JMS transactions seem to need new contexts each time

[ibmmqmet]

This is a simple program using two JMS connections. I'm running it inside Spring Boot, but some of the autowiring is a little bit more explicit to deal with the multiple ConnectionFactories.

UserTransaction utx = com.arjuna.ats.jta.UserTransaction.userTransaction();

XAConnectionFactory xacf2 = context.getBean("qm2", XAConnectionFactory.class);
XAConnectionFactory xacf3 = context.getBean("qm3", XAConnectionFactory.class);
ConnectionFactoryProxy cfp2 = new ConnectionFactoryProxy((XAConnectionFactory) xacf2, new TransactionHelperImpl(tm));
ConnectionFactoryProxy cfp3 = new ConnectionFactoryProxy((XAConnectionFactory) xacf3, new TransactionHelperImpl(tm);

// Should be able to set these up before doing any transactional work.
ctx2 = cfp2.createContext();
ctx3 = cfp3.createContext();
consumer = ctx2.createConsumer(q);
producer = ctx3.createProducer();

for (int i=0;i<2;i++) {
  utx.begin();

... do some work

  if (i==0)  // Alternate between commit/rollback outcomes
    utx.commit();
  else
    utx.rollback():
}

But it does not seem to work - the consumer/producer's operations are not all done inside a coordinated syncpoint. Variations on the code also failed in different ways (such as the two contexts running with independent, non-coordinated transactions.)

The only way I could get this to work was to move the creation of the contexts and the corresponding consumer/producer BELOW the utx.begin() line. So there's a new set of contexts created for each loop or transaction - the TM does not appear to reuse an existing JMSContext. This is very inefficient as creation of a JMSContext can be expensive.

The equivalent program using Atomikos as the TM does appear to work as expected.

Any thoughts?

[graben]

ConnectionFactoryProxy isn't pooled, that's why creation of multiple JMSContexts is expected. Do use the pooled variant using messaginghub dependency instead.