We had the same problem in a high write situation. We have master<->master with each master having 2 replicas. We point writes at A, enable gtid, deployment step and consistency on all 5 replicas and restart. Then repoint writes to B and bounce A to get them all in sync with changes.
Then on the B master, we disable deployment step and every replica of B stopped with binlog corruption. The last event written was the SET @@SESSION.GTID_NEXT event.
We had the same problem in a high write situation. We have master<->master with each master having 2 replicas. We point writes at A, enable gtid, deployment step and consistency on all 5 replicas and restart. Then repoint writes to B and bounce A to get them all in sync with changes.
Then on the B master, we disable deployment step and every replica of B stopped with binlog corruption. The last event written was the SET @@SESSION.GTID_NEXT event.