Building Docker images with GitLab CI/CD

GitLab CI/CD allows you to use Docker Engine to build and test Docker-based projects.

One of the new trends in Continuous Integration/Deployment is to:

  1. Create an application image.
  2. Run tests against the created image.
  3. Push image to a remote registry.
  4. Deploy to a server from the pushed image.

It's also useful when your application already has the Dockerfile that can be used to create and test an image:

docker build -t my-image dockerfiles/
docker run my-image /script/to/run/tests
docker tag my-image my-registry:5000/my-image
docker push my-registry:5000/my-image

This requires special configuration of GitLab Runner to enable docker support during jobs.

Runner Configuration

There are three methods to enable the use of docker build and docker run during jobs, each with their own tradeoffs.

An alternative to using docker build is to use kaniko. This avoids having to execute a runner in privileged mode.

TIP: Tip: To see how Docker and GitLab Runner are configured for shared runners on GitLab.com, see GitLab.com shared runners.

Use shell executor

The simplest approach is to install GitLab Runner in shell execution mode. GitLab Runner then executes job scripts as the gitlab-runner user.

  1. Install GitLab Runner.

  2. During GitLab Runner installation select shell as method of executing job scripts or use command:

    sudo gitlab-runner register -n \
      --url https://gitlab.com/ \
      --registration-token REGISTRATION_TOKEN \
      --executor shell \
      --description "My Runner"
  3. Install Docker Engine on server.

    For more information how to install Docker Engine on different systems, check out the Supported installations.

  4. Add gitlab-runner user to docker group:

    sudo usermod -aG docker gitlab-runner
  5. Verify that gitlab-runner has access to Docker:

    sudo -u gitlab-runner -H docker info

    You can now verify that everything works by adding docker info to .gitlab-ci.yml:

    before_script:
      - docker info
    
    build_image:
      script:
        - docker build -t my-docker-image .
        - docker run my-docker-image /script/to/run/tests
  6. You can now use docker command (and install docker-compose if needed).

By adding gitlab-runner to the docker group you are effectively granting gitlab-runner full root permissions. For more information please read On Docker security: docker group considered harmful.

Use Docker-in-Docker workflow with Docker executor

The second approach is to use the special Docker-in-Docker (dind) Docker image with all tools installed (docker) and run the job script in context of that image in privileged mode.

docker-compose is not part of Docker-in-Docker (dind). To use docker-compose in your CI builds, follow the docker-compose installation instructions.

DANGER: Warning: By enabling --docker-privileged, you are effectively disabling all of the security mechanisms of containers and exposing your host to privilege escalation which can lead to container breakout. For more information, check out the official Docker documentation on Runtime privilege and Linux capabilities.

Docker-in-Docker works well, and is the recommended configuration, but it is not without its own challenges:

  • When using Docker-in-Docker, each job is in a clean environment without the past history. Concurrent jobs work fine because every build gets its own instance of Docker engine so they don't conflict with each other. But this also means that jobs can be slower because there's no caching of layers.

  • By default, Docker 17.09 and higher uses --storage-driver overlay2 which is the recommended storage driver. See Using the overlayfs driver for details.

  • Since the docker:19.03.12-dind container and the runner container don't share their root file system, the job's working directory can be used as a mount point for child containers. For example, if you have files you want to share with a child container, you may create a subdirectory under /builds/$CI_PROJECT_PATH and use it as your mount point (for a more thorough explanation, check issue #41227):

    variables:
      MOUNT_POINT: /builds/$CI_PROJECT_PATH/mnt
    
    script:
      - mkdir -p "$MOUNT_POINT"
      - docker run -v "$MOUNT_POINT:/mnt" my-docker-image

An example project using this approach can be found here: https://gitlab.com/gitlab-examples/docker.

In the examples below, we are using Docker images tags to specify a specific version, such as docker:19.03.12. If tags like docker:stable are used, you have no control over what version is used. This can lead to unpredictable behavior, especially when new versions are released.

TLS enabled

The Docker daemon supports connection over TLS and it's done by default for Docker 19.03.12 or higher. This is the suggested way to use the Docker-in-Docker service and GitLab.com shared runners support this.

GitLab Runner 11.11 or later is required, but it is not supported if GitLab Runner is installed using the Helm chart. See the related issue for details.

