Using Docker images

GitLab CI/CD in conjunction with GitLab Runner can use Docker Engine to test and build any application.

Docker is an open-source project that allows you to use predefined images to run applications in independent "containers" that are run within a single Linux instance. Docker Hub has a rich database of pre-built images that can be used to test and build your applications.

When used with GitLab CI/CD, Docker runs each job in a separate and isolated container using the predefined image that's set up in .gitlab-ci.yml.

This makes it easier to have a simple and reproducible build environment that can also run on your workstation. The added benefit is that you can test all the commands that we explore later from your shell, rather than having to test them on a dedicated CI server.

Register Docker Runner

To use GitLab Runner with Docker you need to register a new runner to use the docker executor.

An example can be seen below. First we set up a temporary template to supply the services:

cat > /tmp/test-config.template.toml << EOF
[[runners]]
[runners.docker]
[[runners.docker.services]]
name = "postgres:latest"
[[runners.docker.services]]
name = "mysql:latest"
EOF

Then we register the runner using the template that was just created:

sudo gitlab-runner register \
  --url "https://gitlab.example.com/" \
  --registration-token "PROJECT_REGISTRATION_TOKEN" \
  --description "docker-ruby:2.6" \
  --executor "docker" \
  --template-config /tmp/test-config.template.toml \
  --docker-image ruby:2.6

The registered runner uses the ruby:2.6 Docker image and runs two services, postgres:latest and mysql:latest, both of which are accessible during the build process.

What is an image

The image keyword is the name of the Docker image the Docker executor runs to perform the CI tasks.

By default, the executor only pulls images from Docker Hub, however this can be configured in the gitlab-runner/config.toml by setting the Docker pull policy to allow using local images.

For more information about images and Docker Hub, please read the Docker Fundamentals documentation.

What is a service

The services keyword defines just another Docker image that's run during your job and is linked to the Docker image that the image keyword defines. This allows you to access the service image during build time.

The service image can run any application, but the most common use case is to run a database container, for example, mysql. It's easier and faster to use an existing image and run it as an additional container than install mysql every time the project is built.

You're not limited to have only database services. You can add as many services you need to .gitlab-ci.yml or manually modify config.toml. Any image found at Docker Hub or your private Container Registry can be used as a service.

Services inherit the same DNS servers, search domains, and additional hosts as the CI container itself.

You can see some widely used services examples in the relevant documentation of CI services examples.

How services are linked to the job

To better understand how the container linking works, read Linking containers together.

To summarize, if you add mysql as service to your application, the image is then used to create a container that's linked to the job container.

The service container for MySQL is accessible under the hostname mysql. So, in order to access your database service you have to connect to the host named mysql instead of a socket or localhost. Read more in accessing the services.

How the health check of services works

Services are designed to provide additional functionality which is network accessible. It may be a database like MySQL, or Redis, and even docker:stable-dind which allows you to use Docker in Docker. It can be practically anything that's required for the CI/CD job to proceed and is accessed by network.

To make sure this works, the runner:

1. Checks which ports are exposed from the container by default.
2. Starts a special container that waits for these ports to be accessible.

When the second stage of the check fails, either because there is no opened port in the service, or the service was not started properly before the timeout and the port is not responding, it prints the warning: *** WARNING: Service XYZ probably didn't start properly.

In most cases it affects the job, but there may be situations when the job still succeeds even if that warning was printed. For example:

• The service was started a little after the warning was raised, and the job is not using the linked service from the beginning. In that case, when the job needed to access the service, it may have been already there waiting for connections.
• The service container is not providing any networking service, but it's doing something with the job's directory (all services have the job directory mounted as a volume under /builds). In that case, the service does its job, and since the job is not trying to connect to it, it does not fail.

What services are not for

As it was mentioned before, this feature is designed to provide network accessible services. A database is the simplest example of such a service.

The services feature is not designed to, and does not add any software from the defined services image(s) to the job's container.

