Scenario-Based OpenShift Interview Questions (2027)

44 scenario-based OpenShift interview questions by level: CrashLoopBackOff, no endpoints, Route timeouts, GitOps OutOfSync, blocked upgrades, OADP.

Sep 13, 2025 - 12:01
Nov 8, 2026 - 04:48
 0  144
Scenario-Based OpenShift Interview Questions (2027)

Scenario-based OpenShift interview questions describe a broken application or cluster and ask how you would find and fix the cause with oc and the web console. Typical cases are pods in CrashLoopBackOff, a Service with no endpoints, Routes timing out on long requests, Argo CD stuck OutOfSync, a cluster upgrade blocked by removed APIs and an image registry running out of space. This page has 44 scenarios grouped by experience level.

Commands and behaviour were checked against the Red Hat OpenShift Container Platform 4 documentation on docs.redhat.com in November 2026. Commands work with the oc client; most also work with kubectl.

Key takeaways

  • Most OpenShift troubleshooting starts with three commands: oc get, oc describe (read the Events section) and oc logs, including --previous for crashed containers.
  • Freshers get application-level scenarios: crashes, image pulls, probes, Services, PVCs and config changes.
  • Engineers with 2 to 5 years get platform-adjacent scenarios: rollouts, autoscaling, Routes, OpenShift Pipelines, OpenShift GitOps and monitoring.
  • Senior scenarios cover upgrades, certificates, registry storage, node maintenance, audit and backup with OADP.
  • Explain how you confirmed the cause, not only the fix. Interviewers want to hear which output told you.

How this page differs from our other OpenShift guides

The 120+ OpenShift interview questions guide is our main reference and its scenario section covers SCC failures, Route "Application is not available", build push failures, stuck updates and Operators. OpenShift administration questions covers routine admin tasks, and advanced OpenShift questions covers architecture, MachineConfig and etcd incidents. The scenarios here are different cases, mostly the ones application and platform teams meet every week.

Pod or object statusFirst commandQuestion
CrashLoopBackOffoc logs pod --previous1
ImagePullBackOffoc describe pod2
Pendingoc describe pod (Events)3, 9
OOMKilledoc get pod -o yaml (lastState)7
Service with no endpointsoc get endpointslices4
Namespace Terminatingoc get namespace -o yaml (conditions)10

OpenShift scenarios for freshers

1. Your pod is in CrashLoopBackOff. What do you do?

Read the logs of the crashed container with oc logs pod-name --previous, because the current container may have just restarted and shows nothing. Then oc describe pod for exit codes and events. Common causes are a missing environment variable or config file, a failed database connection at startup, or a wrong command. oc debug deployment/name starts a copy of the pod with a shell so you can look inside.

2. A pod shows ImagePullBackOff for an image in a private registry. How do you fix it?

oc describe pod shows the pull error. If it is an authentication error, create a pull secret and link it to the service account the pod uses.

oc create secret docker-registry regcred \
  --docker-server=registry.example.com \
  --docker-username=robot --docker-password='***'
oc secrets link default regcred --for=pull

If the error says "not found", check the image name and tag.

3. A new pod stays Pending with "0/6 nodes are available: insufficient cpu". What does it mean?

No node has enough unrequested CPU for the pod's requests. Lower the request if it is too high, check whether a quota or LimitRange set a large default, or ask the platform team for capacity. oc describe node shows allocated requests per node.

4. The app works with oc port-forward to the pod, but the Route returns errors and the Service has no endpoints. Why?

The Service selector does not match the pod labels. Compare oc get svc name -o yaml (selector) with oc get pods --show-labels, and check oc get endpointslices -l kubernetes.io/service-name=name. Fix the labels or the selector. Also check that pods are Ready, because unready pods are not added as endpoints.

5. You changed a ConfigMap, but the application still uses the old value. Why?

Values injected as environment variables are read only when the container starts. Restart the pods with oc rollout restart deployment/name. Files from a mounted ConfigMap are updated after a short delay, unless mounted with subPath, and the application must re-read them.

6. oc new-app https://github.com/org/app fails to detect the language. What do you do?

Tell it which builder image to use with the builder~source syntax, for example oc new-app nodejs~https://github.com/org/app, or check that the repository has the files S2I detection looks for, such as package.json. List available builders with oc new-app --list or the Developer catalog.

7. A pod restarts and shows OOMKilled. What do you check?

oc get pod name -o yaml shows lastState.terminated.reason: OOMKilled. The container used more memory than its limit. Check the memory limit, the application's real usage in the monitoring dashboards, and runtime settings such as JVM heap size relative to the limit. Raise the limit or fix the leak; do not remove limits.

