How to configure distributed multi-actor system

[technomage]

I see how I can have multiple actors in a group in a single executable. It is not clear if I can have the same group in multiple executables to share the workload of messages. When the topology is built it appears to separate groups into local or remote and no way to have both? Ideally I would like each actor in a group spread over many executables (pods in Kubernetes) process messages for the keys the manage.

[loyd]

I'm sorry for not getting back to you sooner.

An actor group cannot contain actors across multiple nodes, but each node can have a group with the same name and routing between them.

// Service A (nodes 1, 2, 3)
let consumers = topology.local("consumers");

// Service B (nodes 4, 5)
let remote_consumers = topology.remote("consumers");
let producers = topology.local("producers");

producers.route_to(&remote_consumers, |e, nd| {
    msg!(match e {
        SomeEvent => ...,
    });
});

The second argument nd (node discovery) is supposed to be used to choose node among the connected ones to avoid explicit routing (SomeEvent { node_no } => Outcome::Unicast(node_no)).

However, it doesn't provide now any methods to do it because we need to decide which syntax to use for it. In the simplest form it can be

SomeEvent => Outcome::Unicast(nd.random()), // nd.nearest_by_rtt() and so on

But how should it look when sent to multiple nodes (Outcome::Multicast(nd.by_tag("something")))? It's not very flexible.

This, it's not the question of implementation, but more of API design =(