shall cluster-client expose sender in replies?
roland:
Another thing I wonder is whether our choice of exposing the responding ActorRef is the right one: an ActorRef signifies the capability to access a resource and passing an ActorRef along proliferates that capability. Maybe the receptionist should send on the reply with its own sender address and if an actor wants to introduce itself to external clients it needs to do so by passing the ActorRef as part of the message, opt-in instead of opt-out.
patrik:
One thing to consider is that the messages sent via the client/receptionist inwards must be wrapped, which must be remembered when replying back to such a reference. I think I prefer deadLetters as sender if we don't want to expose the cluster destination ref as sender. Please create a ticket and we can think more about it.
Another thing I wonder is whether our choice of exposing the responding ActorRef is the right one: an ActorRef signifies the capability to access a resource and passing an ActorRef along proliferates that capability. Maybe the receptionist should send on the reply with its own sender address and if an actor wants to introduce itself to external clients it needs to do so by passing the ActorRef as part of the message, opt-in instead of opt-out.
patrik:
One thing to consider is that the messages sent via the client/receptionist inwards must be wrapped, which must be remembered when replying back to such a reference. I think I prefer deadLetters as sender if we don't want to expose the cluster destination ref as sender. Please create a ticket and we can think more about it.
Leave a comment