Skip to content
Hands-on IT IT Drills

Day 3: Choosing the Right Workload

Goal: You can tell from the task whether a Deployment, DaemonSet, StatefulSet, Job or CronJob is called for, and you build each one in a few minutes. · Duration: approx. 75 min · Prerequisite: Day 2

Briefing

CKAD: Application Design and Build (20 %) – “Choose and use the right workload resource”. In practice you almost never create pods directly; a controller creates them and keeps them alive. The only question is which one:

  runs continuously? ── yes ──> one per node? ── yes ──> DaemonSet
        |                         └─ no ─> stable ID/data? ── yes ──> StatefulSet
        |                                             └─ no ─> Deployment
        └─ no ──> once ──> Job · on a schedule ──> CronJob

Drill 1: Overview and decision guide

k create ns tag03
k config set-context --current --namespace=tag03
mkdir -p ~/ckad/tag03 && cd ~/ckad/tag03
k api-resources --api-group=apps
k api-resources --api-group=batch

Expected:

NAME                  SHORTNAMES   APIVERSION   NAMESPACED   KIND
controllerrevisions                apps/v1      true         ControllerRevision
daemonsets            ds           apps/v1      true         DaemonSet
deployments           deploy       apps/v1      true         Deployment
replicasets           rs           apps/v1      true         ReplicaSet
statefulsets          sts          apps/v1      true         StatefulSet
NAME       SHORTNAMES   APIVERSION   NAMESPACED   KIND
cronjobs   cj           batch/v1     true         CronJob
jobs                    batch/v1     true         Job
Resource Typical use Pod names k create?
Deployment stateless API, web frontend random yes
DaemonSet log agent, node exporter random, 1/node no
StatefulSet database, queue, cluster software name-0, -1 … no
Job migration, batch, export random yes
CronJob report, backup, cleanup name-<time>-… yes

Save your progress? Sign up for free

Drill 2: Deployment

notes-api from Day 2, three replicas. k create deployment has no --env, so go via the file:

k create deployment notes --image=172.16.0.126:30500/notes-api:1.0 --replicas=3 \
  --port=8080 $do > notes.yaml
vim notes.yaml

Add below name: notes-api:

        env:
        - name: NOTES_API_KEY
          value: lab-key-123
k apply -f notes.yaml
k rollout status deployment/notes
k get rs,pods -o wide

Expected (shortened):

deployment "notes" successfully rolled out
NAME                               DESIRED   CURRENT   READY   AGE
replicaset.apps/notes-7b6866d9dd   3         3         3       4s

NAME                         READY   STATUS    RESTARTS   AGE   IP               NODE
pod/notes-7b6866d9dd-97tnm   1/1     Running   0          4s    192.168.235.142  worker1
pod/notes-7b6866d9dd-d8l8h   1/1     Running   0          4s    192.168.235.139  worker1
pod/notes-7b6866d9dd-rj5gb   1/1     Running   0          4s    192.168.189.112  worker2

A pod gets lost – the ReplicaSet replaces it immediately:

k delete pod notes-7b6866d9dd-97tnm
k get pods -l app=notes

Expected:

pod "notes-7b6866d9dd-97tnm" deleted from tag03 namespace
NAME                     READY   STATUS              RESTARTS   AGE
notes-7b6866d9dd-d8l8h   1/1     Running             0          8s
notes-7b6866d9dd-fqrws   0/1     ContainerCreating   0          3s
notes-7b6866d9dd-rj5gb   1/1     Running             0          8s

Tip: Had you run k create deployment first and then k set env, there would be two rollouts – with pods briefly crashing without an API key in between.

