When rmid calls Process.destroy to terminate an child activation group, it does not call Process.waitFor to wait until the child is really terminated. On platforms that support graceful process shutdown, including execution of Runtime shutdown hooks, this means that rmid might terminate before its children have, or that rmid might attempt to start a new group instance before the previous instance has really terminated. In general, this seems undesirable; it seems better for rmid to explicitly waitFor the process. (Except in the case of destroying children from rmid's own shutdown hook [added in merlin], which should finish quickly.)