Is Using Ephemeral Namespaces for Feature Branches a Good Strategy?

In my experience, we ran into scaling problems with having a namespace for every feature branch. We switched it up to use an ephemeral namespace for each pull request, which seems to manage resources better. Also, I'm curious about your idea of setting up a database per namespace because that can drastically increase costs. Perhaps it would make more sense for developers to share a non-production database across those ephemeral namespaces.

We've implemented a similar concept using ArgoCD alongside some bash scripting to handle the setup. We utilize Argo Workflows to kick things off and employ CloudNativePG to maintain an ephemeral state within the cluster. Eventually, I had to create some automation to clean up and garbage collect old namespaces once they were no longer needed.

I use this in some of my work projects with shared managed clusters. Even though our official tools don’t support it well, we still manage to make it work, and it’s been fantastic! Instead of a namespace for each branch, we create a new namespace for each automated build and for each developer. Each PR build generates a unique namespace, allowing complete freedom for developers. We deploy a CronJob to clean up the namespace two hours after it’s created, and if tests succeed, we destroy the namespace immediately. This method has dramatically reduced resource usage while helping every developer test in a realistic, production-like environment.

If you're not tied to using CRDs, I recommend checking out Project Capsule. It fits your needs perfectly, allowing you to apply labels and enforce a tenant prefix on namespace names. It simplifies management as you can associate your merge requests or pull requests as tenants and create the namespaces accordingly. Once you're finished with a feature branch, just clean up the tenant, and all related namespaces will be deleted automatically.