I have a suggestion for this.

In the case of Galera Cluster, there are two GTID domains--gtid_domain_id and wsrep_gtid_domain_id. Only the domain defined by wsrep_gtid_domain_id is used for transactions that are replicated by Galera Cluster. The domain defined by gtid_domain_id is only used for transactions that are local to the node. This means that gtid_domain_id is likely to get out of sync between cluster nodes for all sorts of reasons.

I would suggest updating the documentation, so that DO_DOMAIN_IDS = (<wsrep_gtid_domain_id>) is set in the CHANGE MASTER command.

https://mariadb.com/kb/en/library/change-master-to/#do_domain_ids

However, it looks like MaxScale 2.3.8 doesn't support the DO_DOMAIN_IDS option yet, so it would have to be implemented first. I created MXS-2581 for this.

i.e. if wsrep_gtid_domain_id=2, then the command for the primary master would be:

# mysql -h $MAXSCALE_HOST -P $MAXCALE_PORT

MariaDB> SET @@global.gtid_slave_pos='0-198-123';

MariaDB> CHANGE MASTER TO

         MASTER_HOST='192.168.10.5',

         MASTER_PORT=3306,

         MASTER_USER='repl',

         MASTER_PASSWORD='repl',

         MASTER_USE_GTID=Slave_pos,

         DO_DOMAIN_IDS=(2);

And the command for the secondary master would be:

MariaDB> CHANGE MASTER ':2' TO

         MASTER_HOST='192.168.10.6',

         MASTER_PORT=3306,

         MASTER_USER='repl',

         MASTER_PASSWORD='repl',

         MASTER_USE_GTID=Slave_pos,

         DO_DOMAIN_IDS=(2);

If MariaDB Server supported something like MDEV-20715, then that could make failover between secondary masters a lot safer with Galera Cluster.

The new binlogrouter has an option "select_master" which allows it to use a master of a cluster monitored by maridbmon or galeramon.