Development guide for GitLab CI/CD templates (FREE)

This document explains how to develop GitLab CI/CD templates.

Requirements for CI/CD templates

Before submitting a merge request with a new or updated CI/CD template, you must:

Place the template in the correct directory.
Follow the CI/CD template authoring guidelines.
Name the template following the *.gitlab-ci.yml format.
Use valid .gitlab-ci.yml syntax. Verify it's valid with the CI/CD lint tool.
Add template metrics.
Include a changelog if the merge request introduces a user-facing change.
Follow the template review process.
(Optional but highly recommended) Test the template in an example GitLab project that reviewers can access. Reviewers might not be able to create the data or configuration that the template requires, so an example project helps the reviewers ensure the template is correct. The example project pipeline should succeed before submitting the merge request for review.

Template directories

All template files are saved in lib/gitlab/ci/templates. Save general templates in this directory, but certain template types have a specific directory reserved for them. The ability to select a template in new file UI is determined by the directory it is in:

Sub-directory	Selectable in UI	Template type
`/*` (root)	Yes	General templates.
`/AWS/*`	No	Templates related to Cloud Deployment (AWS).
`/Jobs/*`	No	Templates related to Auto DevOps.
`/Pages/*`	Yes	Sample templates for using Static site generators with GitLab Pages.
`/Security/*`	Yes	Templates related to Security scanners.
`/Terraform/*`	No	Templates related to infrastructure as Code (Terraform).
`/Verify/*`	Yes	Templates related to Testing features.
`/Workflows/*`	No	Sample templates for using the `workflow:` keyword.

Template authoring guidelines

Use the following guidelines to ensure your template submission follows standards:

Template types

Templates have two different types that impact the way the template should be written and used. The style in a template should match one of these two types:

A pipeline template provides an end-to-end CI/CD workflow that matches a project's structure, language, and so on. It usually should be used by itself in projects that don't have any other .gitlab-ci.yml files.

When authoring pipeline templates:

Place any global keywords like image or before_script in a default section at the top of the template.
Note clearly in the code comments if the template is designed to be used with the includes keyword in an existing .gitlab-ci.yml file or not.

A job template provides specific jobs that can be added to an existing CI/CD workflow to accomplish specific tasks. It usually should be used by adding it to an existing .gitlab-ci.yml file by using the includes keyword. You can also copy and paste the contents into an existing .gitlab-ci.yml file.

Configure job templates so that users can add them to their current pipeline with very few or no modifications. It must be configured to reduce the risk of conflicting with other pipeline configuration.

When authoring job templates:

Do not use global or default keywords. When a root .gitlab-ci.yml includes a template, global or default keywords might be overridden and cause unexpected behavior. If a job template requires a specific stage, explain in the code comments that users must manually add the stage to the main .gitlab-ci.yml configuration.
Note clearly in code comments that the template is designed to be used with the includes keyword or copied into an existing configuration.
Consider versioning the template with latest and stable versions to avoid backward compatibility problems. Maintenance of this type of template is more complex, because changes to templates imported with includes can break pipelines for all projects using the template.

Additional points to keep in mind when authoring templates:

Template design points	Pipeline templates	Job templates
Can use global keywords, including `stages`.	Yes	No
Can define jobs.	Yes	Yes
Can be selected in the new file UI	Yes	No
Can include other job templates with `include`	Yes	No
Can include other pipeline templates with `include`.	No	No

Syntax guidelines

To make templates easier to follow, templates should all use clear syntax styles, with a consistent format.

The before_script, script, and after_script keywords of every job are linted using ShellCheck and should follow the Shell scripting standards and style guidelines as much as possible.

ShellCheck assumes that the script is designed to run using Bash. Templates which use scripts for shells that aren't compatible with the Bash ShellCheck rules can be excluded from ShellCheck linting. To exclude a script, add it to the EXCLUDED_TEMPLATES list in scripts/lint_templates_bash.rb.

