/** * Associate a new session with this instance of the router. * * @param instance The router instance data * @param session The session itself * @return Session specific data for this session */ static void * newSession(ROUTER *instance, SESSION *session) { ROUTER_INSTANCE *inst = (ROUTER_INSTANCE *)instance; ROUTER_CLIENT_SES *client_rses; BACKEND *candidate = NULL; int i; BACKEND *master_host = NULL; LOGIF(LD, (skygw_log_write_flush( LOGFILE_DEBUG, "%lu [newSession] new router session with session " "%p, and inst %p.", pthread_self(), session, inst))); client_rses = (ROUTER_CLIENT_SES *)calloc(1, sizeof(ROUTER_CLIENT_SES)); if (client_rses == NULL) { return NULL; } #if defined(SS_DEBUG) client_rses->rses_chk_top = CHK_NUM_ROUTER_SES; client_rses->rses_chk_tail = CHK_NUM_ROUTER_SES; #endif /** * Find the Master host from available servers */ master_host = get_root_master(inst->servers); /** * Find a backend server to connect to. This is the extent of the * load balancing algorithm we need to implement for this simple * connection router. */ /* * Loop over all the servers and find any that have fewer connections * than the candidate server. * * If a server has less connections than the current candidate we mark this * as the new candidate to connect to. * * If a server has the same number of connections currently as the candidate * and has had less connections over time than the candidate it will also * become the new candidate. This has the effect of spreading the * connections over different servers during periods of very low load. */ for (i = 0; inst->servers[i]; i++) { if(inst->servers[i]) { LOGIF(LD, (skygw_log_write( LOGFILE_DEBUG, "%lu [newSession] Examine server in port %d with " "%d connections. Status is %s, " "inst->bitvalue is %d", pthread_self(), inst->servers[i]->server->port, inst->servers[i]->current_connection_count, STRSRVSTATUS(inst->servers[i]->server), inst->bitmask))); } if (SERVER_IN_MAINT(inst->servers[i]->server)) continue; if (inst->servers[i]->weight == 0) continue; /* Check server status bits against bitvalue from router_options */ if (inst->servers[i] && SERVER_IS_RUNNING(inst->servers[i]->server) && (inst->servers[i]->server->status & inst->bitmask & inst->bitvalue)) { if (master_host) { if (inst->servers[i] == master_host && (inst->bitvalue & SERVER_SLAVE)) { /* skip root Master here, as it could also be slave of an external server * that is not in the configuration. * Intermediate masters (Relay Servers) are also slave and will be selected * as Slave(s) */ continue; } if (inst->servers[i] == master_host && (inst->bitvalue & SERVER_MASTER)) { /* If option is "master" return only the root Master as there * could be intermediate masters (Relay Servers) * and they must not be selected. */ candidate = master_host; break; } } else { /* master_host is NULL, no master server. * If requested router_option is 'master' * candidate wll be NULL. */ if (inst->bitvalue & SERVER_MASTER) { candidate = NULL; break; } } /* If no candidate set, set first running server as our initial candidate server */ if (candidate == NULL) { candidate = inst->servers[i]; } else if ((inst->servers[i]->current_connection_count * 1000) / inst->servers[i]->weight < (candidate->current_connection_count * 1000) / candidate->weight) { /* This running server has fewer connections, set it as a new candidate */ candidate = inst->servers[i]; } else if ((inst->servers[i]->current_connection_count * 1000) / inst->servers[i]->weight == (candidate->current_connection_count * 1000) / candidate->weight && inst->servers[i]->server->stats.n_connections < candidate->server->stats.n_connections) { /* This running server has the same number of connections currently as the candidate but has had fewer connections over time than candidate, set this server to candidate*/ candidate = inst->servers[i]; } } } /* There is no candidate server here! * With router_option=slave a master_host could be set, so route traffic there. * Otherwise, just clean up and return NULL */ if (!candidate) { if (master_host) { candidate = master_host; } else { LOGIF(LE, (skygw_log_write_flush( LOGFILE_ERROR, "Error : Failed to create new routing session. " "Couldn't find eligible candidate server. Freeing " "allocated resources."))); free(client_rses); return NULL; } } client_rses->rses_capabilities = RCAP_TYPE_PACKET_INPUT; /* * We now have the server with the least connections. * Bump the connection count for this server */ atomic_add(&candidate->current_connection_count, 1); client_rses->backend = candidate; LOGIF(LD, (skygw_log_write( LOGFILE_DEBUG, "%lu [newSession] Selected server in port %d. " "Connections : %d\n", pthread_self(), candidate->server->port, candidate->current_connection_count))); /* * Open a backend connection, putting the DCB for this * connection in the client_rses->backend_dcb */ client_rses->backend_dcb = dcb_connect(candidate->server, session, candidate->server->protocol); if (client_rses->backend_dcb == NULL) { atomic_add(&candidate->current_connection_count, -1); free(client_rses); return NULL; } dcb_add_callback( client_rses->backend_dcb, DCB_REASON_NOT_RESPONDING, &handle_state_switch, client_rses); inst->stats.n_sessions++; /** * Add this session to the list of active sessions. */ spinlock_acquire(&inst->lock); client_rses->next = inst->connections; inst->connections = client_rses; spinlock_release(&inst->lock); CHK_CLIENT_RSES(client_rses); skygw_log_write( LOGFILE_TRACE, "Readconnroute: New session for server %s. " "Connections : %d", candidate->server->unique_name, candidate->current_connection_count); return (void *)client_rses; }
