This repo contains a connection plugin for Ansible that lets you configure Docker containers without needing to install an SSH server or Ansible itself into the container.

This is a work-in-progress.

Requires Docker 1.3.1. On Linux, you could probably get away with 1.3.0, but On OS X, you'll need Docker 1.3.1 and boot2docker 1.3.1.

• can't explicitly configure docker connection parameters
• when using boot2docker, TLS must be disabled (see boot2docker issues)
• sudo isn't supported
• su isn't supported

1. Create a connection_plugins directory next to your playbooks.
2. Copy the connection_plugins/docker.py file to the directory.

You may also need to modify the remote temporary directory (I had to do this with the ubuntu base image). Add the following to your ansible.cfg:

[defaults]
remote_tmp = /tmp

Because of a docker issue, if you are using boot2docker, you must disable TLS or the connection plugin will hang.

To disable TLS in boot2docker:

boot2docker ssh
sudo -i
/etc/init.d/docker stop
echo DOCKER_TLS="no" > /var/lib/boot2docker/profile
/etc/init.d/docker start

Your DOCKER_HOST should now point to port 2376 (the IP may not match the one below), and the DOCKER_TLS_VERIFY and DOCKER_CERT_PATH environment variables should not be deifned.

export DOCKER_HOST=tcp:https://192.168.59.103:2375
unset DOCKER_TLS_VERIFY
unset DOCKER_CERT_PATH

In your plays, add connection: docker. For example:

- name: configure my container
  connection: docker
  hosts: webcontainer

  tasks:
    - ...

You need a running container first, and you'll need to add the hostname to Ansible's inventory. One way to do it is to write an initial play that runs a container using the docker module and then uses the add_host module to add to the inventory, like this:

- name: start up a docker container
  hosts: localhost
  vars:
    base_image: ubuntu
    docker_hostname: webcontainer

  tasks:
    - name: start up a docker container by running bash
      local_action: docker image={{ base_image }} name={{ docker_hostname }} detach=yes tty=yes command=bash
    - name: add the host
      add_host: name={{ docker_hostname }}

- name: configure the web container
  hosts: webcontainer
  connection: docker
  tasks:
   - ...

Check out the ipython.yml playbook for an example that builds a container that runs an IPython notebook.

To build the example image, just do:

ansible-playbook ipython.yml

It will build a docker image called lorin/ipython

This will start the notebook and mount /notebooks to your current directory on the host.

docker run -d -p 8888:8888 -v `pwd`:/notebooks lorin/ipython

If you're running docker locally on Linux, browse to https://localhost:8888.

If you're using boot2docker, browse to the IP address associated with the DOCKER_HOST environment variable, e.g.: https://192.168.59.104:8888

Q: Does it work with boot2docker?

A: It does, but you need to disable TLS. See above.

Q: What's with the pre_image and final_image in the ipython.yml playbook?

A: The playbook creates the pre-image by starting a container from a base ubuntu image, configuring it with ansible, and then committing it. The playbook creates the final image from the pre-image using a Dockerfile to set things like the working directory, exposed ports, and command to run.

If you know how to do this one step, please let me know.

Q: Why does this repo contain custom docker and docker_image modules?

A: The upstream modules don't have support for boot2docker yet. I've submitted a pull request to get these added.