CKA Exercises mirrors the exam's six clusters, including the one with a missing worker

Practice organised by the five weighted exam domains

CKA Exercises is a set of exercises to help you prepare for the Certified Kubernetes Administrator exam, and the repository is organised the same way the exam is scored.

The Contents section links one file per domain, each with the exam's weight. Cluster Architecture, Installation and Configuration is 25 percent. Workloads and Scheduling is 15 percent. Services and Networking is 20 percent. Storage is 10 percent. Troubleshooting is 30 percent.

That distribution is the single most useful fact in the repository, because it tells you where the hours should go. Troubleshooting at 30 percent is the heaviest domain and the one where partial failures are most common, and Services and Networking at 20 percent is the second. Cluster Architecture at 25 percent is nearly as heavy as troubleshooting but is more memorisation than diagnosis, since it covers installation and configuration decisions that you either know or you do not.

The README also points at the most current exam curriculum in the cncf/curriculum repository, which is the right thing to follow if the weights here have drifted from the live exam.

So the structure is: a practice task set that mirrors the scoring rubric, with the heaviest domains carrying the most exercises to work through.

Six clusters, and one of them is deliberately broken

The most useful part of the README is the cluster table, because it tells you the exact topologies you get on exam day, and the point is that they are not all the same.

| Cluster | Members | CNI | Description | | :------ | :--------------------- | :------- | :--------------------------------- | | k8s | 1 master\, 2 worker | flannel | k8s cluster | | hk8s | 1 master\, 2 worker | calico | k8s cluster | | bk8s | 1 master\, 1 worker | flannel | k8s cluster | | wk8s | 1 master\, 2 worker | flannel | k8s cluster | | ek8s | 1 master\, 2 worker | flannel | k8s cluster | | ik8s | 1 master\, 1 base node | loopback | k8s cluster \- missing worker node |

Five of the six are a plain single control plane with two workers, flannel except for hk8s which uses calico, and bk8s which has only one worker. The sixth, ik8s, is the one to pay attention to: it has a single base node and is described as missing a worker node.

That is a deliberate fault, and it is the design of the whole set. An exam environment where every cluster is healthy teaches you that every cluster is healthy, and then the real exam gives you a node in a failed state and you have no idea which command tells you that. Practising on ik8s is how you learn to check node status before you start diagnosing a pod.

The CNI variation matters for the same reason. If you only ever see flannel, you will not recognise a calico network policy problem when you meet one.

The exam rules, restated because they shape how you practise

The README spends a surprising amount of its length on the mechanics of the exam, and that is deliberate, because the mechanics decide whether your practice is worth anything.

The format of the exam is entirely in a hands-on, command-line environment. You can take the exam at home or in a testing center, and you must complete it in 180 minutes. The cost is US $395.00.

The rule that matters most for practice is the tab restriction. During the exam you may have one and ONLY one of a specific set of tabs open at all times: kubernetes.io/docs, kubernetes.io/blog, the Gateway API site, and helm.sh/docs. Not the Kubernetes API reference, not GitHub, not a search engine. Everything else has to come from your own knowledge and from the cluster in front of you.

That is worth thinking about when you practise, because it determines what you can realistically look up while you learn. If your practice environment has a browser with search open, you are training a skill you will not be allowed to use.

The README also links the official exam handbook for important instructions and a separate FAQ page, plus a system check link for anyone unsure whether their equipment is right to take the exam at home.

And there is a free option for more practice: killercoda.com/chadmcrowell is described as a free Kubernetes lab environment for exam-like scenarios, which is useful precisely because it can be set up to match the exam topologies.

Grading is on outcomes, so practise outcomes

The grading information in the README is the part that changes how you should use the exercises, and it is easy to skim past because it sits under a heading about additional information.

The CKA exam is graded for outcome only, meaning the end state of the system. The path that an exam taker may have taken to get to the outcome is not evaluated, so a candidate can take any path they want as long as it achieves the correct outcome. Incomplete work, meaning work that was input but did not lead to the correct outcome, will not be evaluated.