For example, if you have the following services defined in your job, the php, node or go commands are not available for your script, and thus the job fails:

job:
  services:
    - php:7
    - node:latest
    - golang:1.10
  image: alpine:3.7
  script:
    - php -v
    - node -v
    - go version

If you need to have php, node and go available for your script, you should either:

• Choose an existing Docker image that contains all required tools.
• Create your own Docker image, with all the required tools included and use that in your job.

Accessing the services

Let's say that you need a Wordpress instance to test some API integration with your application.

You can then use for example the tutum/wordpress image in your .gitlab-ci.yml:

services:
  - tutum/wordpress:latest

If you don't specify a service alias, when the job is run, tutum/wordpress is started and you have access to it from your build container under two hostnames to choose from:

• tutum-wordpress
• tutum__wordpress

Hostnames with underscores are not RFC valid and may cause problems in 3rd party applications.

The default aliases for the service's hostname are created from its image name following these rules:

• Everything after the colon (:) is stripped.
• Slash (/) is replaced with double underscores (__) and the primary alias is created.
• Slash (/) is replaced with a single dash (-) and the secondary alias is created (requires GitLab Runner v1.1.0 or higher).

To override the default behavior, you can specify a service alias.

Define image and services from .gitlab-ci.yml

You can simply define an image that's used for all jobs and a list of services that you want to use during build time:

default:
  image: ruby:2.6

  services:
    - postgres:11.7

  before_script:
    - bundle install

test:
  script:
    - bundle exec rake spec

The image name must be in one of the following formats:

• image: <image-name> (Same as using <image-name> with the latest tag)
• image: <image-name>:<tag>
• image: <image-name>@<digest>

It's also possible to define different images and services per job:

default:
  before_script:
    - bundle install

test:2.6:
  image: ruby:2.6
  services:
    - postgres:11.7
  script:
    - bundle exec rake spec

test:2.7:
  image: ruby:2.7
  services:
    - postgres:12.2
  script:
    - bundle exec rake spec

Or you can pass some extended configuration options for image and services:

default:
  image:
    name: ruby:2.6
    entrypoint: ["/bin/bash"]

  services:
    - name: my-postgres:11.7
      alias: db-postgres
      entrypoint: ["/usr/local/bin/db-postgres"]
      command: ["start"]

  before_script:
    - bundle install

test:
  script:
    - bundle exec rake spec

Passing environment variables to services

You can also pass custom environment variables to fine tune your Docker images and services directly in the .gitlab-ci.yml file. For more information, see custom environment variables

# The following variables are automatically passed down to the Postgres container
# as well as the Ruby container and available within each.
variables:
  HTTPS_PROXY: "https://10.1.1.1:8090"
  HTTP_PROXY: "https://10.1.1.1:8090"
  POSTGRES_DB: "my_custom_db"
  POSTGRES_USER: "postgres"
  POSTGRES_PASSWORD: "example"
  PGDATA: "/var/lib/postgresql/data"
  POSTGRES_INITDB_ARGS: "--encoding=UTF8 --data-checksums"

services:
  - name: postgres:11.7
    alias: db
    entrypoint: ["docker-entrypoint.sh"]
    command: ["postgres"]

image:
  name: ruby:2.6
  entrypoint: ["/bin/bash"]

before_script:
  - bundle install

test:
  script:
    - bundle exec rake spec

Extended Docker configuration options

Introduced in GitLab and GitLab Runner 9.4.

When configuring the image or services entries, you can use a string or a map as options:

• when using a string as an option, it must be the full name of the image to use (including the Registry part if you want to download the image from a Registry other than Docker Hub)
• when using a map as an option, then it must contain at least the name option, which is the same name of the image as used for the string setting

For example, the following two definitions are equal:

1. Using a string as an option to image and services:

   image: "registry.example.com/my/image:latest"
   
   services:
     - postgresql:9.4
     - redis:latest

2. Using a map as an option to image and services. The use of image:name is required:

   image:
     name: "registry.example.com/my/image:latest"
   
   services:
     - name: postgresql:9.4
     - name: redis:latest

Available settings for image

Introduced in GitLab and GitLab Runner 9.4.

