In the case a session hangs because of a problem somewhere, it may help to reset the user's account on the cluster. There are two possibilities: soft reset (default) and hard reset. One typically starts with a soft request and then, if it does not work, try a hard request. It may be needed to Ctrl-C (and exit) the ROOT session between attempts.
A request for a soft reset is issued by the following command:
root  TProof::Reset("master")
where "master" is the hostname of the PROOF master, the entry point in the system. A soft request fully cleans the users instances on the master. This should trigger automatic cleaning of the users session also on the workers. Note that, while the sessions are invalidate as soon as the command is issued, the real processes can be cleaned only after a few minutes, depending on the frequency of the session checks done on the cluster machines.
A request for a hard reset is issued by the following command:
root  TProof::Reset("master", kTRUE)
A hard request fully cleans the users instances on the master and forwards the same request also to the worker machines via an independent channel. This should solve all the cases where for some reason a soft request does not help.