Connection pool leak if transaction marked for rollback? / 2

[msquance]

Hi,

I'm experiencing what appears to be a bug in the XADataSource pool when trying to grab a connection in a transaction that has been marked for rollback. It seems to hang onto it within the pool, even though it was not successful in issuing the connection to the caller.

I am using JBoss 2.4.9 with Oracle 8i.

In looking at the ObjectPool code, I can see that the getObject() method does not seem to release anything if an exception is thrown after marking the object as in use.

I haven't found what package ObjectPool belongs to in JBoss 3.0 so I'm not sure if it has been fixed there already.

Here is some sample code that I reproduced this with. It's a message driven bean getting invoked. After the 50 time in the loop, I get No ManagedConnections Available (my pool size is 50).

public void onMessage( Message inMessage )
{
try {
UserTransaction userTrans = myContext_.getUserTransaction();
userTrans.setTransactionTimeout( 1 );

// start the transaction
userTrans.begin();

// wait for the timeout
Thread.sleep( 1500 );

InitialContext initCtx = new InitialContext();
javax.sql.DataSource dataSource =
(javax.sql.DataSource) initCtx.lookup( "java:ProfilerDS" );

for ( int i = 0; i < 55; i++ ) {
try {
java.sql.Connection con = dataSource.getConnection();

} catch ( Exception ex ) {
trace.error( "Caught Exception: " + ex );
}

}
userTrans.rollback();

} catch ( Exception ex ) {
trace.error( ex );
}
}

Am I missing something, or is this a bug? I haven't tried enabling the idle timeouts or garbage collection yet, but I'm hoping that might be a workaround to return these objects to the pool.

Any thoughts would be appreciated.

Thanks,
Mike.

[joehung]

I too am seeing this problem in 3.2.1. I'm not sure if it's the same one but it sounds too familiar. Have you found a solution yet?

-joe

[amayingenta]

We had this problem with 3.2.1 (with a LocalTx Datasource). Most of the problem turned out to be our application doing something dumb :) but there also appears to be a bug in the connection pooling.

See http://www.mail-archive.com/[e-mail]/msg30247.html

I'm afraid I never raised a bug report because the problem largely went away when we changed our application.

-Andrew

[joehung]

I've seen exactly the same exception sequence you mentioned in the archive message.

the problem appears when we keep stop/start the datastore and *intentionally* cause the connection to fail to see if the connection pool can recover.

We are forced to work around this problem by restarting JBoss once we detected this exception... No Managed Connection Available!