• How services are linked to the job
• How the health check of services works
• What services are not for
• Define services in the .gitlab-ci.yml file
• Accessing the services
  • Connecting services
• Passing CI/CD variables to services
• Available settings for services
• Starting multiple services from the same image
• Setting a command for the service
• Using services with docker run (Docker-in-Docker) side-by-side
• How Docker integration works
• Capturing service container logs
• Debug a job locally
• Security when using services containers

Services

When you configure CI/CD, you specify an image, which is used to create the container where your jobs run. To specify this image, you use the image keyword.

You can specify an additional image by using the services keyword. This additional image is used to create another container, which is available to the first container. The two containers have access to one another and can communicate when running the job.

The service image can run any application, but the most common use case is to run a database container, for example:

• MySQL
• PostgreSQL
• Redis
• GitLab as an example for a microservice offering a JSON API

It’s easier and faster to use an existing image and run it as an additional container than to install mysql, for example, every time the project is built.

You’re not limited to 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.

How services are linked to the job

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

If you add mysql as service to your application, the image is used to create a container that’s linked to the job container.

The service container for MySQL is accessible under the hostname mysql. To access your database service, 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 features which are network accessible. They 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.

If the second stage of the check fails, it prints the warning: *** WARNING: Service XYZ probably didn't start properly. This issue can occur because:

• There is no opened port in the service.
• The service was not started properly before the timeout, and the port is not responding.

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 shortly 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 because the job is not trying to connect to it, it does not fail.

If the services start successfully, they start before the before_script runs. This means you can write a before_script that queries the service.

Services stop at the end of the job, even if the job fails.

What services are not for

As 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 images 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 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.

Define services in the .gitlab-ci.yml file

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

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 file:

services:
  - tutum/wordpress:latest

If you don’t specify a service alias, when the job runs, tutum/wordpress is started. You have access to it from your build container under two hostnames:

• tutum-wordpress
• tutum__wordpress

Hostnames with underscores are not RFC valid and may cause problems in third-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.

Connecting services

You can use inter-dependent services with complex jobs, like end-to-end tests where an external API needs to communicate with its own database.

For example, for an end-to-end test for a front-end application that uses an API, and where the API needs a database:

end-to-end-tests:
  image: node:latest
  services:
    - name: selenium/standalone-firefox:${FIREFOX_VERSION}
      alias: firefox
    - name: registry.gitlab.com/organization/private-api:latest
      alias: backend-api
    - postgres:14.3
  variables:
    FF_NETWORK_PER_BUILD: 1
    POSTGRES_PASSWORD: supersecretpassword
    BACKEND_POSTGRES_HOST: postgres
  script:
    - npm install
    - npm test

For this solution to work, you must use the networking mode that creates a new network for each job.

Passing CI/CD variables to services

You can also pass custom CI/CD variables to fine tune your Docker images and services directly in the .gitlab-ci.yml file. For more information, read about .gitlab-ci.yml defined 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

Available settings for services

(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.

(2) Service variables support for the Docker and the Kubernetes executor was introduced in GitLab Runner 14.8.

Starting multiple services from the same image

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, each that uses the mysql:latest image. However, 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

Let’s assume you have a super/sql:latest image with some SQL database in it. 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. 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.

• Use the image in the job’s configuration:

  # 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 set a command in the .gitlab-ci.yml file instead:

# .gitlab-ci.yml

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

The syntax of command is similar to Dockerfile CMD.

Using services with docker run (Docker-in-Docker) side-by-side

Containers started with docker run can also connect to services provided by GitLab.

When booting the service is expensive or time consuming, you can use this technique to run tests from different client environments, while only booting up the tested service once.

access-service:
  stage: build
  image: docker:20.10.16
  services:
    - docker:dind                    # necessary for docker run
    - tutum/wordpress:latest
  variables:
    FF_NETWORK_PER_BUILD: "true"     # activate container-to-container networking
  script: |
    docker run --rm --name curl \
      --volume  "$(pwd)":"$(pwd)"    \
      --workdir "$(pwd)"             \
      --network=host                 \
      curlimages/curl:7.74.0 curl "http://tutum-wordpress"

For this solution to work, you must:

• Use the networking mode that creates a new network for each job.
• Not use the Docker executor with Docker socket binding. If you must, then in the above example, instead of host, use the dynamic network name created for this job.

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 a 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 a build container and link any service container to build container.
4. Start the build container, and send a job script to the container.
5. Run the job script.
6. Checkout code in: /builds/group-name/project-name/.
7. Run any step defined in .gitlab-ci.yml.
8. Check the exit status of build script.
9. Remove the build container and all created service containers.

Capturing service container logs

Logs generated by applications running in service containers can be captured for subsequent examination and debugging. You might want to look at service container’s logs when the service container has started successfully, but is not behaving es expected, leading to job failures. The logs can indicate missing or incorrect configuration of the service within the container.

CI_DEBUG_SERVICES should only by enabled when service containers are being actively debugged as there are both storage and performance consequences to capturing service container logs.

To enable service logging, add the CI_DEBUG_SERVICES variable to the project’s .gitlab-ci.yml file:

variables:
    CI_DEBUG_SERVICES: "true"

Accepted values are:

• Enabled: TRUE, true, True
• Disabled: FALSE, false, False

Any other values result in an error message and effectively disable the feature.

When enabled, logs for all service containers are captured and streamed into the jobs trace log concurrently with other logs. Logs from each container are prefixed with the container’s aliases, and displayed in a different color.

See Mask a CI/CD Variable

Debug a job locally

The following commands are run without root privileges. You should be able to run Docker with your 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. 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, and all volumes (-v) that were created with the container creation.

Security when using services containers

Docker privileged mode applies to services. This means that the service image container can access the host system. You should use container images from trusted sources only.