Skip to content
Hands-on IT IT Drills

Day 2: Exam Craft

Goal: You work with kubectl so fast that the clock is not your opponent in the exam. · Duration: approx. 45 min · Prerequisite: Day 1

Briefing

The exam has about 17 tasks in 120 minutes – roughly seven minutes per task. If you type YAML by hand, you lose. Today you train the moves you need on every following day:

  • Alias and autocompletion
  • Generate YAML instead of writing it (--dry-run=client -o yaml)
  • Switch contexts and namespaces
  • kubectl explain instead of searching the docs
  • Filter output with jsonpath, custom-columns and --sort-by

Drill 1: Alias and autocompletion

On cp1, into ~/.bashrc:

cat <<'EOF' >> ~/.bashrc
source <(kubectl completion bash)
alias k=kubectl
complete -o default -F __start_kubectl k
export do="--dry-run=client -o yaml"
EOF
source ~/.bashrc
type k

Expected:

k is aliased to `kubectl'

Now try k get po -n kube-<TAB> – the namespace is completed.

Tip: In the exam environment, k and completion are usually already set up. Check it in the first minute with type k.

Save your progress? Sign up for free

Drill 2: Generate YAML instead of writing it

k create ns tag02
k create deployment api --image=nginx:1.29 --replicas=3 $do

Expected (shortened):

apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: api
  name: api
spec:
  replicas: 3
  selector:
    matchLabels:
      app: api
  ...
      containers:
      - image: nginx:1.29
        name: nginx

A pod with its own command – everything after -- becomes the container command:

k run tmp --image=busybox:1.37 --restart=Never --command $do -- sleep 3600

Expected (excerpt):

  containers:
  - command:
    - sleep
    - "3600"
    image: busybox:1.37
    name: tmp

Remember: k create knows deployment, service, configmap, secret, job, cronjob, role, rolebinding, serviceaccount, ingress, quota, priorityclass – almost everything that comes up in the exam.

Save your progress? Sign up for free

Drill 3: Contexts and namespaces

Every exam task starts with “Set the context”. If you forget it, you solve the task in the wrong cluster – zero points.

k config get-contexts
k config set-context dev --cluster=kubernetes --user=kubernetes-admin --namespace=tag02
k config use-context dev
k config get-contexts

Expected:

CURRENT   NAME                          CLUSTER      AUTHINFO           NAMESPACE
*         dev                           kubernetes   kubernetes-admin   tag02
          kubernetes-admin@kubernetes   kubernetes   kubernetes-admin   default

Change only the namespace of the current context and check it:

k config set-context --current --namespace=default
k config view --minify -o jsonpath='{..namespace}{"\n"}'
k config use-context kubernetes-admin@kubernetes

Expected:

Context "dev" modified.
default
Switched to context "kubernetes-admin@kubernetes".

Save your progress? Sign up for free

Drill 4: kubectl explain

Which fields does an update strategy have?

k explain deployment.spec.strategy --recursive | sed -n '/FIELDS/,$p'

Expected:

FIELDS:
  rollingUpdate <RollingUpdateDeployment>
    maxSurge    <IntOrString>
    maxUnavailable  <IntOrString>
  type  <string>
  enum: Recreate, RollingUpdate

A single field with its description:

k explain pod.spec.containers.livenessProbe.httpGet | head -9

Expected:

KIND:       Pod
VERSION:    v1

FIELD: httpGet <HTTPGetAction>


DESCRIPTION:
    HTTPGet specifies an HTTP GET request to perform.
    HTTPGetAction describes an action based on HTTP Get requests.

Save your progress? Sign up for free

Drill 5: Filter output

Name and IP of all nodes, one line per node:

J='{range .items[*]}{.metadata.name}{" "}{.status.addresses[0].address}{"\n"}{end}'
k get nodes -o jsonpath="$J"

Expected:

cp1 172.16.0.126
worker1 172.16.0.228
worker2 172.16.0.83

Custom columns, sorted by node:

k get pods -A -o custom-columns=NAME:.metadata.name,NODE:.spec.nodeName \
  --sort-by=.spec.nodeName | head -4

Expected (shortened):

NAME                                      NODE
etcd-cp1                                  cp1
kube-apiserver-cp1                        cp1
tigera-operator-85dfd45867-dwbmx          cp1

The kubelet versions of all nodes in one line:

k get nodes -o jsonpath='{.items[*].status.nodeInfo.kubeletVersion}{"\n"}'

Expected: v1.36.6 v1.36.6 v1.36.6

Tip: Exam tasks often ask you to “write the result to the file /opt/…”. Then append > /opt/… and verify with cat.

Save your progress? Sign up for free

Drill 6: Immutable fields and replace –force

Many pod fields can no longer be changed after startup. Create a pod and try to give it a memory limit afterwards:

mkdir -p ~/cka/tag02 && cd ~/cka/tag02
k run web --image=nginx:1.29 -n tag02 $do > web.yaml
k apply -f web.yaml
k set resources pod web -n tag02 --limits=memory=64Mi $do > web2.yaml
k apply -f web2.yaml

Expected (shortened):

The Pod "web" is invalid: spec: Forbidden: pod updates may not change fields other than
`spec.containers[*].image`,`spec.initContainers[*].image`,`spec.activeDeadlineSeconds`,
`spec.tolerations` ...

The exam approach: delete the pod and recreate it in one step.

k replace --force -f web2.yaml
k get pod web -n tag02 -o jsonpath='{.spec.containers[0].resources}{"\n"}'

Expected:

pod "web" deleted from tag02 namespace
pod/web replaced
{"limits":{"memory":"64Mi"},"requests":{"memory":"64Mi"}}

Why are requests there too? If a request is missing, Kubernetes copies the limit.

Save your progress? Sign up for free

Drill 7: Delete fast, read events

A normal delete waits for the grace period (default 30 s). In the exam, every second counts:

k delete pod web -n tag02 --force --grace-period=0

Expected:

Warning: Immediate deletion does not wait for confirmation that the running resource
has been terminated. ...
pod "web" force deleted from tag02 namespace

Events in chronological order:

k get events -n tag02 --sort-by=.lastTimestamp | tail -3

Expected (shortened):

5s          Normal   Started     pod/web   Container started
3s          Normal   Killing     pod/web   Stopping container web

Warning: Use --force --grace-period=0 only for practice pods. In production the container may keep running on the node.

Save your progress? Sign up for free

Exercise: Speed Drills

  1. Generate the YAML for a Job once with busybox:1.37 that runs date, and save it to job.yaml. Target time: 30 s
  2. List all resource types that are not bound to a namespace. Target time: 20 s
  3. Print the names of all pods in kube-system, one line per pod, without a header. Target time: 45 s
  4. Set the current context to the namespace tag02 and back to default. Target time: 30 s

Solution

k create job once --image=busybox:1.37 $do -- date > job.yaml
k api-resources --namespaced=false
k get pods -n kube-system -o jsonpath='{range .items[*]}{.metadata.name}{"\n"}{end}'
k config set-context --current --namespace=tag02
k config set-context --current --namespace=default

Alternative for 3: k get pods -n kube-system --no-headers -o custom-columns=:.metadata.name.

Save your progress? Sign up for free

Checkpoint

type k

Expected: k is aliased tokubectl’`

k config current-context

Expected: kubernetes-admin@kubernetes

echo $do

Expected: --dry-run=client -o yaml

Challenge

For all pods in the cluster, print namespace, name and image of the first container, sorted by namespace, and write the result to ~/cka/tag02/images.txt.

Hint: custom-columns can address any field, including inside lists.

Solution

C=NS:.metadata.namespace,NAME:.metadata.name,IMAGE:.spec.containers[0].image
cd ~/cka/tag02
k get pods -A --sort-by=.metadata.namespace -o custom-columns=$C > images.txt
head -3 images.txt

Expected (shortened):

NS              NAME                                IMAGE
calico-system   calico-apiserver-79f749f5bc-c8w8d   quay.io/calico/calico:v3.33.1
calico-system   calico-apiserver-79f749f5bc-nltj4   quay.io/calico/calico:v3.33.1

Clean up:

k delete ns tag02
k config delete-context dev

Save your progress? Sign up for free

Exam Tip

  • First copy and run the task’s context command. Always.
  • First $do > datei.yaml, then edit it selectively, then k apply -f. No YAML from scratch.
  • In vim: :set paste before pasting, otherwise the indentation gets mangled.

Enterprise Note

  • In daily work, tools such as k9s, kubectx/kubens or Lens replace many of these moves.
  • Several clusters in one kubeconfig: show the context in the prompt (e.g. kube-ps1) so that nobody accidentally works in production.
  • --force --grace-period=0 is usually forbidden in production runbooks.