Do not hardcode the default branch

Use $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH instead of a hardcoded main branch, and never use master:

job1:
  rules:
    if: $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
  script:
    echo "example job 1"

job2:
  only:
    variables:
      - $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
  script:
    echo "example job 2"

Use `rules` instead of `only` or `except`

Avoid using only or except if possible. Only and except is not being developed any more, and rules is now the preferred syntax:

job2:
  script:
    - echo
  rules:
    - if: $CI_COMMIT_BRANCH

Break up long commands

If a command is very long, or has many command line flags, like -o or --option:

Split these up into a multi-line command to make it easier to see every part of the command.
Use the long name for the flags, when available.

For example, with a long command with short CLI flags like docker run --e SOURCE_CODE="$PWD" -v "$PWD":/code -v /var/run/docker.sock:/var/run/docker.sock "$CODE_QUALITY_IMAGE" /code:

job1:
  script:
    - docker run
        --env SOURCE_CODE="$PWD"
        --volume "$PWD":/code
        --volume /var/run/docker.sock:/var/run/docker.sock
        "$CODE_QUALITY_IMAGE" /code

You can also use the | and > YAML operators to split up multi-line commands.

Explain the template with comments

You can access template contents from the new file menu, and this might be the only place users see information about the template. It's important to clearly document the behavior of the template directly in the template itself.

The following guidelines cover the basic comments expected in all template submissions. Add additional comments as needed if you think the comments can help users or template reviewers.

Explain requirements and expectations

Give the details on how to use the template in # comments at the top of the file. This includes:

Repository/project requirements.
Expected behavior.
Any places that must be edited by users before using the template.
If the template should be used by copy pasting it into a configuration file, or by using it with the include keyword in an existing pipeline.
If any variables must be saved in the project's CI/CD settings.

# Use this template to publish an application that uses the ABC server.
# You can copy and paste this template into a new `.gitlab-ci.yml` file.
# You should not add this template to an existing `.gitlab-ci.yml` file by using the `include:` keyword.
#
# Requirements:
# - An ABC project with content saved in /content and tests in /test
# - A CI/CD variable named ABC-PASSWORD saved in the project CI/CD settings. The value
#   should be the password used to deploy to your ABC server.
# - An ABC server configured to listen on port 12345.
#
# You must change the URL on line 123 to point to your ABC server and port.
#
# For more information, see https://gitlab.com/example/abcserver/README.md

job1:
  ...

Explain how variables affect template behavior

If the template uses variables, explain them in # comments where they are first defined. You can skip the comment when the variable is trivially clear:

variables:                        # Good to have a comment here, for example:
  TEST_CODE_PATH: <path/to/code>  # Update this variable with the relative path to your Ruby specs

job1:
  variables:
    ERROR_MESSAGE: "The $TEST_CODE_PATH path is invalid"  # (No need for a comment here, it's already clear)
  script:
    - echo ${ERROR_MESSAGE}

Use all-caps naming for non-local variables

If you are expecting a variable to be provided via the CI/CD settings, or via the variables keyword, that variable must use all-caps naming with underscores (_) separating words.

.with_login:
  before_script:
    # SECRET_TOKEN should be provided via the project settings
    - docker login -u my-user -p "$SECRET_TOKEN my-registry

Lower-case naming can optionally be used for variables which are defined locally in one of the script keywords:

job1:
  script:
    - response="$(curl "https://example.com/json")"
    - message="$(echo "$response" | jq -r .message)"
    - 'echo "Server responded with: $message"'

Backward compatibility

A template might be dynamically included with the include:template: keyword. If you make a change to an existing template, you must make sure that it doesn't break CI/CD in existing projects.

