I'm trying to attack some of the waiting patches, and I'm a bit perplexed about ZOOKEEPER-1100. The patch is to address the case where somehow a runtime exception causes the SendThread to escape the while loop inside of the run method, and the thread then dies. Why don't we change the state to CLOSED after this happens, if for some reason it is not either CLOSED or AUTH_FAILED? We queue a disconnected event, but the send thread is dead as a doornail, so disconnected is a rather misleading indicator. There's no way you are reusing the client cnxn after the send thread dies.
Any input? Thanks, C
