diff --git a/daemons/controld/controld_callbacks.c b/daemons/controld/controld_callbacks.c index a2947ed53f..0688ea956c 100644 --- a/daemons/controld/controld_callbacks.c +++ b/daemons/controld/controld_callbacks.c @@ -1,385 +1,387 @@ /* * Copyright 2004-2022 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include /* From join_dc... */ extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); void crmd_ha_msg_filter(xmlNode * msg) { if (AM_I_DC) { const char *sys_from = crm_element_value(msg, F_CRM_SYS_FROM); if (pcmk__str_eq(sys_from, CRM_SYSTEM_DC, pcmk__str_casei)) { const char *from = crm_element_value(msg, F_ORIG); if (!pcmk__str_eq(from, controld_globals.our_nodename, pcmk__str_casei)) { int level = LOG_INFO; const char *op = crm_element_value(msg, F_CRM_TASK); /* make sure the election happens NOW */ if (controld_globals.fsa_state != S_ELECTION) { ha_msg_input_t new_input; level = LOG_WARNING; new_input.msg = msg; register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, &new_input, __func__); } do_crm_log(level, "Another DC detected: %s (op=%s)", from, op); goto done; } } } else { const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO); if (pcmk__str_eq(sys_to, CRM_SYSTEM_DC, pcmk__str_casei)) { return; } } /* crm_log_xml_trace(msg, "HA[inbound]"); */ route_message(C_HA_MESSAGE, msg); done: controld_trigger_fsa(); } /*! * \internal * \brief Check whether a node is online * * \param[in] node Node to check * * \retval -1 if completely dead * \retval 0 if partially alive * \retval 1 if completely alive */ static int node_alive(const crm_node_t *node) { if (pcmk_is_set(node->flags, crm_remote_node)) { // Pacemaker Remote nodes can't be partially alive return pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei) ? 1: -1; } else if (crm_is_peer_active(node)) { // Completely up cluster node: both cluster member and peer return 1; } else if (!pcmk_is_set(node->processes, crm_get_cluster_proc()) && !pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei)) { // Completely down cluster node: neither cluster member nor peer return -1; } // Partially up cluster node: only cluster member or only peer return 0; } #define state_text(state) ((state)? (const char *)(state) : "in unknown state") void peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data) { uint32_t old = 0; bool appeared = FALSE; bool is_remote = pcmk_is_set(node->flags, crm_remote_node); /* The controller waits to receive some information from the membership * layer before declaring itself operational. If this is being called for a * cluster node, indicate that we have it. */ if (!is_remote) { controld_set_fsa_input_flags(R_PEER_DATA); } if (type == crm_status_processes && pcmk_is_set(node->processes, crm_get_cluster_proc()) && !AM_I_DC && !is_remote) { /* * This is a hack until we can send to a nodeid and/or we fix node name lookups * These messages are ignored in crmd_ha_msg_filter() */ xmlNode *query = create_request(CRM_OP_HELLO, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); crm_debug("Sending hello to node %u so that it learns our node name", node->id); send_cluster_message(node, crm_msg_crmd, query, FALSE); free_xml(query); } if (node->uname == NULL) { return; } switch (type) { case crm_status_uname: /* If we've never seen the node, then it also won't be in the status section */ crm_info("%s node %s is now %s", (is_remote? "Remote" : "Cluster"), node->uname, state_text(node->state)); return; case crm_status_nstate: /* This callback should not be called unless the state actually * changed, but here's a failsafe just in case. */ CRM_CHECK(!pcmk__str_eq(data, node->state, pcmk__str_casei), return); crm_info("%s node %s is now %s (was %s)", (is_remote? "Remote" : "Cluster"), node->uname, state_text(node->state), state_text(data)); if (pcmk__str_eq(CRM_NODE_MEMBER, node->state, pcmk__str_casei)) { appeared = TRUE; if (!is_remote) { remove_stonith_cleanup(node->uname); } } else { + controld_remove_failed_sync_node(node->uname); controld_remove_voter(node->uname); } crmd_alert_node_event(node); break; case crm_status_processes: CRM_CHECK(data != NULL, return); old = *(const uint32_t *)data; appeared = pcmk_is_set(node->processes, crm_get_cluster_proc()); { const char *dc_s = controld_globals.dc_name; if ((dc_s == NULL) && AM_I_DC) { dc_s = "true"; } crm_info("Node %s is %s a peer " CRM_XS " DC=%s old=%#07x new=%#07x", node->uname, (appeared? "now" : "no longer"), pcmk__s(dc_s, ""), old, node->processes); } if (!pcmk_is_set((node->processes ^ old), crm_get_cluster_proc())) { /* Peer status did not change. This should not be possible, * since we don't track process flags other than peer status. */ crm_trace("Process flag %#7x did not change from %#7x to %#7x", crm_get_cluster_proc(), old, node->processes); return; } if (!appeared) { + controld_remove_failed_sync_node(node->uname); controld_remove_voter(node->uname); } if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { crm_trace("Ignoring peer status change because not connected to CIB"); return; } else if (controld_globals.fsa_state == S_STOPPING) { crm_trace("Ignoring peer status change because stopping"); return; } if (!appeared && pcmk__str_eq(node->uname, controld_globals.our_nodename, pcmk__str_casei)) { /* Did we get evicted? */ crm_notice("Our peer connection failed"); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL); } else if (pcmk__str_eq(node->uname, controld_globals.dc_name, pcmk__str_casei) && !crm_is_peer_active(node)) { /* Did the DC leave us? */ crm_notice("Our peer on the DC (%s) is dead", controld_globals.dc_name); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL); /* @COMPAT DC < 1.1.13: If a DC shuts down normally, we don't * want to fence it. Newer DCs will send their shutdown request * to all peers, who will update the DC's expected state to * down, thus avoiding fencing. We can safely erase the DC's * transient attributes when it leaves in that case. However, * the only way to avoid fencing older DCs is to leave the * transient attributes intact until it rejoins. */ if (compare_version(controld_globals.dc_version, "3.0.9") > 0) { controld_delete_node_state(node->uname, controld_section_attrs, cib_scope_local); } } else if (AM_I_DC || pcmk_is_set(controld_globals.flags, controld_dc_left) || (controld_globals.dc_name == NULL)) { /* This only needs to be done once, so normally the DC should do * it. However if there is no DC, every node must do it, since * there is no other way to ensure some one node does it. */ if (appeared) { te_trigger_stonith_history_sync(FALSE); } else { controld_delete_node_state(node->uname, controld_section_attrs, cib_scope_local); } } break; } if (AM_I_DC) { xmlNode *update = NULL; int flags = node_update_peer; int alive = node_alive(node); pcmk__graph_action_t *down = match_down_event(node->uuid); crm_trace("Alive=%d, appeared=%d, down=%d", alive, appeared, (down? down->id : -1)); if (appeared && (alive > 0) && !is_remote) { register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } if (down) { const char *task = crm_element_value(down->xml, XML_LRM_ATTR_TASK); if (pcmk__str_eq(task, CRM_OP_FENCE, pcmk__str_casei)) { /* tengine_stonith_callback() confirms fence actions */ crm_trace("Updating CIB %s fencer reported fencing of %s complete", (pcmk_is_set(down->flags, pcmk__graph_action_confirmed)? "after" : "before"), node->uname); } else if (!appeared && pcmk__str_eq(task, CRM_OP_SHUTDOWN, pcmk__str_casei)) { // Shutdown actions are immediately confirmed (i.e. no_wait) if (!is_remote) { flags |= node_update_join | node_update_expected; crmd_peer_down(node, FALSE); check_join_state(controld_globals.fsa_state, __func__); } if (alive >= 0) { crm_info("%s of peer %s is in progress " CRM_XS " action=%d", task, node->uname, down->id); } else { crm_notice("%s of peer %s is complete " CRM_XS " action=%d", task, node->uname, down->id); pcmk__update_graph(controld_globals.transition_graph, down); trigger_graph(); } } else { crm_trace("Node %s is %s, was expected to %s (op %d)", node->uname, ((alive > 0)? "alive" : ((alive < 0)? "dead" : "partially alive")), task, down->id); } } else if (appeared == FALSE) { if ((controld_globals.transition_graph == NULL) || (controld_globals.transition_graph->id == -1)) { crm_info("Stonith/shutdown of node %s is unknown to the " "current DC", node->uname); } else { crm_warn("Stonith/shutdown of node %s was not expected", node->uname); } if (!is_remote) { crm_update_peer_join(__func__, node, crm_join_none); check_join_state(controld_globals.fsa_state, __func__); } abort_transition(INFINITY, pcmk__graph_restart, "Node failure", NULL); fail_incompletable_actions(controld_globals.transition_graph, node->uuid); } else { crm_trace("Node %s came up, was not expected to be down", node->uname); } if (is_remote) { /* A pacemaker_remote node won't have its cluster status updated * in the CIB by membership-layer callbacks, so do it here. */ flags |= node_update_cluster; /* Trigger resource placement on newly integrated nodes */ if (appeared) { abort_transition(INFINITY, pcmk__graph_restart, "Pacemaker Remote node integrated", NULL); } } /* Update the CIB node state */ update = create_node_state_update(node, flags, NULL, __func__); if (update == NULL) { crm_debug("Node state update not yet possible for %s", node->uname); } else { fsa_cib_anon_update(XML_CIB_TAG_STATUS, update); } free_xml(update); } controld_trigger_fsa(); } void crmd_cib_connection_destroy(gpointer user_data) { CRM_LOG_ASSERT(user_data == controld_globals.cib_conn); controld_trigger_fsa(); controld_globals.cib_conn->state = cib_disconnected; if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { crm_info("Connection to the CIB manager terminated"); return; } // @TODO This should trigger a reconnect, not a shutdown crm_crit("Lost connection to the CIB manager, shutting down"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_CIB_CONNECTED); return; } gboolean crm_fsa_trigger(gpointer user_data) { crm_trace("Invoked (queue len: %d)", g_list_length(controld_globals.fsa_message_queue)); s_crmd_fsa(C_FSA_INTERNAL); crm_trace("Exited (queue len: %d)", g_list_length(controld_globals.fsa_message_queue)); return TRUE; } diff --git a/daemons/controld/controld_control.c b/daemons/controld/controld_control.c index ef8b472983..77f174257b 100644 --- a/daemons/controld/controld_control.c +++ b/daemons/controld/controld_control.c @@ -1,838 +1,839 @@ /* * Copyright 2004-2022 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include static qb_ipcs_service_t *ipcs = NULL; static crm_trigger_t *config_read_trigger = NULL; #if SUPPORT_COROSYNC extern gboolean crm_connect_corosync(crm_cluster_t * cluster); #endif void crm_shutdown(int nsig); static gboolean crm_read_options(gpointer user_data); /* 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 = pcmk_cluster_new(); } if (action & A_HA_DISCONNECT) { crm_cluster_disconnect(cluster); crm_info("Disconnected from the cluster"); controld_set_fsa_input_flags(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) { controld_election_init(cluster->uname); controld_globals.our_nodename = cluster->uname; controld_globals.our_uuid = cluster->uuid; if(cluster->uuid == NULL) { crm_err("Could not obtain local uuid"); registered = FALSE; } } if (!registered) { controld_set_fsa_input_flags(R_HA_DISCONNECTED); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } populate_cib_nodes(node_update_none, __func__); controld_clear_fsa_input_flags(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), __func__); } } /* 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 */ controld_set_fsa_input_flags(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; controld_set_fsa_input_flags(R_SHUTDOWN); //controld_set_fsa_input_flags(R_STAYDOWN); crm_info("Sending shutdown request to all peers (DC is %s)", pcmk__s(controld_globals.dc_name, "not set")); msg = create_request(CRM_OP_SHUTDOWN_REQ, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } free_xml(msg); } void crmd_fast_exit(crm_exit_t exit_code) { if (pcmk_is_set(controld_globals.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) && pcmk_is_set(controld_globals.fsa_input_register, R_IN_RECOVERY)) { crm_err("Could not recover from internal error"); exit_code = CRM_EX_ERROR; } if (controld_globals.logger_out != NULL) { controld_globals.logger_out->finish(controld_globals.logger_out, exit_code, true, NULL); pcmk__output_free(controld_globals.logger_out); controld_globals.logger_out = NULL; } crm_exit(exit_code); } crm_exit_t crmd_exit(crm_exit_t exit_code) { GMainLoop *mloop = controld_globals.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 */ controld_set_fsa_input_flags(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(); controld_shutdown_schedulerd_ipc(); controld_disconnect_fencer(TRUE); if ((exit_code == CRM_EX_OK) && (controld_globals.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 (GList *iter = controld_globals.fsa_message_queue; iter != NULL; iter = iter->next) { fsa_data_t *fsa_data = (fsa_data_t *) iter->data; crm_info("Dropping %s: [ state=%s cause=%s origin=%s ]", fsa_input2string(fsa_data->fsa_input), fsa_state2string(controld_globals.fsa_state), fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin); delete_fsa_input(fsa_data); } controld_clear_fsa_input_flags(R_MEMBERSHIP); g_list_free(controld_globals.fsa_message_queue); controld_globals.fsa_message_queue = NULL; 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. */ controld_disconnect_cib_manager(); verify_stopped(controld_globals.fsa_state, LOG_WARNING); controld_clear_fsa_input_flags(R_LRM_CONNECTED); lrm_state_destroy_all(); mainloop_destroy_trigger(config_read_trigger); config_read_trigger = NULL; controld_destroy_fsa_trigger(); controld_destroy_transition_trigger(); pcmk__client_cleanup(); crm_peer_destroy(); controld_free_fsa_timers(); te_cleanup_stonith_history_sync(NULL, TRUE); controld_free_sched_timer(); free(controld_globals.our_nodename); controld_globals.our_nodename = NULL; free(controld_globals.our_uuid); controld_globals.our_uuid = NULL; free(controld_globals.dc_name); controld_globals.dc_name = NULL; free(controld_globals.dc_version); controld_globals.dc_version = NULL; free(controld_globals.cluster_name); controld_globals.cluster_name = NULL; free(controld_globals.te_uuid); controld_globals.te_uuid = NULL; free_max_generation(); + controld_destroy_failed_sync_table(); 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(controld_globals.mainloop); /* Don't re-enter this block */ controld_globals.