Now that we are supposed to have working replication, the most logic thing to do is to close replication together with the JournalStorageManager. In that way, during a normal server.stop(), anything persisted on the live server would be persisted on the backup as well.
Turns out that during server.stop() we were still persisting the notifications for ACCEPTOR_CLOSED.
The commit Set server as NOT-active after server sessions are closed marks the server as inactive as soon as all server sessions as closed. It will have the side effect of not persisting these notifications, because the server will be marked as inactive as we stop the RemotingService.
Does anyone have a valid reason to persist these ACCEPTOR_CLOSED during server.stop()?
The acceptor_closed is not suppsoed to be persisted. Perhaps there's an ID being generated which is using the IDGenerator which will turn to the journal?