I learned a lot of things by building a project from scratch: a backend DevOps tool that facilitates managing the fast update cycle in git dependencies.

Overall, the system works by running two concurrent tasks: one to check dependencies, and the other to trigger workflows. When a dependency is updated, a workflow of the dependent GitHub repository is triggered. Initially, to pass data between tasks, I used an MPSC channel, but subsequently I transitioned to a database-backed queue because the channel was not resilient enough to crashes and network errors.

In the article you can find more examples of unexpected quirks that I needed to iron out after the first iteration.