We have a 2 node (plus an arbiter) Percona XtraDB cluster. It is used for data indexing/archiving. It has it's own database called archive
but it is also the replication slave for a database called primary
. This comes from a standard master->slave Percona Server pair.
One of the clustered servers segfaulted (reason still being investigated...). This node was, specifically, the one that was doing replication. On restarting it, it deleted the contents of the MySQL data directory (/var/lib/mysql/
) and did an SST from the other node. This went fine, but it now has a partially incomplete dataset on the primary
database and no master binlog position information.
How should replication be restored?
- It is not feasible to export the
primary
database as a mysql dump since it is circa 1TB in size when looking at the raw innodb files - If we export the
primary
DB with xtrabackup and prepare it as a whole on the clustered node, it is assumed it will get deleted as soon as the node starts SST - It is difficult but not infeasible to allow the node to do SST and then restore an xtrabackup of
primary
DB table by table... but it will take a while given 500+ tables
So, the question is... What is the correct way to set up replication into a multi-master cluster?