Cordon, Drain and Node Recovery

Taking a node down for maintenance safely follows a specific three-step sequence, and skipping steps is where most maintenance-related incidents come from. kubectl cordon marks a node as unschedulable, meaning the scheduler will never place a new Pod there going forward - but critically, it does nothing at all to the Pods already running on that node; they simply keep running undisturbed. kubectl drain is the step that actually evicts the existing ordinary workloads from that node, and it does so respecting PodDisruptionBudgets (from the scheduling module) and controller ownership - it will refuse to proceed past a budget it cannot honor, rather than forcing through and breaking an intentional safety guarantee. Once maintenance is complete, kubectl uncordon marks the node schedulable again, returning its capacity to the pool. A few categories of Pods need explicit, deliberate handling during a drain rather than the default behavior: DaemonSet Pods (one-per-node agents, from the Pods module) are typically skipped by drain entirely by design, since they are meant to run on every node including this one once it returns; Pods using local, node-tied storage need an explicit decision about whether that data can be safely abandoned; and any unmanaged "bare" Pod (recall from the Pods module why those are discouraged) has no controller to recreate it elsewhere at all. When a node instead goes NotReady unexpectedly rather than through planned maintenance, investigate methodically through kubelet's own status and logs, the container runtime, disk pressure, network reachability, and certificate validity - in roughly that order of likelihood.

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: During an active database incident, a responder impatient with a hanging drain command force-evicts Pods straight through a PodDisruptionBudget that was specifically protecting the database's remaining quorum, in an effort to move faster. That forced eviction removes exactly the safety guarantee the PDB existed to enforce, and pushes the database below the minimum number of members needed for quorum - turning a contained incident into a much more serious one. The speed of completing maintenance should never be allowed to outrank the workload invariants (like quorum) that a PDB was specifically written to protect; a hanging drain is a signal to investigate why it is hanging, not an obstacle to force past.

Drain with workload invariants intact

Cordon first, inspect Pods and PDBs on the node, then use drain and read every refused eviction. DaemonSet Pods are skipped with the appropriate flag; mirror/static Pods and local data need separate decisions. After maintenance, verify kubelet, runtime, CNI, storage attachments and node conditions before uncordon. For NotReady nodes, gather journal, disk, certificate and network evidence if reachable.

Scenario: Drain evicts a Pod using emptyDir after an engineer adds delete-emptydir-data casually. The controller recreates compute but the Pod-scoped dataset is gone. Flags encode data-loss choices, not cleanup convenience.
Warning: Uncordoning a node because its Ready condition briefly turns true can reintroduce flapping capacity. Observe sustained health and a canary workload first.

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 taints-pdb-maintenance. Open /labs/kubernetes, choose taints-pdb-maintenance, 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.