target(strtof(value, NULL) / 100.0); } value = crmd_pref(config_hash, "no-quorum-policy"); if (safe_str_eq(value, "suicide") && pcmk_locate_sbd()) { no_quorum_suicide_escalation = TRUE; } set_fence_reaction(crmd_pref(config_hash, XML_CONFIG_ATTR_FENCE_REACTION)); value = crmd_pref(config_hash,"stonith-max-attempts"); update_stonith_max_attempts(value); value = crmd_pref(config_hash, XML_CONFIG_ATTR_FORCE_QUIT); shutdown_escalation_timer->period_ms = crm_parse_interval_spec(value); crm_debug("Shutdown escalation occurs if DC has not responded to request in %ums", shutdown_escalation_timer->period_ms); value = crmd_pref(config_hash, XML_CONFIG_ATTR_ELECTION_FAIL); controld_set_election_period(value); value = crmd_pref(config_hash, XML_CONFIG_ATTR_RECHECK); recheck_timer->period_ms = crm_parse_interval_spec(value); crm_debug("Checking for expired actions every %ums", recheck_timer->period_ms); value = crmd_pref(config_hash, "transition-delay"); transition_timer->period_ms = crm_parse_interval_spec(value); value = crmd_pref(config_hash, "join-integration-timeout"); integration_timer->period_ms = crm_parse_interval_spec(value); value = crmd_pref(config_hash, "join-finalization-timeout"); finalization_timer->period_ms = crm_parse_interval_spec(value); free(fsa_cluster_name); fsa_cluster_name = NULL; value = g_hash_table_lookup(config_hash, "cluster-name"); if (value) { fsa_cluster_name = strdup(value); } alerts = first_named_child(output, XML_CIB_TAG_ALERTS); crmd_unpack_alerts(alerts); set_bit(fsa_input_register, R_READ_CONFIG); crm_trace("Triggering FSA: %s", __FUNCTION__); mainloop_set_trigger(fsa_source); g_hash_table_destroy(config_hash); bail: crm_time_free(now); } gboolean crm_read_options(gpointer user_data) { int call_id = fsa_cib_conn->cmds->query(fsa_cib_conn, "//" XML_CIB_TAG_CRMCONFIG " | //" XML_CIB_TAG_ALERTS, NULL, cib_xpath | cib_scope_local); fsa_register_cib_callback(call_id, FALSE, NULL, config_query_callback); crm_trace("Querying the CIB... call %d", call_id); return TRUE; } /* A_READCONFIG */ void do_read_config(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) { throttle_init(); mainloop_set_trigger(config_read); } void crm_shutdown(int nsig) { if (crmd_mainloop != NULL && g_main_loop_is_running(crmd_mainloop)) { if (is_set(fsa_input_register, R_SHUTDOWN)) { crm_err("Escalating the shutdown"); register_fsa_input_before(C_SHUTDOWN, I_ERROR, NULL); } else { set_bit(fsa_input_register, R_SHUTDOWN); register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); if (shutdown_escalation_timer->period_ms == 0) { const char *value = crmd_pref(NULL, XML_CONFIG_ATTR_FORCE_QUIT); shutdown_escalation_timer->period_ms = crm_parse_interval_spec(value); } /* can't rely on this... */ crm_notice("Shutting down cluster resource manager " CRM_XS " limit=%ums", shutdown_escalation_timer->period_ms); - crm_timer_start(shutdown_escalation_timer); + controld_start_timer(shutdown_escalation_timer); } } else { crm_info("exit from shutdown"); crmd_exit(CRM_EX_OK); } } diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c index 2ebfc52519..0f9d8026e7 100644 --- a/daemons/controld/controld_execd.c +++ b/daemons/controld/controld_execd.c @@ -1,2803 +1,2803 @@ /* * Copyright 2004-2019 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include // lrmd_event_data_t, lrmd_rsc_info_t, etc. #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); + controld_start_timer(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 = 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) { /* Replace newline escape pattern with actual newline (and a space so we * don't have to shuffle the rest of the buffer) */ pch[0] = '\n'; pch[1] = ' '; 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_fsa.c b/daemons/controld/controld_fsa.c index 0074ed1a8a..4645d1f220 100644 --- a/daemons/controld/controld_fsa.c +++ b/daemons/controld/controld_fsa.c @@ -1,648 +1,648 @@ /* * Copyright 2004-2019 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include char *fsa_our_dc = NULL; cib_t *fsa_cib_conn = NULL; char *fsa_our_dc_version = NULL; char *fsa_our_uuid = NULL; char *fsa_our_uname = NULL; char *fsa_cluster_name = NULL; gboolean do_fsa_stall = FALSE; long long fsa_input_register = 0; long long fsa_actions = A_NOTHING; enum crmd_fsa_state fsa_state = S_STARTING; extern uint highest_born_on; extern uint num_join_invites; extern void initialize_join(gboolean before); #define DOT_PREFIX "actions:trace: " #define do_dot_log(fmt, args...) crm_trace( fmt, ##args) long long do_state_transition(long long actions, enum crmd_fsa_state cur_state, enum crmd_fsa_state next_state, fsa_data_t * msg_data); void s_crmd_fsa_actions(fsa_data_t * fsa_data); void log_fsa_input(fsa_data_t * stored_msg); void init_dotfile(void); void init_dotfile(void) { do_dot_log(DOT_PREFIX "digraph \"g\" {"); do_dot_log(DOT_PREFIX " size = \"30,30\""); do_dot_log(DOT_PREFIX " graph ["); do_dot_log(DOT_PREFIX " fontsize = \"12\""); do_dot_log(DOT_PREFIX " fontname = \"Times-Roman\""); do_dot_log(DOT_PREFIX " fontcolor = \"black\""); do_dot_log(DOT_PREFIX " bb = \"0,0,398.922306,478.927856\""); do_dot_log(DOT_PREFIX " color = \"black\""); do_dot_log(DOT_PREFIX " ]"); do_dot_log(DOT_PREFIX " node ["); do_dot_log(DOT_PREFIX " fontsize = \"12\""); do_dot_log(DOT_PREFIX " fontname = \"Times-Roman\""); do_dot_log(DOT_PREFIX " fontcolor = \"black\""); do_dot_log(DOT_PREFIX " shape = \"ellipse\""); do_dot_log(DOT_PREFIX " color = \"black\""); do_dot_log(DOT_PREFIX " ]"); do_dot_log(DOT_PREFIX " edge ["); do_dot_log(DOT_PREFIX " fontsize = \"12\""); do_dot_log(DOT_PREFIX " fontname = \"Times-Roman\""); do_dot_log(DOT_PREFIX " fontcolor = \"black\""); do_dot_log(DOT_PREFIX " color = \"black\""); do_dot_log(DOT_PREFIX " ]"); do_dot_log(DOT_PREFIX "// special nodes"); do_dot_log(DOT_PREFIX " \"S_PENDING\" "); do_dot_log(DOT_PREFIX " ["); do_dot_log(DOT_PREFIX " color = \"blue\""); do_dot_log(DOT_PREFIX " fontcolor = \"blue\""); do_dot_log(DOT_PREFIX " ]"); do_dot_log(DOT_PREFIX " \"S_TERMINATE\" "); do_dot_log(DOT_PREFIX " ["); do_dot_log(DOT_PREFIX " color = \"red\""); do_dot_log(DOT_PREFIX " fontcolor = \"red\""); do_dot_log(DOT_PREFIX " ]"); do_dot_log(DOT_PREFIX "// DC only nodes"); do_dot_log(DOT_PREFIX " \"S_INTEGRATION\" [ fontcolor = \"green\" ]"); do_dot_log(DOT_PREFIX " \"S_POLICY_ENGINE\" [ fontcolor = \"green\" ]"); do_dot_log(DOT_PREFIX " \"S_TRANSITION_ENGINE\" [ fontcolor = \"green\" ]"); do_dot_log(DOT_PREFIX " \"S_RELEASE_DC\" [ fontcolor = \"green\" ]"); do_dot_log(DOT_PREFIX " \"S_IDLE\" [ fontcolor = \"green\" ]"); } static void do_fsa_action(fsa_data_t * fsa_data, long long an_action, void (*function) (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)) { fsa_actions &= ~an_action; crm_trace(DOT_PREFIX "\t// %s", fsa_action2string(an_action)); function(an_action, fsa_data->fsa_cause, fsa_state, fsa_data->fsa_input, fsa_data); } static long long startup_actions = A_STARTUP | A_CIB_START | A_LRM_CONNECT | A_HA_CONNECT | A_READCONFIG | A_STARTED | A_CL_JOIN_QUERY; // A_LOG, A_WARN, A_ERROR void do_log(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) { unsigned log_type = LOG_TRACE; if (action & A_LOG) { log_type = LOG_INFO; } else if (action & A_WARN) { log_type = LOG_WARNING; } else if (action & A_ERROR) { log_type = LOG_ERR; } do_crm_log(log_type, "Input %s received in state %s from %s", fsa_input2string(msg_data->fsa_input), fsa_state2string(cur_state), msg_data->origin); if (msg_data->data_type == fsa_dt_ha_msg) { ha_msg_input_t *input = fsa_typed_data(msg_data->data_type); crm_log_xml_debug(input->msg, __FUNCTION__); } else if (msg_data->data_type == fsa_dt_xml) { xmlNode *input = fsa_typed_data(msg_data->data_type); crm_log_xml_debug(input, __FUNCTION__); } else if (msg_data->data_type == fsa_dt_lrm) { lrmd_event_data_t *input = fsa_typed_data(msg_data->data_type); do_crm_log(log_type, "Resource %s: Call ID %d returned %d (%d)." " New status if rc=0: %s", input->rsc_id, input->call_id, input->rc, input->op_status, (char *)input->user_data); } } enum crmd_fsa_state s_crmd_fsa(enum crmd_fsa_cause cause) { fsa_data_t *fsa_data = NULL; long long register_copy = fsa_input_register; long long new_actions = A_NOTHING; enum crmd_fsa_state last_state; crm_trace("FSA invoked with Cause: %s\tState: %s", fsa_cause2string(cause), fsa_state2string(fsa_state)); fsa_dump_actions(fsa_actions, "Initial"); do_fsa_stall = FALSE; if (is_message() == FALSE && fsa_actions != A_NOTHING) { /* fake the first message so we can get into the loop */ fsa_data = calloc(1, sizeof(fsa_data_t)); fsa_data->fsa_input = I_NULL; fsa_data->fsa_cause = C_FSA_INTERNAL; fsa_data->origin = __FUNCTION__; fsa_data->data_type = fsa_dt_none; fsa_message_queue = g_list_append(fsa_message_queue, fsa_data); fsa_data = NULL; } while (is_message() && do_fsa_stall == FALSE) { crm_trace("Checking messages (%d remaining)", g_list_length(fsa_message_queue)); fsa_data = get_message(); if(fsa_data == NULL) { continue; } log_fsa_input(fsa_data); /* add any actions back to the queue */ fsa_actions |= fsa_data->actions; fsa_dump_actions(fsa_data->actions, "Restored actions"); /* get the next batch of actions */ new_actions = crmd_fsa_actions[fsa_data->fsa_input][fsa_state]; fsa_actions |= new_actions; fsa_dump_actions(new_actions, "New actions"); if (fsa_data->fsa_input != I_NULL && fsa_data->fsa_input != I_ROUTER) { crm_debug("Processing %s: [ state=%s cause=%s origin=%s ]", fsa_input2string(fsa_data->fsa_input), fsa_state2string(fsa_state), fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin); } /* logging : *before* the state is changed */ if (is_set(fsa_actions, A_ERROR)) { do_fsa_action(fsa_data, A_ERROR, do_log); } if (is_set(fsa_actions, A_WARN)) { do_fsa_action(fsa_data, A_WARN, do_log); } if (is_set(fsa_actions, A_LOG)) { do_fsa_action(fsa_data, A_LOG, do_log); } /* update state variables */ last_state = fsa_state; fsa_state = crmd_fsa_state[fsa_data->fsa_input][fsa_state]; /* * Remove certain actions during shutdown */ if (fsa_state == S_STOPPING || ((fsa_input_register & R_SHUTDOWN) == R_SHUTDOWN)) { clear_bit(fsa_actions, startup_actions); } /* * Hook for change of state. * Allows actions to be added or removed when entering a state */ if (last_state != fsa_state) { fsa_actions = do_state_transition(fsa_actions, last_state, fsa_state, fsa_data); } else { do_dot_log(DOT_PREFIX "\t// FSA input: State=%s \tCause=%s" " \tInput=%s \tOrigin=%s() \tid=%d", fsa_state2string(fsa_state), fsa_cause2string(fsa_data->fsa_cause), fsa_input2string(fsa_data->fsa_input), fsa_data->origin, fsa_data->id); } /* start doing things... */ s_crmd_fsa_actions(fsa_data); delete_fsa_input(fsa_data); fsa_data = NULL; } if (g_list_length(fsa_message_queue) > 