22.6.9.2 Point-In-Time Recovery Using NDB Cluster Replication

Back up all NDB databases in the cluster, using the START BACKUP command in the ndb_mgm client (see Section 22.5.3, “Online Backup of NDB Cluster”).

At some later point, prior to restoring the cluster, make a backup of the mysql.ndb_binlog_index table. It is probably simplest to use mysqldump for this task. Also back up the binary log files at this time.

This backup should be updated regularly—perhaps even hourly—depending on your needs.

(Catastrophic failure or error occurs.)

Locate the last known good backup.

Clear the data node file systems (using ndbd --initial or ndbmtd --initial).

Use DROP TABLE or TRUNCATE TABLE with the mysql.ndb_binlog_index table.

Execute ndb_restore, restoring all data. You must include the --restore-epoch option when you run ndb_restore, so that the ndb_apply_status table is populated correctly. (See Section 22.4.23, “ndb_restore — Restore an NDB Cluster Backup”, for more information.)

Restore the ndb_binlog_index table from the output of mysqldump and restore the binary log files from backup, if necessary.

Find the epoch applied most recently—that is, the maximum epoch column value in the ndb_apply_status table—as the user variable @LATEST_EPOCH (emphasized):

Find the latest binary log file (@FIRST_FILE) and position (Position column value) within this file that correspond to @LATEST_EPOCH in the ndb_binlog_index table:

Using mysqlbinlog, replay the binary log events from the given file and position up to the point of the failure. (See Section 4.6.8, “mysqlbinlog — Utility for Processing Binary Log Files”.)