Setting	Required	GitLab version	Description
name	yes, when used with any other option	9.4	Full name of the image that should be used. It should contain the Registry part if needed.
entrypoint	no	9.4	Command or script that should be executed as the container's entrypoint. It's translated to Docker's --entrypoint option while creating the container. The syntax is similar to Dockerfile's ENTRYPOINT directive, where each shell token is a separate string in the array.

Available settings for services

Introduced in GitLab and GitLab Runner 9.4.

Setting	Required	GitLab version	Description
name	yes, when used with any other option	9.4	Full name of the image that should be used. It should contain the Registry part if needed.
entrypoint	no	9.4	Command or script that should be executed as the container's entrypoint. It's translated to Docker's --entrypoint option while creating the container. The syntax is similar to Dockerfile's ENTRYPOINT directive, where each shell token is a separate string in the array.
command	no	9.4	Command or script that should be used as the container's command. It's translated to arguments passed to Docker after the image's name. The syntax is similar to Dockerfile's CMD directive, where each shell token is a separate string in the array.
alias (1)	no	9.4	Additional alias that can be used to access the service from the job's container. Read Accessing the services for more information.

(1) Alias support for the Kubernetes executor was introduced in GitLab Runner 12.8, and is only available for Kubernetes version 1.7 or later.

Starting multiple services from the same image

Introduced in GitLab and GitLab Runner 9.4. Read more about the extended configuration options.

Before the new extended Docker configuration options, the following configuration would not work properly:

services:
  - mysql:latest
  - mysql:latest

The runner would start two containers using the mysql:latest image, but both of them would be added to the job's container with the mysql alias based on the default hostname naming. This would end with one of the services not being accessible.

After the new extended Docker configuration options, the above example would look like:

services:
  - name: mysql:latest
    alias: mysql-1
  - name: mysql:latest
    alias: mysql-2

The runner still starts two containers using the mysql:latest image, however now each of them are also accessible with the alias configured in .gitlab-ci.yml file.

Setting a command for the service

Introduced in GitLab and GitLab Runner 9.4. Read more about the extended configuration options.

Let's assume you have a super/sql:latest image with some SQL database inside it and you would like to use it as a service for your job. Let's also assume that this image does not start the database process while starting the container and the user needs to manually use /usr/bin/super-sql run as a command to start the database.

Before the new extended Docker configuration options, you would need to create your own image based on the super/sql:latest image, add the default command, and then use it in job's configuration, like:

# my-super-sql:latest image's Dockerfile

FROM super/sql:latest
CMD ["/usr/bin/super-sql", "run"]

# .gitlab-ci.yml

services:
  - my-super-sql:latest

After the new extended Docker configuration options, you can now simply set a command in .gitlab-ci.yml, like:

# .gitlab-ci.yml

services:
  - name: super/sql:latest
    command: ["/usr/bin/super-sql", "run"]

As you can see, the syntax of command is similar to Dockerfile's CMD.

Overriding the entrypoint of an image

Introduced in GitLab and GitLab Runner 9.4. Read more about the extended configuration options.

Before showing the available entrypoint override methods, let's describe shortly how the runner starts and uses a Docker image for the containers used in the CI jobs:

1. The runner starts a Docker container using the defined entrypoint (default from Dockerfile that may be overridden in .gitlab-ci.yml)
2. The runner attaches itself to a running container.
3. The runner prepares a script (the combination of before_script, script, and after_script).
4. The runner sends the script to the container's shell STDIN and receives the output.

