I am trying to upgrade our JBoss Messaging application to use HornetQ. I am experiencing my MDBs getting 'stuck' and I have managed to untangle a few different scenarios that cause this. I will try and put together test cases, but I would appreciate your help as to where to look.
We have a cluster of 2 nodes. One of these nodes is responsible for picking up some FTP files, parsing them, and sending messages to the queue. Because the nodes are varying speeds, I have set ConsumerWindowSize to 0. What I am seeing is that if both nodes are up, everything is fine.
But if I take 1 node down, wait for a few minutes, and then try sending some messages to the queue, exactly 1/2 of the messages get processed. If I look in my logs it is a very clean split - like every odd number gets processed and every even numbered message disappears. It's almost as if HornetQ is still routing the messages to the dead node?
Is this possible, is there routing code that is meant to detect whether a node is dead and, if so, could if still be accidentally routing messages there?
|Retrieving data ...|