Free to read · CKA · Cluster architecture and kubectl fluency
Exam setup, contexts and time strategy
How the control plane and the nodes of a kubeadm cluster work together, how the API is organised, and the kubectl and exam habits that make every later module faster.
Lesson · 10 min. This module is free on this site. The rest of the course, the quizzes and the labs open with a free account.
The exam in one paragraph
The CKA is a remotely proctored, performance-based exam. You get 2 hours for roughly 15 to 20 tasks on real clusters running the current Kubernetes minor version (v1.37 at the time of writing). The pass mark is 66%. There are no multiple-choice questions: a grader checks the state you leave behind. During the exam you may open the Kubernetes documentation (kubernetes.io/docs and kubernetes.io/blog), the Helm docs (helm.sh/docs) and the Gateway API docs (gateway-api.sigs.k8s.io), in the browser inside the exam desktop. Always check the Linux Foundation's current exam pages before booking; rules and tools change.
The environment
You work in a remote Linux desktop with a terminal and a browser. Each task names the host (or context) it must be solved on, and you connect to it, typically with ssh <host>. The tools are on those task hosts, not on the base machine you land on.
kubectlis installed with thekalias and bash completion.- You may need root for node-level work: use
sudo -ior prefix commands withsudo. - When you finish a task,
exitback to the base machine before you start the next one. Nested ssh (ssh from one task host to another) is not supported. - Copy and paste in the terminal use
Ctrl+Shift+CandCtrl+Shift+V.
The one habit that matters most
Every task starts by connecting to the right place. Each task's header gives you the command to reach its environment, an ssh or a kubectl config use-context. Run it every time, even if you think you are already there. A perfect answer on the wrong cluster scores zero.
If a task gives you a kubeconfig context, check before you act:
kubectl config get-contexts
kubectl config use-context cluster2-admin@cluster2
kubectl config current-context
A kubeconfig file holds three lists, clusters, users and contexts, plus a current-context. A context is just a named pair of (cluster, user) with an optional default namespace. You will build kubeconfigs by hand in the security module.
Set up your shell in thirty seconds
The alias and completion are already there. A few extra lines help with YAML editing in vim. Add them on each task host where you will edit files:
cat >> ~/.vimrc <<'EOF'
set expandtab tabstop=2 shiftwidth=2 autoindent
EOF
export do="--dry-run=client -o yaml"
export now="--force --grace-period=0"
With those, k run web --image=nginx $do > web.yaml generates a skeleton and k delete pod web $now deletes at once. Do not spend more than a minute on setup; the gain is small after that.
In vim, :set paste before pasting a block keeps the indentation intact. Shift+V then > or < shifts selected lines by one level.
Reading a task
Tasks are short and precise. Every noun in them is a requirement the grader may check.
- Read the whole task once. Note names, namespaces, file paths and exact values.
- Connect to the right host or context.
- Find the relevant docs page if you need YAML you cannot generate.
- Do the work, imperatively where possible.
- Verify:
get,describe,logs,auth can-i,curl, whatever proves the result. Check spelling of every name against the task. - Leave answer files exactly in the requested format.
When a task asks for a file, the grader reads the file, not your terminal. When it asks for an object, the grader reads the cluster. Do not leave "work in progress" objects with wrong names lying around if the task forbids extra changes.
Time strategy
Two hours for up to twenty tasks is tight but fair. Plan for it.
- First pass: solve everything you can finish quickly. If a task is not converging after about eight minutes, flag it and move on.
- Second pass: return to flagged tasks with the time left.
- Weights differ. A long cluster upgrade or etcd restore may be worth much more than a one-line label task. When weights are shown, use them to decide what to skip.
- Do not chase perfection. A task is graded by checking several conditions, so partial work may still earn something; an untouched task certainly earns nothing.
- Keep ten minutes at the end to re-check answer files and the context of your riskiest tasks.
Using the docs efficiently
The docs are allowed, but searching costs time. Know where the YAML you cannot generate lives:
| You need | Search the docs for |
|---|---|
| PersistentVolume, PVC, pod using it | "Configure a Pod to Use a PersistentVolume" |
| NetworkPolicy examples | "Network Policies" |
| Ingress with paths and TLS | "Ingress" |
| Gateway and HTTPRoute | gateway-api.sigs.k8s.io "HTTP routing" |
| etcd backup and restore | "Operating etcd clusters" |
| kubeadm upgrade steps | "Upgrading kubeadm clusters" |
| CSR for a user | "Certificate Signing Requests" (normal user section) |
| Static pod | "Create static Pods" |
| Probes | "Configure Liveness, Readiness and Startup Probes" |
Copy YAML from a docs page, paste it into a file, then edit. Do not type long manifests from memory.
Breaking things safely
Some tasks ask you to edit control plane manifests or node configuration. Before you do:
cp /etc/kubernetes/manifests/kube-apiserver.yaml /root/kube-apiserver.yaml.orig
Keep the backup outside the manifests directory. If your change breaks the API server, you can put the original back and try again.
How our mock exams match this
Our full mock exams give you about the same number of tasks, the same two hours and the same 66% pass mark. Some mocks run every task on a fresh cluster of its own, so practise the "connect first" habit there too. Start one from the Mock exams page, or from the CLI:
./labctl.py exam list
./labctl.py exam start cka-full-01
./labctl.py exam grade cka-full-01
Common mistakes
- Solving a task on the previous task's cluster.
- Writing an answer file to the base machine instead of the task host.
- Forgetting
-n <namespace>and creating the object indefault. - Spending twenty minutes on one hard task while three easy ones wait.
- Leaving vim with unsaved changes and moving on.
Practise it
A broken kubeconfig is a quick way to drill contexts, clusters and users:
Practise this on a real cluster in your account: T32.