When the broker is stopped, we need to ensure that no left overs leak so that it can be restarted without problem in the same Java process. That way, it should be easy to stop and restart the broker Even on some critical  IO error, once he error is fixed.

In theory this should be provided by enhancing the existing test suite.