Then I restored another copy from the same minute as the split, and after a few hundred HA_ERR_FOUND_DUPP_KEY skips, it started catching up. An hour later, it was 0 seconds behind and has stayed in sync.

• Those errors shouldn't have happened. Are you sure you are running the master with synv_binlog=1 and the slave with sync_master_info=1, and both with innodb_flush_log_at_trx_commit=1 ?

sync_binlog is indeed set to 0.   The others are fine.

The description at https://mariadb.com/kb/en/replication-and-binary-log-system-variables/#sync_binlog mentions battery backed ram, which we do have, and a full UPS system. I’m guessing that’s why it was never enabled on our estate.

However, Veeam doing a VM image backup is just a disk snapshot, and no different to having the power pulled and it’s not so surprising that data can be lost.

I will make sure it is turned on soonest.

Thanks again,

S