For example, changing a job name in a template could break pipelines in an existing project. Let's say there is a template named Performance.gitlab-ci.yml with the following content:

performance:
  image: registry.gitlab.com/gitlab-org/verify-tools/performance:v0.1.0
  script: ./performance-test $TARGET_URL

and users include this template with passing an argument to the performance job. This can be done by specifying the CI/CD variable TARGET_URL in their .gitlab-ci.yml:

include:
  template: Performance.gitlab-ci.yml

performance:
  variables:
    TARGET_URL: https://awesome-app.com

If the job name performance in the template is renamed to browser-performance, the user's .gitlab-ci.yml immediately causes a lint error because there are no such jobs named performance in the included template anymore. Therefore, users have to fix their .gitlab-ci.yml that could annoy their workflow.

Please read versioning section for introducing breaking change safely.

Versioning

Versioning allows you to introduce a new template without modifying the existing one. This process is useful when we need to introduce a breaking change, but don't want to affect the existing projects that depends on the current template.

Stable version

A stable CI/CD template is a template that only introduces breaking changes in major release milestones. Name the stable version of a template as <template-name>.gitlab-ci.yml, for example Jobs/Deploy.gitlab-ci.yml.

You can make a new stable template by copying the latest template available in a major milestone release of GitLab like 13.0. All breaking changes must be announced in a blog post before the official release, for example GitLab.com is moving to 13.0, with narrow breaking changes

You can change a stable template version in a minor GitLab release like 13.1 if:

The change is not a breaking change.
The change is ported to the latest template, if one exists.

Latest version

Templates marked as latest can be updated in any release, even with breaking changes. Add .latest to the template name if it's considered the latest version, for example Jobs/Deploy.latest.gitlab-ci.yml.

When you introduce a breaking change, you must test and document the upgrade path. In general, we should not promote the latest template as the best option, as it could surprise users with unexpected problems.

If the latest template does not exist yet, you can copy the stable template.

How to include an older stable template

Users may want to use an older stable template that is not bundled in the current GitLab package. For example, the stable templates in GitLab 13.0 and GitLab 14.0 could be so different that a user wants to continue using the GitLab 13.0 template even after upgrading to GitLab 14.0.

You can add a note in the template or in documentation explaining how to use include:remote to include older template versions. If other templates are included with include: template, they can be combined with the include: remote:

# To use the v13 stable template, which is not included in v14, fetch the specific
# template from the remote template repository with the `include:remote:` keyword.
# If you fetch from the GitLab canonical project, use the following URL format:
# https://gitlab.com/gitlab-org/gitlab/-/raw/<version>/lib/gitlab/ci/templates/<template-name>
include:
  - template: Auto-DevOps.gitlab-ci.yml
  - remote: https://gitlab.com/gitlab-org/gitlab/-/raw/v13.0.1-ee/lib/gitlab/ci/templates/Jobs/Deploy.gitlab-ci.yml

Use a feature flag to roll out a `latest` template

With a major version release like 13.0 or 14.0, stable templates must be updated with their corresponding latest template versions. It may be hard to gauge the impact of this change, so use the redirect_to_latest_template_<name> feature flag to test the impact on a subset of users. Using a feature flag can help reduce the risk of reverts or rollbacks on production.

For example, to redirect the stable Jobs/Deploy template to its latest template in 25% of projects on gitlab.com:

/chatops run feature set redirect_to_latest_template_jobs_deploy 25 --actors

After you're confident the latest template can be moved to stable:

Update the stable template with the content of the latest version.
Remove the migration template from Gitlab::Template::GitlabCiYmlTemplate::TEMPLATES_WITH_LATEST_VERSION const.
Remove the corresponding feature flag.

NOTE: Feature flags are enabled by default in RSpec, so all tests are performed against the latest templates. You should also test the stable templates with stub_feature_flags(redirect_to_latest_template_<name>: false).

Testing

Each CI/CD template must be tested to make sure that it's safe to be published.

