Fair share

Last week a search experiment scaled to forty pods and checkout couldn't schedule a single one. Search gets a budget: a ResourceQuota caps what the pods in a namespace may request in total.

Once a quota counts CPU or memory, the API server refuses any pod that doesn't say how much it needs — and search's pods say nothing. A LimitRange fills in defaults for containers that don't. quota.yaml and limits.yaml are ready in your directory.