DeathWatch: hook up with failure detector
although it is not clear to me what should happen: if we signal unreachable actors as Terminated, they must also terminate before the node becomes reachable again, otherwise everything will be quite confusing …
Leave a comment
Just discussed with Viktor:
- there should be a timeout of unreachability after which all watched refs are considered and signaled dead
- upon connect, the client presents a UUID together with the other credentials; this UUID is created at ActorSystem start-up
- if that client has been determined dead by virtue of the first point, it will be denied reconnects with the same old UUID
- if a client is rejected, it will publish an appropriate message on the eventStream to which user code can then react by restarting, stopping, logging, etc.
- there should be a timeout of unreachability after which all watched refs are considered and signaled dead
- upon connect, the client presents a UUID together with the other credentials; this UUID is created at ActorSystem start-up
- if that client has been determined dead by virtue of the first point, it will be denied reconnects with the same old UUID
- if a client is rejected, it will publish an appropriate message on the eventStream to which user code can then react by restarting, stopping, logging, etc.