Undocker

Converts a Docker image (a bunch of layers) to a flattened "rootfs" tarball.

Why?

Docker images became a popular way to distribute applications with their dependencies. However, Docker itself is not the best runtime environment. At least not for everyone.

Undocker bridges the gap between application images (in docker image format) and container runtimes: now you can run a Docker image with old-fashioned tools: lxc, systemd-nspawn or systemd itself.

Usage -- extract docker image

Download busybox docker image from docker hub and convert it to a rootfs:

skopeo copy docker://docker.io/busybox:latest docker-archive:busybox.tar
undocker busybox.tar - | tar -xv

Almost the same can be done with a combination of docker pull and docker save.

Usage -- systemd-nspawn example

Start with systemd-nspawn:

systemd-nspawn -D $PWD busybox httpd -vfp 8080

Usage -- plain old systemd

systemd-run \
  --wait --pty --collect --service-type=exec \
  -p PrivateUsers=true \
  -p DynamicUser=yes \
  -p ProtectProc=invisible \
  -p RootDirectory=$PWD \
  -- busybox httpd -vfp 8080

Good things like PrivateUsers, DynamicUser, ProtectProc and other systemd protections are available, just like to any systemd unit.

Notes & gotchas

unocker does not magically enable you to run containers from the internet. Many will need significant tuning or not work at all; one will still need to understand what's inside.

Contributions

I want this project to be useful for others, but not become a burden for me. If undocker fails for you (for example, you found a container that undocker cannot extract, or extracts incorrectly), you are on the hook to triage and fix it.

Therefore, the following contributions are welcome:

• Pull rquests (diffs) with accompanying tests.
• Documentation.

Issues without accompanying patches will most likely be rejected, with one exception: reports about regressions do not have to contain patches, but a failing commit is mandatory, and a failing test case is highly appreciated.