Akka HTTP Actor Per Request

This project provides an example akka application that uses the Actor per request model. The project is based on NET-A-PORTER spray-actor-per-request repository.

Why would you want to spin up an Actor for each HTTP request?

• Easily manage a tree of request scoped Actors in the application core
• The per request actor can clean them up in the event of a timeouts and failures
• Leverage the actor supervision hierarchy to propogate failures up to the RequestContext, so you can return useful error responses
• Promote Tell, Don't Ask
• Using request scoped Actors in the application core can make it easier to use tell (!) over ask (?)

Resources:

• Scala Exchange Presentation (video)
• Mathias describes the actor per request approach against others. (mailing list)

Example App

Overview

This example application provides an API to get a list of pets with their owners. There are already two actors that provide a list of pets and a list of animals and the responsibility of this application is to simply aggregate these two together.

Our application is made up of three modules:

• Application Core - The core module contains the business logic for our application. In this example this is how we aggregate pets with their owners.
• Routing - The routing module contains our akka routing which describes our RESTful endpoints. It also contains our PerRequest actor which bridges the gap between the routing and the core modules and contains the piece of code this example project aims to demonstrate.
• Clients - Our code to consume two existing actors that provide us with a list of pets and a list of owners. These actors could use databases, RESTful APIs, etc. It doesn't really matter for the purposes of this example.

Ideally modules these would be in separate sub-projects to prevent unnecessary compile time dependencies, however for simplicity purposes they are just kept in separate packages in this example.

Running

sbt run

Successful request

If we request the pet Lassie:

GET http://localhost:38080/pets?names=Lassie

We get a successful response:

{
  "pets": [
    {
      "name": "Lassie",
      "owner": {
        "name": "Jeff Morrow"
      }
    }
  ]
}

In this scenario, any request scoped actors in the application core are stopped by the PerRequest actor.

Request Timeouts

Tortoises are slow. If we request a tortoise the PetClient will not reply to our application core quick enough. The timeout of 2 seconds in our PerRequest actor will happen first.

GET http://localhost:38080/pets?names=Tortoise

{
  "message": "Request timeout"
}

In this scenario, any request scoped actors in the application core are stopped by the PerRequest actor.

Validation

You shouldn't keep a Lion as a pet. Quite frankly they are too dangerous.

GET http://localhost:38080/pets?names=Lion

{
  "message": "Lions are too dangerous!"
}

In this scenario, our application core returns a generic Validation message to our PerRequest actor to complete. Any request scoped actors in the application core are stopped by the PerRequest actor.

Failures

What about unexpected failures? There is a "bug" in our application core that throws a PetOverflowException if we request too many pets:

GET http://localhost:38080/pets?names=Lassie,Tweety,Tom

{
  "message": "PetOverflowException: OMG. Pets. Everywhere."
}

Any failures that not handled by the application core can be escalated up to the supervision strategy in our PerRequest actor. The PerRequest actor is too generic to recover from any business logic failures, so it will simply handle all failures by completing the request with an error response. Any request scoped actors in the application core are stopped by the PerRequest actor.