To override the entrypoint of a Docker image, the recommended solution is to define an empty entrypoint in .gitlab-ci.yml, so the runner does not start a useless shell layer. However, that does not work for all Docker versions, and you should check which one your runner is using. Specifically:

• If Docker 17.06 or later is used, the entrypoint can be set to an empty value.
• If Docker 17.03 or previous versions are used, the entrypoint can be set to /bin/sh -c, /bin/bash -c or an equivalent shell available in the image.

The syntax of image:entrypoint is similar to Dockerfile's ENTRYPOINT.

Let's assume you have a super/sql:experimental image with some SQL database inside it and you would like to use it as a base image for your job because you want to execute some tests with this database binary. Let's also assume that this image is configured with /usr/bin/super-sql run as an entrypoint. That means that when starting the container without additional options, it runs the database's process, while the runner expects that the image has no entrypoint or that the entrypoint is prepared to start a shell command.

With the extended Docker configuration options, instead of creating your own image based on super/sql:experimental, setting the ENTRYPOINT to a shell, and then using the new image in your CI job, you can now simply define an entrypoint in .gitlab-ci.yml.

For Docker 17.06+:

image:
  name: super/sql:experimental
  entrypoint: [""]

For Docker =< 17.03:

image:
  name: super/sql:experimental
  entrypoint: ["/bin/sh", "-c"]

Define image and services in config.toml

Look for the [runners.docker] section:

[runners.docker]
  image = "ruby:latest"
  services = ["mysql:latest", "postgres:latest"]

The image and services defined this way are added to all job run by that runner.

Define an image from a private Container Registry

To access private container registries, the GitLab Runner process can use:

• Statically defined credentials. That is, a username and password for a specific registry.
• Credentials Store. For more information, see the relevant Docker documentation.
• Credential Helpers. For more information, see the relevant Docker documentation.

To define which should be used, the GitLab Runner process reads the configuration in the following order:

• DOCKER_AUTH_CONFIG variable provided as either:
  • A variable in .gitlab-ci.yml.
  • A project's variables stored on the projects Settings > CI/CD page.
• DOCKER_AUTH_CONFIG variable provided as environment variable in config.toml of the runner.
• config.json file placed in $HOME/.docker directory of the user running GitLab Runner process. If the --user flag is provided to run the GitLab Runner child processes as unprivileged user, the home directory of the main GitLab Runner process user is used.

GitLab Runner reads this configuration only from config.toml and ignores it if it's provided as an environment variable. This is because GitLab Runner uses only config.toml configuration and does not interpolate ANY environment variables at runtime.

Requirements and limitations

• This feature requires GitLab Runner 1.8 or higher.
• For GitLab Runner versions >= 0.6, <1.8 there was a partial support for using private registries, which required manual configuration of credentials on runner's host. We recommend to upgrade your runner to at least version 1.8 if you want to use private registries.
• Available for Kubernetes executor in GitLab Runner 13.1 and later.
• Credentials Store and Credential Helpers require binaries to be added to the GitLab Runner's $PATH, and require access to do so. Therefore, these features are not available on shared runners or any other runner where the user does not have access to the environment where the runner is installed.

Using statically-defined credentials

There are two approaches that you can take in order to access a private registry. Both require setting the environment variable DOCKER_AUTH_CONFIG with appropriate authentication information.

1. Per-job: To configure one job to access a private registry, add DOCKER_AUTH_CONFIG as a job variable.
2. Per-runner: To configure a runner so all its jobs can access a private registry, add DOCKER_AUTH_CONFIG to the environment in the runner's configuration.

See below for examples of each.

Determining your DOCKER_AUTH_CONFIG data

As an example, let's assume you want to use the registry.example.com:5000/private/image:latest image, which is private and requires you to sign in to a private container registry.

Let's also assume that these are the sign-in credentials:

Key	Value
registry	registry.example.com:5000
username	my_username
password	my_password

Use one of the following methods to determine the value of DOCKER_AUTH_CONFIG:

• Do a docker login on your local machine:

  docker login registry.example.com:5000 --username my_username --password my_password

  Then copy the content of ~/.docker/config.json.

