The image shows three situations: normal operation, failure of master, recovered master.

Normal operation: Users are connected to applications that update Database A and Database B, which are both masters. The masters replicate to each other.

Failure of master: All users now connect to the applications that update Database B. No replication occurs.

Recovered master: Some users reconnect to applications that update Database A. Some applications send updates to Database A and some applications send updates to Database B. The masters replicate to each other.

End of description.