This is a deployment bug that affects inflow messages.

JCA 1.5 specification states states that "Before a rsource adapter is undeployed, the application server must deactivate all active enpoints consuming messages from that specific resource adapter". This doesn't seem to be the case:

1 - start the server with a RAR and an MDB endpoint setup (the test connector accept messages from network and the mdb merely print them on the console after wainting 10s to easealy test enpoint concurrency issues).
OK: I see the start() and endpointActivation(...) callback called
2 - run the client test several times (in a way that enables concurrent message consumption): OK (client OK, console prints all messages)
3 - undeploy RAR: I can see the stop callback, but no MDB undeploy or endpointDeactivation(...) callback on resource adapter.
4 - deploy RAR again: deployment OK, I can see the start() callback but no endpointActivation(...) callback called
5 - run the client test again: client OK, demonstrating the network endpoint is working, but no message printed on the console (no endpoint consumption).
6 - undeploy/deploy MDB: everything works again as in 2 (plus, messages pending from step 5 are obviously consumed)

I think we should have, according to spec, an undeployment of MDBs that depend on the RAR service when the service is undeployed, and a furter restart should trigger a new endpointActivation callback.