That has a sharp practical consequence. A partially correct manifest scores zero for that item, however much of it was right.

The second point is about partial credit. Some exam items have multiple parts and therefore multiple checks, one for each verifiable component of the answer, and candidates are given credit for each successful check, so partial credit is possible on such items.

The third is about independence. Exam items are set up to be independent of one another. As long as the candidate does exactly what the questions ask, there will be no dependencies or conflicts, and correctly achieving the outcome earns the points for that item.

So the exercise set is worth practising by checking your end state rather than by comparing your commands to a reference answer. If you reach the desired state by a route the exercise set does not expect, that is the exam's own criterion, and it is also why an exercise set that only gave you one command sequence would be teaching you the wrong thing.

Container runtime: the detail that breaks otherwise-correct answers

The Updates section, as of October 2022, covers the single most common way a candidate loses points for the right reasoning.

The CKA exam is currently on v1.31 of k8s. The removal of dockershim happened in v1.24, so expect the containerd container runtime if you are taking the exam today and into the future. You can view the container runtime in use with the command `k get no -o wide`.

The output will look similar to this.

NAME           STATUS   ROLES           AGE   VERSION   INTERNAL-IP   EXTERNAL-IP   OS-IMAGE             KERNEL-VERSION      CONTAINER-RUNTIME
controlplane   Ready    control-plane   34d   v1.28.1   172.30.1.2    <none>        Ubuntu 20.04.5 LTS   5.4.0-131-generic   containerd://1.6.12
node01         Ready    <none>          34d   v1.28.1   172.30.2.2    <none>        Ubuntu 20.04.5 LTS   5.4.0-131-generic   containerd://1.6.12

The reasoning is straightforward and the failure mode is not. If a question involves a container image and you reason about docker in a cluster that runs containerd, the reasoning is fine and the answer is wrong. The image reference format differs, and a `docker` command does not exist on the node.

The fix is cheap and you should do it before you start the practice exercises rather than discovering it mid-question: run the check, see the CONTAINER-RUNTIME column, and let that inform every image-related answer for the rest of the session.

The README also covers the exam release cycle, noting that the exams are upgraded to the latest version of k8s within 4 to 6 weeks of the version being released, and links the Dockershim FAQ.

Where CKA Exercises is the wrong resource

Three cases, and they are about what you want out of it.

The first is if you are learning Kubernetes rather than certifying for it. These are exam exercises: the tasks assume you know kubectl, they assume you know what a Pod, a Deployment and a Service are, and the goal is to reproduce a specific end state under time pressure. If you do not yet have that vocabulary, a book or a course will get you further, and these tasks will read as an arbitrary list of commands.

The second is if you want to build or operate Kubernetes in production. Nothing here teaches you about clusters, upgrades, networking design or capacity. It teaches you the subset of cluster operations the exam happens to sample, which is a much narrower thing than the job.

The third is if your exam version has moved on. The exam is upgraded to the latest Kubernetes version within 4 to 6 weeks of that version's release, and the curriculum weights live in the cncf/curriculum repository and change. The repository also carries an IMPORTANT EXAM CHANGES notice at the top of the README pointing at a cka-changes-2024 directory, and the Updates section is dated October 2022. The last push to the repository was on 2026-08-20, so the content is being touched, but the individual notes carry their own dates and you should read them as dated rather than current.

One practical gap to plan around: the exercises assume an environment. There is no script here that provisions the six clusters. The free lab on killercoda is the obvious route, and matching its cluster shapes to the table above is the part worth doing carefully.

Editorial conclusion

Adopt CKA-Exercises if you are preparing for the CKA and want practice in the same cluster topologies the exam hands you, including the missing worker node. Do not adopt it as a Kubernetes learning resource in general, because the tasks assume you already know the commands and the goal is to pass a timed exam rather than to understand the system. Verify first which Kubernetes version your exam will run, since the curriculum and the exam environment both move, and work through the container runtime check before you start, because a question about a pod image will fail if you assume docker where the cluster runs containerd.

Frequently asked questions

Official sources

1. chadmcrowell/CKA-Exercises on GitHub
2. Issues
3. README