Use NixOS configurations to provision Docker containers.
Read about the what and why in this blog post
DISCLAIMER: This project is no longer actively maintained and probably broken, if you're interested in fixing it, please fork and contact me: [email protected]
The easy way to do this is to use Vagrant.
When you have Vagrant installed:
git clone https://github.com/zefhemel/nix-docker.git
cd nix-docker
vagrant up
vagrant ssh
If all went well, you're now in a VM that has both Docker and Nix installed
and nix-docker
in its path.
At this point you need to connect to the VM and have nix setup the vagrant users own custom package stores. execute the follow
nix-channel --update
nix-env -i hello
You can now cd into the nix-docker/samples
directory to try to build some of the examples. Note that the ~/nix-docker
directory is mounted from your host machine, so you can edit your files with
your favorite editor and have them available within the VM.
To use nix-docker you need Nix installed as well as
Docker. Realistically, your best way to do this on
an Ubuntu (12.04 or 13.04) box. Once these are installed, installing
nix-docker
is as simple as:
git clone https://github.com/zefhemel/nix-docker.git
nix-env -f nix-docker/default.nix -i nix-docker
To build a stand-alone Docker image:
nix-docker -b -t my-image configuration.nix
This will build the configuration specified in configuration.nix
, have a look
in the samples/
directory for examples. It will produce a docker image named
my-image
which you can then run anywhere. Use username/my-image
to be able
to push them to the Docker index.
To build a host-mounted package:
nix-docker -t my-image configuration.nix
This will produce a Nix package (symlinked in the current directory in result
)
containing a script you can use to spawn the container using Docker, e.g.:
sudo -E ./result/sbin/docker-run
to run the container in the foreground, or:
sudo -E ./result/sbin/docker-run -d
to daemonize it. What the docker-run
script will do is check if there's
already a docker image available with the current image name and tag based on
the Nix build hash. If not, it will quickly build it first (these images take up
barely any space on disk). Then, it will boot up the container.
Distributing host-mounted packages is done by first copying the Nix closure resulting from the build to the target machine (when you do the build it will give you example commands to run):
nix-copy-closure root@targetmachine /nix/store/....
Then, you can spawn the container remotely with the script path provided in the output of the build command.