8. A pod is Running but receives no traffic and shows 0/1 Ready. Why?

The readiness probe is failing. oc describe pod shows probe failures in events. Check that the probe path and port match the application and that the initial delay is long enough for startup. Change it with oc set probe deployment/name --readiness --get-url=http://:8080/healthz or in the YAML. Slow starters should use a startup probe.

9. A PersistentVolumeClaim stays Pending. What do you look at?

oc describe pvc. Causes include no default StorageClass, a requested access mode such as ReadWriteMany that the storage does not support, or a StorageClass with WaitForFirstConsumer, where the PVC stays Pending until a pod uses it. That last case is normal.

10. You deleted a project and it has been Terminating for an hour. What do you do?

Look at the namespace's status conditions with oc get namespace name -o yaml; they list resources that could not be removed and finalizers still pending. Often a custom resource's Operator was removed first, so nothing processes its finalizer. Fix the cause (reinstall the Operator or clean the resource correctly). Removing finalizers by force can leave orphaned external resources such as cloud disks.

11. oc login fails with "x509: certificate signed by unknown authority". What is the right fix?

The API certificate is signed by a CA your machine does not trust, often an internal CA. Pass --certificate-authority=ca.crt or add the CA to your system trust store. --insecure-skip-tls-verify works for a lab but should not be used against production clusters.

12. You pushed a new image to the same tag, but the Deployment did not roll out. Why?

A Deployment only rolls out when its pod template changes. Options are an ImageStream trigger (oc set triggers deployment/name --from-image=app:latest -c app), deploying by new tag or digest from CI, or oc rollout restart. Unique tags or digests also make rollbacks clear.

OpenShift scenarios for 2 to 5 years of experience

13. Every rolling update causes a few seconds of 503 errors. How do you fix it?

Pods are taking traffic before they are ready or being killed while still serving. Add a correct readiness probe, set maxUnavailable: 0 with maxSurge so new pods are ready first, and handle SIGTERM gracefully. A short preStop sleep gives the router time to remove the endpoint before the app stops.

14. A HorizontalPodAutoscaler never scales and shows <unknown> for CPU. Why?

The pods have no CPU requests, so utilisation cannot be calculated, or metrics are not available. Set CPU requests on the containers and check oc adm top pods. oc describe hpa shows the exact condition.

15. A worker node shows NotReady. What do you check?

oc describe node for conditions such as MemoryPressure, DiskPressure or a kubelet that stopped posting status. Read kubelet logs with oc adm node-logs node-name -u kubelet, and use oc debug node/node-name then chroot /host to inspect the host. Check for disk full, a failed CRI-O service or networking problems.

16. Pods on one node are being evicted with DiskPressure. What do you do?

Find what fills the node's disk: container logs, unused images, or pods writing heavily to emptyDir. The kubelet garbage-collects images, but heavy writers need ephemeral-storage requests and limits. Fix the workload, then let the node recover. Also check node disk size against the workload type.

17. A CronJob sometimes runs two copies at once and corrupts data. How do you stop it?

Set concurrencyPolicy: Forbid so a new run is skipped while one is active, or Replace to stop the old one. Make the job idempotent anyway, and set startingDeadlineSeconds so missed schedules do not pile up after a controller outage.

18. Report downloads fail with a 504 after about 30 seconds through the Route. Why?

The router's default server timeout is 30 seconds. Set a longer timeout on that Route with an annotation, rather than changing it for every Route.

oc annotate route reports haproxy.router.openshift.io/timeout=120s

For very long jobs, move to an asynchronous design that returns a job ID.

19. Users are logged out randomly after a scale-up. What is going on?

The application keeps sessions in memory and requests are reaching different pods. The router uses cookies for session affinity by default, but clients or caches may drop the cookie. The real fix is to store sessions externally (for example in Redis) so any pod can serve any request. Route annotations can change the balancing algorithm if needed.

20. An OpenShift Pipelines run fails at the image push step with "unauthorized". What do you check?

The service account the PipelineRun uses must be allowed to push. For the internal registry, grant the image pusher role on the target project with oc policy add-role-to-user system:image-pusher system:serviceaccount:ci:pipeline -n app (adjust names). For an external registry, link a push secret to the service account or reference it in the task's workspace.

21. OpenShift GitOps shows an application as OutOfSync right after every sync. Why?

A controller or webhook changes fields after Argo CD applies them, such as replica counts set by an HPA or defaults added by an admission webhook. Remove those fields from Git (for example omit replicas when an HPA manages it) or configure ignoreDifferences for them in the Application.

