Common OpenShift Interview Questions on Failure Diagnosis
For common OpenShift troubleshooting questions, identify the failing stage before proposing a remedy. A rejected request, an unscheduled Pod, a restarting container and an unavailable Route need different evidence. Explain what the supplied observation proves, which alternatives remain, and the next authorized check that would separate them.
Topic: OpenShift in IT Infrastructure · Interview preparation
Quick answer: For common OpenShift troubleshooting questions, identify the failing stage before proposing a remedy. A rejected request, an unscheduled Pod, a restarting container and an unavailable Route need different evidence. Explain what the supplied observation proves, which alternatives remain, and the next authorized check that would separate them. These original interview cases use a fictional image-pull failure that someone has mistaken for an application crash. Product references follow OpenShift Container Platform 4.22, with upstream documentation supporting the specified stable lifecycle and networking concepts. OpenShift Container Platform 5.0 became generally available on 9 November 2026, so check the documentation for the version your cluster runs.
Key takeaways
- Identify whether the workload was admitted, scheduled and started.
- Treat displayed status as an observation rather than a complete diagnosis.
- Separate image retrieval from application execution.
- Choose the next check that distinguishes plausible explanations.
- Retain sanitized evidence before making an approved change.
Table of Contents
- What should you ask when an interviewer says the application is down?
- How would you answer an ImagePullBackOff question?
- How is CrashLoopBackOff different from an image pull failure?
- What evidence separates scheduling and storage problems for Pending Pods?
- How would you diagnose an unavailable Route when Pods look healthy?
- How do you distinguish Forbidden from an application error?
- What would a concise complete answer to the image-pull case sound like?
- References
What should you ask when an interviewer says the application is down?
Ask what failed from the user's perspective, when it began and which scope is affected. A failed external request can arise before traffic reaches the application. A single unready replica can coexist with a working service. Establish the actual impact instead of assigning a cause from the phrase down.
Then identify the object and its current condition. The Pod lifecycle guide distinguishes Pod phases, container states and restart behavior. Familiar status text such as CrashLoopBackOff describes a particular condition; it is not a universal Pod phase or root cause.
In the fictional case, the supplied observation is ImagePullBackOff for a replacement container, while an older replica still serves some requests. The relevant first branch is image retrieval. There is not yet evidence that the replacement application started and crashed.
How would you answer an ImagePullBackOff question?
Explain that the container image could not be retrieved and retries are being delayed. Request the specific pull failure, image reference and applicable registry access context. Possible causes include an unavailable reference, denied access or connectivity problems, but the displayed summary alone does not choose among them.
The OpenShift image stream guide explains image references and their relation to actual registry content. An image stream entry does not mean every node can retrieve the referenced content under the workload's permissions.
Question: should the team increase the container's memory limit? That does not address a supplied registry-denied failure before execution. Verify the intended image and approved pull-access arrangement. Keep registry credentials out of the interview worksheet and any shared error report.
| Supplied observation | What it suggests | Discriminating next evidence |
|---|---|---|
| Request rejected by quota | Admission did not permit the proposed resource | Exact quota rejection and aggregate use |
| Pod unassigned with placement failure | Scheduling cannot find an eligible placement | Scheduling reason and relevant node constraints |
| Image pull denied | Image retrieval failed before startup | Image identity and sanitized registry denial |
| Container repeatedly exits | Execution or a health check is failing | Previous termination reason and relevant logs |
How is CrashLoopBackOff different from an image pull failure?
CrashLoopBackOff indicates repeated container failures with a restart delay. The lifecycle guide describes application errors, configuration problems and failed health checks among possible causes. Investigate the termination reason and the relevant container's logs, including its previous execution where appropriate.
The probe guide distinguishes startup, readiness and liveness behavior. A failed liveness check can cause restart behavior, while readiness failure can leave the container running but unready. Ask which check failed before changing a timeout.
Original question: a process exits immediately after reporting a missing configuration key. What does that support? Investigating the key and its consumption path is more direct than increasing replica count. Preserve the exact sanitized message and compare it with the intended configuration before changing the workload.
What evidence separates scheduling and storage problems for Pending Pods?
Pending can cover time before scheduling and preparation before containers run. Determine whether the Pod has been assigned to a node. If not, the scheduling event can identify resource fit, required placement rules or a storage-related prerequisite.
The resource guide explains why low observed usage does not prove scheduling capacity. The persistent volume guide describes claims, volumes and access modes; inspect the actual claim and storage configuration instead of assuming more CPU will help.
Question: a claim is pending because its requested storage class is not available. Would raising a CPU limit solve it? No evidence supports that change. Confirm the intended class and provisioner with the storage owner. A storage decision also needs the application's capacity, topology and data-retention requirements.
How would you diagnose an unavailable Route when Pods look healthy?
Trace the specific request through DNS, routing, the Service and selected suitable backends. The OpenShift Route guide describes targets and TLS modes. The Service guide explains port and targetPort mapping.
Question: all application containers are Running, but the Route has no usable backend. Ask whether the relevant Pods are Ready, whether the Service selects them, and whether the target port matches the listener. Running alone does not settle those questions.
If the supplied error concerns certificate validation, identify which connection and certificate failed. Changing application readiness would not explain a hostname mismatch. Keep the observed response, its source and the relevant connection boundary in the answer.
| Interviewer prompt | Answer cue | Avoid this unsupported leap |
|---|---|---|
| Registry denies the replacement image | Check intended image and approved pull access | Call it a memory-driven application crash |
| Running but unready | Inspect the readiness check and dependency | Assume the process must be restarted |
| External certificate name mismatch | Review requested name and presented certificate | Assume Service labels caused certificate validation |
| Forbidden reading a project resource | Confirm identity, action and effective grants | Grant cluster-wide administration immediately |
How do you distinguish Forbidden from an application error?
Identify who returned the response and which identity made the request. An API authorization denial concerns a resource action and scope. An application can separately return an access-denied response under its own authorization logic.
The OpenShift authorization guide describes project and cluster role bindings. Review the effective grants for the actual identity, including group or service account membership. A developer's successful interactive request does not establish the application's service account permissions.
For an original interview case, a reviewer can inspect amber but receives Forbidden in blue. Ask whether blue access is required before proposing a change. A denied action can be the intended boundary, so diagnosis should include the requirement as well as the error.
What would a concise complete answer to the image-pull case sound like?
The replacement container has not started because its image retrieval is failing. The supplied registry-denied message supports checking the intended reference and approved pull-access configuration. The older serving replica explains why impact is partial. No supplied evidence points to the replacement application's memory use or runtime crash.
After an approved correction, verify successful image retrieval, container startup, readiness and the intended application request. Record the relevant workload identity and timing so a later reviewer can distinguish the correction from an unrelated recovery.
If the interviewer changes the evidence to a successfully pulled image followed by repeated exits, change the diagnosis branch. Reusing the earlier pull explanation would ignore the new stage. Strong troubleshooting answers adapt to the supplied evidence and state what would disprove the current hypothesis.
Related reading: OpenShift capacity interview questions; OpenShift rollout acceptance; OpenShift availability and recovery questions.
References
- Kubernetes Pod lifecycle and restart behavior
- OpenShift 4.22 image stream guide
- Kubernetes container probe behavior
- Kubernetes container resource management
- Kubernetes persistent volume concepts
- OpenShift 4.22 Route guide
- Kubernetes Service port and endpoint concepts
- OpenShift 4.22 authentication and authorization guide
Frequently Asked Questions
What's Your Reaction?
Like
0
Dislike
0
Love
0
Funny
0
Wow
0
Sad
0
Angry
0