oci-build task

A Concourse task for building OCI images. Currently uses buildkit for building.

A stretch goal of this is to support running without privileged: true, though it currently still requires it.

• usage
  • image_resource
  • params
  • inputs
  • outputs
  • caches
  • run
• migrating from the docker-image resource
• differences from builder task
• example

usage

The task implementation is available as an image on Docker Hub at concourse/oci-build-task. (This image is built from Dockerfile using the oci-build task itself.)

This task implementation started as a spike to explore patterns around reusable tasks to hopefully lead to a proper RFC. Until that RFC is written and implemented, configuration is still done by way of providing your own task config as follows:

image_resource

First, your task needs to point to the oci-build-task image:

image_resource:
  type: registry-image
  source:
    repository: concourse/oci-build-task

params

Next, any of the following optional parameters may be specified:

(As a convention in the list below, all task parameters are specified with a leading $, in order to remind their environment variable nature, just like shell variables that one would use with the $VAR syntax. When specifying those in the params: YAML dictionary of a task definition though, the leading $ is irrelevant, as readers will notice in the examples below.)

• $CONTEXT (default .): the path to the directory to provide as the context for the build.

• $DOCKERFILE (default $CONTEXT/Dockerfile): the path to the Dockerfile to build.

• $BUILDKIT_SSH your ssh key location that is mounted in your Dockerfile. This is generally used for pulling dependencies from private repositories.

  For Example. In your Dockerfile, you can mount a key as

  RUN --mount=type=ssh,id=github_ssh_key pip install -U -r ./hats/requirements-test.txt

  Then in your Concourse YAML configuration:

  params:
  BUILDKIT_SSH: github_ssh_key=<PATH-TO-YOUR-KEY>

  Read more about ssh mount here.

• $BUILD_ARG_*: params prefixed with BUILD_ARG_ will be provided as build args. For example BUILD_ARG_foo=bar, will set the foo build arg as bar.

• $BUILD_ARGS_FILE (default empty): path to a file containing build args in the form foo=bar, one per line. Empty lines are skipped.

  Example file contents:

  EMAIL=me@yopmail.com
  HOW_MANY_THINGS=1
  DO_THING=false

• $BUILDKIT_SECRET_*: files with extra secrets which are made available via --mount=type=secret,id=.... See New Docker Build secret information for more information on build secrets.

  For example, running with BUILDKIT_SECRET_config=my-repo/config will allow you to do the following...

  RUN --mount=type=secret,id=config cat /run/secrets/config

• $BUILDKIT_SECRETTEXT_*: literal text of extra secrets to be made available via the same mechanism described for $BUILDKIT_SECRET_* above. The difference is that this is easier to use with credential managers:

  BUILDKIT_SECRETTEXT_mysecret=(( mysecret )) puts the content that (( mysecret )) expands to in /run/secrets/mysecret.

• $IMAGE_ARG_*: params prefixed with IMAGE_ARG_* point to image tarballs (i.e. docker save format) to preload so that they do not have to be fetched during the build. An image reference will be provided as the given build arg name. For example, IMAGE_ARG_base_image=ubuntu/image.tar will set base_image to a local image reference for using ubuntu/image.tar.

  This must be accepted as an argument for use; for example:

  ARG base_image
  FROM ${base_image}

• $IMAGE_PLATFORM: Specify the target platform to build the image for. For example IMAGE_PLATFORM=linux/arm64 will build the image for the Linux OS and arm64 architecture. By default, images will be built for the current worker's platform that the task is running on.

• $LABEL_*: params prefixed with LABEL_ will be set as image labels. For example LABEL_foo=bar, will set the foo label to bar.

• $LABELS_FILE (default empty): path to a file containing labels in the form foo=bar, one per line. Empty lines are skipped.

• $TARGET (default empty): a target build stage to build, as named with the FROM … AS <NAME> syntax in your Dockerfile.

• $TARGET_FILE (default empty): path to a file containing the name of the target build stage to build.

• $ADDITIONAL_TARGETS (default empty): a comma-separated (,) list of additional target build stages to build.

• $REGISTRY_MIRRORS (default empty): registry mirrors to use for docker.io.

• $UNPACK_ROOTFS (default false): unpack the image as Concourse's image format (rootfs/, metadata.json) for use with the image task step option.

• $OUTPUT_OCI (default false): outputs an OCI compliant image, allowing for multi-arch image builds when setting IMAGE_PLATFORM to multiple platforms. The image output format will be a directory when this flag is set to true.

