lazy-docker-build-push-action

Lazily build Docker images to content-identifying-hash based tags.

This aims to speed up build times by avoiding expensive rebuilding of slow-moving layers, such as those containing external dependencies. While Docker layer caching can achieve this in some cases, doing so requires storing and loading the relevant caches which can prove almost as time consuming as rebuilding the layers. This Action uses stable hashing of inputs to achieve speed-ups without full caching.

The actual Docker build is handled by docker/build-push-action and the majority of the action's arguments are passed directly to that action.

The following are automatically included in the identifying hash:

• Annotations
• Attestation, provenance and SBOM
• Build args
• Build cgroup-parent and ulimit
• Build contexts
• Build target
• Labels
• the Dockerfile itself (no attempt is made to parse it)

Any other inputs (including e.g: files referenced in COPY statements) must be specified manually.

Tags for the built image are inferred automatically from those passed. For each input tag, the image name is extracted and a tag name of the form content-hash-<hash> is appended, so user/app:latest would additionally be tagged as user/app:content-hash-1234abc....

Hashes are stable for given inputs when using a pinned version of this Action. Counterexamples to this are considered bugs. While effort will be made to maintain hash stability, new releases may include changes to the hash. Since this Action is in effect a cache, such changes will not be considered breaking.

Arguments:

• tags (required): A list of image names (user/app) or full tags (user/app:latest) to apply to the resulting image. Passed through to docker/build-push-action with the content-identifying tags applied.

• pull or load: If either of these are set to true then any existing image matching the computed hash will be made available locally. Otherwise the action will attempt to avoid fetching images unless a build is needed.

  TODO: implement this.

• extra-inputs (optional): A list of globs of files to include in the content hash.

Other arguments to docker/build-push-action are passed through unchanged.

Outputs:

• image-name: The name of the image (e.g: user/app), reflected from the parsed tags input for convenience. Where several names are encountered, this is the first from the list for which the desired image (now) exists.

  TODO: implement handling for one of the names not existing

• image-tag: The tag of the image (e.g: content-hash-1234abc...), as constructed from the hashed inputs.

• image-name-tag: The full name of the image, <image-name>:<image-tag> (e.g: user/app:content-hash-1234abc...), for convenience.

• tag-existed: Whether or not the tag already existed (for any of the input names) and thus whether or not a build occurred.

Example usage:

on:
  push:
    branches:
      - main

jobs:
  docker:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Login to Docker Hub
        uses: docker/login-action@v3
        with:
          username: ${{ secrets.DOCKERHUB_USERNAME }}
          password: ${{ secrets.DOCKERHUB_TOKEN }}

      - name: Build base layer
        uses: PeterJCLaw/lazy-docker-build-push-action@v0.2
        id: build-base
        with:
          context: .
          push: true
          tags: user/base:latest
          extra-inputs: requirements*.txt

      - name: Build and push
        uses: docker/build-push-action@v5
        with:
          context: .
          push: true
          tags: user/app:latest
          build-args: |
            BASE_IMAGE=${{ steps.build-base.outputs.image-name-tag }}

See also the examples directory.