  1. Install GitLab Runner.

  2. Register GitLab Runner from the command line to use docker and privileged mode:

    sudo gitlab-runner register -n \
      --url https://gitlab.com/ \
      --registration-token REGISTRATION_TOKEN \
      --executor docker \
      --description "My Docker Runner" \
      --docker-image "docker:19.03.12" \
      --docker-privileged \
      --docker-volumes "/certs/client"

    The above command registers a new runner to use the special docker:19.03.12 image, which is provided by Docker. Notice that it's using the privileged mode to start the build and service containers. If you want to use Docker-in-Docker mode, you always have to use privileged = true in your Docker containers.

    This also mounts /certs/client for the service and build container, which is needed for the Docker client to use the certificates inside of that directory. For more information on how Docker with TLS works, check https://hub.docker.com/_/docker/#tls.

    The above command creates a config.toml entry similar to this:

    [[runners]]
      url = "https://gitlab.com/"
      token = TOKEN
      executor = "docker"
      [runners.docker]
        tls_verify = false
        image = "docker:19.03.12"
        privileged = true
        disable_cache = false
        volumes = ["/certs/client", "/cache"]
      [runners.cache]
        [runners.cache.s3]
        [runners.cache.gcs]
  3. You can now use docker in the build script (note the inclusion of the docker:19.03.12-dind service):

    image: docker:19.03.12
    
    variables:
      # When using dind service, we need to instruct docker to talk with
      # the daemon started inside of the service. The daemon is available
      # with a network connection instead of the default
      # /var/run/docker.sock socket. Docker 19.03 does this automatically
      # by setting the DOCKER_HOST in
      # https://github.com/docker-library/docker/blob/d45051476babc297257df490d22cbd806f1b11e4/19.03/docker-entrypoint.sh#L23-L29
      #
      # The 'docker' hostname is the alias of the service container as described at
      # https://docs.gitlab.com/ee/ci/docker/using_docker_images.html#accessing-the-services.
      #
      # If you're using GitLab Runner 12.7 or earlier with the Kubernetes executor and Kubernetes 1.6 or earlier,
      # the variable must be set to tcp://localhost:2376 because of how the
      # Kubernetes executor connects services to the job container
      # DOCKER_HOST: tcp://localhost:2376
      #
      # Specify to Docker where to create the certificates, Docker will
      # create them automatically on boot, and will create
      # `/certs/client` that will be shared between the service and job
      # container, thanks to volume mount from config.toml
      DOCKER_TLS_CERTDIR: "/certs"
    
    services:
      - docker:19.03.12-dind
    
    before_script:
      - docker info
    
    build:
      stage: build
      script:
        - docker build -t my-docker-image .
        - docker run my-docker-image /script/to/run/tests

TLS disabled

Sometimes there are legitimate reasons why you might want to disable TLS. For example, you have no control over the GitLab Runner configuration that you are using.

Assuming that the runner's config.toml is similar to:

[[runners]]
  url = "https://gitlab.com/"
  token = TOKEN
  executor = "docker"
  [runners.docker]
    tls_verify = false
    image = "docker:19.03.12"
    privileged = true
    disable_cache = false
    volumes = ["/cache"]
  [runners.cache]
    [runners.cache.s3]
    [runners.cache.gcs]

You can now use docker in the build script (note the inclusion of the docker:19.03.12-dind service):

image: docker:19.03.12

variables:
  # When using dind service we need to instruct docker, to talk with the
  # daemon started inside of the service. The daemon is available with
  # a network connection instead of the default /var/run/docker.sock socket.
  #
  # The 'docker' hostname is the alias of the service container as described at
  # https://docs.gitlab.com/ee/ci/docker/using_docker_images.html#accessing-the-services
  #
  # If you're using GitLab Runner 12.7 or earlier with the Kubernetes executor and Kubernetes 1.6 or earlier,
  # the variable must be set to tcp://localhost:2375 because of how the
  # Kubernetes executor connects services to the job container
  # DOCKER_HOST: tcp://localhost:2375
  #
  DOCKER_HOST: tcp://docker:2375
  #
  # This will instruct Docker not to start over TLS.
  DOCKER_TLS_CERTDIR: ""

services:
  - docker:19.03.12-dind

before_script:
  - docker info

build:
  stage: build
  script:
    - docker build -t my-docker-image .
    - docker run my-docker-image /script/to/run/tests

Enable registry mirror for docker:dind service

When the Docker daemon starts inside of the service container, it uses the default configuration. You may want to configure a registry mirror for performance improvements and ensuring you don't reach DockerHub rate limits.