  If you don't need access to the registry from your computer, you can do a docker logout:

  docker logout registry.example.com:5000

• In some setups, it's possible that Docker client uses the available system key store to store the result of docker login. In that case, it's impossible to read ~/.docker/config.json, so you need to prepare the required base64-encoded version of ${username}:${password} and create the Docker configuration JSON manually. Open a terminal and execute the following command:

  # The use of "-n" - prevents encoding a newline in the password.
  echo -n "my_username:my_password" | base64
  
  # Example output to copy
  bXlfdXNlcm5hbWU6bXlfcGFzc3dvcmQ=

  Create the Docker JSON configuration content as follows:

  {
      "auths": {
          "registry.example.com:5000": {
              "auth": "(Base64 content from above)"
          }
      }
  }

Configuring a job

To configure a single job with access for registry.example.com:5000, follow these steps:

1. Create a variable DOCKER_AUTH_CONFIG with the content of the Docker configuration file as the value:

   {
       "auths": {
           "registry.example.com:5000": {
               "auth": "bXlfdXNlcm5hbWU6bXlfcGFzc3dvcmQ="
           }
       }
   }

2. You can now use any private image from registry.example.com:5000 defined in image and/or services in your .gitlab-ci.yml file:

   image: registry.example.com:5000/namespace/image:tag

   In the example above, GitLab Runner looks at registry.example.com:5000 for the image namespace/image:tag.

You can add configuration for as many registries as you want, adding more registries to the "auths" hash as described above.

The full hostname:port combination is required everywhere for the runner to match the DOCKER_AUTH_CONFIG. For example, if registry.example.com:5000/namespace/image:tag is specified in .gitlab-ci.yml, then the DOCKER_AUTH_CONFIG must also specify registry.example.com:5000. Specifying only registry.example.com does not work.

Configuring a runner

If you have many pipelines that access the same registry, it is probably better to set up registry access at the runner level. This allows pipeline authors to have access to a private registry just by running a job on the appropriate runner. It also makes registry changes and credential rotations much simpler.

Of course this means that any job on that runner can access the registry with the same privilege, even across projects. If you need to control access to the registry, you need to be sure to control access to the runner.

To add DOCKER_AUTH_CONFIG to a runner:

1. Modify the runner's config.toml file as follows:

   [[runners]]
     environment = ["DOCKER_AUTH_CONFIG={\"auths\":{\"registry.example.com:5000\":{\"auth\":\"bXlfdXNlcm5hbWU6bXlfcGFzc3dvcmQ=\"}}}"]

   • The double quotes included in the DOCKER_AUTH_CONFIG data must be escaped with backslashes. This prevents them from being interpreted as TOML.
   • The environment option is a list. Your runner may have existing entries and you should add this to the list, not replace it.

2. Restart the runner service.

Using Credentials Store

Support for using Credentials Store was added in GitLab Runner 9.5.

To configure credentials store, follow these steps:

1. To use a credentials store, you need an external helper program to interact with a specific keychain or external store. Make sure helper program is available in GitLab Runner $PATH.

2. Make GitLab Runner use it. There are two ways to accomplish this. Either:

   • Create a variable DOCKER_AUTH_CONFIG with the content of the Docker configuration file as the value:

       {
         "credsStore": "osxkeychain"
       }

   • Or, if you're running self-managed runners, add the above JSON to ${GITLAB_RUNNER_HOME}/.docker/config.json. GitLab Runner reads this configuration file and uses the needed helper for this specific repository.

credsStore is used to access ALL the registries. If you want to use both images from private registry and public images from Docker Hub, pulling from Docker Hub would fail, because Docker daemon tries to use the same credentials for ALL the registries.

Using Credential Helpers

Support for using Credential Helpers was added in GitLab Runner 12.0

As an example, let's assume that you want to use the aws_account_id.dkr.ecr.region.amazonaws.com/private/image:latest image which is private and requires you to log in into a private container registry.

