Skip to main content
Warning This document has not been updated for a while now. It may be out of date.
Last updated: 11 Dec 2023

Test & build a project with GitHub Actions

GitHub Actions is an automated workflow system that GOV.UK uses for Continuous Integration (CI). In GOV.UK RFC 123 we decided that GitHub Actions is the preferred platform for GOV.UK CI usage where the wider platform integration of Jenkins is not required.

If your workflow requires the use of secrets, please talk to GOV.UK senior tech before deploying it. This is to help GOV.UK establish consistent and effective secret management for GitHub Actions across the wider alphagov GitHub organisation.

GOV.UK Conventions for GitHub Actions

Name of CI workflow file

You should name the file used for configuring the CI workflow ci.yml. It should live in the .github/workflows directory.

Name of CI workflow and jobs

The CI workflow configured above should have a name property with a value of CI. The workflow should have a number of self-contained jobs configured, for actions such as running tests, linting and security scans. These should use reusable workflows where possible. For example:

    uses: alphagov/govuk-infrastructure/.github/workflows/brakeman.yml@main

    uses: alphagov/govuk-infrastructure/.github/workflows/standardx.yml@main
      files: "'app/assets/javascripts/**/*.js'"

    uses: alphagov/govuk-infrastructure/.github/workflows/rubocop.yml@main

The jobs and steps do not need to be given a name attribute as the job/step key should be sufficiently descriptive. In contrast, the ci.yml must have a name property called CI, as this is relied upon by our automated tooling.

We previously recommended that repos should always define a job called test. Historically, this job would be responsible for running the repo’s tests, linter and so on, usually via something like bundle exec rake. In January 2023, a new convention was established, giving each task its own ‘job’ (e.g. ‘test-ruby’ for running tests and ‘lint-ruby’ for running the linter), within an overall ‘workflow’ called CI. Whilst this decoupling does make it harder to run the entire check suite locally, it allows us to take advantage of GitHub Action parallelisation, as well as better code reuse through reusable workflows.

When the CI workflow should run

The workflow should be configured to run when there is a push to the default branch (typically “main”) and when a pull_request opened. This means CI runs when a branch is merged (the push to main) and it runs against any changes introduced in a pull request. We prefer to run against a pull_request event to all push events, as this allows external contributors to have pull requests tested and because pull_request runs against a version of the codebase that is merged with the repository default branch.

Example configuration:

# ./github/workflows/ci.yml
name: CI

      - main

We previously recommended CI jobs to run on [push, pull_request] as per RFC-123 however this caused duplicate runs of the CI jobs of a pull request (one for the push another for the pull_request) which were both superfluous for our testing, a point of confusion and were depleting our allowance quotas.

Running CI on demand

To run CI on-demand, outside a pull request, for example before opening a PR, you may configure a repository to have a workflow_dispatch event so you can run it from the GitHub Actions user interface. If you do this you will need to configure the checkout action to reference the appropriate commit.

Example configuration:

name: CI

      - main
        description: 'The branch, tag or SHA to checkout'
        default: main
        type: string

  # for example
    runs-on: ubuntu-latest
      - uses: actions/checkout@v3
          ref: ${{ inputs.ref || github.ref }}

Branch protection rules

Pull Requests must not be merged until the jobs defined in the CI workflow have passed. You should always define a workflow called CI.

Reuse workflows where possible

We define a number of reusable workflows in the govuk-infrastructure repo.

For example, here’s how Content Publisher uses a reusable workflow that is defined in govuk-infrastructure:

name: CI

    name: Lint Ruby
    uses: alphagov/govuk-infrastructure/.github/workflows/rubocop.yml@main

CI workflow for Ruby gems

This differs to a simple Ruby application because Ruby gems are not tied to particular Ruby or Rails versions. Therefore we should test against various combinations of supported versions. See example in govspeak.

To do this, we use a build matrix. In this example, we test against three Ruby versions (2.7, 3.0, 3.1) and two Rails versions (6, 7). Combined, this results in 6 variations for the workflow to test against.

# .github/workflows/ci.yml
on: [push, pull_request]
  # Run the test suite against multiple Ruby and Rails versions
      fail-fast: false
        # Due to, we have to use quotes for '3.0'
        ruby: [2.7, '3.0', 3.1]
        # Test against multiple Rails versions
        gemfile: [rails_6, rails_7]
    runs-on: ubuntu-latest
      BUNDLE_GEMFILE: gemfiles/${{ matrix.gemfile }}.gemfile
    - uses: actions/checkout@v3
    - uses: ruby/setup-ruby@v1
        ruby-version: ${{ matrix.ruby }}
        bundler-cache: true
    - run: bundle exec rake

  # Branch protection rules cannot directly depend on status checks from matrix jobs.
  # So instead we define `test` as a dummy job which only runs after the preceding `test_matrix` checks have passed.
  # Solution inspired by:
    needs: test_matrix
    runs-on: ubuntu-latest
      - run: echo "All matrix tests have passed 🚀"

  # We have a shared workflow that can be used for most gem publishing needs. You may have to write
  # your own if you have a gem that is released in a complex way.
    needs: test
    if: ${{ github.ref == 'refs/heads/main' }}
      contents: write
    uses: alphagov/govuk-infrastructure/.github/workflows/publish-rubygem.yaml@main

For each Rails version, a *.gemfile should exist in a top-level directory called gemfiles.

# gemfiles/rails_7.gemfile
source ""

gem "rails", "~> 7"

gemspec path: "../"