I know that proper clustering using Akka is the focus of Akka 2.1, however need to build something with what's available now.
I have a multi node Akka setup and want to gracefully handle remote actors dying. However, as the system is symmetric and should have no single point of failure, I'm not sure how to properly set up a supervisor hierarchy. The supervisor cannot be on the same Akka instance as the actor being supervised, as I need to deal with the whole Akka node terminating, but if I put the supervisor on a separate node, I'm now risking orphaning or terminating the supervised actors of the node running the supervisor dies.
What I really want is a clustered supervisor, where an actor on any of the Akka nodes (or n of them) can handle supervision of a set of actors to avoid any single point of failure. Is such a thing possible in Akka 2.0? If not, what would be the best way to get the same effect?