0 || fsa_actions != A_NOTHING || do_fsa_stall) { crm_debug("Exiting the FSA: queue=%d, fsa_actions=0x%llx, stalled=%s", g_list_length(fsa_message_queue), fsa_actions, do_fsa_stall ? "true" : "false"); } else { crm_trace("Exiting the FSA"); } /* cleanup inputs? */ if (register_copy != fsa_input_register) { long long same = register_copy & fsa_input_register; fsa_dump_inputs(LOG_DEBUG, "Added", fsa_input_register ^ same); fsa_dump_inputs(LOG_DEBUG, "Removed", register_copy ^ same); } fsa_dump_actions(fsa_actions, "Remaining"); fsa_dump_queue(LOG_DEBUG); return fsa_state; } void s_crmd_fsa_actions(fsa_data_t * fsa_data) { /* * Process actions in order of priority but do only one * action at a time to avoid complicating the ordering. */ CRM_CHECK(fsa_data != NULL, return); while (fsa_actions != A_NOTHING && do_fsa_stall == FALSE) { /* regular action processing in order of action priority * * Make sure all actions that connect to required systems * are performed first */ if (fsa_actions & A_ERROR) { do_fsa_action(fsa_data, A_ERROR, do_log); } else if (fsa_actions & A_WARN) { do_fsa_action(fsa_data, A_WARN, do_log); } else if (fsa_actions & A_LOG) { do_fsa_action(fsa_data, A_LOG, do_log); /* get out of here NOW! before anything worse happens */ } else if (fsa_actions & A_EXIT_1) { do_fsa_action(fsa_data, A_EXIT_1, do_exit); /* sub-system restart */ } else if ((fsa_actions & O_LRM_RECONNECT) == O_LRM_RECONNECT) { do_fsa_action(fsa_data, O_LRM_RECONNECT, do_lrm_control); } else if ((fsa_actions & O_CIB_RESTART) == O_CIB_RESTART) { do_fsa_action(fsa_data, O_CIB_RESTART, do_cib_control); } else if ((fsa_actions & O_PE_RESTART) == O_PE_RESTART) { do_fsa_action(fsa_data, O_PE_RESTART, do_pe_control); } else if ((fsa_actions & O_TE_RESTART) == O_TE_RESTART) { do_fsa_action(fsa_data, O_TE_RESTART, do_te_control); /* essential start tasks */ } else if (fsa_actions & A_STARTUP) { do_fsa_action(fsa_data, A_STARTUP, do_startup); } else if (fsa_actions & A_CIB_START) { do_fsa_action(fsa_data, A_CIB_START, do_cib_control); } else if (fsa_actions & A_HA_CONNECT) { do_fsa_action(fsa_data, A_HA_CONNECT, do_ha_control); } else if (fsa_actions & A_READCONFIG) { do_fsa_action(fsa_data, A_READCONFIG, do_read_config); /* sub-system start/connect */ } else if (fsa_actions & A_LRM_CONNECT) { do_fsa_action(fsa_data, A_LRM_CONNECT, do_lrm_control); } else if (fsa_actions & A_TE_START) { do_fsa_action(fsa_data, A_TE_START, do_te_control); } else if (fsa_actions & A_PE_START) { do_fsa_action(fsa_data, A_PE_START, do_pe_control); /* Timers */ /* else if(fsa_actions & O_DC_TIMER_RESTART) { do_fsa_action(fsa_data, O_DC_TIMER_RESTART, do_timer_control) */ ; } else if (fsa_actions & A_DC_TIMER_STOP) { do_fsa_action(fsa_data, A_DC_TIMER_STOP, do_timer_control); } else if (fsa_actions & A_INTEGRATE_TIMER_STOP) { do_fsa_action(fsa_data, A_INTEGRATE_TIMER_STOP, do_timer_control); } else if (fsa_actions & A_INTEGRATE_TIMER_START) { do_fsa_action(fsa_data, A_INTEGRATE_TIMER_START, do_timer_control); } else if (fsa_actions & A_FINALIZE_TIMER_STOP) { do_fsa_action(fsa_data, A_FINALIZE_TIMER_STOP, do_timer_control); } else if (fsa_actions & A_FINALIZE_TIMER_START) { do_fsa_action(fsa_data, A_FINALIZE_TIMER_START, do_timer_control); /* * Highest priority actions */ } else if (fsa_actions & A_MSG_ROUTE) { do_fsa_action(fsa_data, A_MSG_ROUTE, do_msg_route); } else if (fsa_actions & A_RECOVER) { do_fsa_action(fsa_data, A_RECOVER, do_recover); } else if (fsa_actions & A_CL_JOIN_RESULT) { do_fsa_action(fsa_data, A_CL_JOIN_RESULT, do_cl_join_finalize_respond); } else if (fsa_actions & A_CL_JOIN_REQUEST) { do_fsa_action(fsa_data, A_CL_JOIN_REQUEST, do_cl_join_offer_respond); } else if (fsa_actions & A_SHUTDOWN_REQ) { do_fsa_action(fsa_data, A_SHUTDOWN_REQ, do_shutdown_req); } else if (fsa_actions & A_ELECTION_VOTE) { do_fsa_action(fsa_data, A_ELECTION_VOTE, do_election_vote); } else if (fsa_actions & A_ELECTION_COUNT) { do_fsa_action(fsa_data, A_ELECTION_COUNT, do_election_count_vote); } else if (fsa_actions & A_LRM_EVENT) { do_fsa_action(fsa_data, A_LRM_EVENT, do_lrm_event); /* * High priority actions */ } else if (fsa_actions & A_STARTED) { do_fsa_action(fsa_data, A_STARTED, do_started); } else if (fsa_actions & A_CL_JOIN_QUERY) { do_fsa_action(fsa_data, A_CL_JOIN_QUERY, do_cl_join_query); } else if (fsa_actions & A_DC_TIMER_START) { do_fsa_action(fsa_data, A_DC_TIMER_START, do_timer_control); /* * Medium priority actions * - Membership */ } else if (fsa_actions & A_DC_TAKEOVER) { do_fsa_action(fsa_data, A_DC_TAKEOVER, do_dc_takeover); } else if (fsa_actions & A_DC_RELEASE) { do_fsa_action(fsa_data, A_DC_RELEASE, do_dc_release); } else if (fsa_actions & A_DC_JOIN_FINAL) { do_fsa_action(fsa_data, A_DC_JOIN_FINAL, do_dc_join_final); } else if (fsa_actions & A_ELECTION_CHECK) { do_fsa_action(fsa_data, A_ELECTION_CHECK, do_election_check); } else if (fsa_actions & A_ELECTION_START) { do_fsa_action(fsa_data, A_ELECTION_START, do_election_vote); } else if (fsa_actions & A_DC_JOIN_OFFER_ALL) { do_fsa_action(fsa_data, A_DC_JOIN_OFFER_ALL, do_dc_join_offer_all); } else if (fsa_actions & A_DC_JOIN_OFFER_ONE) { do_fsa_action(fsa_data, A_DC_JOIN_OFFER_ONE, do_dc_join_offer_one); } else if (fsa_actions & A_DC_JOIN_PROCESS_REQ) { do_fsa_action(fsa_data, A_DC_JOIN_PROCESS_REQ, do_dc_join_filter_offer); } else if (fsa_actions & A_DC_JOIN_PROCESS_ACK) { do_fsa_action(fsa_data, A_DC_JOIN_PROCESS_ACK, do_dc_join_ack); } else if (fsa_actions & A_DC_JOIN_FINALIZE) { do_fsa_action(fsa_data, A_DC_JOIN_FINALIZE, do_dc_join_finalize); } else if (fsa_actions & A_CL_JOIN_ANNOUNCE) { do_fsa_action(fsa_data, A_CL_JOIN_ANNOUNCE, do_cl_join_announce); /* * Low(er) priority actions * Make sure the CIB is always updated before invoking the * PE, and the PE before the TE */ } else if (fsa_actions & A_TE_HALT) { do_fsa_action(fsa_data, A_TE_HALT, do_te_invoke); } else if (fsa_actions & A_TE_CANCEL) { do_fsa_action(fsa_data, A_TE_CANCEL, do_te_invoke); } else if (fsa_actions & A_LRM_INVOKE) { do_fsa_action(fsa_data, A_LRM_INVOKE, do_lrm_invoke); } else if (fsa_actions & A_PE_INVOKE) { do_fsa_action(fsa_data, A_PE_INVOKE, do_pe_invoke); } else if (fsa_actions & A_TE_INVOKE) { do_fsa_action(fsa_data, A_TE_INVOKE, do_te_invoke); /* Shutdown actions */ } else if (fsa_actions & A_DC_RELEASED) { do_fsa_action(fsa_data, A_DC_RELEASED, do_dc_release); } else if (fsa_actions & A_PE_STOP) { do_fsa_action(fsa_data, A_PE_STOP, do_pe_control); } else if (fsa_actions & A_TE_STOP) { do_fsa_action(fsa_data, A_TE_STOP, do_te_control); } else if (fsa_actions & A_SHUTDOWN) { do_fsa_action(fsa_data, A_SHUTDOWN, do_shutdown); } else if (fsa_actions & A_LRM_DISCONNECT) { do_fsa_action(fsa_data, A_LRM_DISCONNECT, do_lrm_control); } else if (fsa_actions & A_HA_DISCONNECT) { do_fsa_action(fsa_data, A_HA_DISCONNECT, do_ha_control); } else if (fsa_actions & A_CIB_STOP) { do_fsa_action(fsa_data, A_CIB_STOP, do_cib_control); } else if (fsa_actions & A_STOP) { do_fsa_action(fsa_data, A_STOP, do_stop); /* exit gracefully */ } else if (fsa_actions & A_EXIT_0) { do_fsa_action(fsa_data, A_EXIT_0, do_exit); /* Error checking and reporting */ } else { crm_err("Action %s not supported "CRM_XS" 0x%llx", fsa_action2string(fsa_actions), fsa_actions); register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, fsa_data, NULL, __FUNCTION__); } } } void log_fsa_input(fsa_data_t * stored_msg) { CRM_ASSERT(stored_msg); crm_trace("Processing queued input %d", stored_msg->id); if (stored_msg->fsa_cause == C_LRM_OP_CALLBACK) { crm_trace("FSA processing LRM callback from %s", stored_msg->origin); } else if (stored_msg->data == NULL) { crm_trace("FSA processing input from %s", stored_msg->origin); } else { ha_msg_input_t *ha_input = fsa_typed_data_adv(stored_msg, fsa_dt_ha_msg, __FUNCTION__); crm_trace("FSA processing XML message from %s", stored_msg->origin); crm_log_xml_trace(ha_input->xml, "FSA message data"); } } long long do_state_transition(long long actions, enum crmd_fsa_state cur_state, enum crmd_fsa_state next_state, fsa_data_t * msg_data) { int level = LOG_INFO; long long tmp = actions; gboolean clear_recovery_bit = TRUE; enum crmd_fsa_cause cause = msg_data->fsa_cause; enum crmd_fsa_input current_input = msg_data->fsa_input; const char *state_from = fsa_state2string(cur_state); const char *state_to = fsa_state2string(next_state); const char *input = fsa_input2string(current_input); CRM_LOG_ASSERT(cur_state != next_state); do_dot_log(DOT_PREFIX "\t%s -> %s [ label=%s cause=%s origin=%s ]", state_from, state_to, input, fsa_cause2string(cause), msg_data->origin); if (cur_state == S_IDLE || next_state == S_IDLE) { level = LOG_NOTICE; } else if (cur_state == S_NOT_DC || next_state == S_NOT_DC) { level = LOG_NOTICE; } else if (cur_state == S_ELECTION) { level = LOG_NOTICE; } else if (cur_state == S_STARTING) { level = LOG_NOTICE; } else if (next_state == S_RECOVERY) { level = LOG_WARNING; } do_crm_log(level, "State transition %s -> %s " CRM_XS " input=%s cause=%s origin=%s", state_from, state_to, input, fsa_cause2string(cause), msg_data->origin); if (next_state != S_ELECTION && cur_state != S_RELEASE_DC) { controld_stop_election_timer(); } #if 0 if ((fsa_input_register & R_SHUTDOWN)) { set_bit(tmp, A_DC_TIMER_STOP); } #endif if (next_state == S_INTEGRATION) { set_bit(tmp, A_INTEGRATE_TIMER_START); } else { set_bit(tmp, A_INTEGRATE_TIMER_STOP); } if (next_state == S_FINALIZE_JOIN) { set_bit(tmp, A_FINALIZE_TIMER_START); } else { set_bit(tmp, A_FINALIZE_TIMER_STOP); } if (next_state != S_PENDING) { set_bit(tmp, A_DC_TIMER_STOP); } if (next_state != S_ELECTION) { highest_born_on = 0; } if (next_state != S_IDLE) { - crm_timer_stop(recheck_timer); + controld_stop_timer(recheck_timer); } if (cur_state == S_FINALIZE_JOIN && next_state == S_POLICY_ENGINE) { populate_cib_nodes(node_update_quick|node_update_all, __FUNCTION__); } switch (next_state) { case S_PENDING: fsa_cib_conn->cmds->set_slave(fsa_cib_conn, cib_scope_local); /* fall through */ case S_ELECTION: crm_trace("Resetting our DC to NULL on transition to %s", fsa_state2string(next_state)); update_dc(NULL); break; case S_NOT_DC: election_trigger->counter = 0; purge_stonith_cleanup(); if (is_set(fsa_input_register, R_SHUTDOWN)) { crm_info("(Re)Issuing shutdown request now" " that we have a new DC"); set_bit(tmp, A_SHUTDOWN_REQ); } CRM_LOG_ASSERT(fsa_our_dc != NULL); if (fsa_our_dc == NULL) { crm_err("Reached S_NOT_DC without a DC" " being recorded"); } break; case S_RECOVERY: clear_recovery_bit = FALSE; break; case S_FINALIZE_JOIN: CRM_LOG_ASSERT(AM_I_DC); if (cause == C_TIMER_POPPED) { crm_warn("Progressed to state %s after %s", fsa_state2string(next_state), fsa_cause2string(cause)); } if (crmd_join_phase_count(crm_join_welcomed) > 