static void reap_dead_nodes(gpointer key, gpointer value, gpointer user_data) { crm_node_t *node = value; if (crm_is_peer_active(node) == FALSE) { crm_update_peer_join(__FUNCTION__, node, crm_join_none); if(node && node->uname) { election_remove(fsa_election, node->uname); if (safe_str_eq(fsa_our_uname, node->uname)) { crm_err("We're not part of the cluster anymore"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); } else if (AM_I_DC == FALSE && safe_str_eq(node->uname, fsa_our_dc)) { crm_warn("Our DC node (%s) left the cluster", node->uname); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } } if (fsa_state == S_INTEGRATION || fsa_state == S_FINALIZE_JOIN) { check_join_state(fsa_state, __FUNCTION__); } fail_incompletable_actions(transition_graph, node->uuid); } }
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)); CRM_CHECK(graph->complete, graph->complete = TRUE); switch (graph->completion_action) { case tg_stop: type = "stop"; /* fall through */ case tg_done: type = "done"; if (fsa_state == S_TRANSITION_ENGINE) { event = I_TE_SUCCESS; } break; case tg_restart: type = "restart"; if (fsa_state == S_TRANSITION_ENGINE) { if (transition_timer->period_ms > 0) { crm_timer_stop(transition_timer); crm_timer_start(transition_timer); } else if (too_many_st_failures() == FALSE) { 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 our" " PE is telling us too."); 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); } }
/* A_DC_RELEASE */ void do_dc_release(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) { if (action & A_DC_RELEASE) { crm_debug("Releasing the role of DC"); clear_bit(fsa_input_register, R_THE_DC); } else if (action & A_DC_RELEASED) { crm_info("DC role released"); #if 0 if (are there errors) { /* we cant stay up if not healthy */ /* or perhaps I_ERROR and go to S_RECOVER? */ result = I_SHUTDOWN; } #endif register_fsa_input(C_FSA_INTERNAL, I_RELEASE_SUCCESS, NULL); } else { crm_err("Unknown action %s", fsa_action2string(action)); } crm_trace("Am I still the DC? %s", AM_I_DC ? XML_BOOLEAN_YES : XML_BOOLEAN_NO); }
/* A_ELECTION_CHECKアクション CRM_OP_VOTEメッセージを受信してから流れる処理 voteハッシュテーブルに全てのアクティブなクラスタ構成メンバーからの受信をおこなったかどうかチェックし 全ての受信が行われている場合は、I_ELECTION_DCへ遷移する */ */ void do_election_check(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 voted_size = 0; /* 現在のアクティブなメンバー数を取得する */ int num_members = crm_active_members(); if(voted) { /* votedハッシュテーブルが存在する場合は、テーブルサイズを取得する */ voted_size = g_hash_table_size(voted); } /* in the case of #voted > #members, it is better to * wait for the timeout and give the cluster time to * stabilize */ if(fsa_state != S_ELECTION) { /* S_ELECTION状態の場合はチェックしない */ crm_debug("Ignore election check: we not in an election"); } else if(voted_size >= num_members) { /* voteハッシュテーブルのサイズ(メンバー数)が現在のアクティブなメンバー数以上の場合 */ /* we won and everyone has voted */ /* メンバーが揃ったのでelection_timeoutタイマーを止める */ crm_timer_stop(election_timeout); /* アクティブメンバー数とvoteサイズが一致もしくは、voteサイズが大きい状態になったので、 */ /* メンバーは揃ったので、内部メッセージにI_ELECTION_DCをセットする */ register_fsa_input(C_FSA_INTERNAL, I_ELECTION_DC, NULL); if(voted_size > num_members) { char *data = NULL; data = crm_strdup("member"); g_hash_table_foreach(crm_peer_cache, log_member_uname, data); crm_free(data); data = crm_strdup("voted"); g_hash_table_foreach(voted, log_node, data); crm_free(data); } crm_debug("Destroying voted hash"); /* votedハッシュテーブルを破棄する */ g_hash_table_destroy(voted); voted = NULL; } else { crm_debug("Still waiting on %d non-votes (%d total)", num_members - voted_size, num_members); } return; }
gboolean crm_timer_popped(gpointer data) { fsa_timer_t *timer = (fsa_timer_t *) data; if (timer == wait_timer || timer == recheck_timer || timer == transition_timer || timer == finalization_timer || timer == election_trigger) { crm_info("%s (%s) just popped (%dms)", get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms); timer->counter++; } else { crm_err("%s (%s) just popped in state %s! (%dms)", get_timer_desc(timer), fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state), timer->period_ms); } 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; } if (timer->repeat == FALSE) { crm_timer_stop(timer); /* make it _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 ourself, 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; }
static void do_cib_replaced(const char *event, HA_Message *msg) { crm_debug("Updating the CIB after a replace"); populate_cib_nodes(fsa_cluster_conn, FALSE); do_update_cib_nodes(AM_I_DC, __FUNCTION__); if(AM_I_DC) { /* start the join process again so we get everyone's LRM status */ register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } }
void do_election_check(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 voted_size = 0; int num_members = crm_active_peers(); if (voted) { voted_size = g_hash_table_size(voted); } /* in the case of #voted > #members, it is better to * wait for the timeout and give the cluster time to * stabilize */ if (fsa_state != S_ELECTION) { crm_debug("Ignore election check: we not in an election"); } else if (voted_size >= num_members) { /* we won and everyone has voted */ crm_timer_stop(election_timeout); register_fsa_input(C_FSA_INTERNAL, I_ELECTION_DC, NULL); if (voted_size > num_members) { GHashTableIter gIter; const crm_node_t *node; char *key = NULL; g_hash_table_iter_init(&gIter, crm_peer_cache); while (g_hash_table_iter_next(&gIter, NULL, (gpointer *) & node)) { if (crm_is_peer_active(node)) { crm_err("member: %s proc=%.32x", node->uname, node->processes); } } g_hash_table_iter_init(&gIter, voted); while (g_hash_table_iter_next(&gIter, (gpointer *) & key, NULL)) { crm_err("voted: %s", key); } } crm_debug("Destroying voted hash"); g_hash_table_destroy(voted); voted = NULL; } else { crm_debug("Still waiting on %d non-votes (%d total)", num_members - voted_size, num_members); } return; }
void crmd_ha_connection_destroy(gpointer user_data) { crm_trace("Invoked"); if (is_set(fsa_input_register, R_HA_DISCONNECTED)) { /* we signed out, so this is expected */ crm_info("Heartbeat disconnection complete"); return; } crm_crit("Lost connection to heartbeat service!"); register_fsa_input(C_HA_DISCONNECT, I_ERROR, NULL); trigger_fsa(fsa_source); }
gboolean crm_timer_popped(gpointer data) { fsa_timer_t *timer = (fsa_timer_t *) data; if (timer == wait_timer || timer == recheck_timer || timer == transition_timer || timer == finalization_timer || timer == election_trigger) { crm_info("%s (%s) just popped (%dms)", get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms); } else { crm_err("%s (%s) just popped in state %s! (%dms)", get_timer_desc(timer), fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state), timer->period_ms); } if (timer->repeat == FALSE) { crm_timer_stop(timer); /* make it _not_ go off again */ } if (timer->fsa_input == I_INTEGRATED) { crm_info("Welcomed: %d, Integrated: %d", g_hash_table_size(welcomed_nodes), g_hash_table_size(integrated_nodes)); if (g_hash_table_size(welcomed_nodes) == 0) { /* If we don't even have ourself, 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; }
int crmd_init(void) { int exit_code = 0; enum crmd_fsa_state state; fsa_state = S_STARTING; fsa_input_register = 0; /* zero out the regester */ init_dotfile(); crm_debug("Starting %s", crm_system_name); register_fsa_input(C_STARTUP, I_STARTUP, NULL); crm_peer_init(); state = s_crmd_fsa(C_STARTUP); if (state == S_PENDING || state == S_STARTING) { /* Create the mainloop and run it... */ crmd_mainloop = g_main_new(FALSE); crm_trace("Starting %s's mainloop", crm_system_name); #ifdef REALTIME_SUPPORT static int crm_realtime = 1; if (crm_realtime == 1) { cl_enable_realtime(); } else if (crm_realtime == 0) { cl_disable_realtime(); } cl_make_realtime(SCHED_RR, 5, 64, 64); #endif g_main_run(crmd_mainloop); if (is_set(fsa_input_register, R_STAYDOWN)) { crm_info("Inhibiting respawn by Heartbeat"); exit_code = 100; } } else { crm_err("Startup of %s failed. Current state: %s", crm_system_name, fsa_state2string(state)); exit_code = 1; } crm_info("[%s] stopped (%d)", crm_system_name, exit_code); qb_log_fini(); return exit_code; }
void do_election_check(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 voted_size = 0; int num_members = crm_active_peers(); if (voted) { voted_size = g_hash_table_size(voted); } /* in the case of #voted > #members, it is better to * wait for the timeout and give the cluster time to * stabilize */ if (fsa_state != S_ELECTION) { crm_debug("Ignore election check: we not in an election"); } else if (voted_size >= num_members) { /* we won and everyone has voted */ crm_timer_stop(election_timeout); register_fsa_input(C_FSA_INTERNAL, I_ELECTION_DC, NULL); if (voted_size > num_members) { char *data = NULL; data = strdup("member"); g_hash_table_foreach(crm_peer_cache, log_member_uname, data); free(data); data = strdup("voted"); g_hash_table_foreach(voted, log_node, data); free(data); } crm_debug("Destroying voted hash"); g_hash_table_destroy(voted); voted = NULL; } else { crm_debug("Still waiting on %d non-votes (%d total)", num_members - voted_size, num_members); } return; }
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(FALSE); do_update_cib_nodes(TRUE, __FUNCTION__); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, 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); } else if (action & A_FINALIZE_TIMER_STOP) { timer_op_ok = crm_timer_stop(finalization_timer); } else if (action & A_INTEGRATE_TIMER_STOP) { timer_op_ok = crm_timer_stop(integration_timer); /* } else if(action & A_ELECTION_TIMEOUT_STOP) { */ /* timer_op_ok = crm_timer_stop(election_timeout); */ } /* dont start a timer that wasnt already running */ if (action & A_DC_TIMER_START && timer_op_ok) { crm_timer_start(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); } else if (action & A_INTEGRATE_TIMER_START) { crm_timer_start(integration_timer); /* } else if(action & A_ELECTION_TIMEOUT_START) { */ /* crm_timer_start(election_timeout); */ } }
/* A_DC_RELEASE */ void do_dc_release(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) { if (action & A_DC_RELEASE) { GListPtr gIter = NULL; crm_debug("Releasing the role of DC"); clear_bit(fsa_input_register, R_THE_DC); for (gIter = stonith_cleanup_list; gIter != NULL; gIter = gIter->next) { char *target = gIter->data; crm_debug("Purging %s from stonith cleanup list", target); free(target); } g_list_free(stonith_cleanup_list); stonith_cleanup_list = NULL; } else if (action & A_DC_RELEASED) { crm_info("DC role released"); #if 0 if (are there errors) { /* we cant stay up if not healthy */ /* or perhaps I_ERROR and go to S_RECOVER? */ result = I_SHUTDOWN; } #endif register_fsa_input(C_FSA_INTERNAL, I_RELEASE_SUCCESS, NULL); } else { crm_err("Unknown action %s", fsa_action2string(action)); } crm_trace("Am I still the DC? %s", AM_I_DC ? XML_BOOLEAN_YES : XML_BOOLEAN_NO); }
void route_message(enum crmd_fsa_cause cause, xmlNode * input) { ha_msg_input_t fsa_input; enum crmd_fsa_input result = I_NULL; fsa_input.msg = input; CRM_CHECK(cause == C_IPC_MESSAGE || cause == C_HA_MESSAGE, return); /* try passing the buck first */ if (relay_message(input, cause == C_IPC_MESSAGE)) { return; } /* handle locally */ result = handle_message(input); /* done or process later? */ switch (result) { case I_NULL: case I_CIB_OP: case I_ROUTER: case I_NODE_JOIN: case I_JOIN_REQUEST: case I_JOIN_RESULT: break; default: /* Defering local processing of message */ register_fsa_input_later(cause, result, &fsa_input); return; } if (result != I_NULL) { /* add to the front of the queue */ register_fsa_input(cause, result, &fsa_input); } }
void crmd_init(void) { crm_exit_t exit_code = CRM_EX_OK; enum crmd_fsa_state state; log_deprecation_warnings(); fsa_state = S_STARTING; fsa_input_register = 0; /* zero out the regester */ init_dotfile(); register_fsa_input(C_STARTUP, I_STARTUP, NULL); crm_peer_init(); state = s_crmd_fsa(C_STARTUP); if (state == S_PENDING || state == S_STARTING) { /* Create the mainloop and run it... */ crm_trace("Starting %s's mainloop", crm_system_name); g_main_loop_run(crmd_mainloop); if (is_set(fsa_input_register, R_STAYDOWN)) { crm_info("Inhibiting automated respawn"); exit_code = CRM_EX_FATAL; } } else { crm_err("Startup of %s failed. Current state: %s", crm_system_name, fsa_state2string(state)); exit_code = CRM_EX_ERROR; } crm_info("%s[%lu] exiting with status %d (%s)", crm_system_name, (unsigned long) getpid(), exit_code, crm_exit_str(exit_code)); crmd_fast_exit(exit_code); }
void abort_transition_graph(int abort_priority, enum transition_action abort_action, const char *abort_text, xmlNode * reason, const char *fn, int line) { int log_level = LOG_INFO; const char *magic = NULL; CRM_CHECK(transition_graph != NULL, return); if (reason) { int diff_add_updates = 0; int diff_add_epoch = 0; int diff_add_admin_epoch = 0; int diff_del_updates = 0; int diff_del_epoch = 0; int diff_del_admin_epoch = 0; xmlNode *diff = get_xpath_object("//" F_CIB_UPDATE_RESULT "//diff", reason, LOG_DEBUG_2); magic = crm_element_value(reason, XML_ATTR_TRANSITION_MAGIC); if (diff) { cib_diff_version_details(diff, &diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates, &diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates); if (crm_str_eq(TYPE(reason), XML_CIB_TAG_NVPAIR, TRUE)) { do_crm_log(log_level, "%s:%d - Triggered transition abort (complete=%d, tag=%s, id=%s, name=%s, value=%s, magic=%s, cib=%d.%d.%d) : %s", fn, line, transition_graph->complete, TYPE(reason), ID(reason), NAME(reason), VALUE(reason), magic ? magic : "NA", diff_add_admin_epoch, diff_add_epoch, diff_add_updates, abort_text); } else { do_crm_log(log_level, "%s:%d - Triggered transition abort (complete=%d, tag=%s, id=%s, magic=%s, cib=%d.%d.%d) : %s", fn, line, transition_graph->complete, TYPE(reason), ID(reason), magic ? magic : "NA", diff_add_admin_epoch, diff_add_epoch, diff_add_updates, abort_text); } } else { do_crm_log(log_level, "%s:%d - Triggered transition abort (complete=%d, tag=%s, id=%s, magic=%s) : %s", fn, line, transition_graph->complete, TYPE(reason), ID(reason), magic ? magic : "NA", abort_text); } } else { do_crm_log(log_level, "%s:%d - Triggered transition abort (complete=%d) : %s", fn, line, transition_graph->complete, abort_text); } 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: do_crm_log(log_level, "Abort suppressed: state=%s (complete=%d)", fsa_state2string(fsa_state), transition_graph->complete); return; default: break; } if (magic == NULL && reason != NULL) { crm_log_xml(log_level + 1, "Cause", reason); } /* Make sure any queued calculations are discarded ASAP */ crm_free(fsa_pe_ref); fsa_pe_ref = NULL; if (transition_graph->complete) { if (transition_timer->period_ms > 0) { crm_timer_start(transition_timer); } else { register_fsa_input(C_FSA_INTERNAL, I_PE_CALC, NULL); } return; } update_abort_priority(transition_graph, abort_priority, abort_action, abort_text); mainloop_set_trigger(transition_trigger); }
/* A_CIB_INVOKE, A_CIB_BUMPGEN, A_UPDATE_NODESTATUS */ void do_cib_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) { HA_Message *answer = NULL; ha_msg_input_t *cib_msg = fsa_typed_data(fsa_dt_ha_msg); const char *sys_from = cl_get_string(cib_msg->msg, F_CRM_SYS_FROM); if(fsa_cib_conn->state == cib_disconnected) { if(cur_state != S_STOPPING) { crm_err("CIB is disconnected"); crm_log_message_adv(LOG_WARNING, "CIB Input", cib_msg->msg); return; } crm_info("CIB is disconnected"); crm_log_message_adv(LOG_DEBUG, "CIB Input", cib_msg->msg); return; } if(action & A_CIB_INVOKE) { if(safe_str_eq(sys_from, CRM_SYSTEM_CRMD)) { action = A_CIB_INVOKE_LOCAL; } else if(safe_str_eq(sys_from, CRM_SYSTEM_DC)) { action = A_CIB_INVOKE_LOCAL; } } if(action & A_CIB_INVOKE || action & A_CIB_INVOKE_LOCAL) { int call_options = 0; enum cib_errors rc = cib_ok; crm_data_t *cib_frag = NULL; const char *section = NULL; const char *op = cl_get_string(cib_msg->msg, F_CRM_TASK); section = cl_get_string(cib_msg->msg, F_CIB_SECTION); ha_msg_value_int(cib_msg->msg, F_CIB_CALLOPTS, &call_options); crm_log_message(LOG_MSG, cib_msg->msg); crm_log_xml_debug_3(cib_msg->xml, "[CIB update]"); if(op == NULL) { crm_err("Invalid CIB Message"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } cib_frag = NULL; rc = fsa_cib_conn->cmds->variant_op( fsa_cib_conn, op, NULL, section, cib_msg->xml, &cib_frag, call_options); if(rc < cib_ok || (action & A_CIB_INVOKE)) { answer = create_reply(cib_msg->msg, cib_frag); ha_msg_add(answer,XML_ATTR_RESULT,cib_error2string(rc)); } if(action & A_CIB_INVOKE) { if(relay_message(answer, TRUE) == FALSE) { crm_err("Confused what to do with cib result"); crm_log_message(LOG_ERR, answer); crm_msg_del(answer); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); return; } } else if(rc < cib_ok) { ha_msg_input_t *input = NULL; crm_err("Internal CRM/CIB command from %s() failed: %s", msg_data->origin, cib_error2string(rc)); crm_log_message_adv(LOG_WARNING, "CIB Input", cib_msg->msg); crm_log_message_adv(LOG_WARNING, "CIB Reply", answer); input = new_ha_msg_input(answer); register_fsa_input(C_FSA_INTERNAL, I_ERROR, input); crm_msg_del(answer); delete_ha_msg_input(input); } } else { crm_err("Unexpected action %s in %s", fsa_action2string(action), __FUNCTION__); } }
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; } /* Make sure any queued calculations are discarded ASAP */ free(fsa_pe_ref); fsa_pe_ref = 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 aborted: %s (source=%s:%d, %d)", abort_text, fn, line, transition_graph->complete); } else if(change == NULL) { char *local_path = xml_get_path(reason); do_crm_log(level, "Transition aborted by %s.%s: %s (cib=%d.%d.%d, source=%s:%d, path=%s, %d)", TYPE(reason), ID(reason), abort_text, add[0], add[1], add[2], fn, line, local_path, transition_graph->complete); 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 aborted by deletion of %s: %s (cib=%d.%d.%d, source=%s:%d, path=%s, %d)", shortpath?shortpath+1:path, abort_text, add[0], add[1], add[2], fn, line, path, transition_graph->complete); } else if (safe_str_eq(XML_CIB_TAG_NVPAIR, kind)) { do_crm_log(level, "Transition aborted by %s, %s=%s: %s (%s cib=%d.%d.%d, source=%s:%d, path=%s, %d)", crm_element_value(reason, XML_ATTR_ID), crm_element_value(reason, XML_NVPAIR_ATTR_NAME), crm_element_value(reason, XML_NVPAIR_ATTR_VALUE), abort_text, op, add[0], add[1], add[2], fn, line, path, transition_graph->complete); } 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 aborted by %s '%s' on %s: %s (magic=%s, cib=%d.%d.%d, source=%s:%d, %d)", 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); } 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 aborted by %s '%s' on %s: %s (cib=%d.%d.%d, source=%s:%d, %d)", kind, op, uname ? uname : ID(reason), abort_text, add[0], add[1], add[2], fn, line, transition_graph->complete); } else { do_crm_log(level, "Transition aborted by %s.%s '%s': %s (cib=%d.%d.%d, source=%s:%d, path=%s, %d)", TYPE(reason), ID(reason), op?op:"change", abort_text, add[0], add[1], add[2], fn, line, path, transition_graph->complete); } } if (transition_graph->complete) { if (transition_timer->period_ms > 0) { crm_timer_stop(transition_timer); crm_timer_start(transition_timer); } else { register_fsa_input(C_FSA_INTERNAL, I_PE_CALC, NULL); } return; } mainloop_set_trigger(transition_trigger); }
/* crmd起動初期処理 C_で表現さえる原因は以下が用意されている enum crmd_fsa_cause { C_UNKNOWN = 0, C_STARTUP, C_IPC_MESSAGE, C_HA_MESSAGE, C_CCM_CALLBACK, C_CRMD_STATUS_CALLBACK, C_LRM_OP_CALLBACK, C_LRM_MONITOR_CALLBACK, C_TIMER_POPPED, C_SHUTDOWN, C_HEARTBEAT_FAILED, C_SUBSYSTEM_CONNECT, C_HA_DISCONNECT, C_FSA_INTERNAL, C_ILLEGAL }; */ int crmd_init(void) { int exit_code = 0; enum crmd_fsa_state state; /* 最初のfsa_stateをS_STARTINGにセット */ fsa_state = S_STARTING; fsa_input_register = 0; /* zero out the regester */ init_dotfile(); crm_info("Starting %s", crm_system_name); /* Start時の自処理入力データを作成する */ /* fsa_data->id = last_data_id; fsa_data->fsa_input = I_STARTUP; fsa_data->fsa_cause = C_STARTUP; fsa_data->origin = raised_from; fsa_data->data = NULL; fsa_data->data_type = fsa_dt_none; fsa_data->actions = with_actions; #define register_fsa_input(cause, input, data) register_fsa_input_adv(cause, input, data, A_NOTHING, FALSE, __FUNCTION__) */ register_fsa_input(C_STARTUP, I_STARTUP, NULL); crm_peer_init(); /* 最初のS_STARTING/C_STARTUP/I_STARTUP状態を処理する */ state = s_crmd_fsa(C_STARTUP); if (state == S_PENDING || state == S_STARTING) { /* Create the mainloop and run it... */ crmd_mainloop = g_main_new(FALSE); crm_info("Starting %s's mainloop", crm_system_name); #ifdef REALTIME_SUPPORT static int crm_realtime = 1; if (crm_realtime == 1){ cl_enable_realtime(); }else if (crm_realtime == 0){ cl_disable_realtime(); } cl_make_realtime(SCHED_RR, 5, 64, 64); #endif /* メインループ開始 */ g_main_run(crmd_mainloop); if(is_set(fsa_input_register, R_STAYDOWN)) { crm_info("Inhibiting respawn by Heartbeat"); exit_code = 100; } } else { crm_err("Startup of %s failed. Current state: %s", crm_system_name, fsa_state2string(state)); exit_code = 1; } crm_info("[%s] stopped (%d)", crm_system_name, exit_code); return exit_code; }
void do_election_count_vote(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 election_id = -1; int log_level = LOG_INFO; gboolean done = FALSE; gboolean we_loose = FALSE; const char *op = NULL; const char *vote_from = NULL; const char *your_version = NULL; const char *election_owner = NULL; const char *reason = "unknown"; crm_node_t *our_node = NULL, *your_node = NULL; ha_msg_input_t *vote = fsa_typed_data(fsa_dt_ha_msg); static time_t last_election_win = 0; static time_t last_election_loss = 0; /* if the membership copy is NULL we REALLY shouldnt be voting * the question is how we managed to get here. */ CRM_CHECK(msg_data != NULL, return); CRM_CHECK(crm_peer_cache != NULL, return); CRM_CHECK(vote != NULL, crm_err("Bogus data from %s", msg_data->origin); return); CRM_CHECK(vote->msg != NULL, crm_err("Bogus data from %s", msg_data->origin); return); /* 受信メッセージデータを取り出す */ op = crm_element_value(vote->msg, F_CRM_TASK); vote_from = crm_element_value(vote->msg, F_CRM_HOST_FROM); your_version = crm_element_value(vote->msg, F_CRM_VERSION); election_owner = crm_element_value(vote->msg, F_CRM_ELECTION_OWNER); crm_element_value_int(vote->msg, F_CRM_ELECTION_ID, &election_id); CRM_CHECK(vote_from != NULL, vote_from = fsa_our_uname); /* CRM_OP_VOTEメッセージの送信元のノード情報を取得する */ your_node = crm_get_peer(0, vote_from); /* 自ノードのノード情報を取得する */ our_node = crm_get_peer(0, fsa_our_uname); if(voted == NULL) { crm_debug("Created voted hash"); /* votedハッシュテーブルが未作成の場合は作成する */ voted = g_hash_table_new_full( g_str_hash, g_str_equal, g_hash_destroy_str, g_hash_destroy_str); } if(cur_state == S_STARTING) { /* 自ノードの状態が、まだ、S_STARTING状態の場合は、DCになれないのでCRM_OP_NOVOTEメッセージを送信する */ reason = "Still starting"; we_loose = TRUE; } else if(our_node == NULL || crm_is_member_active(our_node) == FALSE) { /* 自ノードがまだクラスタ構成として認識されていないか、アクティブでない場合は */ /* DCになれないのでCRM_OP_NOVOTEメッセージを送信する */ reason = "We are not part of the cluster"; log_level = LOG_ERR; we_loose = TRUE; } else if(your_node == NULL || crm_is_member_active(your_node) == FALSE) { /* CRM_OP_VOTEメッセージの送信元のノードがクラスタ構成として認識されていないか、アクティブでない場合は */ /* ログのみを出力する */ reason = "Peer is not part of our cluster"; log_level = LOG_WARNING; done = TRUE; } else if(election_id != current_election_id && crm_str_eq(fsa_our_uuid, election_owner, TRUE)) { /* 現在のelection_idと受信したelection_idが違う場合も、ログのみ出力する */ log_level = LOG_DEBUG_2; reason = "Superceeded"; done = TRUE; } else if(crm_str_eq(op, CRM_OP_NOVOTE, TRUE)) { /* DCになれないと思ったノードが送信したCRM_OP_NOVOTEメッセージの場合 */ char *op_copy = crm_strdup(op); char *uname_copy = crm_strdup(vote_from); CRM_ASSERT(crm_str_eq(fsa_our_uuid, election_owner, TRUE)); /* update the list of nodes that have voted */ /* votedハッシュテーブルにノードデータをセットする */ g_hash_table_replace(voted, uname_copy, op_copy); reason = "Recorded"; done = TRUE; } else if(crm_str_eq(vote_from, fsa_our_uname, TRUE)) { /* 自ノードが送信したCRM_OP_VOTEメッセージを処理する場合 */ char *op_copy = crm_strdup(op); char *uname_copy = crm_strdup(vote_from); CRM_ASSERT(crm_str_eq(fsa_our_uuid, election_owner, TRUE)); /* update ourselves in the list of nodes that have voted */ /* votedハッシュテーブルにノードデータをセットする */ g_hash_table_replace(voted, uname_copy, op_copy); reason = "Recorded"; done = TRUE; } else if(compare_version(your_version, CRM_FEATURE_SET) < 0) { /* 受信したCRM_OP_VOTEメッセージの送り元のversionがCRM_FEATURE_SETよりも小さい場合 */ /* 自ノードは、DCになれない */ reason = "Version"; we_loose = TRUE; } else if(compare_version(your_version, CRM_FEATURE_SET) > 0) { /* 受信したCRM_OP_VOTEメッセージの送り元のversionがCRM_FEATURE_SETよりも大きい場合、ログのみ出力する */ reason = "Version"; } else if(your_node->born < our_node->born) { /* 受信したCRM_OP_VOTEメッセージの送り元の方がbornが自ノードよりも小さい場合 */ reason = "Age"; /* 自ノードは、DCになれない */ we_loose = TRUE; } else if(your_node->born > our_node->born) { /* 受信したCRM_OP_VOTEメッセージの送り元の方がbornが自ノードよりも大きい場合 */ /* 自ノードは、DCの候補 */ reason = "Age"; } else if(fsa_our_uname == NULL) { /* 自ノードのノード名称がセットされていない場合 */ /* 自ノードは、DCになれない */ reason = "Unknown host name"; we_loose = TRUE; } else if(strcasecmp(fsa_our_uname, vote_from) > 0) { /* 自ノードのノード名が送信元...*/ /* 自ノードは、DCになれない */ reason = "Host name"; we_loose = TRUE; } else { /* その他の場合 */ reason = "Host name"; CRM_ASSERT(strcmp(fsa_our_uname, vote_from) != 0); /* cant happen... * } else if(strcasecmp(fsa_our_uname, vote_from) == 0) { * * default... * } else { // strcasecmp(fsa_our_uname, vote_from) < 0 * we win */ } if(done) { do_crm_log(log_level+1, "Election %d (current: %d, owner: %s): Processed %s from %s (%s)", election_id, current_election_id, election_owner, op, vote_from, reason); } else if(we_loose) { /* born値の比較などから、DCノードになれないと判断した場合 */ /* CRM_OP_NOVOTEメッセージをCRMD宛に生成する */ xmlNode *novote = create_request( CRM_OP_NOVOTE, NULL, vote_from, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); do_crm_log(log_level+1, "Election %d (owner: %s) lost: %s from %s (%s)", election_id, election_owner, op, vote_from, reason); /* DCノードにNULLをセットする */ update_dc(NULL); /* election_timeoutタイマーを止める */ crm_timer_stop(election_timeout); if(fsa_input_register & R_THE_DC) { crm_debug_3("Give up the DC to %s", vote_from); register_fsa_input(C_FSA_INTERNAL, I_RELEASE_DC, NULL); } else if(cur_state != S_STARTING) { crm_debug_3("We werent the DC anyway"); register_fsa_input(C_FSA_INTERNAL, I_PENDING, NULL); } /* CRM_OP_NOVOTEメッセージのF_CRM_ELECTION_OWNERに受信メッセージのelection_ownerをセットする */ crm_xml_add(novote, F_CRM_ELECTION_OWNER, election_owner); /* CRM_OP_NOVOTEメッセージのF_CRM_ELECTION_IDに受信メッセージのelection_idをセットする */ crm_xml_add_int(novote, F_CRM_ELECTION_ID, election_id); /* CRM_OP_VOTEメッセージの送信元にCRM_OP_NOVOTEメッセージを送信する */ send_cluster_message(vote_from, crm_msg_crmd, novote, TRUE); /* 送信メッセージを解放する */ free_xml(novote); /* CIBのset_slave処理を実行する */ fsa_cib_conn->cmds->set_slave(fsa_cib_conn, cib_scope_local); last_election_loss = time(NULL); last_election_win = 0; } else { do_crm_log(log_level, "Election %d (owner: %s) pass: %s from %s (%s)", election_id, election_owner, op, vote_from, reason); if(last_election_loss) { time_t tm_now = time(NULL); if(tm_now - last_election_loss < (time_t)loss_dampen) { crm_info("Election %d ignore: We already lost an election less than %ds ago", election_id, loss_dampen); update_dc(NULL); return; } last_election_loss = 0; } #if 0 /* Enabling this code can lead to multiple DCs during SimulStart. * Specifically when a node comes up after our last 'win' vote. * * Fixing and enabling this functionality might become important when * we start running realy big clusters, but for now leave it disabled. */ if(last_election_win) { time_t tm_now = time(NULL); if(tm_now - last_election_win < (time_t)win_dampen) { crm_info("Election %d ignore: We already won an election less than %ds ago", election_id, win_dampen); return; } } last_election_win = time(NULL); #endif /* I_ELECTIONへ */ register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); g_hash_table_destroy(voted); voted = NULL; } }
