Add Docker executor support for a Windows version

GitLab supports specific versions of Windows.

To add support for a new Windows version for the Docker executor, you must release a helper image with the same Windows version. Then you can run the helper image on the Windows host OS.

To build the helper image for the version, you need GitLab Runner installed on that Windows version, because Windows requires your host OS and container OS versions to match.

Infrastructure

We need to build the helper image for it to be used for the user job.

Create a base image for infrastructure to use

Windows requires us to have the host OS version match the container OS, so if we are building Windows Server Core 2004 image we need to have gitlab-runner installed on Windows Server Core 2004.

To do this we need to update the windows-containers repository to build a base image. The base image will be used by the autoscaler for our CI. The new base image will be used to build the GitLab Runner helper image.

For example, if we want to add support for Windows Server Core 2004 in the 13.7 milestone we can see the following merge request. Depending on the base image provided by GCP, we might have to install Docker as part of the build process or not. In this MR we update the following files:

  1. .gitlab-ci.yml
  2. .gitlab/ci/build.gitlab-ci.yml

Publish the image

After we merge the merge request created from the previous step, we need to run the publish job manually for the image to be published to our production GCP project.

Take note of the image name that is created from the publish job, for example in this job we created an image called runners-windows-2019-core-containers-2020-07-17. This will be used for the install part.

Add two new runner managers

At this point we should have a base image ready in our production environment, so we can use it inside the CI pipeline for the GitLab Runner project. The only thing that is left is to set up the Runner Managers.

Register

Run gitlab-runner register to register the two new runners. These should be project-specific runners, so we need to use the registration token from the project settings. The name of the runner should follow the same naming convention as the existing ones.

For example, for Windows Server Core 2004 we should name the Runner Managers the following:

  1. windows-2004-private-runner-manager-1
  2. windows-2004-private-runner-manager-2

Once registered, make sure you safely store the runner tokens found in the config.toml file since we are going to need these for the installation step.

Finally, we’ll need to assign the new Runner Managers to the security fork project and to the ‘liveness’ test support project. So for each of the new Runner Managers:

  1. Go to the Runners section of the Runner project CI/CD settings page;
  2. Unlock the new Runner by editing its properties and unchecking Lock to current projects;
  3. For the security fork project:
    1. Go to the Runners section of the project’s CI/CD settings page;
    2. Scroll down to the Other available runners section and enable the runner for this project;
  4. For the ‘liveness’ test support project:
    1. Go to the Runners section of the project’s CI/CD settings page;
    2. Scroll down to the Other available runners section and enable the runner for this project;
  5. Lock the Runner back again in the Runner project CI/CD settings page.

Install

Install a new instance of autoscaler to have a specific config.toml for that Windows version. We need to update our Ansible repository to include the new Windows version.

For example, if we want to add support for Windows Server Core 2004 in the 13.7 milestone we can see the following merge request, where we update the following files:

  1. ansible/roles/runner/tasks/main.yml
  2. ansible/roles/runner/tasks/autoscaler.yml
  3. ansible/group_vars/gcp_role_runner_manager.yml
  4. ansible/host_vars/windows-shared-runners-manager-1.yml
  5. ansible/host_vars/windows-shared-runners-manager-2.yml

When opening a merge request make sure that the maintainer is aware that they need to register 2 new runners and save them inside the CI/CD variables with the keys defined in ansible/host_vars.

Publish registry.gitlab.com/gitlab-org/ci-cd/tests/liveness

The image registry.gitlab.com/gitlab-org/ci-cd/tests/liveness is used as part of the CI process for GitLab Runner. Make sure that an image based on the new Windows version is published.

For example, if we want to add support for Windows Server Core 2004 in the 13.7 milestone we can see the following merge request, where we update the following files:

  1. .gitlab-ci.yml
  2. Makefile

Update GitLab Runner to support specific Windows version

Since we need to provide a helper image for users to be able to use the Docker executor we have specific checks inside the code base, we need to allow the new Windows version.

We should update the following:

  1. List of support versions, and tests surrounding it.
  2. List of base images, and tests surrounding it.
  3. Update GitLab CI to run tests on the default branch.
  4. Update the release stage.

For example, if we want to add support for Windows Server Core 2004 in the 13.7 milestone we can see the following merge request, where we update the following files:

  1. helpers/container/helperimage/windows_info.go
  2. helpers/container/helperimage/windows_info_test.go
  3. helpers/container/windows/version.go
  4. helpers/container/windows/version_test.go
  5. .gitlab/ci/test.gitlab-ci.yml
  6. .gitlab/ci/coverage.gitlab-ci.yml
  7. .gitlab/ci/_common.gitlab-ci.yml
  8. .gitlab/ci/release.gitlab-ci.yml
  9. ci/.test-failures.servercore2004.txt
  10. docs/executors/docker.md