Inside .gitlab-ci.yml

You can append extra CLI flags to the dind service to set the registry mirror:

services:
   - name: docker:19.03.13-dind
     command: ["--registry-mirror", "https://registry-mirror.example.com"] # Specify the registry mirror to use.
DinD service defined inside of GitLab Runner configuration

Introduced in GitLab Runner 13.6.

If you are an administrator of GitLab Runner and you have the dind service defined for the Docker executor, or the Kubernetes executor you can specify the command to configure the registry mirror for the Docker daemon.

Docker:

[[runners]]
  ...
  executor = "docker"
  [runners.docker]
    ...
    privileged = true
    [[runners.docker.services]]
      name = "docker:19.03.13-dind"
      command = ["--registry-mirror", "https://registry-mirror.example.com"]

Kubernetes:

[[runners]]
  ...
  name = "kubernetes"
  [runners.kubernetes]
    ...
    privileged = true
    [[runners.kubernetes.services]]
      name = "docker:19.03.13-dind"
      command = ["--registry-mirror", "https://registry-mirror.example.com"]
Docker executor inside GitLab Runner configuration

If you are an administrator of GitLab Runner and you want to use the mirror for every dind service, update the configuration to specify a volume mount.

For example, if you have a /opt/docker/daemon.json file with the following content:

{
  "registry-mirrors": [
    "https://registry-mirror.example.com"
  ]
}

Update the config.toml file to mount the file to /etc/docker/daemon.json. This would mount the file for every container that is created by GitLab Runner. The configuration will be picked up by the dind service.

[[runners]]
  ...
  executor = "docker"
  [runners.docker]
    image = "alpine:3.12"
    privileged = true
    volumes = ["/opt/docker/daemon.json:/etc/docker/daemon.json:ro"]
Kubernetes executor inside GitLab Runner configuration

Introduced in GitLab Runner 13.6.

If you are an administrator of GitLab Runner and you want to use the mirror for every dind service, update the configuration to specify a ConfigMap volume mount.

For example, if you have a /tmp/daemon.json file with the following content:

{
  "registry-mirrors": [
    "https://registry-mirror.example.com"
  ]
}

Create a ConfigMap with the content of this file. You can do this with a command like:

kubectl create configmap docker-daemon --namespace gitlab-runner --from-file /tmp/daemon.json

NOTE: Note: Make sure to use the namespace that GitLab Runner Kubernetes executor uses to create job pods in.

After the ConfigMap is created, you can update the config.toml file to mount the file to /etc/docker/daemon.json. This update mounts the file for every container that is created by GitLab Runner. The configuration is picked up by the dind service.

[[runners]]
  ...
  executor = "kubernetes"
  [runners.kubernetes]
    image = "alpine:3.12"
    privileged = true
    [[runners.kubernetes.volumes.config_map]]
      name = "docker-daemon"
      mount_path = "/etc/docker/daemon.json"
      sub_path = "daemon.json"

Use Docker socket binding

The third approach is to bind-mount /var/run/docker.sock into the container so that Docker is available in the context of that image.

NOTE: Note: If you bind the Docker socket when using GitLab Runner 11.11 or newer, you can no longer use docker:19.03.12-dind as a service because volume bindings are done to the services as well, making these incompatible.

In order to do that, follow the steps:

  1. Install GitLab Runner.

  2. Register GitLab Runner from the command line to use docker and share /var/run/docker.sock:

    sudo gitlab-runner register -n \
      --url https://gitlab.com/ \
      --registration-token REGISTRATION_TOKEN \
      --executor docker \
      --description "My Docker Runner" \
      --docker-image "docker:19.03.12" \
      --docker-volumes /var/run/docker.sock:/var/run/docker.sock

    The above command registers a new runner to use the special docker:19.03.12 image which is provided by Docker. Notice that it's using the Docker daemon of the runner itself, and any containers spawned by Docker commands are siblings of the runner rather than children of the runner. This may have complications and limitations that are unsuitable for your workflow.

    The above command creates a config.toml entry similar to this:

    [[runners]]
      url = "https://gitlab.com/"
      token = REGISTRATION_TOKEN
      executor = "docker"
      [runners.docker]
        tls_verify = false
        image = "docker:19.03.12"
        privileged = false
        disable_cache = false
        volumes = ["/var/run/docker.sock:/var/run/docker.sock", "/cache"]
      [runners.cache]
        Insecure = false
  3. You can now use docker in the build script (note that you don't need to include the docker:19.03.12-dind service as when using the Docker in Docker executor):

    image: docker:19.03.12
    
    before_script:
      - docker info
    
    build:
      stage: build
      script:
        - docker build -t my-docker-image .
        - docker run my-docker-image /script/to/run/tests

While the above method avoids using Docker in privileged mode, you should be aware of the following implications:

  • By sharing the Docker daemon, you are effectively disabling all the security mechanisms of containers and exposing your host to privilege escalation which can lead to container breakout. For example, if a project ran docker rm -f $(docker ps -a -q) it would remove the GitLab Runner containers.

