Bug 865728

Description of problem:
Platform BZ for https://issues.jboss.org/browse/JBESB-3772

I have a number of chained async, one-way services, which all use transacted JCA (with HornetQ tech preview). I set a generic "FaultTo" EPR in a custom composer-class set on the FS-Provider of the first service (entry point into the ESB), with the intention that, if any of async, one-way services fail, the Fault Message will be sent to my FaultService.

One of my one-way services uses a SyncServiceInvoker to do a request-response call, in the middle of its action pipeline to a synchronous request-response service for message validation. The result is then used in the same pipeline to do some CBR. The problem is that, when I pass the message to the SyncServiceInvoker (which is configured to suspend the transaction), the ReplyTo and FaultTo headers get set to 'null' (as expected, as the synchronous request-response service needs to pass its reply and fault message to the caller), however, these ReplyTo and FaultTo headers never get set back to their original value if the call succeeds. In the exception flow of SyncServiceInvoker, the ReplyTo and FaultTo do get reset to their original value (with the remark that this is not actually necessary, which is wrong). So, when my CBR routes the message to the next service, the ReplyTo and FaultTo headers are null, implying that the async, one-way services will not sent a possible Fault Message to the FaultService.
In my opinion, resetting the ReplyTo and FaultTo headers needs to be done in the 'finally' block, not the 'catch' block.

How reproducible:
Easily by making a reproducer ESB service

Steps to Reproduce:
1. Make a ESB service which invokes SyncServiceInvoker action and a custom action in order
2. In the second custom action, write a logic which confirms that ReplyTo/FaultTo headers are erased after the invocation to SyncServiceInvoker
3. Run the ESB service to confirm the reported behaviour
  
Actual results:
ReplyTo/FaultTo headers are erased

Expected results:
ReplyTo/FaultTo headers should be retained