0) { crm_warn("%u cluster nodes failed to respond" " to the join offer.", crmd_join_phase_count(crm_join_welcomed)); crmd_join_phase_log(LOG_NOTICE); } else { crm_debug("All %d cluster nodes responded to the join offer.", crmd_join_phase_count(crm_join_integrated)); } break; case S_POLICY_ENGINE: election_trigger->counter = 0; CRM_LOG_ASSERT(AM_I_DC); if (cause == C_TIMER_POPPED) { crm_info("Progressed to state %s after %s", fsa_state2string(next_state), fsa_cause2string(cause)); } if (crmd_join_phase_count(crm_join_finalized) > 0) { crm_err("%u cluster nodes failed to confirm their join.", crmd_join_phase_count(crm_join_finalized)); crmd_join_phase_log(LOG_NOTICE); } else if (crmd_join_phase_count(crm_join_confirmed) == crm_active_peers()) { crm_debug("All %u cluster nodes are" " eligible to run resources.", crm_active_peers()); } else if (crmd_join_phase_count(crm_join_confirmed) > crm_active_peers()) { crm_err("We have more confirmed nodes than our membership does: %d vs. %d", crmd_join_phase_count(crm_join_confirmed), crm_active_peers()); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } else if (saved_ccm_membership_id != crm_peer_seq) { crm_info("Membership changed: %llu -> %llu - join restart", saved_ccm_membership_id, crm_peer_seq); register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } else { crm_warn("Only %u of %u cluster " "nodes are eligible to run resources - continue %d", crmd_join_phase_count(crm_join_confirmed), crm_active_peers(), crmd_join_phase_count(crm_join_welcomed)); } /* initialize_join(FALSE); */ break; case S_STOPPING: case S_TERMINATE: /* possibly redundant */ set_bit(fsa_input_register, R_SHUTDOWN); break; case S_IDLE: CRM_LOG_ASSERT(AM_I_DC); if (is_set(fsa_input_register, R_SHUTDOWN)) { crm_info("(Re)Issuing shutdown request now" " that we are the DC"); set_bit(tmp, A_SHUTDOWN_REQ); } if (recheck_timer->period_ms > 0) { crm_debug("Starting %s", get_timer_desc(recheck_timer)); - crm_timer_start(recheck_timer); + controld_start_timer(recheck_timer); } break; default: break; } if (clear_recovery_bit && next_state != S_PENDING) { tmp &= ~A_RECOVER; } else if (clear_recovery_bit == FALSE) { tmp |= A_RECOVER; } if (tmp != actions) { /* fsa_dump_actions(actions ^ tmp, "New actions"); */ actions = tmp; } return actions; } diff --git a/daemons/controld/controld_te_actions.c b/daemons/controld/controld_te_actions.c index 505310bcf5..17f737b711 100644 --- a/daemons/controld/controld_te_actions.c +++ b/daemons/controld/controld_te_actions.c @@ -1,635 +1,635 @@ /* * 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 // lrmd_event_data_t, lrmd_free_event() #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 = 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); + controld_stop_timer(transition_timer); + controld_start_timer(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/daemons/controld/controld_te_utils.c b/daemons/controld/controld_te_utils.c index a37d6a3ec1..abd45a6597 100644 --- a/daemons/controld/controld_te_utils.c +++ b/daemons/controld/controld_te_utils.c @@ -1,294 +1,294 @@ /* * 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 gboolean stop_te_timer(crm_action_timer_t * timer) { if (timer == NULL) { return FALSE; } if (timer->source_id != 0) { crm_trace("Stopping action timer"); g_source_remove(timer->source_id); timer->source_id = 0; } else { crm_trace("Action timer was already stopped"); return FALSE; } return TRUE; } gboolean te_graph_trigger(gpointer user_data) { enum transition_status graph_rc = -1; if (transition_graph == NULL) { crm_debug("Nothing to do"); return TRUE; } crm_trace("Invoking graph %d in state %s", transition_graph->id, fsa_state2string(fsa_state)); switch (fsa_state) { case S_STARTING: case S_PENDING: case S_NOT_DC: case S_HALT: case S_ILLEGAL: case S_STOPPING: case S_TERMINATE: return TRUE; break; default: break; } if (transition_graph->complete == FALSE) { int limit = transition_graph->batch_limit; transition_graph->batch_limit = throttle_get_total_job_limit(limit); graph_rc = run_graph(transition_graph); transition_graph->batch_limit = limit; /* Restore the configured value */ /* significant overhead... */ /* print_graph(LOG_TRACE, transition_graph); */ if (graph_rc == transition_active) { crm_trace("Transition not yet complete"); return TRUE; } else if (graph_rc == transition_pending) { crm_trace("Transition not yet complete - no actions fired"); return TRUE; } if (graph_rc != transition_complete) { crm_warn("Transition failed: %s", transition_status(graph_rc)); print_graph(LOG_NOTICE, transition_graph); } } crm_debug("Transition %d is now complete", transition_graph->id); transition_graph->complete = TRUE; notify_crmd(transition_graph); return TRUE; } void trigger_graph_processing(const char *fn, int line) { crm_trace("%s:%d - Triggered graph processing", fn, line); mainloop_set_trigger(transition_trigger); } static struct abort_timer_s { bool aborted; guint id; int priority; enum transition_action action; const char *text; } abort_timer = { 0, }; static gboolean abort_timer_popped(gpointer data) { if (AM_I_DC && (abort_timer.aborted == FALSE)) { abort_transition(abort_timer.priority, abort_timer.action, abort_timer.text, NULL); } abort_timer.id = 0; return FALSE; // do not immediately reschedule timer } /*! * \internal * \brief Abort transition after delay, if not already aborted in that