On a shared cluster, platform teams cap each namespace. A ResourceQuota limits the total a namespace may use: number of Pods, Services or claims, and the sum of CPU and memory requests and limits. Once a quota covers requests, every new container must state them, or its Pod is refused.
Refusals are easy to miss. A Pod you apply directly fails with a Forbidden error, but a Deployment's Pods are created by its ReplicaSet, so the only trace is a FailedCreate event and a READY column stuck below the target. kubectl describe resourcequota shows Used next to Hard.
A LimitRange works per container: it fills in a default request and limit when a manifest has none, and can enforce a minimum and maximum. Both act when a Pod is admitted, so changing them never edits running Pods; restart a Deployment to apply new defaults.
A namespace cap and the trail it leaves when it says no.