/** * The entry point for the monitoring module thread * * @param arg The handle of the monitor */ static void monitorMain(void *arg) { MYSQL_MONITOR *handle = (MYSQL_MONITOR *)arg; MONITOR_SERVERS *ptr; size_t nrounds = 0; MONITOR_SERVERS *candidate_master = NULL; int master_stickiness = handle->disableMasterFailback; int is_cluster=0; int log_no_members = 1; if (mysql_thread_init()) { LOGIF(LE, (skygw_log_write_flush( LOGFILE_ERROR, "Fatal : mysql_thread_init failed in monitor " "module. Exiting.\n"))); return; } handle->status = MONITOR_RUNNING; while (1) { if (handle->shutdown) { handle->status = MONITOR_STOPPING; mysql_thread_end(); handle->status = MONITOR_STOPPED; return; } /** Wait base interval */ thread_millisleep(MON_BASE_INTERVAL_MS); /** * Calculate how far away the monitor interval is from its full * cycle and if monitor interval time further than the base * interval, then skip monitoring checks. Excluding the first * round. */ if (nrounds != 0 && ((nrounds*MON_BASE_INTERVAL_MS)%handle->interval) >= MON_BASE_INTERVAL_MS) { nrounds += 1; continue; } nrounds += 1; /* reset cluster members counter */ is_cluster=0; ptr = handle->databases; while (ptr) { monitorDatabase(handle, ptr); /* clear bits for non member nodes */ if ( ! SERVER_IN_MAINT(ptr->server) && (ptr->server->node_id < 0 || ! SERVER_IS_JOINED(ptr->server))) { ptr->server->depth = -1; /* clear M/S status */ server_clear_status(ptr->server, SERVER_SLAVE); server_clear_status(ptr->server, SERVER_MASTER); /* clear master sticky status */ server_clear_status(ptr->server, SERVER_MASTER_STICKINESS); } /* Log server status change */ if (mon_status_changed(ptr)) { LOGIF(LD, (skygw_log_write_flush( LOGFILE_DEBUG, "Backend server %s:%d state : %s", ptr->server->name, ptr->server->port, STRSRVSTATUS(ptr->server)))); } if (SERVER_IS_DOWN(ptr->server)) { /** Increase this server'e error count */ ptr->mon_err_count += 1; } else { /** Reset this server's error count */ ptr->mon_err_count = 0; } ptr = ptr->next; } /* * Let's select a master server: * it could be the candidate master following MIN(node_id) rule or * the server that was master in the previous monitor polling cycle * Decision depends on master_stickiness value set in configuration */ /* get the candidate master, followinf MIN(node_id) rule */ candidate_master = get_candidate_master(handle->databases); /* Select the master, based on master_stickiness */ handle->master = set_cluster_master(handle->master, candidate_master, master_stickiness); ptr = handle->databases; while (ptr && handle->master) { if (!SERVER_IS_JOINED(ptr->server) || SERVER_IN_MAINT(ptr->server)) { ptr = ptr->next; continue; } if (ptr != handle->master) { /* set the Slave role */ server_set_status(ptr->server, SERVER_SLAVE); server_clear_status(ptr->server, SERVER_MASTER); /* clear master stickyness */ server_clear_status(ptr->server, SERVER_MASTER_STICKINESS); } else { /* set the Master role */ server_set_status(handle->master->server, SERVER_MASTER); server_clear_status(handle->master->server, SERVER_SLAVE); if (candidate_master && handle->master->server->node_id != candidate_master->server->node_id) { /* set master stickyness */ server_set_status(handle->master->server, SERVER_MASTER_STICKINESS); } else { /* clear master stickyness */ server_clear_status(ptr->server, SERVER_MASTER_STICKINESS); } } is_cluster++; ptr = ptr->next; } if (is_cluster == 0 && log_no_members) { LOGIF(LE, (skygw_log_write_flush( LOGFILE_ERROR, "Error: there are no cluster members"))); log_no_members = 0; } else { if (is_cluster > 0 && log_no_members == 0) { LOGIF(LE, (skygw_log_write_flush( LOGFILE_ERROR, "Info: found cluster members"))); log_no_members = 1; } } } }