  • Concurrent jobs may not work; if your tests create containers with specific names, they may conflict with each other.

  • Sharing files and directories from the source repository into containers may not work as expected since volume mounting is done in the context of the host machine, not the build container. For example:

    docker run --rm -t -i -v $(pwd)/src:/home/app/src test-image:latest run_app_tests

Making Docker-in-Docker builds faster with Docker layer caching

When using Docker-in-Docker, Docker downloads all layers of your image every time you create a build. Recent versions of Docker (Docker 1.13 and above) can use a pre-existing image as a cache during the docker build step, considerably speeding up the build process.

How Docker caching works

When running docker build, each command in Dockerfile results in a layer. These layers are kept around as a cache and can be reused if there haven't been any changes. Change in one layer causes all subsequent layers to be recreated.

You can specify a tagged image to be used as a cache source for the docker build command by using the --cache-from argument. Multiple images can be specified as a cache source by using multiple --cache-from arguments. Keep in mind that any image that's used with the --cache-from argument must first be pulled (using docker pull) before it can be used as a cache source.

Using Docker caching

Here's a .gitlab-ci.yml file showing how Docker caching can be used:

image: docker:19.03.12

services:
  - docker:19.03.12-dind

variables:
  # Use TLS https://docs.gitlab.com/ee/ci/docker/using_docker_build.html#tls-enabled
  DOCKER_HOST: tcp://docker:2376
  DOCKER_TLS_CERTDIR: "/certs"

before_script:
  - docker login -u $CI_REGISTRY_USER -p $CI_REGISTRY_PASSWORD $CI_REGISTRY

build:
  stage: build
  script:
    - docker pull $CI_REGISTRY_IMAGE:latest || true
    - docker build --cache-from $CI_REGISTRY_IMAGE:latest --tag $CI_REGISTRY_IMAGE:$CI_COMMIT_SHA --tag $CI_REGISTRY_IMAGE:latest .
    - docker push $CI_REGISTRY_IMAGE:$CI_COMMIT_SHA
    - docker push $CI_REGISTRY_IMAGE:latest

The steps in the script section for the build stage can be summed up to:

  1. The first command tries to pull the image from the registry so that it can be used as a cache for the docker build command.
  2. The second command builds a Docker image using the pulled image as a cache (notice the --cache-from $CI_REGISTRY_IMAGE:latest argument) if available, and tags it.
  3. The last two commands push the tagged Docker images to the container registry so that they may also be used as cache for subsequent builds.

Use the OverlayFS driver

NOTE: Note: The shared runners on GitLab.com use the overlay2 driver by default.

By default, when using docker:dind, Docker uses the vfs storage driver which copies the filesystem on every run. This is a disk-intensive operation which can be avoided if a different driver is used, for example overlay2.

Requirements

  1. Make sure a recent kernel is used, preferably >= 4.2.

  2. Check whether the overlay module is loaded:

    sudo lsmod | grep overlay

    If you see no result, then it isn't loaded. To load it use:

    sudo modprobe overlay

    If everything went fine, you need to make sure module is loaded on reboot. On Ubuntu systems, this is done by editing /etc/modules. Just add the following line into it:

    overlay

Use the OverlayFS driver per project

You can enable the driver for each project individually by using the DOCKER_DRIVER environment variable in .gitlab-ci.yml:

variables:
  DOCKER_DRIVER: overlay2

Use the OverlayFS driver for every project

If you use your own GitLab Runners, you can enable the driver for every project by setting the DOCKER_DRIVER environment variable in the [[runners]] section of config.toml:

environment = ["DOCKER_DRIVER=overlay2"]

If you're running multiple runners, you have to modify all configuration files.

Read more about the runner configuration and using the OverlayFS storage driver.

Using the GitLab Container Registry

After you've built a Docker image, you can push it up to the built-in GitLab Container Registry.

Troubleshooting

docker: Cannot connect to the Docker daemon at tcp://docker:2375. Is the docker daemon running?

This is a common error when you are using Docker in Docker v19.03 or higher.

This occurs because Docker starts on TLS automatically, so you need to do some setup. If: