Real-Time Harness Interview Questions and Answers (2027)

42 real-time Harness interview questions on live pipeline failures: delegates, null expressions, rollback, verification, approvals and deployment freezes.

Sep 25, 2025 - 14:03
Nov 9, 2026 - 20:01
 0  105
Real-Time Harness Interview Questions and Answers (2027)

Real-time Harness interview questions ask what you do when a live pipeline run goes wrong: a delegate goes offline, an expression resolves to null, a canary fails verification, an approval waits forever, a rollback does not trigger, or a deployment freeze blocks a hotfix. This page has 42 questions, each answered with the checks you would run, in order, on a real Harness execution.

Harness behaviour described here was checked against developer.harness.io in November 2026.

Key takeaways

  • Most live failures trace back to four places: delegate connectivity, connector credentials, expressions, and failure strategy settings.
  • The delegate only makes outbound HTTPS connections on port 443. It needs no inbound ports, so "delegate offline" is usually egress, proxy, token or resource trouble.
  • An expression resolves to null if the value it points to is not available yet at that point in the run. Check timing before syntax.
  • Rollback only happens if a failure strategy says so. Know where strategies are set (step, step group, stage) and which wins.
  • Interviewers grade the order of your checks: read the execution logs, find the failed step, check the delegate that ran it, then fix and prevent.

How this page fits with our other Harness guides

Each Harness page covers a different job. 75+ Harness CD interview questions explains pipelines, services, environments and deployment strategies. Advanced Harness interview questions covers pipeline design at scale. Harness engineer questions covers CI, STO, feature management and cost, and Harness certification questions covers the exams. This page is about debugging runs that are already in progress or have just failed.

SymptomLook first atQuestions
Step stuck or "no eligible delegate"Delegate status, selectors, egress8, 15 to 19
Null or wrong valuesExpression timing and scope20 to 23
Deployment fails or hangsStep logs, Kubernetes events, manifests24 to 29
Bad release reached usersVerification, failure strategy, rollback30 to 34
Run blockedApprovals, freeze windows, policies, barriers35 to 37

Real-time Harness questions for freshers

1. Where do you start when a Harness pipeline run fails?

Open the execution, find the first step marked failed, and read its log and error summary. Note which delegate ran the step (shown in the step details) and the inputs it received. Fix the first failure first; later failures are often side effects.

2. What is a delegate and why does it matter for troubleshooting?

The delegate is a service you run in your own network that executes tasks for Harness: connecting to clusters, pulling artifacts, running scripts. Harness SaaS sends work to it, and it reports results back. If the delegate cannot reach a target or Harness, the step fails or waits.

3. What is a connector?

A connector stores how Harness reaches an external system, such as a Kubernetes cluster, Git provider, artifact registry or cloud account, including credentials (as secret references) and, often, which delegates to use. Many live failures are connector failures: expired tokens, wrong URLs or missing permissions.

4. How do you test a connector?

Use the connection test on the connector page. It runs through an eligible delegate and reports whether it could authenticate and reach the endpoint. If the test passes but the pipeline fails, compare the delegate the test used with the one that ran the pipeline step.

5. What does the execution view tell you that logs do not?

It shows the graph of stages and steps with status and duration, which step is waiting and on what (approval, barrier, resource constraint), the resolved inputs, and output variables. Waiting steps usually explain "the pipeline is stuck" faster than logs.

6. What is a failure strategy?

A failure strategy tells Harness what to do when a step or stage fails, by error type. Actions include Retry, Manual Intervention, Mark as Success, Ignore Failure, Abort, Rollback Stage and Rollback Pipeline. Without the right strategy, a failed deployment step may simply fail without rolling back.

7. How do you rerun only the failed part of a pipeline?

Harness lets you retry a failed pipeline from a failed stage or rerun from a chosen stage, reusing earlier successful stages where allowed. This saves time when, for example, the build succeeded and only the deployment step failed because of a temporary cluster issue.

8. What does "no eligible delegate" mean?

No connected delegate matches what the task needs. Either no delegate is online in that scope, the step or connector uses delegate selectors (tags) that no running delegate has, or the delegate lacks the capability (for example, it cannot reach the target). Check delegate status and selectors first.

9. How do you find which version of an artifact was deployed?

Look at the service step outputs in the execution, which show the artifact source and tag, or use expressions such as <+artifacts.primary.tag> in a log step. The service dashboard also lists deployed versions per environment.

10. How do you see secrets in logs?

You should not. Harness masks secret values in logs when they are referenced with <+secrets.getValue("name")>. If a secret appears, it was probably transformed (for example base64 encoded) by a script before printing, which defeats masking. Rotate it and fix the script.

11. What is the difference between a pipeline variable and a runtime input?

A pipeline variable has a value defined in the pipeline or an input set. A runtime input, marked <+input>, is provided when the run starts, through the UI, an input set, a trigger or the API. Many "wrong value" problems come from a trigger passing a different input set than expected.

12. How do you abort a run safely?

Use Abort on the execution. Harness stops current steps, but changes already applied to a cluster stay as they are, so check the target after aborting. If the deployment was half applied, roll back or redeploy the last good version rather than leaving a mixed state.

Real-time Harness questions for 2 to 5 years of experience

13. A step keeps running far longer than usual. What do you check?

Look at the live log for where it stopped, check the step timeout, and check the target: a Kubernetes rollout may be waiting for pods that never become ready. Look at the delegate's CPU and memory, since a starved delegate slows every task. Set step timeouts so hangs fail clearly instead of waiting for hours.

14. How do you read Kubernetes rollout failures in Harness?

The Kubernetes steps print rollout status and events. Look for image pull errors, failed readiness probes, CrashLoopBackOff or quota limits. Then confirm with kubectl describe pod and kubectl logs on the cluster. Harness reports the symptom; the cluster shows the cause.

15. A delegate shows as disconnected. How do you troubleshoot?

Check the delegate pod or container is running and not restarting, read its logs, confirm outbound HTTPS on 443 to Harness is allowed through any proxy or firewall, confirm the delegate token is valid and not revoked, and check CPU and memory. The delegate needs no inbound ports, so do not open any.

16. Why might a Docker delegate break after weeks without changes?

Kubernetes and Helm delegates can auto-upgrade through an upgrader job, but Docker delegates do not, and delegate images expire. An old Docker delegate eventually stops being supported. Plan regular upgrades or move to a Kubernetes delegate with auto-upgrade.

17. Two delegates are online but only one works for a cluster. Why?

Harness picks any eligible delegate unless selectors narrow it down. If only one can reach the cluster network, tasks on the other fail randomly. Add selectors to the connector or step so tasks go to delegates that can reach the target, and make the capability check meaningful.

18. How do you scale delegates when many pipelines run at once?

Run several delegate replicas, give them enough CPU and memory, and use the delegate's task limits so one replica does not take too much work. Spread replicas so one node failure does not take all of them out. Watch delegate metrics during peak times.

19. A script step fails with "command not found" on the delegate. What happened?

The delegate image does not include the tool. Install it at startup with the delegate's init script, build a custom delegate image with the tool, or run the step in a container step with an image that has it. Do not assume the delegate image has kubectl plugins, cloud CLIs or jq.

20. An expression resolved to null. How do you debug it?

First check timing: the value must exist when the expression is evaluated. A step output referenced before that step runs is null. Then check the path and scope (stage, step group, or pipeline) and the exact identifier, not the display name. Add a temporary shell step that echoes the expression.

21. Why does an expression work in one stage but not another?

Expressions are scoped. Inside the same stage you can use a short path, but from another stage you need the full path, such as <+pipeline.stages.build.spec.execution.steps.compile.output.outputVariables.version>. Identifiers change when a stage is in a looping strategy, because each iteration gets an indexed identifier.