22. The default OpenShift GitOps instance cannot create resources in a new namespace. Why?

The Argo CD instance only manages namespaces it has been given permission for. Label the namespace with argocd.argoproj.io/managed-by=openshift-gitops (the name of the Argo CD instance's namespace) so the GitOps Operator grants it access, or grant the RBAC explicitly.

23. A team wants to store application secrets in Git for GitOps. What do you recommend?

Do not store plain Secrets. Use the External Secrets Operator to sync values from a secrets manager such as HashiCorp Vault or a cloud secrets service, or encrypt secrets before committing with a tool such as Sealed Secrets. Both keep Git as the source of truth without exposing values.

24. Pods cannot resolve external hostnames but internal Service names work. What do you check?

The DNS Operator's upstream resolvers (oc get dns.operator/default -o yaml), whether the nodes' upstream DNS servers work, and egress network policies or firewalls that block DNS to upstream. oc debug into a pod and test with dig to compare internal and external names.

25. A ServiceMonitor for your application is ignored and no metrics appear. Why?

Monitoring for user-defined projects is not enabled, or the ServiceMonitor's selector does not match the Service labels or port name. Enable it by setting enableUserWorkload: true in the cluster-monitoring-config ConfigMap in openshift-monitoring, then check the target in the console's Observe, Targets page.

26. After a pod restarts, its logs are gone and the team needs them for an incident. What should be in place?

Container logs are kept on the node only for the current and previous container. Install the Red Hat OpenShift Logging stack, which in current versions uses a Loki-based log store and the cluster log forwarder, or forward logs to an external system. Then logs survive restarts and node replacement.

27. A developer needs a database for testing in their project. What is the quickest safe option?

Deploy a database from the Developer catalog or a certified Operator into their project with a PVC, resource limits and a Secret for credentials. Make clear it is for testing; production databases need backups, monitoring and an owner.

28. A deployment rollout is stuck with "ProgressDeadlineExceeded". What do you do?

New pods are not becoming ready within the deadline. Check them with oc get pods and oc describe, fix the cause (image, probe, config), or undo with oc rollout undo deployment/name to restore the previous ReplicaSet while you investigate.

Senior OpenShift scenarios

29. A lab cluster was shut down for two months and, after restart, nodes do not become Ready. What do you check?

Certificates rotated while the cluster was off, and nodes are waiting for new kubelet certificates. Check oc get csr for Pending requests, verify they come from your nodes, and approve them with oc adm certificate approve. Red Hat documents recovery from expired control plane certificates for longer outages.

30. A minor update is blocked because the cluster reports Upgradeable=False about removed APIs. What do you do?

The next Kubernetes version removes API versions that something in the cluster still calls. Use oc get apirequestcounts to find which removed APIs are used and by which users or service accounts, update those manifests, Helm charts or Operators, then provide the administrator acknowledgement in the admin-acks ConfigMap in openshift-config as the documentation describes.

31. The internal image registry storage is nearly full. How do you manage it?

Prune old images. The ImagePruner custom resource runs pruning on a schedule, and oc adm prune images can be run manually with options for how many revisions to keep. Check that builds push to the registry only what is needed and that teams delete unused ImageStreams.

32. Draining nodes for hardware replacement keeps hanging. What usually blocks it?

PodDisruptionBudgets that allow zero disruptions, for example minAvailable equal to the replica count, or single-replica workloads with a PDB. Work with the owners to set PDBs that allow one pod down, scale them up temporarily, and then drain with oc adm drain node --ignore-daemonsets --delete-emptydir-data.

33. One team's batch jobs starve other projects of CPU. What controls do you apply?

ResourceQuotas and LimitRanges on that team's projects, PriorityClasses so important services win scheduling, and dedicated nodes with taints and tolerations for batch work if the load justifies it. Show the team their usage in dashboards so the limits are understood.

34. The identity provider is broken and kubeadmin was removed. How do you get back in?

Use the kubeconfig file with the system:admin certificate generated at install time, which does not depend on OAuth. Keep that file in a secure vault as break-glass access. Fix the OAuth configuration, then confirm normal logins work.

35. Someone deleted a production Deployment and nobody admits it. How do you find out who?

Read the API server audit logs. oc adm node-logs --role=master --path=kube-apiserver/ lists log files, and you can fetch them and search for delete events on that resource. The audit entry has the user, time and source IP. Forward audit logs to a central system so they are kept long enough.

36. A vendor's monitoring agent needs privileged access on every node. How do you evaluate the request?

Prefer a certified Operator from the vendor, which ships the SCC and RBAC it needs. Otherwise, run it in a dedicated namespace with its own service account bound to the narrowest SCC that works, never the default service account, and review exactly which host access it needs. Document the exception.