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(controld_globals.cib_conn); controld_globals.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); config_read_trigger = mainloop_add_trigger(G_PRIORITY_HIGH, crm_read_options, NULL); controld_init_fsa_trigger(); controld_init_transition_trigger(); crm_debug("Creating CIB manager and executor objects"); controld_globals.cib_conn = cib_new(); lrm_state_init_local(); if (controld_init_fsa_timers() == FALSE) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } // \return libqb error code (0 on success, -errno on error) static int32_t accept_controller_client(qb_ipcs_connection_t *c, uid_t uid, gid_t gid) { crm_trace("Accepting new IPC client connection"); if (pcmk__new_client(c, uid, gid) == NULL) { return -EIO; } return 0; } // \return libqb error code (0 on success, -errno on error) static int32_t dispatch_controller_ipc(qb_ipcs_connection_t * c, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; pcmk__client_t *client = pcmk__find_client(c); xmlNode *msg = pcmk__client_data2xml(client, data, &id, &flags); if (msg == NULL) { pcmk__ipc_send_ack(client, id, flags, "ack", NULL, CRM_EX_PROTOCOL); return 0; } pcmk__ipc_send_ack(client, id, flags, "ack", NULL, CRM_EX_INDETERMINATE); CRM_ASSERT(client->user != NULL); pcmk__update_acl_user(msg, F_CRM_USER, client->user); crm_xml_add(msg, F_CRM_SYS_FROM, client->id); if (controld_authorize_ipc_message(msg, client, NULL)) { crm_trace("Processing IPC message from client %s", pcmk__client_name(client)); route_message(C_IPC_MESSAGE, msg); } controld_trigger_fsa(); free_xml(msg); return 0; } static int32_t ipc_client_disconnected(qb_ipcs_connection_t *c) { pcmk__client_t *client = pcmk__find_client(c); if (client) { crm_trace("Disconnecting %sregistered client %s (%p/%p)", (client->userdata? "" : "un"), pcmk__client_name(client), c, client); free(client->userdata); pcmk__free_client(client); controld_trigger_fsa(); } return 0; } static void ipc_connection_destroyed(qb_ipcs_connection_t *c) { crm_trace("Connection %p", c); ipc_client_disconnected(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 = accept_controller_client, .connection_created = NULL, .msg_process = dispatch_controller_ipc, .connection_closed = ipc_client_disconnected, .connection_destroyed = ipc_connection_destroyed }; if (cur_state != S_STARTING) { crm_err("Start cancelled... %s", fsa_state2string(cur_state)); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_MEMBERSHIP)) { crm_info("Delaying start, no membership data (%.16llx)", R_MEMBERSHIP); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_LRM_CONNECTED)) { crm_info("Delaying start, not connected to executor (%.16llx)", R_LRM_CONNECTED); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { crm_info("Delaying start, CIB not connected (%.16llx)", R_CIB_CONNECTED); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_READ_CONFIG)) { crm_info("Delaying start, Config not read (%.16llx)", R_READ_CONFIG); crmd_fsa_stall(TRUE); return; } else if (!pcmk_is_set(controld_globals.fsa_input_register, R_PEER_DATA)) { crm_info("Delaying start, No peer data (%.16llx)", R_PEER_DATA); crmd_fsa_stall(TRUE); return; } crm_debug("Init server comms"); ipcs = pcmk__serve_controld_ipc(&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(); controld_clear_fsa_input_flags(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) { controld_set_fsa_input_flags(R_IN_RECOVERY); crm_warn("Fast-tracking shutdown in response to errors"); register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL); } static pcmk__cluster_option_t controller_options[] = { /* name, old name, type, allowed values, * default value, validator, * short description, * long description */ { "dc-version", NULL, "string", NULL, PCMK__VALUE_NONE, NULL, N_("Pacemaker version on cluster node elected Designated Controller (DC)"), N_("Includes a hash which identifies the exact changeset the code was " "built from. Used for diagnostic purposes.") }, { "cluster-infrastructure", NULL, "string", NULL, "corosync", NULL, N_("The messaging stack on which Pacemaker is currently running"), N_("Used for informational and diagnostic purposes.") }, { "cluster-name", NULL, "string", NULL, NULL, NULL, N_("An arbitrary name for the cluster"), N_("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", pcmk__valid_interval_spec, N_("How long to wait for a response from other nodes during start-up"), N_("The optimal value will depend on the speed and load of your network " "and the type of switches used.") }, { XML_CONFIG_ATTR_RECHECK, NULL, "time", N_("Zero disables polling, while positive values are an interval in seconds" "(unless other units are specified, for example \"5min\")"), "15min", pcmk__valid_interval_spec, N_("Polling interval to recheck cluster state and evaluate rules " "with date specifications"), N_("Pacemaker is primarily event-driven, and looks ahead to know when to " "recheck cluster state for failure timeouts and most time-based " "rules. However, it will also recheck the cluster after this " "amount of inactivity, to evaluate rules with date specifications " "and serve as a fail-safe for certain types of scheduler bugs.") }, { "load-threshold", NULL, "percentage", NULL, "80%", pcmk__valid_percentage, N_("Maximum amount of system load that should be used by cluster nodes"), N_("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", pcmk__valid_number, N_("Maximum number of jobs that can be scheduled per node " "(defaults to 2x cores)") }, { XML_CONFIG_ATTR_FENCE_REACTION, NULL, "string", NULL, "stop", NULL, N_("How a cluster node should react if notified of its own fencing"), N_("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", pcmk__valid_interval_spec, "*** Advanced Use Only ***", N_("Declare an election failed if it is not decided within this much " "time. If you need to adjust this value, it probably indicates " "the presence of a bug.") }, { XML_CONFIG_ATTR_FORCE_QUIT, NULL, "time", NULL, "20min", pcmk__valid_interval_spec, "*** Advanced Use Only ***", N_("Exit immediately if shutdown does not complete within this much " "time. If you need to adjust this value, it probably indicates " "the presence of a bug.") }, { "join-integration-timeout", "crmd-integration-timeout", "time", NULL, "3min", pcmk__valid_interval_spec, "*** Advanced Use Only ***", N_("If you need to adjust this value, it probably indicates " "the presence of