22. A string comparison in a conditional execution never matches. What do you check?

Check quoting and whitespace in the value, check that the value is a string and not null, and remember that Harness supports only == and != for string comparison in expressions. Print the resolved value in a step to see exactly what you are comparing.

23. How do you pass a value from a shell script to later steps?

Export the variable in the script and declare it as an output variable on the step. Later steps read it with the step's output path. If the variable is not exported or the names differ, later steps get null.

export IMAGE_TAG=$(git rev-parse --short HEAD)
# Declare IMAGE_TAG as an output variable on this Shell Script step

24. A deployment fails with an image pull error. What do you check?

Confirm the artifact tag exists in the registry, the image path is correct, the cluster has pull credentials (an image pull secret or node identity), and the registry is reachable from the nodes, not just from the delegate. Harness resolves the tag, but the nodes pull the image.

25. Manifests render wrongly in a Kubernetes deployment. How do you debug?

Use dry run or review the rendered manifests in the step log, check the values files and overrides that apply to this environment, and check the order in which overrides merge. A missing environment override often explains "it works in dev but not in prod".

26. A Helm deployment fails because a release is stuck in a pending state. What do you do?

Check the release with helm history on the cluster. A previous interrupted run can leave the release in a pending state that blocks new upgrades. Roll back to the last deployed revision with Helm after checking what changed, then rerun the pipeline.

27. A blue-green swap succeeded but users still hit the old version. Why?

The swap step changes which service selectors point at primary and stage, and it does not run a health check itself. Check the primary service's selector after the swap, check any ingress or DNS that points at the wrong service, and check client-side caching. Add a verification step after the swap.

28. A canary deployment does not split traffic by percentage. Why?

The Harness Kubernetes canary scales a canary workload next to the stable one. Without traffic routing through a service mesh or ingress, traffic split follows pod counts, not an exact percentage. If the interviewer asks for 10% exact traffic, mention configuring traffic routing.

29. A step fails only on the first run of the morning. What could it be?

Look for things that expire or scale overnight: short-lived cloud credentials, a delegate scaled down, a cluster autoscaler that has to add nodes, or caches that need warming. Check the timing in the logs, then make the step tolerant with a retry on that specific error type.

Real-time Harness questions for senior engineers

30. A bad release reached production even though you have a Verify step. How do you investigate?

Check that the Verify step ran and what it compared: health source, metrics, baseline, sensitivity and duration. If traffic was low, the analysis may not have had enough data. If the metric did not cover the failure (for example only latency, not errors), add the right signal. Adjust sensitivity only with evidence.

31. Rollback did not run after a failed deployment. Why?

Check the failure strategy on the failed step and its stage. If the strategy was Ignore Failure, Mark as Success or a Retry that eventually succeeded, rollback never starts. Also check whether the error type matched the strategy. Set Rollback Stage for deployment step errors in production stages.

32. Rollback ran but the service is still broken. What now?

Rollback restores the previous release of what Harness deployed, but it does not undo database migrations, config changes made elsewhere, or external dependencies. Check what changed outside the deployment. Make schema changes backward compatible and deploy them separately so app rollback stays safe.

33. How do you design failure strategies so they help during an incident?

Retry only on errors that are transient (connectivity, timeouts), roll back on deployment and verification failures, and use manual intervention with a timeout for production when a human decision is needed. Set defaults at stage level and override at step level only when needed.

34. A pipeline that ran fine for months now fails across many teams at once. Where do you look?

Look for shared dependencies: a template that was updated, a shared delegate pool, a connector whose token expired, a changed OPA policy, or a registry outage. Check template version history and recent account changes in the audit trail. Pin template versions for critical pipelines.

35. An approval step waits forever. How do you handle it?

Check who the approvers are (users or groups), whether they were notified, and whether the minimum approver count is reachable. Set a timeout on approval steps with a clear failure action. For Jira or ServiceNow approvals, check the ticket status matches the approval criteria exactly.