Save your progress? Sign up for free

Drill 3: DaemonSet

There is no k create for DaemonSets. The exam way: generate Deployment YAML, change kind, delete replicas, strategy and status.

k create deployment node-agent --image=busybox:1.37 $do -- sleep infinity > ds.yaml
sed -i 's/^kind: Deployment/kind: DaemonSet/' ds.yaml
k apply -f ds.yaml

Expected (shortened):

Error from server (BadRequest): error when creating "ds.yaml": DaemonSet in version "v1"
cannot be handled as a DaemonSet: strict decoding error: unknown field "spec.replicas",
unknown field "spec.strategy"

So delete the replicas, strategy and status lines with vim ds.yaml. Afterwards the file looks like this:

apiVersion: apps/v1
kind: DaemonSet
metadata:
  labels:
    app: node-agent
  name: node-agent
spec:
  selector:
    matchLabels:
      app: node-agent
  template:
    metadata:
      labels:
        app: node-agent
    spec:
      containers:
      - command:
        - sleep
        - infinity
        image: busybox:1.37
        name: busybox
k apply -f ds.yaml
k get ds node-agent
k get pods -l app=node-agent -o custom-columns=NAME:.metadata.name,NODE:.spec.nodeName

Expected:

daemonset.apps/node-agent created
NAME         DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
node-agent   2         2         2       2            2           <none>          8s
NAME               NODE
node-agent-m976s   worker1
node-agent-xdvk6   worker2

Why only two? The control plane node has a taint:

k describe node cp1 | grep Taints

Expected:

Taints:             node-role.kubernetes.io/control-plane:NoSchedule

If the agent should run there too, add under template.spec in ds.yaml:

      tolerations:
      - key: node-role.kubernetes.io/control-plane
        operator: Exists
        effect: NoSchedule
k apply -f ds.yaml
k rollout status ds/node-agent
k get ds node-agent

Expected (shortened):

daemon set "node-agent" successfully rolled out
NAME         DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
node-agent   3         3         3       3            3           <none>          89s

Save your progress? Sign up for free

Drill 4: StatefulSet with a headless Service

A StatefulSet needs a headless Service (clusterIP: None) for stable DNS names:

k create service clusterip db --clusterip=None --tcp=80:80

Via volumeClaimTemplates, each pod gets its own PVC from the default StorageClass local-path:

cat > sts.yaml <<'EOF'
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: db
spec:
  serviceName: db
  replicas: 2
  selector:
    matchLabels:
      app: db
  template:
    metadata:
      labels:
        app: db
    spec:
      containers:
      - name: web
        image: nginx:1.29
        ports:
        - containerPort: 80
        volumeMounts:
        - name: data
          mountPath: /usr/share/nginx/html
  volumeClaimTemplates:
  - metadata:
      name: data
    spec:
      accessModes: ["ReadWriteOnce"]
      resources:
        requests:
          storage: 100Mi
EOF
k apply -f sts.yaml
k get pods -l app=db -w

Expected (shortened, stop with Ctrl+C):

NAME   READY   STATUS              RESTARTS   AGE
db-0   0/1     Pending             0          0s
db-0   0/1     ContainerCreating   0          6s
db-0   1/1     Running             0          8s
db-1   0/1     Pending             0          0s
db-1   0/1     ContainerCreating   0          7s

db-1 only starts once db-0 is running. The PVCs are named <template>-<pod>:

k get pvc

Expected (shortened):

NAME        STATUS   VOLUME                                     CAPACITY   ACCESS MODES
data-db-0   Bound    pvc-136e1c71-ec46-4556-8ff9-29fe94443ca3   100Mi      RWO
data-db-1   Bound    pvc-e53db033-9fa8-4cd1-bc15-1984f4b96b59   100Mi      RWO

Save your progress? Sign up for free

Drill 5: Identity and data persist

k exec db-0 -- sh -c 'echo "ich bin $(hostname)" > /usr/share/nginx/html/index.html'
k exec db-1 -- sh -c 'echo "ich bin $(hostname)" > /usr/share/nginx/html/index.html'
k run client --image=busybox:1.37 -- sleep 3600
k exec client -- wget -qO- db-1.db
k exec client -- nslookup db.tag03.svc.cluster.local

Expected (shortened):

ich bin db-1
Name:   db.tag03.svc.cluster.local
Address: 192.168.235.149
Name:   db.tag03.svc.cluster.local
Address: 192.168.189.125

The headless Service returns the pod IPs directly, and db-1.db always reaches the same pod. Now delete db-0:

k delete pod db-0
k get pod db-0 -o wide
k exec db-0 -- cat /usr/share/nginx/html/index.html

Expected (shortened):

pod "db-0" deleted from tag03 namespace
NAME   READY   STATUS    RESTARTS   AGE   IP               NODE
db-0   1/1     Running   0          2s    192.168.189.66   worker2
ich bin db-0

Same name, new IP, same data. When you scale down, the PVCs stay:

k scale sts db --replicas=1
k get pods,pvc -l app=db

Expected (shortened):

statefulset.apps/db scaled
NAME       READY   STATUS    RESTARTS   AGE
pod/db-0   1/1     Running   0          8s
NAME                              STATUS   VOLUME
persistentvolumeclaim/data-db-0   Bound    pvc-136e1c71-...
persistentvolumeclaim/data-db-1   Bound    pvc-e53db033-...

