Sync Replication?

[yairogen]

I've experienced an issue where I kill a live server and my client fails over to the backup I identify that messages are handled twice.

I cam to a conclusion that the replication is async. Is that correct?

Although a performance hit - is there a configuration that supports sync replication? I.e. producer won't get ack on sending message till message was replicated to backup server.Similarly - prevent situations where message was processed but the ack on the message didn't get replicated (or something similar).

[jbertram]

Replication should behave in an essentially synchronous manner.

What specifically is the use-case when you see these issues?  Are you using transactions?

[yairogen]

On another thread I've seen that when failing over to another server, my listeners re-connect correctly but log messages that were already handled prior to the fail over. This makes me wonder how could that be. If the replication is syhc - what other explanation can be that the ack on message handling was not "replicated" to the backup server (which would result in same message being handled twice)?

[jbertram]

Perhaps the ack didn't make it to the server before the server crashed?

Got a reproducible test-case?