time * * \param[in] abort_text Must be literal string */ void abort_after_delay(int abort_priority, enum transition_action abort_action, const char *abort_text, guint delay_ms) { if (abort_timer.id) { // Timer already in progress, stop and reschedule g_source_remove(abort_timer.id); } abort_timer.aborted = FALSE; abort_timer.priority = abort_priority; abort_timer.action = abort_action; abort_timer.text = abort_text; abort_timer.id = g_timeout_add(delay_ms, abort_timer_popped, NULL); } void abort_transition_graph(int abort_priority, enum transition_action abort_action, const char *abort_text, xmlNode * reason, const char *fn, int line) { int add[] = { 0, 0, 0 }; int del[] = { 0, 0, 0 }; int level = LOG_INFO; xmlNode *diff = NULL; xmlNode *change = NULL; CRM_CHECK(transition_graph != NULL, return); switch (fsa_state) { case S_STARTING: case S_PENDING: case S_NOT_DC: case S_HALT: case S_ILLEGAL: case S_STOPPING: case S_TERMINATE: crm_info("Abort %s suppressed: state=%s (complete=%d)", abort_text, fsa_state2string(fsa_state), transition_graph->complete); return; default: break; } abort_timer.aborted = TRUE; controld_expect_sched_reply(NULL); if (transition_graph->complete == FALSE) { if(update_abort_priority(transition_graph, abort_priority, abort_action, abort_text)) { level = LOG_NOTICE; } } if(reason) { xmlNode *search = NULL; for(search = reason; search; search = search->parent) { if (safe_str_eq(XML_TAG_DIFF, TYPE(search))) { diff = search; break; } } if(diff) { xml_patch_versions(diff, add, del); for(search = reason; search; search = search->parent) { if (safe_str_eq(XML_DIFF_CHANGE, TYPE(search))) { change = search; break; } } } } if(reason == NULL) { do_crm_log(level, "Transition %d aborted: %s "CRM_XS" source=%s:%d complete=%s", transition_graph->id, abort_text, fn, line, (transition_graph->complete? "true" : "false")); } else if(change == NULL) { char *local_path = xml_get_path(reason); do_crm_log(level, "Transition %d aborted by %s.%s: %s " CRM_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s", transition_graph->id, TYPE(reason), ID(reason), abort_text, add[0], add[1], add[2], fn, line, local_path, (transition_graph->complete? "true" : "false")); free(local_path); } else { const char *kind = NULL; const char *op = crm_element_value(change, XML_DIFF_OP); const char *path = crm_element_value(change, XML_DIFF_PATH); if(change == reason) { if(strcmp(op, "create") == 0) { reason = reason->children; } else if(strcmp(op, "modify") == 0) { reason = first_named_child(reason, XML_DIFF_RESULT); if(reason) { reason = reason->children; } } } kind = TYPE(reason); if(strcmp(op, "delete") == 0) { const char *shortpath = strrchr(path, '/'); do_crm_log(level, "Transition %d aborted by deletion of %s: %s " CRM_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s", transition_graph->id, (shortpath? (shortpath + 1) : path), abort_text, add[0], add[1], add[2], fn, line, path, (transition_graph->complete? "true" : "false")); } else if (safe_str_eq(XML_CIB_TAG_NVPAIR, kind)) { do_crm_log(level, "Transition %d aborted by %s doing %s %s=%s: %s " CRM_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s", transition_graph->id, crm_element_value(reason, XML_ATTR_ID), op, crm_element_value(reason, XML_NVPAIR_ATTR_NAME), crm_element_value(reason, XML_NVPAIR_ATTR_VALUE), abort_text, add[0], add[1], add[2], fn, line, path, (transition_graph->complete? "true" : "false")); } else if (safe_str_eq(XML_LRM_TAG_RSC_OP, kind)) { const char *magic = crm_element_value(reason, XML_ATTR_TRANSITION_MAGIC); do_crm_log(level, "Transition %d aborted by operation %s '%s' on %s: %s " CRM_XS " magic=%s cib=%d.%d.%d source=%s:%d complete=%s", transition_graph->id, crm_element_value(reason, XML_LRM_ATTR_TASK_KEY), op, crm_element_value(reason, XML_LRM_ATTR_TARGET), abort_text, magic, add[0], add[1], add[2], fn, line, (transition_graph->complete? "true" : "false")); } else if (safe_str_eq(XML_CIB_TAG_STATE, kind) || safe_str_eq(XML_CIB_TAG_NODE, kind)) { const char *uname = crm_peer_uname(ID(reason)); do_crm_log(level, "Transition %d aborted by %s '%s' on %s: %s " CRM_XS " cib=%d.%d.%d source=%s:%d complete=%s", transition_graph->id, kind, op, (uname? uname : ID(reason)), abort_text, add[0], add[1], add[2], fn, line, (transition_graph->complete? "true" : "false")); } else { const char *id = ID(reason); do_crm_log(level, "Transition %d aborted by %s.%s '%s': %s " CRM_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s", transition_graph->id, TYPE(reason), (id? id : ""), (op? op : "change"), abort_text, add[0], add[1], add[2], fn, line, path, (transition_graph->complete? "true" : "false")); } } if (transition_graph->complete) { if (transition_timer->period_ms > 0) { - crm_timer_stop(transition_timer); - crm_timer_start(transition_timer); + controld_stop_timer(transition_timer); + controld_start_timer(transition_timer); } else { register_fsa_input(C_FSA_INTERNAL, I_PE_CALC, NULL); } return; } mainloop_set_trigger(transition_trigger); } diff --git a/daemons/controld/controld_timers.c b/daemons/controld/controld_timers.c index 57949f8d13..a67fbb336d 100644 --- a/daemons/controld/controld_timers.c +++ b/daemons/controld/controld_timers.c @@ -1,321 +1,316 @@ /* * 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 // Wait before retrying a failed cib or executor connection fsa_timer_t *wait_timer = NULL; // Periodically