Requests, Limits and QoS

Every container can declare two separate numbers for CPU and memory, and confusing them is one of the most common Kubernetes mistakes: a request is what the scheduler (from the scheduling module) uses to decide which node has room for this Pod - it is a reservation, not a cap - while a limit is what the container runtime actually enforces as a hard ceiling once the Pod is running. CPU and memory behave very differently once a limit is hit, and that difference explains most confusing incidents in this area: CPU is compressible, meaning a container that tries to use more than its CPU limit is simply throttled - slowed down - without being killed, whereas memory is incompressible, so a container that tries to exceed its memory limit gets killed outright (an OOM kill, out-of-memory) because there is no way to "slow down" a memory allocation the way you can slow down a computation. Every Pod is also assigned one of three Quality of Service (QoS) classes purely from the relationship between its requests and limits across all its containers - Guaranteed (requests equal limits everywhere), Burstable (requests are set but below limits, or only some containers set them), or BestEffort (no requests or limits set at all) - and this classification directly determines eviction priority under node pressure, with BestEffort Pods evicted first. Because requests directly influence both scheduling and priority under contention, they should be derived from actually observed usage distributions over real traffic, not a round number picked without data.

Analogy: Kubernetes is a thermostat, not a remote control. You declare the temperature; independent controllers measure reality and keep acting. Debugging means finding which sensor, rule or actuator prevents convergence.

Read the object as evidence

kubectl get RESOURCE NAME -n NAMESPACE -o yaml
kubectl describe RESOURCE NAME -n NAMESPACE
kubectl get events -n NAMESPACE --sort-by=.metadata.creationTimestamp
kubectl explain RESOURCE.spec

Do not memorize these as a ritual. The first command exposes desired and observed state, the second connects conditions and events, the third supplies a timeline, and the fourth checks the server's schema. Compare what the controller was asked to do with what it reports doing. Then test the narrowest hypothesis.

Scenario: A container is given a 128Mi memory limit, but its actual real-world usage routinely peaks at around 160Mi during normal operation - and it keeps getting killed and restarted. This is not a bug or a mysterious lifecycle problem; it is the memory limit enforcement working exactly as designed, correctly killing a container that is trying to use more memory than it was ever allowed. Before assuming something is wrong with the application's lifecycle or restart behavior, always check the container's lastState.terminated.reason field for OOMKilled - it directly confirms whether this is enforcement doing its job, in which case the limit itself (or the application's actual memory usage) is what needs adjusting, not the restart mechanism.

Distinguish reservation from enforcement

CPU and memory requests drive scheduling. CPU limits normally enforce throttling; memory limits can lead to cgroup OOM kills because memory is not compressible. QoS follows the relationship of requests and limits across every container, including sidecars. Size requests from percentiles and service objectives, then reserve rollout and failure headroom rather than setting request equal to a quiet average.

Scenario: An app requests 100m, uses 900m steadily, and has no CPU limit. It runs until contention, then receives far less CPU because scheduling never reserved its real need. Requests influence both placement and relative contention.
Warning: A very high limit is not capacity. If no node can accommodate the request or memory physically runs out, declared limits cannot create resources.

Production reasoning

Ask four questions: Who owns this object? What dependency must become ready next? Which controller reports the blocking condition? What evidence would disprove my current theory? This prevents symptom-driven changes. Record the context, namespace, object generation, image digest and recent rollout before mutation; a recreated Pod may erase the evidence you needed.

Warning: Running is not the same as ready, healthy, durable or correct. Kubernetes status is layered. Confirm the application-level outcome as well as the object state.
Goal: Put this model into practice in the Kubernetes lab resource-requests-oom. Open /labs/kubernetes, choose resource-requests-oom, predict the failure path before changing anything, then use the simulator's check command to validate the finished state.

Deliberate practice

Before the lab, write the expected object relationship and the first three commands you will run. Afterward, explain why the fix converged and name one tempting change that would only mask the symptom. Repeat using an explicit namespace and a structured output format. This prediction-observation-explanation loop is what turns command familiarity into production judgment.

45-minute investigation

  1. Map (5 min): draw the owner-to-child chain and mark every namespace, selector, identity and dependency involved.
  2. Predict (5 min): write one expected status condition, one likely event and one log or metric signal before opening the lab.
  3. Observe (10 min): collect YAML, describe output and ordered events. Do not mutate state. Record which observation disproves your first theory.
  4. Repair (15 min): make the smallest declarative correction, watch the responsible controller converge, and verify the user-facing path rather than stopping at Running.
  5. Stress (10 min): change one relevant constraint - replica count, label, readiness, resource value or placement rule - predict the outcome, observe it, then restore the known-good declaration.
Tip: Keep a short incident note with symptom, evidence, hypothesis, change, result. Across four sections this produces a reusable runbook instead of a pile of remembered commands.