Docker-in-Docker in Vagrant

My continuously-evolving development box.

First time use:

$ git clone https://github.com/themattrix/vagrant-dind.git
$ cd vagrant-dind
$ # Optionally create a .ssh directory here. See "SSH Config" below.
$ vagrant up
$ vagrant ssh

At this point you'll be sitting in a tmux session, which is running in a single-session docker container. Exiting or disconnecting the session will cause the container to be destroyed. When the container is destroyed, only three things survive:

The /repos directory, which is provided as a work area.
The /home/.ssh directory, which allows for persistent changes to things like authorized_keys and known_hosts.
Docker images and containers.

In this way, you can be sure to always start with a clean environment, but your work area and any docker images and containers will persist between sessions.

Architecture

$ vagrant up
$ vagrant ssh

.--[ Vagrant: boot2docker ]----------------------.
|                                                |
| <autorun.sh>                                   |
|                                                |
| .--[ docker (privileged): ubuntu:14.04 ]------.|
| |  _______________________________________[0] ||
| |  root@c23311ce9a9f [/]                      ||
| |  #                                          ||
| '---------------------------------------------'|
'------------------------------------------------'

SSH Config

In order to make SSH operations useful inside the development container, you may optionally provide a .ssh directory to it by placing it at the root of the vagrant-dind directory. This will then be copied into the VM and provided as a volume to the development container.

How do I...

...update my SSH config inside the development container?

Assuming you've updated the vagrant-dind/.ssh directory as described in the SSH Config section above, the solution is to either reload or re-provision the Vagrant box.

...install additional packages into the development container?

In your current session:

apt-get install <package>

For all future sessions:

Modify vagrant-dind/develop/Dockerfile to install the package. It's probably easiest to include it in one of the existing apt-get install commands.

Reload or re-provision the Vagrant box.

...clone a repo and get started changing it?

If you'd like the repo to stick around for longer than the current session, clone the repo into /repos. This directory is mapped to a persistent area in the boot2docker VM and will survive until the VM is deleted.

...add or change the persistent directories?

vagrant-dind/autorun.sh contains the docker run command which lauches the development container when you SSH into the Vagrant box. This command contains several volumes mounted into the container (in the form -v /vm/dir:/container/dir). Feel free to add as many additional volumes as you'd like.

Be sure to consider the following:

Since the VM is a boot2docker instance, only the /mnt/sda1 directory persists across reboots of the VM. If you'd like your volume to persist across reboots of the VM as well, then consider making the source of the mount a subdirectory of /mnt/sda1. (This is exactly what the /repos volume does.)

Like the other permanent changes, this will require Vagrant to either be reloaded or re-provisioned.

...run a one-time command in the boot2docker VM instead of automatically entering the development container?

vagrant ssh -c '<your-command>'

If your development container ever gets messed up, this can be a life saver.

...develop outside of the development container, but run things inside it?

My current technique is to set my IDE (PyCharm, in my case) to deploy files to the Vagrant box in a subdirectory of /mnt/sda1/persist/repos/. I then have a Vagrant ssh session open, from which I can run anything from the command line.