Comment 12 for bug 855030

Hm. I haven't used the multi host network stuff, myself, so I can't comment much on the above right now.

There _is_ a bug filed for the DBError above. Unclear exactly what it is, but sqlalchemy is catching an exception, doing transaction cleanup, and then doing a re-raise via a bare raise. Somewhere in sqlalchemy's transaction cleanup, the original exception is getting eaten, so the bare raise does a raise of NoneType.

BTW, there is 1 case I can think of that could happen in the real world where the queue doesn't exist when doing a reply. This can happen if a service makes an rpc.call that results in a long-running transaction.... and the original service is restarted before the long-running transaction completes. I know this probably isn't what is happening in your case, but we probably should catch the AMQPChannelException for this case...