Running an Akka Cluster on Docker with Fig.sh

Akka got this amazing technology called clustering. It would be nice to simulate a real world with Docker. To orchistrate multiple docker instances fig.sh works well. The example can be found in on github - Akka test/fig.sh.

The different nodes:

seed node - Used for lookup. Two would be nice, but it is not very important for the example.
client 1 - A client for the cluster
client 2 - Could be different than client 1 but this is just for show.
backend 1 - A simple backend to do the work for the frontend clients.
backend 2 - Just like backend 1 but with persistence (because it is another cool technology from the Akka community).

The important parts are are:

fig.sh

seed:
  build: seed/target/docker
  ports:
   - "2551:2551"
  hostname: seed
backend1:
  build: backend1/target/docker
  expose:
   - "10101"
  links:
   - seed:seed
  hostname: backend1
backend2:
  build: backend2/target/docker
  expose:
   - "10101"
  links:
   - seed:seed
  hostname: backend2
client1:
  build: client1/target/docker
  expose:
   - "10101"
  links:
   - seed:seed
  hostname: client1
client2:
  build: client2/target/docker
  expose:
   - "10101"
  links:
   - seed:seed
  hostname: client2

System IP configuration

To fix problems with hostname and IP addresses added this to the system configuration (scala):

val address = InetAddress.getByName(System.getenv.get("HOSTNAME"))
val hostAddress = address.getHostAddress
println("\nHostAddress: " + hostAddress)

val config = ConfigFactory.parseString("akka.remote.netty.tcp.hostname=" + hostAddress).
    withFallback(ConfigFactory.load())

val system = ActorSystem("ClusterSystem", config)

Running the cluster

To run this (as it says in the README file):

install docker.io
install fig.sh

install sbt

run

sbt docker:stage
fig build
fig up

The drawback here is sbt docker:stage. Ideally there should be a DockerFile in every project instead of generating it. It would also be nice to have a super quick turn around for each of these projects. However that is another problem and maybe another blog post.