JMS+HAIL+XA: Pooled Connections Never reconnects on singleto

[arisa]

I'm using JMS with HAIL and XA on JBoss 3.2.3, which seems uncommon from the posts I see. I have a clustered configuration (2 servers), and need to send JMS Messages transactionally with the container-managed transactions. I need to send the JMS Message to the singleton JMS Server (which in 3.2.3 is done via HAIL. This is different in 3.2.4 I believe).

Everything works fine except when the JMS Singleton Server moves. Then I can no longer send messages because I get the IllegalStateTransition error that singletonServerMoved_. The app server never recovers from this problem.

So far, I have diagnosed the following:
1. The connections in the managed connection pool, once created, are not destroyed even in the event of a singletonServerMoved_ exception.
2. The connections are thus stale when the singleton server moves.
3. I can manually fix the problem by invoking "flush" on the managed connection pool, which destroys the connection. Sending a JMS message after this will result in new connections being created in the pool, which works perfectly.
4. I cannot use setExceptionListener on the connections because a) I can't set it on the pre-created pooled connections, and b) I don't know how to force the pooled connection to be recreated.

My questions:
1. Is there a way to solve this problem?
2. If not, is there a way to workaround this problem? My current thought is to listen to the singleton server moved notification event, and flush the managed connection pool at that time.
3. If not, how are other people sending JMS Messages on 3.2.3 with container-managed transactions on a clustered configuration to a singleton? I suppose with singleton MDBs, but I haven't tried that yet.

Many thanks for any direction any of you can give...

Aris

[schrouf]

Have a look into org.jboss.resource.adapter.jdbc.vendor for some vendor specific ExceptionSorter implementations (currently only Oracle/Sybase). I am not sure, but I guess that a propper ExceptionSorter implementation for your database might do the job. By returning true for a fatal error (based on the database specific error code) the associated connection should be removed from the connection pool by the connection manager (at least this is my expectation ?!?) If you follow this guess, please give a feedback on success/failure.

Regards
Ulf

[arisa]

Thanks Ulf for the tip. It wasn't directly applicable as I am using JMS which doesn't have any relations to jdbc and ExceptionSorter. However, it pointed me in the right direction.

I've solved this problem using the following approach:

1. The problem is that the JmsManagedConnection does not register an ExceptionListener for the JMS Connection.
2. I extended the JmsManagedConnection class with my own. My version of setup() sets an exception listener on the JMS Connection.
3. I also extended the JmsManagedConnectionFactory so that it instantiates my extended version of JmsManagedConnection.
4. I then created a new JMS RAR which defines a new Jms Provider which uses my extended JmsManagedConnectionFactory.
5. I then create a new pool of JMS Connections, and use this pool.

It works. There's one bit that's messy: How the exception listener removes the connection from the pool. I couldn't figure this one out. Trying to destroy the ManagedConnection doesn't work, because it will only be flushed from the pool after one user tries to use it (and gets an error that the ManagedConnection has been destroyd).

So, because I was desperate I hacked it so that it calls the flush() method on the ManagedConnectionPool MBean.

If anyone knows how a ManagedConnection can remove itself from a ManagedConnectionPool cleanly without being used anymore, I'd love to hear it.

Thanks to all who considered this problem,
Aris

[adrian.brock]

This was all fixed in 3.2.4 for HAJMS
http://cvs.sourceforge.net/viewcvs.py/jboss/jbosscx/src/main/org/jboss/resource/connectionmanager/InternalManagedConnectionPool.java?r1=1.22&r2=1.23

[arisa]

Excellent. I see now why the ManagedConnections do not get flushed from the pool on a ConnectionError.

Thanks all again.

[wangchen_ca]

I face similiar problems for failover for HAJMS when singleton node moves.

I tried the link posted by [e-mail] at last-two reply
http://cvs.sourceforge.net/viewcvs.py/jboss/jbosscx/src/main/org/jboss/resource/connectionmanager/InternalManagedConnectionPool.java?r1=1.22&r2=1.23

which does not make sense for me.

Arisa, Can you explain what is your point?

Thanks