Save your progress? Sign up for free

Drill 6: Job with completions and parallelism

k create job batch --image=busybox:1.37 $do \
  -- sh -c 'echo Teil fertig auf $(hostname); sleep 3' > job.yaml
vim job.yaml

Add directly below the first spec::

  completions: 6
  parallelism: 2
  backoffLimit: 2
k apply -f job.yaml
k get pods -l job-name=batch
k wait --for=condition=complete job/batch --timeout=120s
k get job batch

Expected (shortened):

NAME          READY   STATUS      RESTARTS   AGE
batch-smhxk   0/1     Completed   0          7s
batch-xx5jx   1/1     Running     0          7s
job.batch/batch condition met
NAME    STATUS     COMPLETIONS   DURATION   AGE
batch   Complete   6/6           26s        27s

Six successful pods, never more than two at a time. k create job sets restartPolicy: Never; for Jobs, only Never and OnFailure are allowed.

Save your progress? Sign up for free

Drill 7: When Jobs fail

backoffLimit limits failed attempts, activeDeadlineSeconds the total runtime:

k apply -f - <<'EOF'
apiVersion: batch/v1
kind: Job
metadata:
  name: fail
spec:
  backoffLimit: 2
  template:
    spec:
      restartPolicy: Never
      containers:
      - name: fail
        image: busybox:1.37
        command: ["sh", "-c", "echo versuche; exit 1"]
---
apiVersion: batch/v1
kind: Job
metadata:
  name: slow
spec:
  activeDeadlineSeconds: 10
  template:
    spec:
      restartPolicy: Never
      containers:
      - name: slow
        image: busybox:1.37
        command: ["sleep", "60"]
EOF

After one minute:

k get jobs fail slow
k get pods -l job-name=fail
k describe job fail | tail -1
k describe job slow | tail -1

Expected (shortened):

NAME   STATUS   COMPLETIONS   DURATION   AGE
fail   Failed   0/1           51s        51s
slow   Failed   0/1           50s        50s
NAME         READY   STATUS   RESTARTS   AGE
fail-2t2np   0/1     Error    0          51s
fail-kz7pd   0/1     Error    0          38s
fail-vgsts   0/1     Error    0          16s
  Warning  BackoffLimitExceeded  12s   job-controller  Job has reached the specified
backoff limit
  Warning  DeadlineExceeded  9s    job-controller  Job was active longer than specified
deadline

Three pods: one attempt plus two retries, with a growing pause in between. The controller deleted the pod of slow once the deadline had passed.

Save your progress? Sign up for free

Drill 8: CronJob

k create cronjob report --image=busybox:1.37 --schedule="*/1 * * * *" $do \
  -- sh -c 'date; sleep 90' > cj.yaml
vim cj.yaml

Add below schedule: (at the same level):

  concurrencyPolicy: Forbid
  successfulJobsHistoryLimit: 2
  failedJobsHistoryLimit: 1
k apply -f cj.yaml
k create job report-manual --from=cronjob/report

A run takes 90 s, the interval is 60 s. After a good two minutes:

k get cj report
k get events --field-selector involvedObject.name=report --sort-by=.lastTimestamp

Expected (shortened):

NAME     SCHEDULE      TIMEZONE   SUSPEND   ACTIVE   LAST SCHEDULE   AGE
report   */1 * * * *   <none>     False     1        56s             2m37s
REASON             OBJECT           MESSAGE
SuccessfulCreate   cronjob/report   Created job report-29916944
UnexpectedJob      cronjob/report   Saw a job that the controller did not create or
                                    forgot: report-manual
JobAlreadyActive   cronjob/report   Not starting job because prior execution is running
                                    and concurrency policy is Forbid
SawCompletedJob    cronjob/report   Saw completed job: report-29916944, condition: ...