/** * 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; } } }
/** * Associate a new session with this instance of the router. * * @param instance The router instance data * @param session The session itself * @return Session specific data for this session */ static void * newSession(ROUTER *instance, SESSION *session) { ROUTER_INSTANCE *inst = (ROUTER_INSTANCE *)instance; ROUTER_CLIENT_SES *client_rses; BACKEND *candidate = NULL; int i; int master_host = -1; LOGIF(LD, (skygw_log_write_flush( LOGFILE_DEBUG, "%lu [newSession] new router session with session " "%p, and inst %p.", pthread_self(), session, inst))); client_rses = (ROUTER_CLIENT_SES *)calloc(1, sizeof(ROUTER_CLIENT_SES)); if (client_rses == NULL) { return NULL; } #if defined(SS_DEBUG) client_rses->rses_chk_top = CHK_NUM_ROUTER_SES; client_rses->rses_chk_tail = CHK_NUM_ROUTER_SES; #endif /** * Find a backend server to connect to. This is the extent of the * load balancing algorithm we need to implement for this simple * connection router. */ /* * Loop over all the servers and find any that have fewer connections * than the candidate server. * * If a server has less connections than the current candidate we mark this * as the new candidate to connect to. * * If a server has the same number of connections currently as the candidate * and has had less connections over time than the candidate it will also * become the new candidate. This has the effect of spreading the * connections over different servers during periods of very low load. */ for (i = 0; inst->servers[i]; i++) { if(inst->servers[i]) { LOGIF(LD, (skygw_log_write( LOGFILE_DEBUG, "%lu [newSession] Examine server in port %d with " "%d connections. Status is %d, " "inst->bitvalue is %d", pthread_self(), inst->servers[i]->server->port, inst->servers[i]->current_connection_count, inst->servers[i]->server->status, inst->bitmask))); } /* * If router_options=slave, get the running master * It will be used if there are no running slaves at all */ if (inst->bitvalue == SERVER_SLAVE) { if (master_host < 0 && (SERVER_IS_MASTER(inst->servers[i]->server))) { master_host = i; } } if (inst->servers[i] && SERVER_IS_RUNNING(inst->servers[i]->server) && (inst->servers[i]->server->status & inst->bitmask) == inst->bitvalue) { /* If no candidate set, set first running server as our initial candidate server */ if (candidate == NULL) { candidate = inst->servers[i]; } else if (inst->servers[i]->current_connection_count < candidate->current_connection_count) { /* This running server has fewer connections, set it as a new candidate */ candidate = inst->servers[i]; } else if (inst->servers[i]->current_connection_count == candidate->current_connection_count && inst->servers[i]->server->stats.n_connections < candidate->server->stats.n_connections) { /* This running server has the same number of connections currently as the candidate but has had fewer connections over time than candidate, set this server to candidate*/ candidate = inst->servers[i]; } } } /* There is no candidate server here! * With router_option=slave a master_host could be set, so route traffic there. * Otherwise, just clean up and return NULL */ if (!candidate) { if (master_host >= 0) { candidate = inst->servers[master_host]; } else { LOGIF(LE, (skygw_log_write_flush( LOGFILE_ERROR, "Error : Failed to create new routing session. " "Couldn't find eligible candidate server. Freeing " "allocated resources."))); free(client_rses); return NULL; } } client_rses->rses_capabilities = RCAP_TYPE_PACKET_INPUT; /* * We now have the server with the least connections. * Bump the connection count for this server */ atomic_add(&candidate->current_connection_count, 1); client_rses->backend = candidate; LOGIF(LD, (skygw_log_write( LOGFILE_DEBUG, "%lu [newSession] Selected server in port %d. " "Connections : %d\n", pthread_self(), candidate->server->port, candidate->current_connection_count))); /* * Open a backend connection, putting the DCB for this * connection in the client_rses->backend_dcb */ client_rses->backend_dcb = dcb_connect(candidate->server, session, candidate->server->protocol); if (client_rses->backend_dcb == NULL) { atomic_add(&candidate->current_connection_count, -1); free(client_rses); return NULL; } inst->stats.n_sessions++; /** * Add this session to the list of active sessions. */ spinlock_acquire(&inst->lock); client_rses->next = inst->connections; inst->connections = client_rses; spinlock_release(&inst->lock); CHK_CLIENT_RSES(client_rses); return (void *)client_rses; }