Quorum, Replication and Kafka

It is tempting to assume that if kubectl get pods shows every StatefulSet replica as Running and Ready, the distributed system they form must be healthy - but Kubernetes' notion of "ready" (defined by whatever probe you configured) and a distributed protocol's own notion of health are genuinely different things, measured by different systems, and they can disagree. Kafka is the clearest example the labs use: a broker Pod can be perfectly Running from Kubernetes' point of view while having fallen out of the ISR (in-sync replicas - the set of replicas actually caught up and eligible to become leader), or while consumer lag (how far behind consumers are from the latest messages) climbs steadily. Kubernetes readiness alone cannot see any of that, because it never looks inside the application protocol - it only knows what your probe told it. Maintaining a healthy distributed system therefore requires two layers of protection working together during any maintenance: Kubernetes-level guardrails like PodDisruptionBudgets and spreading replicas across real failure domains (covered in the scheduling module), and separately watching the application's own protocol-level health metrics, which only the application itself can report.

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: Every Kafka broker Pod shows as Running with a passing readiness probe - by every signal Kubernetes exposes, everything looks healthy. Yet one broker has quietly fallen out of the ISR for several partitions, and consumer lag on those partitions is climbing steadily, both invisible to Kubernetes because neither is something a generic readiness probe checks. Draining or restarting another broker at this exact moment, believing the cluster is fully healthy because Kubernetes says so, could push a partition below its minimum in-sync-replica count and make it refuse writes entirely. Pod phase reflects infrastructure health; it says nothing about application-level correctness, which has to be observed through the application's own metrics.

Protect application quorum

Replica count, Pod readiness and protocol health are different. For Kafka, inspect partition leaders, under-replicated partitions, ISR and consumer lag. For Redis or databases, inspect roles, replication offset and quorum. Spread members across nodes and zones, use a realistic PDB, and sequence maintenance one member at a time while application metrics remain healthy.

Scenario: All three Kafka Pods are Ready, but one broker has fallen out of ISR and partitions have minimum in-sync replicas two. Draining another broker can make writes unavailable despite green Kubernetes status.
Warning: A readiness probe that only checks an open port may admit a recovering member before it has caught up. Probe semantics should reflect whether serving this member preserves protocol guarantees.

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 kafka-isr-consumer-lag. Open /labs/kubernetes, choose kafka-isr-consumer-lag, 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.