Skip to content

The CKAD drill block, as a timed harness

Artifact — a repeatable timed session. Run it until every task lands inside its clock twice in a row, on a cluster that you did not prepare.

This is a different activity from every exercise above it. Everything above optimises for referents and fluency. This block optimises for speed and correctness under a clock. Do not blend the two. Do not read source during this block — the phase's framing says why.

Topology — pair, freshly provisioned. It is not the cluster that you spent three weeks on. That cluster is gone by the incident note, and it is gone deliberately: a drill against a cluster that you built is a drill against your own memory of it. Use two nodes rather than solo, because the exposure tasks need a NodePort that lands somewhere you are not.

The harness
Clock Per task, below. Start it before you read the task, as the exam does.
Docs Only what the tab policy permits. Nothing else open.
Environment A cluster you did not prepare, with the context set to the wrong namespace. Set it wrong on purpose each session. The first thing that every exam task needs is -n.
Pass mark Every task inside its clock, twice in consecutive sessions. One clean run is luck.
Scoring Binary per task. A task that needed a second look did not pass, even if the object is right.

The task list — the timed items are the phase's own checklist, and not a new set.

  1. A Deployment, a Service and a ConfigMap-configured app, exposed through NodePort — 4 minutes, from memory, with no docs.
  2. Diagnose a wedged rollout, and roll it back from kubectl output alone — 3 minutes. Use the triage order that you wrote. Run it against all three break-shapes, and not only against the probe one.
  3. A Pod with a dropped-and-re-added capability and runAsNonRoot, proven with capsh inside the container — 3 minutes, per the capability exercise.
  4. What not to drill is as timed as what to drill. The phase says which topics are wasted effort. The weights, the mechanics and the practice-resource verdicts are in the certs strand. Neither source is repeated here. For the drill list and the tactics, see speed tactics and practice resources.

Expect — task 1 fails first, and it fails on typing rather than on recall. kubectl create deploy --image=... --dry-run=client -o yaml plus an editor beats writing YAML by hand every time. Finding that out under a clock is the point of the clock. Task 2 improves most between sessions, because the triage order is a list that you own, and not a fact that you remember.

Write down — for each session, which tasks passed, and the wall-clock time for each one. The trend is the signal, and no single run is.

Teardown — run just tofu labs destroy when the block ends. This ends P1's cluster for good. P2 provisions etcd-only, and it needs the RAM. The exam result is deliberately not a gate condition.