A simple locking/unlocking mechanism to provide mutual exclusion in Github Actions

To prevent concurrent access to a job:

jobs:
  run_in_mutex:
    runs-on: ubuntu-latest
    name: Simple mutex test
    steps:
      - uses: actions/checkout@v3
      - name: Set up mutex
        uses: ben-z/[email protected]
      - run: |
          echo "I am protected!"
          sleep 5

By default, the gh-mutex branch in the current repo is used to store the state of locks.

To have multiple mutexes, simply specify the branch argument in the workflow config:

jobs:
  two_clients_test_client_1:
    runs-on: ubuntu-latest
    name: Two clients test (client 1)
    needs: [simple_test]
    steps:
      - uses: actions/checkout@v3
      - name: Set up mutex
        uses: ben-z/[email protected]
        with:
          branch: another-mutex
      - run: |
          echo "I am protected by the 'another-mutex' mutex!"
          sleep 5

More options such as using a different repo to store the mutex (which allows sharing a mutex between jobs from arbitrary repos) or using different access tokens can be found in action.yml.

GitHub Action has the concurrency option for preventing running multiple jobs concurrently. However, it has a queue of length 1. When multiple jobs with the same concurrency group get queued, only the currently running job and the latest queued job are kept. Other jobs are simply cancelled. There's more discussion here and it appears that GitHub does not want to add the requested cancel-pending feature any time soon (as of 2022-03-26). This GitHub action solves that issue.

Mutexes are implemented using simple spinlocks. The test-and-set functionality is provided by Git, where a git push can only succeed if the commit to be pushed is a fast-forward of what's on the remote.

1. Install act
2. Populate .github-token with a personal access token with the repo permision.
3. act --rebuild -v -s GITHUB_TOKEN=$(cat .github-token)

• actions/checkout for the authentication logic using GITHUB_TOKEN.