This is docker compose config inspired by [blog post by Sebastien Goasguen] (http://sebgoa.blogspot.com/2015/03/1-command-to-mesos-with-docker-compose.html).

It uses official images from mesosphere and host networking to enable interactions with tasks over real network. It also enables docker containerizer. No state is persisted between runs so feel free to start over if you screwed cluster state or something.

• Mesos 1.0.1
• Marathon 1.3.0
• Chronos 2.4.0 (optional)
• Metronome 0.2.0 (optional)

Note that you need docker-compose 1.6.0 or newer:

• https://github.com/docker/compose

This project should work out of the box with docker machine. It requires host networking to work, so docker for mac won't work.

If you have docker built dynamically, which is the case on most distros, you should download and bind-mount statically linked docker client:

curl -sL https://get.docker.com/builds/Linux/x86_64/docker-1.11.1.tgz | \
  tar xv --strip-components 1 -C /usr/local/bin

This downloads docker 1.11.1 and installs binaries into /usr/local/bin.

You have to specify DOCKER_IP env variable in order to make Mesos work properly. The default value is 127.0.0.1 and it should work if you have Docker daemon running locally.

If you use docker-machine you can do the following, assuming dev is your machine's name:

export DOCKER_IP=$(docker-machine ip dev)

Run your cluster in the background (equivalent to docker-compose up -d):

make start

That's it, use the following URLs:

• http://$DOCKER_IP:5050/ for Mesos master UI
• http://$DOCKER_IP:5051/ for the first Mesos slave UI
• http://$DOCKER_IP:5052/ for the second Mesos slave UI
• http://$DOCKER_IP:8080/ for Marathon UI
• http://$DOCKER_IP:8888/ for Chronos UI
• http://$DOCKER_IP:8081/ for Metronome UI (REST API)

If you want to run your cluster in foreground to see logs and be able to stop it with Ctrl+C, use the following (equivalent to docker-compose up):

make run

To kill your cluster and wipe all state to start fresh:

make destroy

This is equivalent to issuing docker-compose stop && docker-compose rm -f -v.

Since Mesos slaves run in containers and everything is killed when container is stopped, there is no possibility to test task recovery. You might think that docker container recovery is still possible, but you are wrong:

• https://issues.apache.org/jira/browse/MESOS-3573

In docker-compose.yml you can uncomment additional containers.

Uncomment slave-two section to get access to the second mesos-slave. Port ranges of two mesos-slaves are separated so it shouldn't be an issue.

Note that both slaves share same memory so you might see OOM if you allocate and use more memory than your docker host has.

Uncomment chronos section to get access to Chronos framework.

Uncomment metronome section to get access to Metronome framework.

A sample app definition in YAML is included for your convenience in apps dir. Use the following command to deploy it:

make APP=example deploy

Feel free to add your apps to apps and deploy them similarly.