re-run scheduler (for date_spec evaluation and as a failsafe) fsa_timer_t *recheck_timer = NULL; // Wait at start-up, or after an election, for DC to make contact fsa_timer_t *election_trigger = NULL; // Delay start of new transition with expectation something else might happen fsa_timer_t *transition_timer = NULL; // join-integration-timeout fsa_timer_t *integration_timer = NULL; // join-finalization-timeout fsa_timer_t *finalization_timer = NULL; // Wait for DC to stop all resources and give us the all-clear to shut down fsa_timer_t *shutdown_escalation_timer = NULL; /* A_DC_TIMER_STOP, A_DC_TIMER_START, * A_FINALIZE_TIMER_STOP, A_FINALIZE_TIMER_START * A_INTEGRATE_TIMER_STOP, A_INTEGRATE_TIMER_START */ void do_timer_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) { gboolean timer_op_ok = TRUE; if (action & A_DC_TIMER_STOP) { - timer_op_ok = crm_timer_stop(election_trigger); + timer_op_ok = controld_stop_timer(election_trigger); } else if (action & A_FINALIZE_TIMER_STOP) { - timer_op_ok = crm_timer_stop(finalization_timer); + timer_op_ok = controld_stop_timer(finalization_timer); } else if (action & A_INTEGRATE_TIMER_STOP) { - timer_op_ok = crm_timer_stop(integration_timer); + timer_op_ok = controld_stop_timer(integration_timer); } /* don't start a timer that wasn't already running */ if (action & A_DC_TIMER_START && timer_op_ok) { - crm_timer_start(election_trigger); + controld_start_timer(election_trigger); if (AM_I_DC) { /* there can be only one */ register_fsa_input(cause, I_ELECTION, NULL); } } else if (action & A_FINALIZE_TIMER_START) { - crm_timer_start(finalization_timer); + controld_start_timer(finalization_timer); } else if (action & A_INTEGRATE_TIMER_START) { - crm_timer_start(integration_timer); + controld_start_timer(integration_timer); } } const char * get_timer_desc(fsa_timer_t * timer) { if (timer == election_trigger) { return "Election Trigger"; } else if (timer == shutdown_escalation_timer) { return "Shutdown Escalation"; } else if (timer == integration_timer) { return "Integration Timer"; } else if (timer == finalization_timer) { return "Finalization Timer"; } else if (timer == transition_timer) { return "New Transition Timer"; } else if (timer == wait_timer) { return "Wait Timer"; } else if (timer == recheck_timer) { return "Cluster Recheck Timer"; } return "Unknown Timer"; } static gboolean crm_timer_popped(gpointer data) { fsa_timer_t *timer = (fsa_timer_t *) data; if (timer->log_error) { crm_err("%s just popped in state %s! " CRM_XS " input=%s time=%ums", get_timer_desc(timer), fsa_state2string(fsa_state), fsa_input2string(timer->fsa_input), timer->period_ms); } else { crm_info("%s just popped " CRM_XS " input=%s time=%ums", get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms); timer->counter++; } if (timer == election_trigger && election_trigger->counter > 5) { crm_notice("We appear to be in an election loop, something may be wrong"); crm_write_blackbox(0, NULL); election_trigger->counter = 0; } - crm_timer_stop(timer); // Make timer _not_ go off again + controld_stop_timer(timer); // Make timer _not_ go off again if (timer->fsa_input == I_INTEGRATED) { crm_info("Welcomed: %d, Integrated: %d", crmd_join_phase_count(crm_join_welcomed), crmd_join_phase_count(crm_join_integrated)); if (crmd_join_phase_count(crm_join_welcomed) == 0) { // If we don't even have ourselves, start again register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, NULL, __FUNCTION__); } else { register_fsa_input_before(C_TIMER_POPPED, timer->fsa_input, NULL); } } else if (timer == recheck_timer && fsa_state != S_IDLE) { crm_debug("Discarding %s event in state: %s", fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state)); } else if (timer == finalization_timer && fsa_state != S_FINALIZE_JOIN) { crm_debug("Discarding %s event in state: %s", fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state)); } else if (timer->fsa_input != I_NULL) { register_fsa_input(C_TIMER_POPPED, timer->fsa_input, NULL); } crm_trace("Triggering FSA: %s", __FUNCTION__); mainloop_set_trigger(fsa_source); return TRUE; } bool controld_init_fsa_timers() { transition_timer = calloc(1, sizeof(fsa_timer_t)); if (transition_timer == NULL) { return FALSE; } integration_timer = calloc(1, sizeof(fsa_timer_t)); if (integration_timer == NULL) { return FALSE; } finalization_timer = calloc(1, sizeof(fsa_timer_t)); if (finalization_timer == NULL) { return FALSE; } election_trigger = calloc(1, sizeof(fsa_timer_t)); if (election_trigger == NULL) { return FALSE; } shutdown_escalation_timer = calloc(1, sizeof(fsa_timer_t)); if (shutdown_escalation_timer == NULL) { return FALSE; } wait_timer = calloc(1, sizeof(fsa_timer_t)); if (wait_timer == NULL) { return FALSE; } recheck_timer = calloc(1, sizeof(fsa_timer_t)); if (recheck_timer == NULL) { return FALSE; } election_trigger->source_id = 0; election_trigger->period_ms = 0; election_trigger->fsa_input = I_DC_TIMEOUT; election_trigger->callback = crm_timer_popped; election_trigger->log_error = FALSE; transition_timer->source_id = 0; transition_timer->period_ms = 0; transition_timer->fsa_input = I_PE_CALC; transition_timer->callback = crm_timer_popped; transition_timer->log_error = FALSE; integration_timer->source_id = 0; integration_timer->period_ms = 0; integration_timer->fsa_input = I_INTEGRATED; integration_timer->callback = crm_timer_popped; integration_timer->log_error = TRUE; finalization_timer->source_id = 0; finalization_timer->period_ms = 0; finalization_timer->fsa_input = I_FINALIZED; finalization_timer->callback = crm_timer_popped; finalization_timer->log_error = FALSE; /* for possible enabling... a bug in the join protocol left * a slave in S_PENDING while we think it's in S_NOT_DC * * raising I_FINALIZED put us into a transition loop which is * never resolved. * in this loop we continually send probes which the node * NACK's because it's in S_PENDING * * if we have nodes where the cluster layer is active but the * CRM is not... then this will be handled in the * integration phase */ finalization_timer->fsa_input = I_ELECTION; shutdown_escalation_timer->source_id = 0; shutdown_escalation_timer->period_ms = 0; shutdown_escalation_timer->fsa_input = I_STOP; shutdown_escalation_timer->callback = crm_timer_popped; shutdown_escalation_timer->log_error = TRUE; wait_timer->source_id = 0; wait_timer->period_ms = 2000; wait_timer->fsa_input = I_NULL; wait_timer->callback = crm_timer_popped; wait_timer->log_error = FALSE; recheck_timer->source_id = 0; recheck_timer->period_ms = 0; recheck_timer->fsa_input = I_PE_CALC; recheck_timer->callback = crm_timer_popped; recheck_timer->log_error = FALSE; return TRUE; } void controld_free_fsa_timers() { - crm_timer_stop(transition_timer); - crm_timer_stop(integration_timer); - crm_timer_stop(finalization_timer); - crm_timer_stop(election_trigger); - crm_timer_stop(shutdown_escalation_timer); - crm_timer_stop(wait_timer); - crm_timer_stop(recheck_timer); + controld_stop_timer(transition_timer); + controld_stop_timer(integration_timer); + controld_stop_timer(finalization_timer); + controld_stop_timer(election_trigger); + controld_stop_timer(shutdown_escalation_timer); + controld_stop_timer(wait_timer); + controld_stop_timer(recheck_timer); free(transition_timer); transition_timer = NULL; free(integration_timer); integration_timer = NULL; free(finalization_timer); finalization_timer = NULL; free(election_trigger); election_trigger = NULL; free(shutdown_escalation_timer); shutdown_escalation_timer = NULL; free(wait_timer); wait_timer = NULL; free(recheck_timer); recheck_timer = NULL; } gboolean is_timer_started(fsa_timer_t * timer) { return (timer->period_ms > 0) && (timer->source_id != 0); } -gboolean -crm_timer_start(fsa_timer_t * timer) +void +controld_start_timer(fsa_timer_t *timer) { if (timer->source_id == 0 && timer->period_ms > 0) { timer->source_id = g_timeout_add(timer->period_ms, timer->callback, (void *)timer); CRM_ASSERT(timer->source_id != 0); crm_debug("Started %s (inject %s if pops after %ums, source=%d)", get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms, timer->source_id); - return TRUE; + } else { + crm_debug("%s already running (inject %s if pops after %ums, source=%d)", + get_timer_desc(timer), fsa_input2string(timer->fsa_input), + timer->period_ms, timer->source_id); } - - crm_debug("%s already running (inject %s if pops after %ums, source=%d)", - get_timer_desc(timer), fsa_input2string(timer->fsa_input), - timer->period_ms, timer->source_id); - return FALSE; } gboolean -crm_timer_stop(fsa_timer_t * timer) +controld_stop_timer(fsa_timer_t *timer) { - const char *timer_desc = get_timer_desc(timer); - - if (timer == NULL) { - crm_err("Attempted to stop NULL timer"); - return FALSE; + CRM_CHECK(timer != NULL, return FALSE); - } else if (timer->source_id != 0) { + if (timer->source_id != 0) { crm_trace("Stopping %s (would inject %s if popped after %ums, src=%d)", - timer_desc, fsa_input2string(timer->fsa_input), + get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms, timer->source_id); g_source_remove(timer->source_id); timer->source_id = 0; } else { crm_trace("%s already stopped (would inject %s if popped after %ums)", - timer_desc, fsa_input2string(timer->fsa_input), timer->period_ms); + get_timer_desc(timer), fsa_input2string(timer->fsa_input), + timer->period_ms); return FALSE; } return TRUE; } diff --git a/daemons/controld/controld_timers.h b/daemons/controld/controld_timers.h index 2db7edfd5a..8c27919057 100644 --- a/daemons/controld/controld_timers.h +++ b/daemons/controld/controld_timers.h @@ -1,42 +1,42 @@ /* * 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 CONTROLD_TIMERS__H # define CONTROLD_TIMERS__H # include // bool # include // gboolean, gpointer, guint # include // crmd_fsa_input typedef struct fsa_timer_s { guint source_id; // Timer source ID guint period_ms; // Timer period enum crmd_fsa_input fsa_input; // Input to register if timer pops gboolean (*callback) (gpointer data); // What do if timer pops bool log_error; // Timer popping indicates error int counter; // For detecting loops } fsa_timer_t; extern fsa_timer_t *election_trigger; extern fsa_timer_t *shutdown_escalation_timer; extern fsa_timer_t *transition_timer; extern fsa_timer_t *integration_timer; extern fsa_timer_t *finalization_timer; extern fsa_timer_t *wait_timer; extern fsa_timer_t *recheck_timer; bool controld_init_fsa_timers(void); void controld_free_fsa_timers(void); -gboolean crm_timer_stop(fsa_timer_t *timer); -gboolean crm_timer_start(fsa_timer_t *timer); +gboolean controld_stop_timer(fsa_timer_t *timer); +void controld_start_timer(fsa_timer_t *timer); gboolean is_timer_started(fsa_timer_t *timer); const char *get_timer_desc(fsa_timer_t * timer); #endif