There are two different methods by which the backup may be started.

In either case, it is highly advisable to allow any pending transactions to be completed before beginning the backup, and then not to permit any new transactions to begin during the backup process.

Copy the cluster backup files to the slave that is being brought on line. Each system running an ndbd process for the master cluster will have cluster backup files located on it, and all of these files must be copied to the slave to ensure a successful restore. The backup files can be copied into any directory on the computer where the slave management host resides, so long as the MySQL and NDB binaries have read permissions in that directory. In this case, we will assume that these files have been copied into the directory /var/BACKUPS/BACKUP-1.

It is not necessary that the slave cluster have the same number of ndbd processes (data nodes) as the master; however, it is highly recommended this number be the same. It is necessary that the slave be started with the --skip-slave-start option, to prevent premature startup of the replication process.

Create any databases on the slave cluster that are present on the master cluster that are to be replicated to the slave. Important: A CREATE SCHEMA statement corresponding to each database to be replicated must be executed on each data node in the slave cluster.

Reset the slave cluster using this statement in the MySQL Monitor:

mysqlS> RESET SLAVE;

It is important to make sure that the slave's apply_status table does not contain any records prior to running the restore process. You can accomplish this by running this SQL statement on the slave:

mysqlS> DELETE FROM cluster_replication.apply_status;

You can now start the cluster restoration process on the replication slave using the ndb_restore command for each backup file in turn. For the first of these, it is necessary to include the -m option to restore the cluster metadata:

shellS> ndb_restore -c slave_host:port -n node-id \
        -b backup-id -m -r dir

dir is the path to the directory where the backup files have been placed on the replication slave. For the ndb_restore commands corresponding to the remaining backup files, the -m option should not be used.

For restoring from a master cluster with four data nodes (as shown in the figure in Section 16.7, “MySQL Cluster Replication”) where the backup files have been copied to the directory /var/BACKUPS/BACKUP-1, the proper sequence of commands to be executed on the slave might look like this:

shellS> ndb_restore -c rep-slave:1186 -n 2 -b 1 -m \
        -r ./VAR/BACKUPS/BACKUP-1
shellS> ndb_restore -c rep-slave:1186 -n 3 -b 1 \
        -r ./VAR/BACKUPS/BACKUP-1
shellS> ndb_restore -c rep-slave:1186 -n 4 -b 1 \
        -r ./VAR/BACKUPS/BACKUP-1
shellS> ndb_restore -c rep-slave:1186 -n 5 -b 1 \
        -r ./VAR/BACKUPS/BACKUP-1

This sequence of commands causes the most recent epoch records to be written to the slave's apply_status table.

Next, it is necessary to make all nodes in the slave cluster aware of the new tables. (This is due to the fact that the NDB Cluster storage engine does not currently support autodiscovery of schema changes. See Section 16.7.9.2, “Initiating Discovery of Schema Changes”.) You can accomplish this using these commands:

mysqlS*> USE db_name;
mysqlS*> SHOW TABLES;

db_name is the name of the database which was backed up and restored. Where multiple databases have been backed up and then restored, it is necessary to issue the USE and SHOW statements for each database in turn. Note also that these commands must be issued on each host acting as a data node in the slave cluster.

Now you need to obtain the most recent epoch from the binlog_index table on the slave (as discussed in Section 16.7.8, “Implementing Failover with MySQL Cluster”):

mysqlS> SELECT @latest:=MAX(epoch)
        FROM cluster_replication.apply_status;

Using @latest as the epoch value obtained in the previous step, you can obtain the correct starting position @pos in the correct binary logfile @file from the master's cluster_replication.binlog_index table using the query shown here:

mysqlM> SELECT 
     ->     @file:=SUBSTRING_INDEX(File, '/', -1),
     ->     @pos:=Position
     -> FROM cluster_replication.binlog_index
     -> WHERE epoch > @latest
     -> ORDER BY epoch ASC LIMIT 1;

Using the values obtained in the previous step, you can now issue the appropriate CHANGE MASTER TO statement in the slave's mysql client:

mysqlS> CHANGE MASTER TO
     ->     MASTER_LOG_FILE='@file',
     ->     MASTER_LOG_POS=@pos;

Now that the slave “knows” from what point in which binlog file to start reading data from the master, you can cause the slave to begin replicating with this standard MySQL statement:

mysqlS> START SLAVE;