Docker builds images incrementally. Every line in a Dockerfile will generate a new image that builds on top of the last one. This can be really handy if something is not right in your build.

Since version 18.09 Docker has added a new backend for building images, buildkit. Unfortunately, buildkit does not let you view the IDs of the intermediate containers it uses under the hood. To work around that, you can opt out of buildkit by running a build with buildkit disabled:

DOCKER_BUILDKIT=0 docker build --pull --rm -t myproject:latest .

You should now see the IDs of the intermediate containers:

Sending build context to Docker daemon  87.84MB
Step 1/16 : FROM node:16.15.1-alpine3.16 AS development
16.15.1-alpine3.16: Pulling from library/node
Digest: sha256:c785e617c8d7015190c0d41af52cc69be8a16e3d9eb7cb21f0bb58bcfca14d6b
Status: Image is up to date for node:16.15.1-alpine3.16
 ---> e548f8c9983f
Step 2/16 : WORKDIR /usr/src/app
 ---> Using cache
 ---> 34e5c9bdb910
Step 3/16 : COPY package*.json ./
 ---> Using cache
 ---> 626e4ae998fc
Step 4/16 : RUN npm install glob rimraf
 ---> Using cache
 ---> 2d036b8354e0
Step 5/16 : RUN npm install
 ---> Using cache
 ---> 948709b4957f      <-- HERE
Step 6/16 : COPY . .
...

As mentioned, these IDs are valid docker images, so you can just launch them and attach a shell like every other image:

docker run -ti --rm 948709b4957f

If you're not seeing a regular shell, but a Node.js REPL for example, this might be because the ENTRYPOINT of that image was set to the binary of that REPL. To work around that, you can override the entrypoint:

docker run -ti --rm --entrypoint=/bin/sh 948709b4957f

When is this helpful?

If your build fails at a particular step, you can attach a shell to the last working step, inspect the filesystem, and execute the failing command manually.

That's all!

This is post 039 of #100DaysToOffload.


Continue Reading