Diagnose Pressure and OOM

Diagnosing a resource-related problem well requires distinguishing between several distinct failure modes that can all look superficially similar from a distance. An application OOM is the application's own process managing its memory badly and crashing on its own terms (sometimes with its own error message). A container (cgroup) OOM is what was described in the previous section - the container runtime enforcing a memory limit by killing the process outright. Node-pressure eviction is a different mechanism entirely: the kubelet on an overall memory-starved node proactively evicts entire Pods (starting with lowest QoS priority) to protect the node itself from becoming completely unresponsive, which is a node-level decision, not a per-container limit being crossed. On the CPU side, throttling (from the previous section) can produce real, user-visible latency spikes without ever showing high overall node CPU utilization, because throttling happens per-container against its own limit, regardless of how idle the rest of the node is. The practical skill here is comparing usage, requests, and limits together, inspecting lastState, node conditions, and eviction events specifically, and measuring things like working-set memory and throttled-seconds over short, meaningful time windows - not broad averages, which hide exactly the spikes that actually caused user-facing problems.

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 dashboard shows average CPU utilization sitting at a comfortable 30% across the relevant time period, which looks like there is plenty of headroom - yet users are reporting real, frequent request timeouts during that same window. The explanation is that short, intense bursts of CPU demand are repeatedly hitting a tight CPU limit and getting throttled for brief periods, and those brief throttling episodes are exactly when the timeouts happen - but they get completely smoothed away and hidden once averaged over a longer window. Averages, by their nature, erase precisely the short saturation intervals that actual users experienced; short-window and percentile-based views are what expose them.

Identify which memory failure occurred

Inspect lastState.terminated.reason, exit code, restart count, node MemoryPressure, eviction events and application logs. Container OOM means its cgroup crossed a limit; node-pressure eviction means kubelet reclaimed the node based on QoS and usage; an application-managed heap OOM may exit cleanly with its own message. For CPU, inspect throttled seconds and latency at short windows, not only average utilization.

Scenario: A Java process sees a 512Mi limit but configures a 512Mi heap, leaving no room for metaspace, threads and native buffers. It is OOMKilled below apparent heap capacity. Budget total resident memory.
Tip: Compare the same time interval across usage, limit, restarts and request latency before changing values.

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.