Is it the desired behavior to have the RemoteInvoker.invoke(...) method return null when there is a in-only operation invoked? Should I rely on exceptions being thrown as the way to understand if there was a problem in the invocation?
Can you file a JIRA on this against 1.0? Lukasz and I have discussed a generalized mechanism w/r/t fault handling on in-only MEPs that would optionally expose runtime faults to the consumer. That's the missing bit right now to get this back in the RemoteMessage response.