>  We could modify this to unsetSuccessAndSendWriteRequest after each sucessful 
> zk write.

Aren't we doing that now? If there is a change ensemble in flight we are 
putting failed bookies back to delayedWriteFailedBookies(), and leaving from 
here. So when the current ensemble change comes back we call 
unsetSuccessAndSendWriteRequest(). But the problem is we don't attempt to 
change ensemble with delayedWriteFailedBookies() until next write comes in. 
What if the next write will take long time to come in? We may fire 
writeQuorumTimeout without even attempting an ensemble change. 

[ Full content available at: https://github.com/apache/bookkeeper/pull/1646 ]
This message was relayed via gitbox.apache.org for [email protected]

Reply via email to