Skip to content

Surgically remove layers from a Docker image (with a chainsaw)

Notifications You must be signed in to change notification settings

bh185140/layeremove

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 

Repository files navigation

Layeremove

A (very dirty) experiment to remove layers from a Docker image.

How to use

Requirements:

  • Docker Engine
  • Skopeo
  • Python 3

Build a trivial Docker image:

docker build example/ -t figlet

Export it to an OCI directory, so that we can mess with it:

OCI_DIR=src
skopeo copy docker-daemon:figlet:latest oci:$OCI_DIR

Run the horrible script:

./layeremove.py $OCI_DIR

Copy the updated image back to the Docker Engine:

skopeo copy oci:$OCI_DIR docker-daemon:figlet:layeremove

Check that the image still works:

docker run figlet:layeremove

Check the size and layers of the original and new image:

docker images figlet
docker history figlet:latest
docker history figlet:layeremove

How it works

The layeremove.py script will scan an OCI image, and it will remove some layers from that image. It will layers that contain #LAYEREMOVE# in their created_by attribute; so if you put RUN something #LAYEREMOVE# in a Dockerfile, the resulting layer will be removed. You can also pass layer hashes as extra command line arguments, e.g.:

./layeremove.py oci-src-dir sha256:abc123... sha256:def678...

It identifies the layers, then removes them from the image manifest as well as the image configuration. In the image configuration, it leaves a dummy layer in the history, which is why these layers will still show up in docker history.

Caveats

The code worked for me at least once, but there is no guarantee that it will work for you! 🤷🏻

Make sure that the OCI directory contains only one image. Otherwise, adapt things a bit.

The code deserves to be cleaned up a bit, but I think I'll go and have a beer instead! 🍺

About

Surgically remove layers from a Docker image (with a chainsaw)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 96.8%
  • Dockerfile 3.2%