Top DevOps Engineer Interview Questions & Answers

Source control trigger → automated tests (unit, integration, security scan) → build artifact → staging deployment → integration tests → approval gate → production deployment → post-deployment monitoring alerts. Key principles: fast feedback, idempotency, and blue/green or canary deployment to minimize production risk.

Continuous delivery means every change is built, tested, and ready to deploy — but a human approves each production release. Continuous deployment goes one step further: every passing build is automatically deployed to production without human approval.

Never hardcode secrets in code or CI configuration. Use secrets management tools: HashiCorp Vault, AWS Secrets Manager, Azure Key Vault. Rotate secrets regularly, use least-privilege IAM roles, and scan for accidentally committed secrets (GitGuardian, git-secrets).

Terraform: cloud-agnostic, large ecosystem, requires state management. Pulumi: uses general-purpose languages, smaller community. CloudFormation: AWS-native, deep service integration, verbose. CDK: AWS-specific but uses programming languages. Most teams use Terraform for multi-cloud.

kubectl describe pod (check Events), kubectl logs (container logs), kubectl get events (namespace-level events), inspect image pull status, check resource requests vs. node capacity, check ConfigMaps/Secrets mounting. For CrashLoopBackOff: examine logs with --previous flag to see the last crash reason.

Use the RED method (Rate, Errors, Duration) for services and the USE method (Utilization, Saturation, Errors) for resources. Alert on symptoms (user-facing impact), not causes — too many cause-based alerts create alert fatigue.

Detection → triage → incident commander assignment → mitigation (restore service: rollback, scale, disable feature flag) → communication (status page, stakeholder updates) → root cause analysis (blameless postmortem within 48 hours) → prevention (backlog items for long-term fixes).

Requests are what the scheduler uses to place pods; limits are the maximum the container can use. Set requests based on actual measured usage, set limits at 2-3x requests to handle bursts. Use VPA in recommendation mode to gather data first.

Deployments manage stateless pods — they can be replaced, rescheduled, and scaled freely. StatefulSets manage stateful applications (databases, message queues) where pod identity, stable network names, and ordered deployment/scaling matter.

Use contract testing (Pact) to verify service interfaces. Deploy with health checks and circuit breakers. Use feature flags for risky changes. Ensure each service can be deployed independently — tight coupling in deployment pipelines is a sign of architectural debt.