コード例 #1
0
ファイル: recovery.c プロジェクト: CHINA-JD/prestogres
/*
 * Start online recovery.
 * "recovery_node" is the node to be recovered.
 * Master or primary node is chosen in this function.
 */
void start_recovery(int recovery_node)
{
	int node_id;
	BackendInfo *backend;
	BackendInfo *recovery_backend;
	PGconn *conn;
	int failback_wait_count;
#define FAILBACK_WAIT_MAX_RETRY 5		/* 5 seconds should be enough for failback operation */

	ereport(LOG,
		(errmsg("starting recovering node %d", recovery_node)));

	if ( (recovery_node < 0) || (recovery_node >= pool_config->backend_desc->num_backends) )
		ereport(ERROR,
				(errmsg("node recovery failed, node id: %d is not valid", recovery_node)));

	if (VALID_BACKEND(recovery_node))
		ereport(ERROR,
				(errmsg("node recovery failed, node id: %d is alive", recovery_node)));

	/* select master/primary node */
	node_id = MASTER_SLAVE ? PRIMARY_NODE_ID : REAL_MASTER_NODE_ID;
	backend = &pool_config->backend_desc->backend_info[node_id];

	/* get node info to be recovered */
	recovery_backend = &pool_config->backend_desc->backend_info[recovery_node];

	conn = connect_backend_libpq(backend);
	if (conn == NULL)
		ereport(ERROR,
				(errmsg("node recovery failed, unable to connect to master node: %d ", node_id)));

	PG_TRY();
	{
		/* 1st stage */
		if (REPLICATION)
		{
			exec_checkpoint(conn);
			ereport(LOG,
				(errmsg("node recovery, CHECKPOINT in the 1st stage done")));
		}

		exec_recovery(conn, backend, recovery_backend, FIRST_STAGE);

		ereport(LOG,
			(errmsg("node recovery, 1st stage is done")));

		if (REPLICATION)
		{
			ereport(LOG,
				(errmsg("node recovery, starting 2nd stage")));

			/* 2nd stage */
			*InRecovery = RECOVERY_ONLINE;
			if (pool_config->use_watchdog)
			{
				/* announce start recovery */
				if (WD_OK != wd_start_recovery())
					ereport(ERROR,
							(errmsg("node recovery failed, failed to send start recovery packet")));
			}

			if (wait_connection_closed() != 0)
				ereport(ERROR,
						(errmsg("node recovery failed, waiting connection closed in the other pgpools timeout")));

			ereport(LOG,
				(errmsg("node recovery, all connections from clients have been closed")));

			exec_checkpoint(conn);

			ereport(LOG,
				(errmsg("node recovery"),
					 errdetail("CHECKPOINT in the 2nd stage done")));

			exec_recovery(conn, backend, recovery_backend, SECOND_STAGE);
		}

		exec_remote_start(conn, recovery_backend);

		check_postmaster_started(recovery_backend);

		ereport(LOG,
			(errmsg("node recovery, node: %d restarted", recovery_node)));

		/*
		 * reset failover completion flag.  this is necessary since
		 * previous failover/failback will set the flag to 1.
		 */
		pcp_wakeup_request = 0;

		/* send failback request to pgpool parent */
		send_failback_request(recovery_node);

		/* wait for failback */
		failback_wait_count = 0;
		while (!pcp_wakeup_request)
		{
			struct timeval t = {1, 0};
			/* polling SIGUSR2 signal every 1 sec */
			select(0, NULL, NULL, NULL, &t);
			failback_wait_count++;
			if (failback_wait_count >= FAILBACK_WAIT_MAX_RETRY)
			{
				ereport(LOG,
					(errmsg("node recovery"),
						errdetail("waiting for wake up request is timeout(%d seconds)",
							   FAILBACK_WAIT_MAX_RETRY)));

				break;
			}
		}
		pcp_wakeup_request = 0;
	}
	PG_CATCH();
	{
		PQfinish(conn);
		PG_RE_THROW();
	}
	PG_END_TRY();

