Check out my Kubernetes book!

The actor model in 10 minutes


Our CPUs are not getting any faster. What’s happening is that we now have multiple cores on them. If we want to take advantage of all this hardware we have available now, we need a way to run our code concurrently. Decades of untraceable bugs and developers’ depression have shown that threads are not the way to go. But fear not, there are great alternatives out there and today I want to show you one of them: The actor model.

The model

The actor model is a conceptual model to deal with concurrent computation. It defines some general rules for how the system’s components should behave and interact with each other. The most famous language that uses this model is probably Erlang. I’ll try to focus more on the model itself and not in how it’s implemented in different languages or libraries.

Actors

An actor is the primitive unit of computation. It’s the thing that receives a message and does some kind of computation based on it.

The idea is very similar to what we have in object-oriented languages: An object receives a message (a method call) and does something depending on which message it receives (which method we are calling).

The main difference is that actors are completely isolated from each other and they will never share memory. It’s also worth noting that an actor can maintain a private state that can never be changed directly by another actor.

One ant is no ant

And one actor is no actor. They come in systems. In the actor model everything is an actor and they need to have addresses so one actor can send a message to another.

Actors have mailboxes

It’s important to understand that, although multiple actors can run at the same time, an actor will process a given message sequentially. This means that if you send 3 messages to the same actor, it will just execute one at a time. To have these 3 messages being executed concurrently, you need to create 3 actors and send one message each.

Messages are sent asynchronously to an actor, that needs to store them somewhere while it’s processing another message. The mailbox is the place where these messages are stored.

Actors communicate with each other by sending asynchronous messages. Those messages are stored in other actors' mailboxes until they're processed.

What actors do

When an actor receives a message, it can do one of these 3 things:

  • Create more actors
  • Send messages to other actors
  • Designate what to do with the next message

The first two bullet points are pretty straightforward, but the last one is interesting.
I said before that an actor can maintain a private state. “Designating what to do with the next message” basically means defining how this state will look like for the next message it receives. Or, more clearly, it’s how actors mutate state.

Let’s imagine we have an actor that behaves like a calculator and that its initial state is simply the number 0. When this actor receives the add(1) message, instead of mutating its original state, it designates that for the next message it receives, the state will be 1.

Fault tolerance

Erlang introduced the “let it crash” philosophy. The idea is that you shouldn’t need to program defensively, trying to anticipate all the possible problems that could happen and find a way to handle them, simply because there is no way to think about every single failure point.

What Erlang does is simply letting it crash, but make this critical code be supervised by someone whose only responsibility is to know what to do when this crash happens (like resetting this unit of code to a stable state), and what makes it all possible is the actor model.

Every code run inside a process (that is basically how Erlang calls its actors). This process is completely isolated, meaning its state is not going to influence any other process. We have a supervisor, that is basically another process (everything is an actor, remember?), that will be notified when the supervised process crashes and then can do something about it.

This makes it possible to create systems that “self heal”, meaning that if an actor gets to an exceptional state and crashes, by whatever reason, a supervisor can do something about it to try to put it in a consistent state again (and there are multiple strategies to do that, the most common being just to restart the actor with its initial state).

Distribution

Another interesting aspect of the actor model is that it doesn’t matter if the actor that I’m sending a message to is running locally or in another node.

Think about it, if an actor is just this unit of code with a mailbox and an internal state, and it just respond to messages, who cares in which machine it’s actually running? As long as we can make the message get there we are fine.
This allows us to create systems that leverage multiple computers and helps us to recover if one of them fail.

Next steps and other resources

This was a quick overview of the conceptual model that is the base of great languages like Erlang and Elixir and libraries like Akka (for the JVM) and Celluloid (for Ruby).

If I was successful in making you curious about how this model is implemented and used in the real world, this is the list of books that I read or am reading about this topic and can recommend:

And if you are interested in more details about the conceptual idea itself, I can’t recommend this video enough:

Interested in learning Kubernetes?

I just published a new book called Kubernetes in Practice, you can use the discount code blog to get 10% off.

Get fresh articles in your inbox

If you liked this article, you might want to subscribe. If you don't like what you get, unsubscribe with one click.