Distributed Elixir Compatibility

[benwilson512]

Hey!

I'm really just getting started along the process of evaluating CQRS as a pattern for us, and I'm curious if you can speak to how this works in a clustered environment. Using something like Postgres for the actual event log solves the issue of having a consistent view of events but how does that work for any in memory state held?

We're looking at an upcoming project that seems to be a good fit for a CQRS/ES approach but I confess I'm still at the point where it makes sense as a diagram on someone's slide deck but I am unclear about what a project's code actually looks like.

[slashdotdash]

The short answer is that it doesn't. Commanded only supports running on a single node. There's an open issue to support running on a cluster of nodes (#39) and an open pull request with further discussion on this topic (#60).

The approach will likely be to distribute aggregate, event handler, and process manager processes amongst the nodes in the cluster. So they run as singleton instances, but the load is distributed (e.g. user1 aggregate on node1, user2 aggregate on node2). In a net split scenario only one half of the cluster will continue functioning. This is to ensure consistency and prevent duplicate aggregates/handlers running. I'm open to suggestions, and pull requests, for alternative ways of tackling this.

An alternative approach, that you could implement now, is to use a single event store and Commanded instance per bounded context (microservice). Then you can deploy each service to its own node for distribution. You would need a way of message passing between each service for commands and events. This could be a REST API endpoint for commands and an event handler that pushes events onto a message bus for the other services to consume.

Feel free to chat about Commanded on Gitter.

[benwilson512]

Gotcha. Is there any literature on established techniques used in other platforms?

[slashdotdash]

Here are a couple of relevant projects that implement distributed event sourcing:

• Akka is an actor based toolkit for Java, Scala, and .NET [cluster | split brain resolver].
• Eventuate uses causally ordered event streams [overview | repo].

[slashdotdash]

Closing this issue. Refer to related issue "Support running on a cluster of nodes" (#39) instead.