37. Finance says the cluster costs too much. Where do you look?

Compare requests with actual usage per project, because over-requested pods block capacity. Use the Vertical Pod Autoscaler in recommendation mode, enable the cluster autoscaler's scale-down, and review idle projects. Then agree requests guidelines with teams.

38. A team needs to restore a namespace to yesterday's state after a bad change. What do you use?

The OpenShift API for Data Protection (OADP) Operator, which is based on Velero, backs up namespaces' resources and persistent volumes to object storage and restores them. Use scheduled backups for important namespaces and test restores into a new namespace.

Incident drills for all levels

39. All Routes in the cluster stop working at once. What do you check?

The ingress Operator and router pods (oc get co ingress and oc get pods -n openshift-ingress), the nodes they run on (often infra nodes), and the external load balancer or DNS wildcard record in front of them. A full outage across Routes points to the router layer, not to applications.

40. oc commands time out against the API. What are the possible causes?

The API load balancer, expired or changed certificates, control plane nodes down, or etcd without quorum. Check from a control plane node's console or SSH session, review kube-apiserver and etcd pod status, and follow the documented recovery procedures if etcd quorum is lost.

41. After rotating the cluster-wide pull secret, image pulls start failing on new pods. What went wrong?

The global pull secret in openshift-config was updated with wrong or missing registry credentials. The Machine Config Operator rolls the change to nodes. Restore a correct secret with oc set data secret/pull-secret -n openshift-config --from-file=.dockerconfigjson=file and watch nodes pick it up.

42. A namespace can suddenly not create any new pods. What is a quick check?

oc describe quota -n project. If pods or requests.cpu is at the hard limit, new pods are rejected, which shows in the ReplicaSet's events. Clean up completed pods from jobs or adjust the quota.

43. A cluster Operator shows Degraded after a configuration change. How do you approach it?

oc get co to list it, oc describe co name to read the condition message, then the Operator's pods and logs in its namespace. Revert the configuration change if the message points to it. Collect oc adm must-gather before opening a support case.

44. A security scan finds a critical CVE in a base image used by 40 applications. How do you respond?

Update the base image, rebuild applications through their pipelines (ImageStream or build triggers can automate this for BuildConfigs), and track which deployments still run the old image digest. Use image policies to block the vulnerable digest from new deployments once fixes are out.

How to prepare for an OpenShift scenario round

  1. Install OpenShift Local (formerly CodeReady Containers) or use the Developer Sandbox for application scenarios.
  2. Recreate questions 1 to 9 on purpose: bad command, wrong image, high requests, label mismatch, bad probe.
  3. Install OpenShift Pipelines and OpenShift GitOps from OperatorHub and break a push and a sync.
  4. Practise reading oc describe Events quickly; most answers start there.
  5. Read the troubleshooting sections of the OpenShift docs for nodes, networking and updates.

Related reading: 12 must-know Kubernetes architecture components, how RHEL 10 integrates with Kubernetes and OpenShift and what GitOps is. For hands-on labs, the OpenShift training runs in our Kharadi, Pune classroom and live online, and the CKA training covers the Kubernetes troubleshooting underneath.

FAQs

How do I see why an OpenShift pod crashed?

Run oc logs pod-name --previous to read the crashed container's output, then oc describe pod for the exit code and events.

Why does my Service have no endpoints?

Usually the Service selector does not match the pod labels, or the pods are not Ready because their readiness probe fails.

What is the default Route timeout in OpenShift?

The router's default server timeout is 30 seconds. Change it per Route with the haproxy.router.openshift.io/timeout annotation.

How do I fix a namespace stuck in Terminating?

Read the namespace's status conditions to find which resources or finalizers block deletion, then fix the cause. Force-removing finalizers can leave orphaned resources.

Why is OpenShift GitOps always OutOfSync?

Something in the cluster changes fields after Argo CD applies them. Remove those fields from Git or configure ignoreDifferences.

How do I back up a namespace in OpenShift?

Use the OADP Operator, which is based on Velero, to back up resources and persistent volumes to object storage and restore them.

How is this page different from the 120+ OpenShift questions?

The 120+ guide explains concepts and includes core scenarios such as SCC failures and stuck updates. This page has different day-to-day scenarios, grouped by experience level.

Official references

What's Your Reaction?

Like Like 0
Dislike Dislike 0
Love Love 0
Funny Funny 0
Angry Angry 0
Sad Sad 0
Wow Wow 0
Mridul I am a passionate technology enthusiast with a strong focus on DevOps, Cloud Computing, and Cybersecurity. Through my blogs at DevOps Training Institute, I aim to simplify complex concepts and share practical insights for learners and professionals. My goal is to empower readers with knowledge, hands-on tips, and industry best practices to stay ahead in the ever-evolving world of DevOps.