DevOps & Cloud Engineer Interview Questions

For zero-downtime Kubernetes deployments, I implement a Canary release pattern using Argo Rollouts combined with Istio service mesh. First, properly configure `readinessProbe` and `livenessProbe` with graceful `preStop` lifecycle hooks so inflight requests drain safely before pods terminate. During rollout, route 5% of ingress traffic to the canary replica. We run automated Prometheus metric checks for 5 minutes monitoring HTTP 5xx error rates and p99 latency. If error rates remain below 0.05%, traffic steps up to 25%, 50%, and 100%; otherwise, automated instant rollback occurs without user-facing downtime.

First, I check whether the high CPU is impacting user-facing SLA (error rates and latency). Second, I run `kubectl top pods -A --sort-by=cpu` and inspect Grafana dashboards to identify if one specific namespace or rogue container has runaway CPU consumption or an unconstrained thread loop. Third, I verify if Cluster Autoscaler and Horizontal Pod Autoscaler (HPA) are properly spinning up additional compute nodes or if we hit cloud quota limits. If a runaway bad release caused a spin loop, I execute an immediate rollback while pulling thread and heap dumps for post-incident RCA.