Rolling Updates Without Wishful Thinking

A rolling update - the default Deployment strategy - only counts as genuinely safe if three separate things are all true at once: readiness must actually reflect the Pod's real ability to serve traffic (not just "the process started"), the rollout capacity envelope from the previous section must be large enough to cover the surge without starving remaining capacity, and outgoing Pods must be given time to finish in-flight requests before being killed (the graceful termination sequence from the Pods module). You can watch and control this process directly: kubectl rollout status blocks until the rollout finishes or fails, kubectl rollout history shows previous revisions, and pause/resume/undo let you freeze a rollout mid-flight or instantly roll back to the previous ReplicaSet. Two safety fields close important gaps: progressDeadlineSeconds tells the Deployment controller how long to wait for visible progress before marking the rollout as stalled (so a silently stuck release gets flagged rather than hanging forever), and minReadySeconds requires a new Pod to stay ready for a sustained period - not just pass one instantaneous check - before it counts toward rollout progress.

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 new version's health endpoint returns HTTP 200 immediately on startup, before its internal caches have actually warmed up. Kubernetes' readiness probe sees that 200 and happily starts routing real user traffic to the Pod, causing a latency spike as every request now has to do the slow, uncached work - yet the rollout itself reports as perfectly healthy the whole time, because nothing it is measuring detected the problem. Probes have to represent the application's true ability to serve production-quality responses, not merely that the process is alive.

Calculate the rollout envelope

For ten replicas with maxSurge: 25% and maxUnavailable: 20%, the rollout may create up to thirteen total Pods and allow two unavailable after percentage rounding rules. Verify real capacity, quota and disruption constraints before release. Watch kubectl rollout status deploy/web, ReplicaSet counts and endpoint readiness together. A rollout completing means the controller met its conditions, not that latency or errors stayed acceptable.

Scenario: New Pods become ready after one successful probe, then fail under traffic. Add startup protection, representative readiness and minReadySeconds; use application error and latency gates outside Kubernetes.
Scenario: Old Pods receive TERM while a load balancer still sends connections. Coordinate endpoint removal, preStop only when necessary, application drain behavior and grace period.

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 rolling-update-hpa-scaling. Open /labs/kubernetes, choose rolling-update-hpa-scaling, 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.