36. A deployment freeze blocks an urgent hotfix. What are the options?

Deployment freeze windows block deployments in their scope. Users with the right permission can override freezes, and freeze rules can exclude specific services or environments. Follow your change process: get approval, use the authorised override, record why, and avoid editing the freeze window ad hoc.

37. A run fails with a policy evaluation error. How do you respond?

Read which OPA policy failed and why; Harness shows the policy set and message. Either the pipeline breaks a rule (for example missing an approval step in production) and should be fixed, or the policy is wrong and needs a reviewed change. Do not disable the policy set to get one run through.

Scenario-based real-time Harness questions

38. Production deploys fail with "no eligible delegate" right after a network change. Walk through your response.

Confirm delegates are disconnected in the UI, check delegate logs for connection errors, test outbound HTTPS 443 from the delegate host through the proxy, and work with the network team to restore egress. While that happens, pause scheduled triggers so failures do not pile up, then rerun the failed stages.

39. A canary passes verification but errors spike after the rolling phase. What happened, and what do you change?

The canary had too little traffic or too short a window to show the error, or the error only appears at full load. Roll back, then lengthen verification, include error-rate metrics, and verify again after full rollout. Consider a second verify step after the rolling phase.

40. A trigger fires twice for one Git push and deploys twice. How do you fix it?

Check whether two triggers match the same event, or one trigger listens to multiple event types (for example push and pull request merge). Check webhook configuration in the Git provider for duplicate hooks. Tighten trigger conditions and use a resource constraint or queue step so overlapping runs do not deploy the same service at once.

41. Your team's pipeline runs slowly during peak hours. How do you diagnose?

Compare step durations at peak and off-peak. If time is spent waiting before steps start, the delegates are saturated or a resource constraint is queuing runs. If steps themselves are slow, check target systems. Add delegate replicas, split delegate pools by team, and remove serial waits that are not needed.

42. After an incident, management asks how to stop the same failure. What do you propose?

Write a short review: what failed, why the pipeline let it through, and what changes. Typical actions are a verification metric that would have caught it, a failure strategy that rolls back on that error, a policy that requires the step in production templates, and an alert on delegate health.

How to prepare for real-time Harness questions

  1. Install a Kubernetes delegate on a local or cloud cluster and deploy a sample app with a rolling deployment.
  2. Break things on purpose: wrong image tag, expired connector token, missing selector, failing readiness probe.
  3. Write expressions that pass values between steps and stages, then reference them too early to see the null result.
  4. Set different failure strategies and watch which ones roll back.
  5. Practise explaining each failure in symptom, cause, fix and prevention order.

Harness questions often lead into Kubernetes and general CI/CD design. Review 12 Kubernetes architecture components and the CI/CD pipeline examples from big tech companies, plus the 200+ DevOps interview questions hub. Our DevOps training covers pipelines, Kubernetes deployments and rollback practice, in our Kharadi, Pune classroom and live online.

Official references

FAQs

What ports does a Harness delegate need?

Outbound HTTPS on port 443 to Harness. The delegate needs no inbound ports, and it also needs network access to the targets it works with.

Why does a Harness expression return null?

Usually because the value is not available yet when the expression is evaluated, or because the path, scope or identifier is wrong. Print it in a shell step to check.

Why didn't my Harness deployment roll back?

Rollback only runs if the failure strategy for that error says Rollback Stage or Rollback Pipeline. Ignore Failure or Mark as Success prevents it.

Do Docker delegates upgrade automatically?

No. Kubernetes and Helm delegates can auto-upgrade; Docker delegates must be upgraded manually, and delegate images expire.

Can Harness canary deployments split traffic by exact percentage?

Not by themselves. Without a service mesh or ingress traffic routing, traffic follows pod counts.

Does the blue-green swap step check health?

No. Add a verification or health check step after the swap.

How should I answer real-time Harness questions in an interview?

State the symptom, the first place you would look, the likely causes in order, the fix, and how you would prevent a repeat.

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.