a bug.") }, { "join-finalization-timeout", "crmd-finalization-timeout", "time", NULL, "30min", pcmk__valid_interval_spec, "*** Advanced Use Only ***", N_("If you need to adjust this value, it probably indicates " "the presence of a bug.") }, { "transition-delay", "crmd-transition-delay", "time", NULL, "0s", pcmk__valid_interval_spec, N_("*** Advanced Use Only *** Enabling this option will slow down " "cluster recovery under all conditions"), N_("Delay cluster recovery for this much time to allow for additional " "events to occur. Useful if your configuration is sensitive to " "the order in which ping updates arrive.") }, { "stonith-watchdog-timeout", NULL, "time", NULL, "0", controld_verify_stonith_watchdog_timeout, N_("How long before nodes can be assumed to be safely down when " "watchdog-based self-fencing via SBD is in use"), N_("If this is set to a positive value, lost nodes are assumed to " "self-fence using watchdog-based SBD within this much time. This " "does not require a fencing resource to be explicitly configured, " "though a fence_watchdog resource can be configured, to limit use " "to specific nodes. If this is set to 0 (the default), the cluster " "will never assume watchdog-based self-fencing. If this is set to a " "negative value, the cluster will use twice the local value of the " "`SBD_WATCHDOG_TIMEOUT` environment variable if that is positive, " "or otherwise treat this as 0. WARNING: When used, this timeout " "must be larger than `SBD_WATCHDOG_TIMEOUT` on all nodes that use " "watchdog-based SBD, and Pacemaker will refuse to start on any of " "those nodes where this is not true for the local value or SBD is " "not active. When this is set to a negative value, " "`SBD_WATCHDOG_TIMEOUT` must be set to the same value on all nodes " "that use SBD, otherwise data corruption or loss could occur.") }, { "stonith-max-attempts", NULL, "integer", NULL, "10", pcmk__valid_positive_number, N_("How many times fencing can fail before it will no longer be " "immediately re-attempted on a target") }, // Already documented in libpe_status (other values must be kept identical) { "no-quorum-policy", NULL, "select", "stop, freeze, ignore, demote, suicide", "stop", pcmk__valid_quorum, "What to do when the cluster does not have quorum", NULL }, { XML_CONFIG_ATTR_SHUTDOWN_LOCK, NULL, "boolean", NULL, "false", pcmk__valid_boolean, "Whether to lock resources to a cleanly shut down node", "When true, resources active on a node when it is cleanly shut down " "are kept \"locked\" to that node (not allowed to run elsewhere) " "until they start again on that node after it rejoins (or for at " "most shutdown-lock-limit, if set). Stonith resources and " "Pacemaker Remote connections are never locked. Clone and bundle " "instances and the promoted role of promotable clones are currently" " never locked, though support could be added in a future release." }, }; void crmd_metadata(void) { const char *desc_short = "Pacemaker controller options"; const char *desc_long = "Cluster options used by Pacemaker's controller"; gchar *s = pcmk__format_option_metadata("pacemaker-controld", desc_short, desc_long, controller_options, PCMK__NELEM(controller_options)); printf("%s", s); g_free(s); } 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"); controld_set_fsa_input_flags(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 = pcmk__strkey_table(free, free); pe_unpack_nvpairs(crmconfig, crmconfig, XML_CIB_TAG_PROPSET, NULL, config_hash, CIB_OPTIONS_FIRST, FALSE, now, NULL); // Validate all options, and use defaults if not already present in hash pcmk__validate_cluster_options(config_hash, controller_options, PCMK__NELEM(controller_options)); value = g_hash_table_lookup(config_hash, "no-quorum-policy"); if (pcmk__str_eq(value, "suicide", pcmk__str_casei) && pcmk__locate_sbd()) { controld_set_global_flags(controld_no_quorum_suicide); } value = g_hash_table_lookup(config_hash, XML_CONFIG_ATTR_SHUTDOWN_LOCK); if (crm_is_true(value)) { controld_set_global_flags(controld_shutdown_lock_enabled); } else { controld_clear_global_flags(controld_shutdown_lock_enabled); } value = g_hash_table_lookup(config_hash, "cluster-name"); pcmk__str_update(&(controld_globals.cluster_name), value); // Let subcomponents initialize their own static variables controld_configure_election(config_hash); controld_configure_fencing(config_hash); controld_configure_fsa_timers(config_hash); controld_configure_throttle(config_hash); alerts = first_named_child(output, XML_CIB_TAG_ALERTS); crmd_unpack_alerts(alerts); controld_set_fsa_input_flags(R_READ_CONFIG); controld_trigger_fsa(); g_hash_table_destroy(config_hash); bail: crm_time_free(now); } /*! * \internal * \brief Trigger read and processing of the configuration * * \param[in] fn Calling function name * \param[in] line Line number where call occurred */ void controld_trigger_config_as(const char *fn, int line) { if (config_read_trigger != NULL) { crm_trace("%s:%d - Triggered config processing", fn, line); mainloop_set_trigger(config_read_trigger); } } gboolean crm_read_options(gpointer user_data) { cib_t *cib_conn = controld_globals.cib_conn; int call_id = cib_conn->cmds->query(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(); controld_trigger_config(); } void crm_shutdown(int nsig) { const char *value = NULL; guint default_period_ms = 0; if ((controld_globals.mainloop == NULL) || !g_main_loop_is_running(controld_globals.mainloop)) { crmd_exit(CRM_EX_OK); return; } if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { crm_err("Escalating shutdown"); register_fsa_input_before(C_SHUTDOWN, I_ERROR, NULL); return; } controld_set_fsa_input_flags(R_SHUTDOWN); register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); /* If shutdown timer doesn't have a period set, use the default * * @TODO: Evaluate whether this is still necessary. As long as * config_query_callback() has been run at least once, it doesn't look like * anything could have changed the timer period since then. */ value = pcmk__cluster_option(NULL, controller_options, PCMK__NELEM(controller_options), XML_CONFIG_ATTR_FORCE_QUIT); default_period_ms = crm_parse_interval_spec(value); controld_shutdown_start_countdown(default_period_ms); } diff --git a/daemons/controld/controld_join_dc.c b/daemons/controld/controld_join_dc.c index e7964d24dd..30817ba196 100644 --- a/daemons/controld/controld_join_dc.c +++ b/daemons/controld/controld_join_dc.c @@ -1,837 +1,958 @@ /* * Copyright 2004-2022 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include static char *max_generation_from = NULL; static xmlNodePtr max_generation_xml = NULL; +/*! + * \internal + * \brief Nodes from which a CIB sync has failed since the peer joined + * + * This table is of the form (node_name -> join_id). \p node_name is + * the name of a client node from which a CIB \p sync_from() call has failed in + * \p do_dc_join_finalize() since the client joined the cluster as a peer. + * \p join_id is the ID of the join round