Skip to content

Docker Builder on VM

Docker Builder VM

"Docker Builder" tasks are a way to build and publish Docker Images to Docker Registries of your choice using a VM as build environment. In essence, a docker_builder is basically a task that is executed in a VM with pre-installed Docker. A docker_builder can be defined the same way as a task:

docker_builder:
  build_script: docker build --tag myrepo/foo:latest .
docker_builder:
  env:
    CIRRUS_ARCH: arm64
  build_script: docker build --tag myrepo/foo:latest .

Leveraging features such as Task Dependencies, Conditional Execution and Encrypted Variables with a Docker Builder can help building relatively complex pipelines. It can also be used to execute builds which need special privileges.

In the example below, a docker_builder will be only executed on a tag creation, once both test and lint tasks have finished successfully:

test_task: ...
lint_task: ...

docker_builder:
  only_if: $CIRRUS_TAG != ''
  depends_on: 
    - test
    - lint
  env:
    DOCKER_USERNAME: ENCRYPTED[...]
    DOCKER_PASSWORD: ENCRYPTED[...]
  build_script: docker build --tag myrepo/foo:$CIRRUS_TAG .
  login_script: docker login --username $DOCKER_USERNAME --password $DOCKER_PASSWORD
  push_script: docker push myrepo/foo:$CIRRUS_TAG

Example

For more examples please check how we use Docker Builder to build and publish Cirrus CI's Docker Images for Android.

Multi-arch builds

Docker Builder VM has QEMU pre-installed and is able to execute multi-arch builds via buildx. Add the following setup_script to enable buildx and then use docker buildx build instead of the regular docker build:

docker_builder:
  setup_script:
    - docker buildx create --name multibuilder
    - docker buildx use multibuilder
    - docker buildx inspect --bootstrap
  build_script: docker buildx build --platform linux/amd64,linux/arm64 --tag myrepo/foo:$CIRRUS_TAG .

Pre-installed Packages

For your convenience, a Docker Builder VM has some common packages pre-installed:

  • AWS CLI
  • Docker Compose
  • OpenJDK
  • Python
  • Ruby with Bundler

Under the hood

Under the hood a simple integration with Google Compute Engine is used and basically docker_builder is a syntactic sugar for the following compute_engine_instance configuration:

task:
  compute_engine_instance:
    image_project: cirrus-images
    image: family/docker-builder
    platform: linux
    cpu: 4
    memory: 16G
task:
  compute_engine_instance:
    image_project: cirrus-images
    image: family/docker-builder-arm64
    architecture: arm64
    platform: linux
    cpu: 4
    memory: 16G

You can check Packer templates of the VM image in cirruslabs/vm-images repository.

Layer Caching

Docker has the --cache-from flag which allows using a previously built image as a cache source. This way only changed layers will be rebuilt which can drastically improve performance of the build_script. Here is a snippet that uses the --cache-from flag:

# pull an image if available
docker pull myrepo/foo:latest || true
docker build --cache-from myrepo/foo:latest \
  --tag myrepo/foo:$CIRRUS_TAG \
  --tag myrepo/foo:latest .

Dockerfile as a CI environment

With Docker Builder there is no need to build and push custom containers so they can be used as an environment to run CI tasks in. Cirrus CI can do it for you! Just declare a path to a Dockerfile with the dockerfile field for your container or arm_container declarations in your .cirrus.yml like this:

efficient_task:
  container:
    dockerfile: ci/Dockerfile
    docker_arguments:
      foo: bar
  test_script: ...

inefficient_task:
  container:
    image: node:latest
  setup_script:
    - apt-get update
    - apt-get install build-essential
  test_script: ...
efficient_task:
  arm_container:
    dockerfile: ci/Dockerfile
    docker_arguments:
      foo: bar
  test_script: ...

inefficient_task:
  arm_container:
    image: node:latest
  setup_script:
    - apt-get update
    - apt-get install build-essential
  test_script: ...

Cirrus CI will build a container and cache the resulting image based on Dockerfile’s content. On the next build, Cirrus CI will check if a container was already built, and if so, Cirrus CI will instantly start a CI task using the cached image.

Under the hood, for every Dockerfile that is needed to be built, Cirrus CI will create a Docker Builder task as a dependency. You will see such build_docker_image_HASH tasks in the UI.

Danger of using COPY and ADD instructions

Cirrus only includes files directly added or copied into a container image in the cache key. But Cirrus is not recursively waking contents of folders that are being included into the image. This means that for a public repository a potential bad actor can create a PR with malicious scripts included into a container, wait for it to be cached and then reset the PR, so it looks harmless.

Please try to only COPY files by full path, e.g.:

FROM python:3

COPY requirements.txt /tmp/
RUN pip install --requirement /tmp/requirements.txt
Using with private GKE clusters

To use dockerfile with gke_container you first need to create a VM with Docker installed within your GCP project. This image will be used to perform building of Docker images for caching. Once this image is available, for example, by MY_DOCKER_VM name, you can use it like this:

gke_container:
  dockerfile: .ci/Dockerfile
  builder_image_name: MY_DOCKER_VM
  cluster_name: cirrus-ci-cluster
  zone: us-central1-a
  namespace: default

Please make sure your buidler image has gcloud configured as a credential helper.

If your builder image is stored in another project you can also specify it by using builder_image_project field. By default, Cirrus CI assumes builder image is stored within the same project as the GKE cluster.

Using with private EKS clusters

To use dockerfile with eks_container you need three things:

  1. Either create an AMI with Docker installed or use one like ECS-optimized AMIa. For example, MY_DOCKER_AMI.
  2. Create a role which has AmazonEC2ContainerRegistryFullAccess policy. For example, cirrus-builder.
  3. Create cirrus-cache repository in your Elastic Container registry and make sure user that aws_credentials are associated with has ecr:DescribeImages access to it.

Once all of the above requirement are met you can configure eks_container like this:

eks_container:
  region: us-east-2
  cluster_name: my-company-arm-cluster
  dockerfile: .ci/Dockerfile
  builder_image: MY_DOCKER_AMI
  builder_role: cirrus-builder # role for builder instance profile
  builder_instance_type: c7g.xlarge # should match the architecture below
  builder_subnet_ids: # optional, list of subnets from your default VPC to randomly choose from for scheduling the instance
    - ...
  builder_subnet_filters: # optional, map of filters to use for DescribeSubnets API call. Note to make sure Cirrus is given `ec2:DescribeSubnets`  
    - name: tag:Name
      values:
        - subnet1
        - subnet2
  architecture: arm64 # default is amd64

This will make Cirrus CI to check whether cirrus-cache repository in us-east-2 region contains a precached image for .ci/Dockerfile of this repository.

Windows Support

Docker builders also support building Windows Docker containers - use the platform and os_version fields:

docker_builder:
  platform: windows
  ...