Search code examples
dockerdocker-registry

How to push only whats changed with Docker push?

A. Here is how I created the image:

  1. Got latest Ubuntu image
  2. Ran as container and attached to it
  3. Cloned source code from git inside docker container
  4. Tagged and pushed docker image to my registry

B. And from a different machine I pulled, changed and pushed it by doing:

  1. Docker pull from the registry
  2. Start container with the pulled image and attach to it
  3. Change something in the cloned git directory
  4. Stop container, tag and push it to registry

Now the issue I'm seeing is that every time B is repeated it will try to upload ~600MB (which is the public image layer) to the registry which takes a long time in my case.

Is there any way to avoid uploading the whole 600MB and instead pushing the only directory that has changed?

What am I doing wrong? How do you guys use docker for frequent pushes?

Solution

  • Docker will only push changed layers, so it looks as though something in your workflow is not quite right. It will be much clearer if you use a Dockerfile, as each instruction explicitly creates a layer, but even with docker commit the results should be the same.

    Example - run a container from the ubuntu image and run apt-get update and then commit the container to a new image. Now run docker history and you'll see the new images adds a layer on top of the bash image, which has the additional state from running the APT update:

    > docker history sixeyed/temp1

IMAGE               CREATED              CREATED BY SIZE                COMMENT
2d98a4114b7c        About a minute ago   /bin/bash                                       22.2 MB
14b59d36bae0        7 months ago         /bin/sh -c #(nop) CMD ["/bin/bash"]             0 B
<missing>           7 months ago         /bin/sh -c sed -i 's/^#\s*\(deb.*universe\)$/   1.895 kB
<missing>           7 months ago         /bin/sh -c echo '#!/bin/sh' > /usr/sbin/polic   194.5 kB
<missing>           7 months ago         /bin/sh -c #(nop) ADD file:620b1d9842ebe18eaa   187.8 MB

    In this case, the diff between ubuntu and my temp1 image is the 22MB layer 2d98.

    Now if I run a new container from temp1, create an empty file and run docker commit to create a new image, the new layer only has the changed file:

    > docker history sixeyed/temp2
IMAGE               CREATED              CREATED BY SIZE                COMMENT
e9ea4b4963e4        45 seconds ago       /bin/bash                                       0 B
2d98a4114b7c        About a minute ago   /bin/bash                                       22.2 MB
14b59d36bae0        7 months ago         /bin/sh -c #(nop) CMD ["/bin/bash"]             0 B
<missing>           7 months ago         /bin/sh -c sed -i 's/^#\s*\(deb.*universe\)$/   1.895 kB
<missing>           7 months ago         /bin/sh -c echo '#!/bin/sh' > /usr/sbin/polic   194.5 kB
<missing>           7 months ago         /bin/sh -c #(nop) ADD file:620b1d9842ebe18eaa   187.8 MB

    When I push the first image, only the 22MB layer will get uploaded - the others are mounted from ubuntu, which is already in the Hub. If I push the second image, only the changed layer gets pushed - the temp1 layer is mounted from the first push:

    > docker push sixeyed/temp2
The push refers to a repository [docker.io/sixeyed/temp2]
f741d3d3ee9e: Pushed
64f89772a568: Mounted from sixeyed/temp1
5f70bf18a086: Mounted from library/ubuntu
6f32b23ac95d: Mounted from library/ubuntu
14d918629d81: Mounted from library/ubuntu
fd0e26195ab2: Mounted from library/ubuntu

    So if your pushes are uploading 600MB, you're either making 600MB changes to the image, or your workflow is preventing Docker using layers correctly.