void notify_crmd(crm_graph_t *graph) { int log_level = LOG_DEBUG; const char *type = "unknown"; enum crmd_fsa_input event = I_NULL; crm_debug("Processing transition completion in state %s", fsa_state2string(fsa_state)); CRM_CHECK(graph->complete, graph->complete = TRUE); switch(graph->completion_action) { case tg_stop: type = "stop"; /* fall through */ case tg_done: type = "done"; log_level = LOG_INFO; 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_start(transition_timer); } else { event = I_PE_CALC; } } else if(fsa_state == S_POLICY_ENGINE) { /* fsa_actionにA_PE_INVOKEアクションを追加して、fsa_sourceトリガーを叩いてcrmdに通知する */ register_fsa_action(A_PE_INVOKE); } break; case tg_shutdown: type = "shutdown"; if(is_set(fsa_input_register, R_SHUTDOWN)) { event = I_STOP; } else { event = I_TERMINATE; } } te_log_action(log_level, "Transition %d status: %s - %s", graph->id, type, crm_str(graph->abort_reason)); graph->abort_reason = NULL; graph->completion_action = tg_done; clear_bit_inplace(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); } }
/* A_ELECTION_VOTE */ void do_election_vote(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) { struct timeval age; xmlNode *vote = NULL; gboolean not_voting = FALSE; /* don't vote if we're in one of these states or wanting to shut down */ switch (cur_state) { case S_STARTING: case S_RECOVERY: case S_STOPPING: case S_TERMINATE: crm_warn("Not voting in election, we're in state %s", fsa_state2string(cur_state)); not_voting = TRUE; break; default: break; } if (not_voting == FALSE) { if (is_set(fsa_input_register, R_STARTING)) { not_voting = TRUE; } } if (not_voting) { if (AM_I_DC) { register_fsa_input(C_FSA_INTERNAL, I_RELEASE_DC, NULL); } else { register_fsa_input(C_FSA_INTERNAL, I_PENDING, NULL); } return; } vote = create_request(CRM_OP_VOTE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); current_election_id++; crm_xml_add(vote, F_CRM_ELECTION_OWNER, fsa_our_uuid); crm_xml_add_int(vote, F_CRM_ELECTION_ID, current_election_id); crm_uptime(&age); crm_xml_add_int(vote, F_CRM_ELECTION_AGE_S, age.tv_sec); crm_xml_add_int(vote, F_CRM_ELECTION_AGE_US, age.tv_usec); send_cluster_message(NULL, crm_msg_crmd, vote, TRUE); free_xml(vote); crm_debug("Started election %d", current_election_id); if (voted) { g_hash_table_destroy(voted); } voted = NULL; if (cur_state == S_ELECTION || cur_state == S_RELEASE_DC) { crm_timer_start(election_timeout); } else if (cur_state != S_INTEGRATION) { crm_err("Broken? Voting in state %s", fsa_state2string(cur_state)); } return; }
/* A_ELECTION_COUNT */ void do_election_count_vote(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) { struct timeval your_age; int age; int election_id = -1; int log_level = LOG_INFO; gboolean use_born_on = FALSE; gboolean done = FALSE; gboolean we_loose = FALSE; const char *op = NULL; const char *vote_from = NULL; const char *your_version = NULL; const char *election_owner = NULL; const char *reason = "unknown"; crm_node_t *our_node = NULL, *your_node = NULL; ha_msg_input_t *vote = fsa_typed_data(fsa_dt_ha_msg); static time_t last_election_loss = 0; /* if the membership copy is NULL we REALLY shouldnt be voting * the question is how we managed to get here. */ CRM_CHECK(msg_data != NULL, return); CRM_CHECK(crm_peer_cache != NULL, return); CRM_CHECK(vote != NULL, crm_err("Bogus data from %s", msg_data->origin); return); CRM_CHECK(vote->msg != NULL, crm_err("Bogus data from %s", msg_data->origin); return); your_age.tv_sec = 0; your_age.tv_usec = 0; op = crm_element_value(vote->msg, F_CRM_TASK); vote_from = crm_element_value(vote->msg, F_CRM_HOST_FROM); your_version = crm_element_value(vote->msg, F_CRM_VERSION); election_owner = crm_element_value(vote->msg, F_CRM_ELECTION_OWNER); crm_element_value_int(vote->msg, F_CRM_ELECTION_ID, &election_id); crm_element_value_int(vote->msg, F_CRM_ELECTION_AGE_S, (int *)&(your_age.tv_sec)); crm_element_value_int(vote->msg, F_CRM_ELECTION_AGE_US, (int *)&(your_age.tv_usec)); CRM_CHECK(vote_from != NULL, vote_from = fsa_our_uname); your_node = crm_get_peer(0, vote_from); our_node = crm_get_peer(0, fsa_our_uname); if (voted == NULL) { crm_debug("Created voted hash"); voted = g_hash_table_new_full(crm_str_hash, g_str_equal, g_hash_destroy_str, g_hash_destroy_str); } if (is_heartbeat_cluster()) { use_born_on = TRUE; } else if (is_classic_ais_cluster()) { use_born_on = TRUE; } age = crm_compare_age(your_age); if (cur_state == S_STARTING) { reason = "Still starting"; we_loose = TRUE; } else if (our_node == NULL || crm_is_peer_active(our_node) == FALSE) { reason = "We are not part of the cluster"; log_level = LOG_ERR; we_loose = TRUE; } else if (election_id != current_election_id && crm_str_eq(fsa_our_uuid, election_owner, TRUE)) { log_level = LOG_DEBUG_2; reason = "Superceeded"; done = TRUE; } else if (your_node == NULL || crm_is_peer_active(your_node) == FALSE) { /* Possibly we cached the message in the FSA queue at a point that it wasn't */ reason = "Peer is not part of our cluster"; log_level = LOG_WARNING; done = TRUE; } else if (crm_str_eq(op, CRM_OP_NOVOTE, TRUE)) { char *op_copy = strdup(op); char *uname_copy = strdup(vote_from); CRM_ASSERT(crm_str_eq(fsa_our_uuid, election_owner, TRUE)); /* update the list of nodes that have voted */ g_hash_table_replace(voted, uname_copy, op_copy); reason = "Recorded"; done = TRUE; } else if (crm_str_eq(vote_from, fsa_our_uname, TRUE)) { char *op_copy = strdup(op); char *uname_copy = strdup(vote_from); CRM_ASSERT(crm_str_eq(fsa_our_uuid, election_owner, TRUE)); /* update ourselves in the list of nodes that have voted */ g_hash_table_replace(voted, uname_copy, op_copy); reason = "Recorded"; done = TRUE; } else if (compare_version(your_version, CRM_FEATURE_SET) < 0) { reason = "Version"; we_loose = TRUE; } else if (compare_version(your_version, CRM_FEATURE_SET) > 0) { reason = "Version"; } else if (age < 0) { reason = "Uptime"; we_loose = TRUE; } else if (age > 0) { reason = "Uptime"; /* TODO: Check for y(our) born < 0 */ } else if (use_born_on && your_node->born < our_node->born) { reason = "Born"; we_loose = TRUE; } else if (use_born_on && your_node->born > our_node->born) { reason = "Born"; } else if (fsa_our_uname == NULL) { reason = "Unknown host name"; we_loose = TRUE; } else if (strcasecmp(fsa_our_uname, vote_from) > 0) { reason = "Host name"; we_loose = TRUE; } else { reason = "Host name"; CRM_ASSERT(strcmp(fsa_our_uname, vote_from) != 0); /* cant happen... * } else if(strcasecmp(fsa_our_uname, vote_from) == 0) { * * default... * } else { // strcasecmp(fsa_our_uname, vote_from) < 0 * we win */ } if (done) { do_crm_log(log_level + 1, "Election %d (current: %d, owner: %s): Processed %s from %s (%s)", election_id, current_election_id, election_owner, op, vote_from, reason); } else if (we_loose) { xmlNode *novote = create_request(CRM_OP_NOVOTE, NULL, vote_from, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); do_crm_log(log_level, "Election %d (owner: %s) lost: %s from %s (%s)", election_id, election_owner, op, vote_from, reason); update_dc(NULL); crm_timer_stop(election_timeout); if (fsa_input_register & R_THE_DC) { crm_trace("Give up the DC to %s", vote_from); register_fsa_input(C_FSA_INTERNAL, I_RELEASE_DC, NULL); } else if (cur_state != S_STARTING) { crm_trace("We werent the DC anyway"); register_fsa_input(C_FSA_INTERNAL, I_PENDING, NULL); } crm_xml_add(novote, F_CRM_ELECTION_OWNER, election_owner); crm_xml_add_int(novote, F_CRM_ELECTION_ID, election_id); send_cluster_message(crm_get_peer(0, vote_from), crm_msg_crmd, novote, TRUE); free_xml(novote); fsa_cib_conn->cmds->set_slave(fsa_cib_conn, cib_scope_local); last_election_loss = time(NULL); } else { do_crm_log(log_level, "Election %d (owner: %s) pass: %s from %s (%s)", election_id, election_owner, op, vote_from, reason); if (last_election_loss) { time_t tm_now = time(NULL); if (tm_now - last_election_loss < (time_t) loss_dampen) { crm_info("Election %d ignore: We already lost an election less than %ds ago (%s)", election_id, loss_dampen, ctime(&last_election_loss)); update_dc(NULL); return; } last_election_loss = 0; } register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); g_hash_table_destroy(voted); voted = NULL; } }
void peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data) { uint32_t old = 0; uint32_t changed = 0; bool appeared = FALSE; const char *status = NULL; set_bit(fsa_input_register, R_PEER_DATA); if (node->uname == NULL) { return; } switch (type) { case crm_status_uname: /* If we've never seen the node, then it also wont be in the status section */ crm_info("%s is now %s", node->uname, node->state); return; case crm_status_rstate: crm_info("Remote node %s is now %s (was %s)", node->uname, node->state, (const char *)data); /* Keep going */ case crm_status_nstate: crm_info("%s is now %s (was %s)", node->uname, node->state, (const char *)data); if (safe_str_eq(data, node->state)) { /* State did not change */ return; } else if(safe_str_eq(CRM_NODE_MEMBER, node->state)) { appeared = TRUE; } break; case crm_status_processes: if (data) { old = *(const uint32_t *)data; changed = node->processes ^ old; } /* crmd_proc_update(node, proc_flags); */ status = (node->processes & proc_flags) ? ONLINESTATUS : OFFLINESTATUS; crm_info("Client %s/%s now has status [%s] (DC=%s, changed=%6x)", node->uname, peer2text(proc_flags), status, AM_I_DC ? "true" : crm_str(fsa_our_dc), changed); if ((changed & proc_flags) == 0) { /* Peer process did not change */ crm_trace("No change %6x %6x %6x", old, node->processes, proc_flags); return; } else if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) { crm_trace("Not connected"); return; } else if (fsa_state == S_STOPPING) { crm_trace("Stopping"); return; } appeared = (node->processes & proc_flags) != 0; if (safe_str_eq(node->uname, fsa_our_uname) && (node->processes & proc_flags) == 0) { /* Did we get evicted? */ crm_notice("Our peer connection failed"); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL); } else if (safe_str_eq(node->uname, fsa_our_dc) && crm_is_peer_active(node) == FALSE) { /* Did the DC leave us? */ crm_notice("Our peer on the DC (%s) is dead", fsa_our_dc); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL); } else if(AM_I_DC && appeared == FALSE) { crm_info("Peer %s left us", node->uname); /* crm_update_peer_join(__FUNCTION__, node, crm_join_none); */ } break; } if (AM_I_DC) { xmlNode *update = NULL; int flags = node_update_peer; gboolean alive = crm_is_peer_active(node); crm_action_t *down = match_down_event(0, node->uuid, NULL, appeared); crm_trace("Alive=%d, appear=%d, down=%p", alive, appeared, down); if (alive && type == crm_status_processes) { 