If a heartbeat_timeout occurs, then the connection_sup of amqp dies.
Now, this should lead to handling inside turtle_janitor, but the janitor is monitoring the wrong process and will never receive a down message on the connection.
The system now hangs. The connection manager for the connection thinks the connection is fine.
In addition:
The janitor ought to cleanup and remove every channel connected up on the connection as well. This is somewhat mind-boggling because all of the channels should be gone by now and destroyed. Yet, internal testing shows that the channels actually persist in the system.
Only possible conclusion for now: something is not getting cleaned up like it should.
The problem is as follows:
In addition: