CODEWITHSUNDEEP
akkaakka-clusterakka-remote-actorakka.net
Patryk Ćwiek
Patryk Ćwiek

Reputation: 14318

Cross-service calls with Akka

I'm playing with newly released Akka.Net 1.0 (congratulations on the release!) so it's all quite new to me, but I'm pretty sure anyone with JVM Akka experience could also chime in since there's nothing that's runtime-dependent in the question.

Let's consider several (for the sake of example, 2) separate services that are a part of a larger system/application. Those services usually do their own things, but cross-service calls are sometimes needed. Let's say that Service 2 can be standalone and has a GetStuff action. Service 1 has a DoSomething action, which has to get the result of GetStuff action first.

What is preferred way of handling that kind of situation when both services can be deployed separately and to different machines?

As I said, I don't know much about Akka, but digging through examples, docs and source I found two options:

  1. Remoting. Separate actor systems in their own services using Remoting to get ActorSelection from remote host. It would be pretty much the same as Remoting docs example, just that two actor systems would be equal 'clients'.
  2. Clustering. I'm trying to wrap my head around that and the most I can figure out right now would be to set up a separate cluster service that would just set up the cluster system, creating a simple listener actor so that the seed node could be properly initialized (?). Then each and every separate actor system created in their own services would join to said cluster system under different role.

Maybe there's yet another solution that I'm not aware of...?

Personally, clustering solution seems harder to grasp and set up at first glance, but maybe there are some significant advantages that I can't see right now.

To reiterate, what is the preferred way of handling such situation and what should I look out for?

Upvotes: 2

Views: 697

Answers (1)

Aaronontheweb
Aaronontheweb

Reputation: 8394

Akka.Cluster depends on Akka.Remote - here's what's fundamentally different about them:

  • Akka.Remote - allows you to connect and communicate with actor systems running in remote processes. Can be totally separate code-bases running entirely different Akka.NET applications ("services", if you will.) All you need to communicate between the two systems is a shared set of message classes that's visible in both processes.
  • Akka.Cluster - an abstraction on top of Akka.Remote that eliminates the need for each of your service instances to have to know the explicit address of every other possible service instance you might need to connect to. These can be instances of the same services or instances of different services. Enables dynamic discovery of services via a really simple "seed node" strategy.

I recommend you take a look at the Akka.Cluster microservices example that I wrote - it shows how you can use the Akka.Cluster "roles" feature to dynamically make cross-service calls to nodes in a different service without having to explicitly define any of their network addresses. In particular, take a look at how I use "cluster-aware" routers to do this.

Upvotes: 3

Related Questions