/** * The entry point for the monitoring module thread * * @param arg The handle of the monitor */ static void monitorMain(void *arg) { MONITOR* mon = (MONITOR*) arg; GALERA_MONITOR *handle; MONITOR_SERVERS *ptr; size_t nrounds = 0; MONITOR_SERVERS *candidate_master = NULL; int master_stickiness; int is_cluster = 0; int log_no_members = 1; monitor_event_t evtype; spinlock_acquire(&mon->lock); handle = (GALERA_MONITOR *) mon->handle; spinlock_release(&mon->lock); master_stickiness = handle->disableMasterFailback; if (mysql_thread_init()) { MXS_ERROR("mysql_thread_init failed in monitor module. Exiting."); return; } handle->status = MONITOR_RUNNING; while (1) { if (handle->shutdown) { handle->status = MONITOR_STOPPING; mysql_thread_end(); handle->status = MONITOR_STOPPED; return; } /** Wait base interval */ thread_millisleep(MON_BASE_INTERVAL_MS); /** * Calculate how far away the monitor interval is from its full * cycle and if monitor interval time further than the base * interval, then skip monitoring checks. Excluding the first * round. */ if (nrounds != 0 && ((nrounds * MON_BASE_INTERVAL_MS) % mon->interval) >= MON_BASE_INTERVAL_MS) { nrounds += 1; continue; } nrounds += 1; /* reset cluster members counter */ is_cluster = 0; ptr = mon->databases; while (ptr) { ptr->mon_prev_status = ptr->server->status; monitorDatabase(mon, ptr); /* Log server status change */ if (mon_status_changed(ptr)) { MXS_DEBUG("Backend server %s:%d state : %s", ptr->server->name, ptr->server->port, STRSRVSTATUS(ptr->server)); } if (!(SERVER_IS_RUNNING(ptr->server)) || !(SERVER_IS_IN_CLUSTER(ptr->server))) { dcb_hangup_foreach(ptr->server); } if (SERVER_IS_DOWN(ptr->server)) { /** Increase this server'e error count */ dcb_hangup_foreach(ptr->server); ptr->mon_err_count += 1; } else { /** Reset this server's error count */ ptr->mon_err_count = 0; } ptr = ptr->next; } /* * Let's select a master server: * it could be the candidate master following MIN(node_id) rule or * the server that was master in the previous monitor polling cycle * Decision depends on master_stickiness value set in configuration */ /* get the candidate master, following MIN(node_id) rule */ candidate_master = get_candidate_master(mon); /* Select the master, based on master_stickiness */ if (1 == handle->disableMasterRoleSetting) { handle->master = NULL; } else { handle->master = set_cluster_master(handle->master, candidate_master, master_stickiness); } ptr = mon->databases; while (ptr) { const int repl_bits = (SERVER_SLAVE | SERVER_MASTER | SERVER_MASTER_STICKINESS); if (SERVER_IS_JOINED(ptr->server)) { if (handle->master) { if (ptr != handle->master) { /* set the Slave role and clear master stickiness */ server_clear_set_status(ptr->server, repl_bits, SERVER_SLAVE); } else { if (candidate_master && handle->master->server->node_id != candidate_master->server->node_id) { /* set master role and master stickiness */ server_clear_set_status(ptr->server, repl_bits, (SERVER_MASTER | SERVER_MASTER_STICKINESS)); } else { /* set master role and clear master stickiness */ server_clear_set_status(ptr->server, repl_bits, SERVER_MASTER); } } } is_cluster++; } else { server_clear_set_status(ptr->server, repl_bits, 0); } ptr = ptr->next; } if (is_cluster == 0 && log_no_members) { MXS_ERROR("There are no cluster members"); log_no_members = 0; } else { if (is_cluster > 0 && log_no_members == 0) { MXS_NOTICE("Found cluster members"); log_no_members = 1; } } ptr = mon->databases; while (ptr) { /** Execute monitor script if a server state has changed */ if (mon_status_changed(ptr)) { evtype = mon_get_event_type(ptr); if (isGaleraEvent(evtype)) { MXS_NOTICE("Server changed state: %s[%s:%u]: %s", ptr->server->unique_name, ptr->server->name, ptr->server->port, mon_get_event_name(ptr)); if (handle->script && handle->events[evtype]) { monitor_launch_script(mon, ptr, handle->script); } } } ptr = ptr->next; } } }