	PQfinish(conn);

	ereport(LOG,
			(errmsg("recovery done")));
}
コード例 #2
0
ファイル: recovery.c プロジェクト: aerfaman/pgpool2
/*
 * Start online recovery.
 * "recovery_node" is the node to be recovered.
 * Master or primary node is chosen in this function.
 */
int start_recovery(int recovery_node)
{
	int node_id;
	BackendInfo *backend;
	BackendInfo *recovery_backend;
	PGconn *conn;
	int failback_wait_count;
#define FAILBACK_WAIT_MAX_RETRY 5		/* 5 seconds should be enough for failback operation */

	pool_log("starting recovering node %d", recovery_node);

	if (VALID_BACKEND(recovery_node))
	{
		pool_error("start_recovery: backend node %d is alive", recovery_node);
		return 1;
	}

	Req_info->kind = NODE_RECOVERY_REQUEST;

	/* select master/primary node */
	node_id = MASTER_SLAVE ? PRIMARY_NODE_ID : REAL_MASTER_NODE_ID;
	backend = &pool_config->backend_desc->backend_info[node_id];

	/* get node info to be recovered */
	recovery_backend = &pool_config->backend_desc->backend_info[recovery_node];

	conn = connect_backend_libpq(backend);
	if (conn == NULL)
	{
		pool_error("start_recovery: could not connect master node (%d)", node_id);
		return 1;
	}

	/* 1st stage */
	if (REPLICATION)
	{
		if (exec_checkpoint(conn) != 0)
		{
			PQfinish(conn);
			pool_error("start_recovery: CHECKPOINT failed");
			return 1;
		}
		pool_log("CHECKPOINT in the 1st stage done");
	}

	if (exec_recovery(conn, recovery_backend, FIRST_STAGE) != 0)
	{
		PQfinish(conn);
		return 1;
	}

	pool_log("1st stage is done");

	if (REPLICATION)
	{
		pool_log("starting 2nd stage");

		/* 2nd stage */
		*InRecovery = RECOVERY_ONLINE;
		if (pool_config->use_watchdog)
		{
			/* announce start recovery */
			if (WD_OK != wd_start_recovery())
			{
				PQfinish(conn);
				pool_error("start_recovery: timeover for waiting connection closed in the other pgpools");
				return 1;
			}
		}

		if (wait_connection_closed() != 0)
		{
			PQfinish(conn);
			pool_error("start_recovery: timeover for waiting connection closed");
			return 1;
		}

		pool_log("all connections from clients have been closed");

		if (exec_checkpoint(conn) != 0)
		{
			PQfinish(conn);
			pool_error("start_recovery: CHECKPOINT failed");
			return 1;
		}

		pool_log("CHECKPOINT in the 2nd stage done");

		if (exec_recovery(conn, recovery_backend, SECOND_STAGE) != 0)
		{
			PQfinish(conn);
			return 1;
		}
	}

	if (exec_remote_start(conn, recovery_backend) != 0)
	{
		PQfinish(conn);
		pool_error("start_recovery: remote start failed");
		return 1;
	}

	if (check_postmaster_started(recovery_backend))
	{
		PQfinish(conn);
		pool_error("start_recovery: check start failed");
		return 1;
	}

	pool_log("%d node restarted", recovery_node);

	/*
	 * reset failover completion flag.  this is necessary since
	 * previous failover/failback will set the flag to 1.
	 */
	pcp_wakeup_request = 0;

	/* send failback request to pgpool parent */
	send_failback_request(recovery_node);

	/* wait for failback */
	failback_wait_count = 0;
	while (!pcp_wakeup_request)
	{
		struct timeval t = {1, 0};
		/* polling SIGUSR2 signal every 1 sec */
		select(0, NULL, NULL, NULL, &t);
		failback_wait_count++;
		if (failback_wait_count >= FAILBACK_WAIT_MAX_RETRY)
		{
			pool_log("start_recovery: waiting for wake up request is timeout(%d seconds)",
					 FAILBACK_WAIT_MAX_RETRY);
			break;
		}
	}
	pcp_wakeup_request = 0;

	PQfinish(conn);

	pool_log("recovery done");

	return 0;
}