Resource Requests, Limits, and QoS Classes

A Pod stuck Pending with 0/3 nodes available: insufficient cpu. Another OOM-killed at 2 AM, restarting in a loop. Both problems trace back to the same root cause: how you configured resource requests, limits, or didn't configure them at all.

Resource management sits at the intersection of scheduling, enforcement, and eviction. The kube-scheduler uses requests to place Pods. The kubelet uses limits to enforce boundaries. And QoS classes determine who gets evicted first when a node runs out of memory. Get this wrong and you'll either waste capacity or wake up to pager alerts. Get it right and your cluster runs predictably under pressure.