Forbid skips the run as long as the previous one is still running; Replace would cancel it, Allow (the default) would start in parallel. The CronJob does not count the manual Job as one of its own. The history: the controller deletes successful Jobs beyond two (event SuccessfulDelete). To pause it:

k patch cronjob report -p '{"spec":{"suspend":true}}'
k get cj report

Expected:

cronjob.batch/report patched
NAME     SCHEDULE      TIMEZONE   SUSPEND   ACTIVE   LAST SCHEDULE   AGE
report   */1 * * * *   <none>     True      1        71s             2m52s

Save your progress? Sign up for free

Exercise: Speed Drills

  1. Job hello with busybox:1.37 that runs echo hallo successfully three times in a row. Target time: 60 s
  2. CronJob cleanup, echo weg every 15 minutes, suspended right away. Target time: 45 s
  3. Run the CronJob report once by hand right now as Job report-now. Target time: 15 s
  4. Scale the Deployment notes to one replica. Target time: 10 s

Solution

k create job hello --image=busybox:1.37 $do -- echo hallo > hello.yaml
sed -i 's/^spec:$/spec:\n  completions: 3/' hello.yaml
k apply -f hello.yaml
k create cronjob cleanup --image=busybox:1.37 --schedule="*/15 * * * *" -- echo weg
k patch cronjob cleanup -p '{"spec":{"suspend":true}}'
k create job report-now --from=cronjob/report
k scale deployment notes --replicas=1

Expected (shortened):

job.batch/hello created
cronjob.batch/cleanup created
cronjob.batch/cleanup patched
job.batch/report-now created
deployment.apps/notes scaled

Save your progress? Sign up for free

Checkpoint

k get job hello -o jsonpath='{.status.succeeded}'

Expected: 3

k get ds node-agent -o jsonpath='{.status.numberReady}'

Expected: 3

k get pvc -l app=db

Expected: data-db-0 and data-db-1, both Bound – the StatefulSet controller copies the selector labels to the PVCs.

k get cj -o custom-columns=NAME:.metadata.name,SUSPEND:.spec.suspend

Expected: cleanup and report, both true.

Challenge

A Job parts should process four parts in parallel; each pod prints its part number (0 to 3).

Hint: completionMode: Indexed sets the variable JOB_COMPLETION_INDEX in each pod.

Solution

k apply -f - <<'EOF'
apiVersion: batch/v1
kind: Job
metadata:
  name: parts
spec:
  completions: 4
  parallelism: 4
  completionMode: Indexed
  template:
    spec:
      restartPolicy: Never
      containers:
      - name: part
        image: busybox:1.37
        command: ["sh", "-c", "echo Teil $JOB_COMPLETION_INDEX"]
EOF
k wait --for=condition=complete job/parts --timeout=90s
for p in $(k get pods -l job-name=parts -o name); do k logs $p; done | sort

Expected:

job.batch/parts created
job.batch/parts condition met
Teil 0
Teil 1
Teil 2
Teil 3

Clean up:

k config set-context --current --namespace=default
k delete ns tag03

The PVCs, and with them the local-path volumes, disappear together with the namespace.

Save your progress? Sign up for free

Exam Tip

  • DaemonSets and StatefulSets cannot be created with k create – generate Deployment YAML and rework it. For DaemonSets, delete replicas and strategy, otherwise the API reports “strict decoding error: unknown field”.
  • Job fields (completions, parallelism, backoffLimit, activeDeadlineSeconds) belong under spec, not under spec.template.spec. k explain job.spec helps.
  • A CronJob that should run “right now”: k create job <name> --from=cronjob/<cj>.

Enterprise Note

  • Databases as StatefulSets are usually run via an operator (CloudNativePG, Strimzi) that handles backups, failover and upgrades.
  • DaemonSets for logs and monitoring are provided by the platform; application teams rarely build them themselves.
  • CronJobs get timeZone, startingDeadlineSeconds and alerts on failed Jobs – a silent backup job that has been failing for weeks is a classic.