Investigation: Kubernetes cluster scale limitation
Utilize this issue to discover what issues we need to solve when it comes to running GitLab.com at the scale that we need. There is suspicion that one single cluster will not suffice for GitLab.com. While we do utilize a multi-zone configuration in GKE, though, if Google were to loose this region, we'd be down. If something went wrong with, say the Kubernetes API services across all three master nodes in each zone, we'll loose the ability to control what is running in this region.
- Determine if 1 cluster is indeed enough to run GitLab.com - we already know we lack regional redundancy already
- Determine what our tooling shortfalls are to operate against multiple clusters
- Where are there opportunities to dog food our helm chart in a multi-cluster configuration
- Are there systems are available that would allow us to expand our cluster, federation for example