A webhook handler that triggers AWS CodePipelines for mono repos, implemented as a serverless AWS Lambda function.

If you are building a big application consisting of multiple microservices, or you are building a data crunching application with lots of data ingestion pipelines, you probably want to use a mono repo. Until now, this has been a pain to build properly with AWS CodePipeline, because CodePipeline still doesn't support mono repos, even though this feature has been requested for over two years. Let's say your application is structured like this:

mono-repo/
├── microservice-1/
│   ├── buildspec.yml
│   ├── handler.js
│   └── README.md
├── microservice-2/
│   ├── buildspec.yml
│   └── main.py
│ ...

When a commit changes files in microservice-1, you only want to build this specific folder, not the whole repository (this can take a long time and get expensive with a big mono repo). The CodePipeline WebhookFilterRule is insufficient when trying to model this workflow, because it does not allow for regex matching. Strangely, the CodeBuild WebhookFilter does support it.

To solve this problem, create a CodePipeline for every microservice/subrepo that exists in your application and let this webhook figure out which CodePipelines need to be started.

You will need Poetry, serverless and npm. Run

poetry install --no-dev
npm install

to install the required dependencies.

The webhook handler needs a secret string GithubSecret in the AWS Parameter Store to authenticate the GitHub requests.

To deploy to AWS, run

serverless deploy --env prd

Add the URL of the Lambda function to your mono repo on Github via Settings - Webhook - Add webhook. Add the value of GithubSecet to the webhook under Secret.

Check if everything works as expected by looking at the response to the initial ping request sent by GitHub. If the response says 'Ping received', the webhook handler is ready. To debug problems, first look at the response to the webhook request, it'll inform you which CodePipelines have been started and which CodePipelines could not be found. For more in-depth debugging, look at the Cloudwatch logs for the lambda function.

There are two configuration options:

• target_branch: Which branch the webhook handler should listen on. If this is set to master, and GitHub triggers the webhook with a request containing commits on feature/add-logging, the request will be dismissed.
• prefix_repo_name: If set to true, the name of the mono repo is prepended to the names of the subfolders when building the names of CodePipelines to start. Using the example directory structure from above:
  • prefix_repo_name is true: CodePipeline mono-repo-microservice-1 will be started.
  • prefix_repo_name is false: CodePipeline microservice-1 will be started.

Pull requests are welcome. To start development, install the dev dependencies with

poetry install

Run tests with

poetry run pytest

TODO:

• implement a blacklist of directories that do not map to CodePipelines