in which the \p sync_from() failed, + * and is intended for use in nack log messages. + */ +static GHashTable *failed_sync_nodes = NULL; + void finalize_join_for(gpointer key, gpointer value, gpointer user_data); void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data); gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); /* Numeric counter used to identify join rounds (an unsigned int would be * appropriate, except we get and set it in XML as int) */ static int current_join_id = 0; +/*! + * \internal + * \brief Destroy the hash table containing failed sync nodes + */ +void +controld_destroy_failed_sync_table(void) +{ + if (failed_sync_nodes != NULL) { + g_hash_table_destroy(failed_sync_nodes); + failed_sync_nodes = NULL; + } +} + +/*! + * \internal + * \brief Remove a node from the failed sync nodes table if present + * + * \param[in] node_name Node name to remove + */ +void +controld_remove_failed_sync_node(const char *node_name) +{ + if (failed_sync_nodes != NULL) { + g_hash_table_remove(failed_sync_nodes, (gchar *) node_name); + } +} + +/*! + * \internal + * \brief Add to a hash table a node whose CIB failed to sync + * + * \param[in] node_name Name of node whose CIB failed to sync + * \param[in] join_id Join round when the failure occurred + */ +static void +record_failed_sync_node(const char *node_name, gint join_id) +{ + if (failed_sync_nodes == NULL) { + failed_sync_nodes = pcmk__strikey_table(g_free, NULL); + } + + /* If the node is already in the table then we failed to nack it during the + * filter offer step + */ + CRM_LOG_ASSERT(g_hash_table_insert(failed_sync_nodes, g_strdup(node_name), + GINT_TO_POINTER(join_id))); +} + +/*! + * \internal + * \brief Look up a node name in the failed sync table + * + * \param[in] node_name Name of node to look up + * \param[out] join_id Where to store the join ID of when the sync failed + * + * \return Standard Pacemaker return code. Specifically, \p pcmk_rc_ok if the + * node name was found, or \p pcmk_rc_node_unknown otherwise. + * \note \p *join_id is set to -1 if the node is not found. + */ +static int +lookup_failed_sync_node(const char *node_name, gint *join_id) +{ + *join_id = -1; + + if (failed_sync_nodes != NULL) { + gpointer result = g_hash_table_lookup(failed_sync_nodes, + (gchar *) node_name); + if (result != NULL) { + *join_id = GPOINTER_TO_INT(result); + return pcmk_rc_ok; + } + } + return pcmk_rc_node_unknown; +} + void crm_update_peer_join(const char *source, crm_node_t * node, enum crm_join_phase phase) { enum crm_join_phase last = 0; CRM_CHECK(node != NULL, return); /* Remote nodes do not participate in joins */ if (pcmk_is_set(node->flags, crm_remote_node)) { return; } last = node->join; if(phase == last) { crm_trace("Node %s join-%d phase is still %s " CRM_XS " nodeid=%u source=%s", node->uname, current_join_id, crm_join_phase_str(last), node->id, source); } else if ((phase <= crm_join_none) || (phase == (last + 1))) { node->join = phase; crm_trace("Node %s join-%d phase is now %s (was %s) " CRM_XS " nodeid=%u source=%s", node->uname, current_join_id, crm_join_phase_str(phase), crm_join_phase_str(last), node->id, source); } else { crm_warn("Rejecting join-%d phase update for node %s because " "can't go from %s to %s " CRM_XS " nodeid=%u source=%s", current_join_id, node->uname, crm_join_phase_str(last), crm_join_phase_str(phase), node->id, source); } } static void start_join_round(void) { GHashTableIter iter; crm_node_t *peer = NULL; crm_debug("Starting new join round join-%d", current_join_id); g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { crm_update_peer_join(__func__, peer, crm_join_none); } if (max_generation_from != NULL) { free(max_generation_from); max_generation_from = NULL; } if (max_generation_xml != NULL) { free_xml(max_generation_xml); max_generation_xml = NULL; } controld_clear_fsa_input_flags(R_HAVE_CIB|R_CIB_ASKED); } /*! * \internal * \brief Create a join message from the DC * * \param[in] join_op Join operation name * \param[in] host_to Recipient of message */ static xmlNode * create_dc_message(const char *join_op, const char *host_to) { xmlNode *msg = create_request(join_op, NULL, host_to, CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL); /* Identify which election this is a part of */ crm_xml_add_int(msg, F_CRM_JOIN_ID, current_join_id); /* Add a field specifying whether the DC is shutting down. This keeps the * joining node from fencing the old DC if it becomes the new DC. */ pcmk__xe_set_bool_attr(msg, F_CRM_DC_LEAVING, pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)); return msg; } static void join_make_offer(gpointer key, gpointer value, gpointer user_data) { xmlNode *offer = NULL; crm_node_t *member = (crm_node_t *)value; CRM_ASSERT(member != NULL); if (crm_is_peer_active(member) == FALSE) { crm_info("Not making join-%d offer to inactive node %s", current_join_id, (member->uname? member->uname : "with unknown name")); if(member->expected == NULL && pcmk__str_eq(member->state, CRM_NODE_LOST, pcmk__str_casei)) { /* You would think this unsafe, but in fact this plus an * active resource is what causes it to be fenced. * * Yes, this does mean that any node that dies at the same * time as the old DC and is not running resource (still) * won't be fenced. * * I'm not happy about this either. */ pcmk__update_peer_expected(__func__, member, CRMD_JOINSTATE_DOWN); } return; } if (member->uname == NULL) { crm_info("Not making join-%d offer to node uuid %s with unknown name", current_join_id, member->uuid); return; } if (controld_globals.membership_id != crm_peer_seq) { controld_globals.membership_id = crm_peer_seq; crm_info("Making join-%d offers based on membership event %llu", current_join_id, crm_peer_seq); } if(user_data && member->join > crm_join_none) { crm_info("Not making join-%d offer to already known node %s (%s)", current_join_id, member->uname, crm_join_phase_str(member->join)); return; } crm_update_peer_join(__func__, (crm_node_t*)member, crm_join_none); offer = create_dc_message(CRM_OP_JOIN_OFFER, member->uname); // Advertise our feature set so the joining node can bail if not compatible crm_xml_add(offer, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET); crm_info("Sending join-%d offer to %s", current_join_id, member->uname); send_cluster_message(member, crm_msg_crmd, offer, TRUE); free_xml(offer); crm_update_peer_join(__func__, member, crm_join_welcomed); } /* A_DC_JOIN_OFFER_ALL */ void do_dc_join_offer_all(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) { int count; /* Reset everyone's status back to down or in_ccm in the CIB. * Any nodes that are active in the CIB but not in the cluster membership * will be seen as offline by the scheduler anyway. */ current_join_id++; start_join_round(); // do_update_cib_nodes(TRUE, __func__); update_dc(NULL); if (cause == C_HA_MESSAGE && current_input == I_NODE_JOIN) { crm_info("A new node joined the cluster"); } g_hash_table_foreach(crm_peer_cache, join_make_offer, NULL); count = crmd_join_phase_count(crm_join_welcomed); crm_info("Waiting on join-%d requests from %d outstanding node%s", current_join_id, count, pcmk__plural_s(count)); // Don't waste time by invoking the scheduler yet } /* A_DC_JOIN_OFFER_ONE */ void do_dc_join_offer_one(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_node_t *member; ha_msg_input_t *welcome = NULL; int count; const char *join_to = NULL; if (msg_data->data == NULL) { crm_info("Making join-%d offers to any unconfirmed nodes " "because an unknown node joined", current_join_id); g_hash_table_foreach(crm_peer_cache, join_make_offer, &member); check_join_state(cur_state, __func__); return; } welcome = fsa_typed_data(fsa_dt_ha_msg); if (welcome == NULL) { // fsa_typed_data() already logged an error return; } join_to = crm_element_value(welcome->msg, F_CRM_HOST_FROM); if (join_to == NULL) { crm_err("Can't make join-%d offer to unknown node", current_join_id); return; } member = crm_get_peer(0, join_to); /* It is possible that a node will have been sick or starting up when the * original offer was made. However, it will either re-announce itself in * due course, or we can re-store the original offer on the client. */ crm_update_peer_join(__func__, member, crm_join_none); join_make_offer(NULL, member, NULL); /* If the offer isn't to the local node, make an offer to the local node as * well, to ensure the correct value for max_generation_from. */ if (strcasecmp(join_to, controld_globals.our_nodename) != 0) { member = crm_get_peer(0, controld_globals.our_nodename); join_make_offer(NULL, member, NULL); } /* This was a genuine join request; cancel any existing transition and * invoke the scheduler. */ abort_transition(INFINITY, pcmk__graph_restart, "Node join", NULL); count = crmd_join_phase_count(crm_join_welcomed); crm_info("Waiting on join-%d requests from %d outstanding node%s", current_join_id, count, pcmk__plural_s(count)); // Don't waste time by invoking the scheduler yet } static int compare_int_fields(xmlNode * left, xmlNode * right, const char *field) { const char *elem_l = crm_element_value(left, field); const char *elem_r = crm_element_value(right, field); long long int_elem_l; long long int_elem_r; pcmk__scan_ll(elem_l, &int_elem_l, -1LL); pcmk__scan_ll(elem_r, &int_elem_r, -1LL); if (int_elem_l < int_elem_r) { return -1; } else if (int_elem_l > int_elem_r) { return 1; } return 0; } /* A_DC_JOIN_PROCESS_REQ */ void do_dc_join_filter_offer(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 *generation = NULL; int cmp = 0; int join_id = -1; int count = 0; + gint value = 0; gboolean ack_nack_bool = TRUE; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *join_from = crm_element_value(join_ack->msg, F_CRM_HOST_FROM); const char *ref = crm_element_value(join_ack->msg, F_CRM_REFERENCE); const char *join_version = crm_element_value(join_ack->msg, XML_ATTR_CRM_VERSION); crm_node_t *join_node = NULL; if (join_from == NULL) { crm_err("Ignoring invalid join request without node name"); return; } join_node = crm_get_peer(0, join_from); crm_element_value_int(join_ack->msg, F_CRM_JOIN_ID, &join_id); if (join_id != current_join_id) { crm_debug("Ignoring join-%d request from %s because we are on join-%d", join_id, join_from, current_join_id); check_join_state(cur_state, __func__); return; } generation = join_ack->xml; if (max_generation_xml != NULL && generation != NULL) { int lpc = 0; const char *attributes[] = { XML_ATTR_GENERATION_ADMIN, XML_ATTR_GENERATION, XML_ATTR_NUMUPDATES, }; for (lpc = 0; cmp == 0 && lpc < PCMK__NELEM(attributes); lpc++) { cmp = compare_int_fields(max_generation_xml, generation, attributes[lpc]); } } if (ref == NULL) { ref = "none"; // for logging only } - if (!crm_is_peer_active(join_node)) { + if (lookup_failed_sync_node(join_from, &value) == pcmk_rc_ok) { + crm_err("Rejecting join-%d request from node %s because we failed to " + "sync its CIB in join-%d " CRM_XS " ref=%s", + join_id, join_from, value, ref); + ack_nack_bool = FALSE; + + } else if (!crm_is_peer_active(join_node)) { if (match_down_event(join_from) != NULL) { /* The join request was received after the node was fenced or * otherwise shutdown in a way that we're aware of. No need to log * an error in this rare occurrence; we know the client was recently * shut down, and receiving a lingering in-flight request is not * cause for alarm. */ crm_debug("Rejecting join-%d request from inactive node %s " CRM_XS " ref=%s", join_id, join_from, ref); } else { crm_err("Rejecting join-%d request from inactive node %s " CRM_XS " ref=%s", join_id, join_from, ref); } ack_nack_bool = FALSE; } else if (generation == NULL) { crm_err("Rejecting invalid join-%d request from node %s " "missing CIB generation " CRM_XS " ref=%s", join_id, join_from, ref); ack_nack_bool = FALSE; } else if ((join_version == NULL) || !feature_set_compatible(CRM_FEATURE_SET, join_version)) { crm_err("Rejecting join-%d request from node %s because feature set %s" " is incompatible with ours (%s) " CRM_XS " ref=%s", join_id, join_from, (join_version? join_version : "pre-3.1.0"), CRM_FEATURE_SET, ref); ack_nack_bool = FALSE; } else if (max_generation_xml == NULL) { - crm_debug("Accepting join-%d request from %s " - "(with first CIB generation) " CRM_XS " ref=%s", - join_id, join_from, ref); - max_generation_xml = copy_xml(generation); - max_generation_from = strdup(join_from); + const char *validation = crm_element_value(generation, + XML_ATTR_VALIDATION); + + if (get_schema_version(validation) < 0) { + crm_err("Rejecting join-%d request from %s (with first CIB " + "generation) due to unknown schema version %s " + CRM_XS " ref=%s", + join_id, join_from, validation, ref); + ack_nack_bool = FALSE; + + } else { + crm_debug("Accepting join-%d request from %s (with first CIB " + "generation) " CRM_XS " ref=%s", + join_id, join_from, ref); + max_generation_xml = copy_xml(generation); + pcmk__str_update(&max_generation_from, join_from); + } } else if ((cmp < 0) || ((cmp == 0) && pcmk__str_eq(join_from, controld_globals.our_nodename, pcmk__str_casei))) { - crm_debug("Accepting join-%d request from %s (with better " - "CIB generation than current best from %s) " CRM_XS " ref=%s", - join_id, join_from, max_generation_from, ref); - crm_log_xml_debug(max_generation_xml, "Old max generation"); - crm_log_xml_debug(generation, "New max generation"); + const char *validation = crm_element_value(generation, + XML_ATTR_VALIDATION); - pcmk__str_update(&max_generation_from, join_from); + if (get_schema_version(validation) < 0) { + crm_err("Rejecting join-%d request from %s (with better CIB " + "generation than current best from %s) due to unknown " + "schema version %s " CRM_XS " ref=%s", + join_id, join_from, max_generation_from, validation, ref); + ack_nack_bool = FALSE; - free_xml(max_generation_xml); - max_generation_xml = copy_xml(join_ack->xml); + } else { + crm_debug("Accepting join-%d request from %s (with better CIB " + "generation than current best from %s) " CRM_XS " ref=%s", + join_id, join_from, max_generation_from, ref); + crm_log_xml_debug(max_generation_xml, "Old max generation"); + crm_log_xml_debug(generation, "New max generation"); + + free_xml(max_generation_xml); + max_generation_xml = copy_xml(join_ack->xml); + pcmk__str_update(&max_generation_from, join_from); + } } else { crm_debug("Accepting join-%d request from %s " CRM_XS " ref=%s", join_id, join_from, ref); } if (!ack_nack_bool) { if (compare_version(join_version, "3.17.0") < 0) { /* Clients with CRM_FEATURE_SET < 3.17.0 may respawn infinitely * after a nack message, don't send one */ crm_update_peer_join(__func__, join_node, crm_join_nack_quiet); } else { crm_update_peer_join(__func__, join_node, crm_join_nack); } pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_NACK); } else { crm_update_peer_join(__func__, join_node, crm_join_integrated); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER); } count = crmd_join_phase_count(crm_join_integrated); crm_debug("%d node%s currently integrated in join-%d", count, pcmk__plural_s(count), join_id); if (check_join_state(cur_state, __func__) == FALSE) { // Don't waste time by invoking the scheduler yet count = crmd_join_phase_count(crm_join_welcomed); crm_debug("Waiting on join-%d requests from %d outstanding node%s", join_id, count, pcmk__plural_s(count)); } } /* A_DC_JOIN_FINALIZE */ void do_dc_join_finalize(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) { char *sync_from = NULL; int rc = pcmk_ok; int count_welcomed = crmd_join_phase_count(crm_join_welcomed); int count_finalizable = crmd_join_phase_count(crm_join_integrated) + crmd_join_phase_count(crm_join_nack) + crmd_join_phase_count(crm_join_nack_quiet); /* This we can do straight away and avoid clients timing us out * while we compute the latest CIB */ if (count_welcomed != 0) { crm_debug("Waiting on join-%d requests from %d outstanding node%s " "before finalizing join", current_join_id, count_welcomed, pcmk__plural_s(count_welcomed)); crmd_join_phase_log(LOG_DEBUG); /* crmd_fsa_stall(FALSE); Needed? */ return; } else if (count_finalizable == 0) { crm_debug("Finalization not needed for join-%d at the current time", current_join_id); crmd_join_phase_log(LOG_DEBUG); check_join_state(controld_globals.fsa_state, __func__); return; } controld_clear_fsa_input_flags(R_HAVE_CIB); if (pcmk__str_eq(max_generation_from, controld_globals.our_nodename, pcmk__str_null_matches|pcmk__str_casei)) { controld_set_fsa_input_flags(R_HAVE_CIB); } if (pcmk_is_set(controld_globals.fsa_input_register, R_IN_TRANSITION)) { crm_warn("Delaying join-%d finalization while transition in progress", current_join_id); crmd_join_phase_log(LOG_DEBUG); crmd_fsa_stall(FALSE); return; } if ((max_generation_from != NULL) && !pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) { /* ask for the agreed best CIB */ pcmk__str_update(&sync_from, max_generation_from); controld_set_fsa_input_flags(R_CIB_ASKED); crm_notice("Finalizing join-%d for %d node%s (sync'ing CIB from %s)", current_join_id, count_finalizable, pcmk__plural_s(count_finalizable), sync_from); crm_log_xml_notice(max_generation_xml, "Requested CIB version"); } else { /* Send _our_ CIB out to everyone */ pcmk__str_update(&sync_from, controld_globals.our_nodename); crm_debug("Finalizing join-%d for %d node%s (sync'ing from local CIB)", current_join_id, count_finalizable, pcmk__plural_s(count_finalizable)); crm_log_xml_debug(max_generation_xml, "Requested CIB version"); } crmd_join_phase_log(LOG_DEBUG); rc = controld_globals.cib_conn->cmds->sync_from(controld_globals.cib_conn, sync_from, NULL, cib_quorum_override); fsa_register_cib_callback(rc, FALSE, sync_from, finalize_sync_callback); } void free_max_generation(void) { free(max_generation_from); max_generation_from = NULL; free_xml(max_generation_xml); max_generation_xml = NULL; } void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { CRM_LOG_ASSERT(-EPERM != rc); controld_clear_fsa_input_flags(R_CIB_ASKED); if (rc != pcmk_ok) { + const char *sync_from = (const char *) user_data; + do_crm_log(((rc == -pcmk_err_old_data)? LOG_WARNING : LOG_ERR), "Could not sync CIB from %s in join-%d: %s", - (char *) user_data, current_join_id, pcmk_strerror(rc)); + sync_from, current_join_id, pcmk_strerror(rc)); + + record_failed_sync_node(sync_from, current_join_id); /* restart the whole join process */ register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION_DC, NULL, NULL, __func__); } else if (!AM_I_DC) { crm_debug("Sync'ed CIB for join-%d but no longer DC", current_join_id); } else if (controld_globals.fsa_state != S_FINALIZE_JOIN) { crm_debug("Sync'ed CIB for join-%d but no longer in S_FINALIZE_JOIN " "(%s)", current_join_id, fsa_state2string(controld_globals.fsa_state)); } else { controld_set_fsa_input_flags(R_HAVE_CIB); controld_clear_fsa_input_flags(R_CIB_ASKED); /* make sure dc_uuid is re-set to us */ if (!check_join_state(controld_globals.fsa_state, __func__)) { int count_finalizable = 0; count_finalizable = crmd_join_phase_count(crm_join_integrated) + crmd_join_phase_count(crm_join_nack) + crmd_join_phase_count(crm_join_nack_quiet); crm_debug("Notifying %d node%s of join-%d results", count_finalizable, pcmk__plural_s(count_finalizable), current_join_id); g_hash_table_foreach(crm_peer_cache, finalize_join_for, NULL); } } } static void join_update_complete_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { fsa_data_t *msg_data = NULL; if (rc == pcmk_ok) { crm_debug("join-%d node history update (via CIB call %d) complete", current_join_id, call_id); check_join_state(controld_globals.fsa_state, __func__); } else { crm_err("join-%d node history update (via CIB call %d) failed: %s " "(next transition may determine resource status incorrectly)", current_join_id, call_id, pcmk_strerror(rc)); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } /* A_DC_JOIN_PROCESS_ACK */ void do_dc_join_ack(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) { int join_id = -1; int call_id = 0; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); enum controld_section_e section = controld_section_lrm; const char *op = crm_element_value(join_ack->msg, F_CRM_TASK); const char *join_from = crm_element_value(join_ack->msg, F_CRM_HOST_FROM); crm_node_t *peer = NULL; // Sanity checks if (join_from == NULL) { crm_warn("Ignoring message received without node identification"); return; } if (op == NULL) { crm_warn("Ignoring message received from %s without task", join_from); return; } if (strcmp(op, CRM_OP_JOIN_CONFIRM)) { crm_debug("Ignoring '%s' message from %s while waiting for '%s'", op, join_from, CRM_OP_JOIN_CONFIRM); return; } if (crm_element_value_int(join_ack->msg, F_CRM_JOIN_ID, &join_id) != 0) { crm_warn("Ignoring join confirmation from %s without valid join ID", join_from); return; } peer = crm_get_peer(0, join_from); if (peer->join != crm_join_finalized) { crm_info("Ignoring out-of-sequence join-%d confirmation from %s " "(currently %s not %s)", join_id, join_from, crm_join_phase_str(peer->join), crm_join_phase_str(crm_join_finalized)); return; } if (join_id != current_join_id) { crm_err("Rejecting join-%d confirmation from %s " "because currently on join-%d", join_id, join_from, current_join_id); crm_update_peer_join(__func__, peer, crm_join_nack); return; } crm_update_peer_join(__func__, peer, crm_join_confirmed); /* Update CIB with node's current executor state. A new transition will be * triggered later, when the CIB notifies us of the change. */ if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { section = controld_section_lrm_unlocked; } controld_delete_node_state(join_from, section, cib_scope_local); if (pcmk__str_eq(join_from, controld_globals.our_nodename, pcmk__str_casei)) { xmlNode *now_dc_lrmd_state = controld_query_executor_state(); if (now_dc_lrmd_state != NULL) { fsa_cib_update(XML_CIB_TAG_STATUS, now_dc_lrmd_state, cib_scope_local | cib_quorum_override | cib_can_create, call_id, NULL); free_xml(now_dc_lrmd_state); crm_debug("Updating local node history for join-%d " "from query result (via CIB call %d)", join_id, call_id); } else { fsa_cib_update(XML_CIB_TAG_STATUS, join_ack->xml, cib_scope_local | cib_quorum_override | cib_can_create, call_id, NULL); crm_warn("Updating local node history from join-%d confirmation " "because query failed (via CIB call %d)", join_id, call_id); } } else { fsa_cib_update(XML_CIB_TAG_STATUS, join_ack->xml, cib_scope_local | cib_quorum_override | cib_can_create, call_id, NULL); crm_debug("Updating node history for %s from join-%d confirmation " "(via CIB call %d)", join_from, join_id, call_id); } fsa_register_cib_callback(call_id, FALSE, NULL, join_update_complete_callback); } void finalize_join_for(gpointer key, gpointer value, gpointer user_data) { xmlNode *acknak = NULL; xmlNode *tmp1 = NULL; crm_node_t *join_node = value; const char *join_to = join_node->uname; bool integrated = false; switch (join_node->join) { case crm_join_integrated: integrated = true; break; case crm_join_nack: case crm_join_nack_quiet: break; default: crm_trace("Not updating non-integrated and non-nacked node %s (%s) " "for join-%d", join_to, crm_join_phase_str(join_node->join), current_join_id); return; } /* Update the element with the node's name and UUID, in case they * weren't known before */ crm_trace("Updating node name and UUID in CIB for %s", join_to); tmp1 = create_xml_node(NULL, XML_CIB_TAG_NODE); set_uuid(tmp1, XML_ATTR_UUID, join_node); crm_xml_add(tmp1, XML_ATTR_UNAME, join_to); fsa_cib_anon_update(XML_CIB_TAG_NODES, tmp1); free_xml(tmp1); if (join_node->join == crm_join_nack_quiet) { crm_trace("Not sending nack message to node %s with feature set older " "than 3.17.0", join_to); return; } join_node = crm_get_peer(0, join_to); if (!crm_is_peer_active(join_node)) { /* * NACK'ing nodes that the membership layer doesn't know about yet * simply creates more churn * * Better to leave them waiting and let the join restart when * the new membership event comes in * * All other NACKs (due to versions etc) should still be processed */ pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_PENDING); return; } // Acknowledge or nack node's join request crm_debug("%sing join-%d request from %s", integrated? "Acknowledg" : "Nack", current_join_id, join_to); acknak = create_dc_message(CRM_OP_JOIN_ACKNAK, join_to); pcmk__xe_set_bool_attr(acknak, CRM_OP_JOIN_ACKNAK, integrated); if (integrated) { // No change needed for a nacked node crm_update_peer_join(__func__, join_node, crm_join_finalized); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER); } send_cluster_message(join_node, crm_msg_crmd, acknak, TRUE); free_xml(acknak); return; } gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source) { static unsigned long long highest_seq = 0; if (controld_globals.membership_id != crm_peer_seq) { crm_debug("join-%d: Membership changed from %llu to %llu " CRM_XS " highest=%llu state=%s for=%s", current_join_id, controld_globals.membership_id, crm_peer_seq, highest_seq, fsa_state2string(cur_state), source); if(highest_seq < crm_peer_seq) { /* Don't spam the FSA with duplicates */ highest_seq = crm_peer_seq; register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } } else if (cur_state == S_INTEGRATION) { if (crmd_join_phase_count(crm_join_welcomed) == 0) { int count = crmd_join_phase_count(crm_join_integrated); crm_debug("join-%d: Integration of %d peer%s complete " CRM_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); register_fsa_input_before(C_FSA_INTERNAL, I_INTEGRATED, NULL); return TRUE; } } else if (cur_state == S_FINALIZE_JOIN) { if (!pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) { crm_debug("join-%d: Delaying finalization until we have CIB " CRM_XS " state=%s for=%s", current_join_id, fsa_state2string(cur_state), source); return TRUE; } else if (crmd_join_phase_count(crm_join_welcomed) != 0) { int count = crmd_join_phase_count(crm_join_welcomed); crm_debug("join-%d: Still waiting on %d welcomed node%s " CRM_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else if (crmd_join_phase_count(crm_join_integrated) != 0) { int count = crmd_join_phase_count(crm_join_integrated); crm_debug("join-%d: Still waiting on %d integrated node%s " CRM_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else if (crmd_join_phase_count(crm_join_finalized) != 0) { int count = crmd_join_phase_count(crm_join_finalized); crm_debug("join-%d: Still waiting on %d finalized node%s " CRM_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else { crm_debug("join-%d: Complete " CRM_XS " state=%s for=%s", current_join_id, fsa_state2string(cur_state), source); register_fsa_input_later(C_FSA_INTERNAL, I_FINALIZED, NULL); return TRUE; } } return FALSE; } void do_dc_join_final(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("Ensuring DC, quorum and node attributes are up-to-date"); crm_update_quorum(crm_have_quorum, TRUE); } int crmd_join_phase_count(enum crm_join_phase phase) { int count = 0; crm_node_t *peer; GHashTableIter iter; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { if(peer->join == phase) { count++; } } return count; } void crmd_join_phase_log(int level) { crm_node_t *peer; GHashTableIter iter; g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { do_crm_log(level, "join-%d: %s=%s", current_join_id, peer->uname, crm_join_phase_str(peer->join)); } } diff --git a/daemons/controld/controld_membership.h b/daemons/controld/controld_membership.h index 789c330762..cfe8cee951 100644 --- a/daemons/controld/controld_membership.h +++ b/daemons/controld/controld_membership.h @@ -1,26 +1,29 @@ /* * Copyright 2012-2021 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 MEMBERSHIP__H # define MEMBERSHIP__H #ifdef __cplusplus extern "C" { #endif #include void post_cache_update(int instance); extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); +void controld_destroy_failed_sync_table(void); +void controld_remove_failed_sync_node(const char *node_name); + #ifdef __cplusplus } #endif #endif