If a failover is happening, and consumers are stuck for any reason during the process, HornetQ might issue interrupts on wrong threads.
This is because the list of threads is captured before a time wait what could cause issues with the system is the thread doen't belong to the consumer any longer.

How reproducible:

On a MDB that's receiving messages and sending messages to a remote node, pull the cable (or disable network). and make failover happen to a backup.

Steps to Reproduce:
1. Deploy the MDB on that condition
2. Pull the network cable
3. failover that server.

Actual results:

JCA will have connections into invalid state. You will get exception on an infinite loop.

There's a secondary issue here as those connections shouldn't come back to the pool.

Expected results:

Connections to reconnect successfully without