fd.o costs: excessive kube cluster?
When I go look at the fdo-gitlab-prod cluster, it gives a warning about excessive resource allocation. This seems to be because gitlab isn't actually requesting significant CPU in its kube pods, so GKE doesn't know how much demand we really have. But even if we did say that 8 CPUs were used by gitlab-prod, and some by its postgres, and etc., we'd still have idle boxes I think. gke-fdo-gitlab-prod-n16-default-pool-e665a9be-srz2
right now has just two gitlab-prod-gitlab-backup pods, and gke-fdo-gitlab-prod-n16-default-pool-e665a9be-rbsz
has one back up and Mesa's marge bot.
Should we scale down to one node? One node and one spare to migrate to during maintenance? I'm not too sure how our availability is managed in the current deployment.