one possible scenario might be: (1) pants allocates lots of stuff in lmdb (2) machine restart (3) pants maps the same stuff into lmdb immediately when it starts up again after restart