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 (alive && safe_str_eq(task, CRM_OP_FENCE)) { crm_info("Node return implies stonith of %s (action %d) completed", node->uname, down->id); erase_status_tag(node->uname, XML_CIB_TAG_LRM, cib_scope_local); erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local); /* down->confirmed = TRUE; Only stonith-ng returning should imply completion */ down->sent_update = TRUE; /* Prevent tengine_stonith_callback() from calling send_stonith_update() */ } else if (safe_str_eq(task, CRM_OP_FENCE)) { crm_trace("Waiting for stonithd to report the fencing of %s is complete", node->uname); /* via tengine_stonith_callback() */ } else if (alive == FALSE) { crm_notice("%s of %s (op %d) is complete", task, node->uname, down->id); /* down->confirmed = TRUE; Only stonith-ng returning should imply completion */ stop_te_timer(down->timer); flags |= node_update_join | node_update_expected; crmd_peer_down(node, FALSE); check_join_state(fsa_state, __FUNCTION__); update_graph(transition_graph, down); trigger_graph(); } else { crm_trace("Other %p", down); } } else if (appeared == FALSE) { crm_notice("Stonith/shutdown of %s not matched", node->uname); crm_update_peer_join(__FUNCTION__, node, crm_join_none); check_join_state(fsa_state, __FUNCTION__); abort_transition(INFINITY, tg_restart, "Node failure", NULL); fail_incompletable_actions(transition_graph, node->uuid); } else { crm_trace("Other %p", down); } update = do_update_node_cib(node, flags, NULL, __FUNCTION__); fsa_cib_anon_update(XML_CIB_TAG_STATUS, update, cib_scope_local | cib_quorum_override | cib_can_create); free_xml(update); } trigger_fsa(fsa_source); }
void peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data) { uint32_t old = 0; uint32_t changed = 0; bool appeared = FALSE; bool is_remote = is_set(node->flags, crm_remote_node); const char *status = NULL; /* Crmd 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) { set_bit(fsa_input_register, R_PEER_DATA); } 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_rstate: 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(safe_str_neq(data, node->state), 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 (safe_str_eq(CRM_NODE_MEMBER, node->state)) { appeared = TRUE; if (!is_remote) { remove_stonith_cleanup(node->uname); } } crmd_alert_node_event(node); break; case crm_status_processes: if (data) { old = *(const uint32_t *)data; changed = node->processes ^ old; } status = (node->processes & proc_flags) ? ONLINESTATUS : OFFLINESTATUS; crm_info("Client %s/%s now has status [%s] (DC=%s, changed=%6x)", node->uname, peer2text(proc_flags), status, AM_I_DC ? "true" : crm_str(fsa_our_dc), changed); if ((changed & proc_flags) == 0) { /* Peer process did not change */ crm_trace("No change %6x %6x %6x", old, node->processes, proc_flags); return; } else if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) { crm_trace("Not connected"); return; } else if (fsa_state == S_STOPPING) { crm_trace("Stopping"); return; } appeared = (node->processes & proc_flags) != 0; if (safe_str_eq(node->uname, fsa_our_uname) && (node->processes & proc_flags) == 0) { /* Did we get evicted? */ crm_notice("Our peer connection failed"); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL); } else if (safe_str_eq(node->uname, fsa_our_dc) && crm_is_peer_active(node) == FALSE) { /* Did the DC leave us? */ crm_notice("Our peer on the DC (%s) is dead", fsa_our_dc); 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(fsa_our_dc_version, "3.0.9") > 0) { erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local); } } else if(AM_I_DC && appeared == FALSE) { crm_info("Peer %s left us", node->uname); erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local); } break; } if (AM_I_DC) { xmlNode *update = NULL; int flags = node_update_peer; gboolean alive = is_remote? appeared : crm_is_peer_active(node); crm_action_t *down = match_down_event(node->uuid, appeared); crm_trace("Alive=%d, appeared=%d, down=%d", alive, appeared, (down? down->id : -1)); if (alive && type == crm_status_processes) { 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 (safe_str_eq(task, CRM_OP_FENCE)) { /* tengine_stonith_callback() confirms fence actions */ crm_trace("Updating CIB %s stonithd reported fencing of %s complete", (down->confirmed? "after" : "before"), node->uname); } else if ((alive == FALSE) && safe_str_eq(task, CRM_OP_SHUTDOWN)) { crm_notice("%s of peer %s is complete "CRM_XS" op=%d", task, node->uname, down->id); /* down->confirmed = TRUE; */ stop_te_timer(down->timer); if (!is_remote) { flags |= node_update_join | node_update_expected; crmd_peer_down(node, FALSE); check_join_state(fsa_state, __FUNCTION__); } update_graph(transition_graph, down); trigger_graph(); } else { crm_trace("Node %s is %salive, was expected to %s (op %d)", node->uname, (alive? "" : "not "), task, down->id); } } else if (appeared == FALSE) { crm_notice("Stonith/shutdown of %s not matched", node->uname); if (!is_remote) { crm_update_peer_join(__FUNCTION__, node, crm_join_none); check_join_state(fsa_state, __FUNCTION__); } abort_transition(INFINITY, tg_restart, "Node failure", NULL); fail_incompletable_actions(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, tg_restart, "pacemaker_remote node integrated", NULL); } } /* Update the CIB node state */ update = create_node_state_update(node, flags, NULL, __FUNCTION__); fsa_cib_anon_update(XML_CIB_TAG_STATUS, update, cib_scope_local | cib_quorum_override | cib_can_create); free_xml(update); } trigger_fsa(fsa_source); }