To configure access for aws_account_id.dkr.ecr.region.amazonaws.com, follow these steps:

1. Make sure docker-credential-ecr-login is available in GitLab Runner's $PATH.

2. Have any of the following AWS credentials setup. Make sure that GitLab Runner can access the credentials.

3. Make GitLab Runner use it. There are two ways to accomplish this. Either:

   • Create a variable DOCKER_AUTH_CONFIG with the content of the Docker configuration file as the value:

     {
       "credHelpers": {
         "aws_account_id.dkr.ecr.region.amazonaws.com": "ecr-login"
       }
     }

     This configures Docker to use the credential helper for a specific registry.

     or

     {
       "credsStore": "ecr-login"
     }

     This configures Docker to use the credential helper for all Amazon ECR registries.

   • Or, if you're running self-managed runners, add the above JSON to ${GITLAB_RUNNER_HOME}/.docker/config.json. GitLab Runner reads this configuration file and uses the needed helper for this specific repository.

4. You can now use any private image from aws_account_id.dkr.ecr.region.amazonaws.com defined in image and/or services in your .gitlab-ci.yml file:

   image: aws_account_id.dkr.ecr.region.amazonaws.com/private/image:latest

   In the example above, GitLab Runner looks at aws_account_id.dkr.ecr.region.amazonaws.com for the image private/image:latest.

You can add configuration for as many registries as you want, adding more registries to the "credHelpers" hash as described above.

Configuring services

Many services accept environment variables which allow you to easily change database names or set account names depending on the environment.

GitLab Runner 0.5.0 and up passes all YAML-defined variables to the created service containers.

For all possible configuration variables check the documentation of each image provided in their corresponding Docker hub page.

All variables are passed to all services containers. It's not designed to distinguish which variable should go where.

PostgreSQL service example

See the specific documentation for using PostgreSQL as a service.

MySQL service example

See the specific documentation for using MySQL as a service.

How Docker integration works

Below is a high level overview of the steps performed by Docker during job time.

1. Create any service container: mysql, postgresql, mongodb, redis.
2. Create cache container to store all volumes as defined in config.toml and Dockerfile of build image (ruby:2.6 as in above example).
3. Create build container and link any service container to build container.
4. Start build container and send job script to the container.
5. Run job script.
6. Checkout code in: /builds/group-name/project-name/.
7. Run any step defined in .gitlab-ci.yml.
8. Check exit status of build script.
9. Remove build container and all created service containers.

How to debug a job locally

The following commands are run without root privileges. You should be able to run Docker with your regular user account.

First start with creating a file named build_script:

cat <<EOF > build_script
git clone https://gitlab.com/gitlab-org/gitlab-runner.git /builds/gitlab-org/gitlab-runner
cd /builds/gitlab-org/gitlab-runner
make
EOF

Here we use as an example the GitLab Runner repository which contains a Makefile, so running make executes the commands defined in the Makefile. Your mileage may vary, so instead of make you could run the command which is specific to your project.

Then create some service containers:

docker run -d --name service-mysql mysql:latest
docker run -d --name service-postgres postgres:latest

This creates two service containers, named service-mysql and service-postgres which use the latest MySQL and PostgreSQL images respectively. They both run in the background (-d).

Finally, create a build container by executing the build_script file we created earlier:

docker run --name build -i --link=service-mysql:mysql --link=service-postgres:postgres ruby:2.6 /bin/bash < build_script

The above command creates a container named build that's spawned from the ruby:2.6 image and has two services linked to it. The build_script is piped using STDIN to the bash interpreter which in turn executes the build_script in the build container.

When you finish testing and no longer need the containers, you can remove them with:

docker rm -f -v build service-mysql service-postgres

This forcefully (-f) removes the build container, the two service containers as well as all volumes (-v) that were created with the container creation.