Remote death watch apparently not working correctly

Are there any known issues with DeathWatch of remote actors? We’re currently scratching heads looking at a bug where we have:

  • two ActorSystems on different nodes;
  • an actor in system A retrieves an ActorRef of an actor in actor B and does unwatch followed by watchWith ;
  • system B is restarted;
  • the actor in system A receives the death watch message and tries to retrieve the actor ref once again, receives the old ActorRef , again does unwatch && watchWith ;
  • by this time, it must be dead — after all, we’ve already received the death watch message! — but the latter watchWith didn’t produce any death message;
  • now we observe that the actor in system A is stuck with the “old” actor ref — we figured it because any message it tries to send there ends up in dead letters.

Any ideas / known issues? We’re at 2.5.14.

(This was posted to gitter some time ago, but didn’t catch on)

Sounds like a bug, please create an issue.