Hey everyone! I'm starting a new company alongside my current job, where eventually, the goal is to handle all SRE and platform monitoring for the new company. Meanwhile, the development team will remain with the old company. For VPS, it's pretty clear: customers pay a monthly fee for us to be on call and ensure the servers and services stay updated, while developers handle the applications. However, I'm having trouble figuring out how to split responsibilities when it comes to Kubernetes.
I have two potential plans:
**Plan A**: The platform team would take care of maintaining the platform, managing helm charts, setting up CI with a GitOps repo, monitoring the app, and updating any dependencies not included in the developers' Dockerfiles. On the other hand, the developer's role would be limited to creating the Dockerfiles themselves.
**Plan B**: The platform team would still maintain the platform and handle monitoring, but the developers would take on the responsibility for helm charts, CI with GitOps, and updating dependencies.
I've tried plan B a couple of times, but I found that once we stopped having structured sprints, the developers struggled to keep up with the demands of the project. With plan A, while it works for some projects, the developers are often confused by the helm charts and are hesitant to make any changes because they don't fully understand how they work.
Right now, I'm leaning towards plan A, but I want to stay flexible, such as allowing developers to submit merge requests for CI and helm, and helping them create compliant Docker images.
1 Answer
I think a mix of both plans could work best here. Start with plan A until the developers get some hands-on experience with platform changes related to app requests. Once everything is more mature, transition to plan B where minor changes can be managed by the dev team, leveraging existing apps, charts, and Dockerfiles as references.
But doesn’t that lead to issues where the helm charts can become messy? Like, if a developer copies a chart without removing unnecessary bits for their use case, it just turns into a cluttered mess, you know? It’s a challenge when they’re still new to Kubernetes.