• $BUILDKIT_ADD_HOSTS (default empty): extra host definitions for buildkit to properly resolve custom hostnames. The value is as comma-separated (,) list of key-value pairs (using syntax hostname=ip-address), each defining an IP address for resolving some custom hostname.

Note: this is the main pain point with reusable tasks - env vars are kind of an awkward way to configure a task. Once the RFC lands these will turn into a JSON structure similar to configuring params on a resource, and task params will become env instead.

inputs

There are no required inputs - your task should just list each artifact it needs as an input. Typically this is in close correlation with $CONTEXT:

params:
  CONTEXT: my-image

inputs:
- name: my-image

Should your build be dependent on multiple inputs, you may want to leave $CONTEXT as its default (.) and set an explicit path to the $DOCKERFILE:

params:
  DOCKERFILE: my-repo/Dockerfile

inputs:
- name: my-repo
- name: some-dependency

It might also make sense to place one input under another, like so:

params:
  CONTEXT: my-repo

inputs:
- name: my-repo
- name: some-dependency
  path: my-repo/some-dependency

Or, to fully rely on the default behavior and use path to wire up the context accordingly, you could set your primary context as path: . and set up any additional inputs underneath:

inputs:
- name: my-repo
  path: .
- name: some-dependency

outputs

A single output named image may be configured:

outputs:
- name: image

Use output_mapping to map this output to a different name in your build plan. This approach should be used if you're building multiple images in parallel so that they can have distinct names.

The output will contain the following files:

• image.tar: the OCI image tarball. This tarball can be uploaded to a registry using the Registry Image resource.

• digest: the digest of the OCI config. This file can be used to tag the image after it has been loaded with docker load, like so:

  docker load -i image/image.tar
  docker tag $(cat image/digest) my-name

If $UNPACK_ROOTFS is configured, the following additional entries will be created:

• rootfs/*: the unpacked contents of the image's filesystem.

• metadata.json: a JSON file containing the image's env and user configuration.

This is a Concourse-specific format to support using the newly built image for a subsequent task by pointing the task step's image option to the output, like so:

plan:
- task: build-image
  params:
    UNPACK_ROOTFS: true
  output_mapping: {image: my-built-image}
- task: use-image
  image: my-built-image

(The output_mapping here is just for clarity; alternatively you could just set image: image.)

Note: at some point Concourse will likely standardize on OCI instead.

caches

Caching can be enabled by caching the cache path on the task:

caches:
- path: cache

NOTE: the contents of --mount=type=cache directories are not cached, see https://github.com/concourse/oci-build-task/issues/87

run

Your task should run the build executable:

run:
  path: build

migrating from the docker-image resource

The docker-image resource was previously used for building and pushing a Docker image to a registry in one fell swoop.

The oci-build task, in contrast, only supports building images - it does not support pushing or even tagging the image. It can be used to build an image and use it for a subsequent task image without pushing it to a registry, by configuring $UNPACK_ROOTFS.

In order to push the newly built image, you can use a resource like the registry-image resource like so:

resources:
- name: my-image-src
  type: git
  source:
    uri: https://github.com/...

- name: my-image
  type: registry-image
  source:
    repository: my-user/my-repo

jobs:
- name: build-and-push
  plan:
  # fetch repository source (containing Dockerfile)
  - get: my-image-src

  # build using `oci-build` task
  #
  # note: this task config could be pushed into `my-image-src` and loaded using
  # `file:` instead
  - task: build
    privileged: true
    config:
      platform: linux

      image_resource:
        type: registry-image
        source:
          repository: concourse/oci-build-task

      inputs:
      - name: my-image-src
        path: .

      outputs:
      - name: image

      run:
        path: build

  # push using `registry-image` resource
  - put: my-image
    params: {image: image/image.tar}

differences from builder task

The builder task was a stepping stone that led to the oci-build task. It is now deprecated. The transition should be relatively smooth, with the following differences:

• The oci-build task does not support configuring $REPOSITORY or $TAG.
  • for running the image with docker, a digest file is provided which can be tagged with docker tag
  • for pushing the image, the repository and tag are configured in the registry-image resource
• The oci-build task has a more efficient caching implementation. By using buildkit directly we can make use of its local cache exporter/importer, which doesn't require a separate translation step for saving into the task cache.
• This task is written in Go instead of Bash, and has tests!

example

This repo contains an example.yml, which builds the image for the task itself:

fly -t dev execute -c example.yml -o image=. -p
docker load -i image.tar

That -p at the end is not a typo; it runs the task with elevated privileges.