Actor-based continuations with Akka and Swarm

April 17, 2011

Swarm is a Scala continuations framework which allows computation to be moved from one machine to another. This example shows how it can be extended to use Akka actors to transport computation between actors.

Transporter hierarchy
Transporter hierarchy

To use Swarm, Transporter.transport must be implemented to deliver continuations between nodes. For Akka, this means sending each continuation as a message.

case class ActorLocation(val name: String,
                         val host: String,
                         val port: Short) extends Location

object AkkaTransporter extends Transporter {
  override def transport(f: (Unit => Bee), destination: Location) {
    destination match {
      case ActorLocation(name, host, port) =>
        remote.actorFor(name, host, port) ! f

In the Swarm API, continuations are passed as functions of Unit => Bee to destinations represented by implementations of Location. Here, a destination is identified by the name, host, and port of a remote actor.

A sample actor is used to indicate when it has control of the continuation.

class ContinuationActor extends Actor {

  implicit val tx: Transporter = AkkaTransporter

  def receive = {
    case f: (Unit => Bee) =>
      println("Entering actor uuid " + self.uuid)
      println("Extiting actor uuid " + self.uuid)

The actor has an implicit Transporter, which is used by Swarm behind the scenes.

To demonstrate sending continuations between actors, two main classes are used: one to set up the remote actors, and one to kick off the continuation.

object RemoteActorStarter {
  def main(args: Array[String]) {
    remote.start("localhost", 2552)
    remote.register("continuationActor1", actorOf[ContinuationActor])
    remote.register("continuationActor2", actorOf[ContinuationActor])

object AkkaTest {
  def main(args: Array[String]) {
    implicit val tx: Transporter = AkkaTransporter

  def f(u: Unit): Bee@swarm = {
    println("Here I am!")
    Swarm.moveTo(ActorLocation("continuationActor1", "localhost", 2552))
    println("I'm here now!")
    Swarm.moveTo(ActorLocation("continuationActor2", "localhost", 2552))
    println("Over here!")

RemoteActorStarter registers two ContinuationActor instances named continuationActor1 and continuationActor2. AkkaTest then uses Swarm to spawn the continuation defined as the method f. This will print a message, transfer the code to the first actor using the implicit AkkaTransporter, print another message, transfer the code to the second actor, and print a third message.

The flow of execution of this code looks something like the following.

Flow of execution
Flow of execution

The output of the above main methods shows the code moving from actor to actor.


[info] Running swarm.akka.AkkaTest
Here I am!


[info] Running swarm.akka.RemoteActorStarter

Entering actor uuid 079c1a40-6853-11e0-a43c-001e6522d06c
I'm here now!
Extiting actor uuid 079c1a40-6853-11e0-a43c-001e6522d06c

Entering actor uuid 07a25bd0-6853-11e0-a43c-001e6522d06c
Over here!
Extiting actor uuid 07a25bd0-6853-11e0-a43c-001e6522d06c


This project depends on Swarm being installed in your local Maven repository.

  1. Download Swarm:

    $ git clone
  2. Generate a Maven POM:

    $ cd Swarm
    $ sbt update make-pom
  3. Install in your local Maven repository:

    $ mkdir -p ~/.m2/repository/swarm-dpl/swarm_2.8.1/1.0-SNAPSHOT
    $ cp target/scala_2.8.1/swarm_2.8.1-1.0-SNAPSHOT.* \