We are a ‘n’ node system (with one ActorSystem per node) and would like to run ‘m’ ActorSystems on it for load balancing. Each node should only run one ActorSystem. I could use ‘m’ ClusterSingletons (and then leave every other singleton clusters. Sharded cluster might be used where all nodes in a similar way (by leaving). Is there a more natural way where the Leader actively manages ‘m’ Actors? There is no shared state and so that is not a concern here. I didn’t see any generic cluster examples (where I guess the Leader node implements the above functionality). Using Singleton cluster manager to get the ActorRef is nice and would prefer an approach which gives us this functionality.
did you already have a look at Akka Cluster Sharding?
Once you have all your nprocesses (or nodes) running, each has its own ActorSystem and they all become a cluster (I’ll skip the part on forming clusters). Then, when the cluster is ready you can deploy sharded entities which act as distributed singletons where each singleton is bound to a subset of the traffic. A common use case is keeping a memory image of you database so each of these sharded entities is database entity ( a user, a shopping cart, a bank account). Another possible use case is Cluster Sharded Daemon Processes.
Ahh, Sharded Daemon process seems to be what I’m lookin for. So, I’d create ‘m’ daemon processes and use the system receptionist service to either send work to a specific daemon process or to any of them. Thanks much @ignasi35