Manual QA

It's always good practice to test the template in a minimal demo project. To do so, please follow the following steps:

Create a public sample project on https://gitlab.com.
Add a .gitlab-ci.yml to the project with the proposed template.
Run pipelines and make sure that everything runs properly, in all possible cases (merge request pipelines, schedules, and so on).
Link to the project in the description of the merge request that is adding a new template.

This is useful information for reviewers to make sure the template is safe to be merged.

Make sure the new template can be selected in UI

Templates located under some directories are also selectable in the New file UI. When you add a template into one of those directories, make sure that it correctly appears in the dropdown:

Write an RSpec test

You should write an RSpec test to make sure that pipeline jobs are generated correctly:

Add a test file at spec/lib/gitlab/ci/templates/<template-category>/<template-name>_spec.rb
Test that pipeline jobs are properly created via Ci::CreatePipelineService.

Verify breaking changes

When you introduce a breaking change to a latest template, you must:

Test the upgrade path from the stable template.
Verify what kind of errors users encounter.
Document it as a troubleshooting guide.

This information is important for users when a stable template is updated in a major version GitLab release.

Add metrics

Every CI/CD template must also have metrics defined to track their use.

To add a metric definition for a new template:

Install and start the GitLab GDK.
In the gitlab directory in your GDK, check out the branch that contains the new template.

Add the template inclusion event with this Rake task:

bin/rake gitlab:usage_data:generate_ci_template_events

The task adds a section like the following to lib/gitlab/usage_data_counters/known_events/ci_templates.yml:

- name: p_ci_templates_my_template_name
  category: ci_templates
  redis_slot: ci_templates
  aggregation: weekly

Copy the value of name from the new YAML section, and add it to the weekly and monthly CI/CD template total count metrics:
- config/metrics/counts_7d/20210216184557_ci_templates_total_unique_counts_weekly.yml
- config/metrics/counts_28d/20210216184559_ci_templates_total_unique_counts_monthly.yml

Use the same name as above as the last argument in the following command to add new metric definitions:

bundle exec rails generate gitlab:usage_metric_definition:redis_hll ci_templates <template_metric_event_name>

The output should look like:

$ bundle exec rails generate gitlab:usage_metric_definition:redis_hll ci_templates p_ci_templates_my_template_name
      create  config/metrics/counts_7d/20220120073740_p_ci_templates_my_template_name_weekly.yml
      create  config/metrics/counts_28d/20220120073746_p_ci_templates_my_template_name_monthly.yml

Edit both newly generated files as follows:
- name: and performance_indicator_type:: Delete (not needed).
- introduced_by_url:: The URL of the MR adding the template.
- data_source:: Set to redis_hll.
- All other fields that have no values: Set to empty strings ('').
- Add the following to the end of each file:
```
options:
  events:
    - p_ci_templates_my_template_name
```
Commit and push the changes.

For example, these are the metrics configuration files for the 5 Minute Production App template:

The template inclusion event: lib/gitlab/usage_data_counters/known_events/ci_templates.yml#L438-L441
The weekly and monthly metrics definitions:
- config/metrics/counts_7d/20210901223501_p_ci_templates_5_minute_production_app_weekly.yml
- config/metrics/counts_28d/20210901223505_p_ci_templates_5_minute_production_app_monthly.yml
The metrics count totals:
- config/metrics/counts_7d/20210216184557_ci_templates_total_unique_counts_weekly.yml#L19
- config/metrics/counts_28d/20210216184559_ci_templates_total_unique_counts_monthly.yml#L19

Security

A template could contain malicious code. For example, a template that contains the export shell command in a job might accidentally expose secret project CI/CD variables in a job log. If you're unsure if it's secure or not, you must ask security experts for cross-validation.

Contribute CI/CD template merge requests

After your CI/CD template MR is created and labeled with ci::templates, DangerBot suggests one reviewer and one maintainer that can review your code. When your merge request is ready for review, please mention the reviewer and ask them to review your CI/CD template changes. See details in the merge request that added a DangerBot task for CI/CD template MRs.