kdeldycke / workflows

⚙️ Central repository for all workflows used in my projects

Workflows

Maintaining project takes time. This repository contains workflows to automate most of the boring tasks.

These workflows are mostly used for Poetry-based Python CLI and their documentation, but not only. They're all reusable GitHub actions workflows.

Reasons for a centralized workflow repository:

• reusability of course: no need to update dozens of repository where 95% of workflows are the same
• centralize all dependencies pertaining to automation: think of the point-release of an action that triggers dependabot upgrade to all your repositories depending on it

Guidelines

I don't want to copy-n-past, keep in sync and maintain another Nth CI/CD file at the root of my repositories.

So my policy is: move every repository-specific config in a pyproject.toml file, or hide the gory details in a reused workflow.

Why all these requirements/*.txt files?

Let's look for example at the lint-yaml job from .github/workflows/lint.yaml. Here we only need the yamllint CLI. This CLI is distributed on PyPi. So before executing it, we could have simply run the following step:

  - name: Install yamllint
    run: |
      pip install yamllint

Instead, we install it via the requirements/yamllint.txt file.

Why? Because I want the version of yamllint to be pinned. By pinning it, I make the workflow stable, predictable and reproducible.

So why use a dedicated requirements file? Why don't we simply add the version? Like this:

  - name: Install yamllint
    run: |
      pip install yamllint==1.35.1

That would indeed pin the version. But it requires the maintainer (me) to keep track of new release and update manually the version string. That's a lot of work. And I'm lazy. So this should be automated.

To automate that, the only practical way I found was to rely on dependabot. But dependabot cannot update arbitrary versions in run: YAML blocks. It only supports requirements.txt and pyproject.toml files for Python projects.

So to keep track of new versions of dependencies while keeping them stable, we've hard-coded all Python libraries and CLIs in the requirements/*.txt files. All with pinned versions.

And for the case we need to install all dependencies in one go, we have a requirements.txt file at the root that is referencing all files from the requirements/ subfolder.

Permissions and token

This repository updates itself via GitHub actions. It particularly updates its own YAML files in .github/workflows. That's forbidden by default. So we need extra permissions.

Usually, to grant special permissions to some jobs, you use the permissions parameter in workflow files. It looks like this:

on: (...)

jobs:

  my-job:
    runs-on: ubuntu-latest
    permissions:
      contents: write
      pull-requests: write

    steps: (...)

But the contents: write permission doesn't allow write access to the workflow files in the .github subfolder. There is actions: write, but it only covers workflow runs, not their YAML source file. Even a permissions: write-all doesn't work. So you cannot use the permissions parameter to allow a repository's workflow update its own workflow files.

You will always end up with this kind or errors:

   ! [remote rejected] branch_xxx -> branch_xxx (refusing to allow a GitHub App to create or update workflow `.github/workflows/my_workflow.yaml` without `workflows` permission)

  error: failed to push some refs to 'https://github.com/kdeldycke/my-repo'

Note

That's also why the Settings > Actions > General > Workflow permissions parameter on your repository has no effect on this issue, even with the Read and write permissions set:

To bypass the limitation, we rely on a custom access token. By convention, we call it WORKFLOW_UPDATE_GITHUB_PAT. It will be used, in place of the default secrets.GITHUB_TOKEN, in steps in which we need to change the workflow YAML files.

To create this custom WORKFLOW_UPDATE_GITHUB_PAT:

• From your GitHub user, go to Settings > Developer Settings > Personal Access Tokens > Fine-grained tokens
• Click on the Generate new token button
• Choose a good token name like workflow-self-update to make your intention clear
• Choose Only select repositories and the list the repositories in needs of updating their workflow YAML files
• In the Repository permissions drop-down, sets:
  • Contents: Access: **Read and Write**
  • Metadata (mandatory): Access: **Read-only**
  • Pull Requests: Access: **Read and Write**
  • Workflows: Access: **Read and Write**

    [!NOTE] This is the only place where I can have control over the Workflows permission, which is not supported by the permissions: parameter in YAML files.

• Now save these parameters and copy the github_pat_XXXX secret token
• Got to your repo > Settings > Security > Secrets and variables > Actions > Secrets > Repository secrets and click New repository secrets
• Name your secret WORKFLOW_UPDATE_GITHUB_PAT and copy the github_pat_XXXX token in the Secret field

Now re-run your actions and they should be able to update the workflow files in .github folder without the refusing to allow a GitHub App to create or update workflow error.

Release management

It turns out Release Engineering is a full-time job, and full of edge-cases.

Rust has cargo-dist. Go has... ? But there is no equivalent for Python.

So I made up a release.yaml workflow, which:

1. Extracts project metadata from pyproject.toml
2. Generates a build matrix of all commits / os / arch / CLI entry points
3. Build Python wheel with Twine
4. Compile binaries of all CLI with Nuitka
5. Tag the release commit in Git
6. Publish new version to PyPi
7. Publish a GitHub release
8. Attach and rename build artifacts to it

Changelog

A detailed changelog is available.

Used in

Check these projects to get real-life examples of usage and inspiration:

• Awesome Falsehood - Falsehoods Programmers Believe in.
• Awesome Engineering Team Management - How to transition from software development to engineering management.
• Awesome IAM - Identity and Access Management knowledge for cloud platforms.
• Awesome Billing - Billing & Payments knowledge for cloud platforms.
• Meta Package Manager - A unifying CLI for multiple package managers.
• Mail Deduplicate - A CLI to deduplicate similar emails.
• dotfiles - macOS dotfiles for Python developers.
• Click Extra - Extra colorization and configuration loading for Click.
• Wiki bot - A bot which provides features from Wikipedia like summary, title searches, location API etc.
• workflows - Itself. Eat your own dog-food.
• Stock Analysis - Simple to use interfaces for basic technical analysis of stocks.
• GeneticTabler - Time Table Scheduler using Genetic Algorithms.
• Excel Write - Optimised way to write in excel files.

Feel free to send a PR to add your project in this list if you are relying on these scripts.

Release process

All steps of the release process and version management are automated in the changelog.yaml and release.yaml workflows.

All there's left to do is to:

• check the open draft prepare-release PR and its changes,
• click the Ready for review button,
• click the Rebase and merge button,
• let the workflows tag the release and set back the main branch into a development state.