SeniorCI/CD & DevOps for QA
How do you implement canary deployments and test them?
- canary deployment
- progressive delivery
- rollback
Answer it first, then compare. Nothing here is saved or scored.
Model answer
Canary deployment routes a small percentage of traffic (1-5%) to the new version. Testing approach: (1) Automated health checks on the canary (error rate, latency, CPU). (2) Comparison metrics: canary vs baseline (old version). (3) Automatic rollback if canary metrics exceed thresholds. (4) Gradual promotion: 1% to 5% to 25% to 100% over hours/days. (5) Feature flag integration: canary + feature flags for fine-grained control. (6) QA smoke tests against the canary endpoint specifically. Tools: Kubernetes canary with Istio/Argo Rollouts, AWS CodeDeploy, Vercel preview deployments.
What interviewers listen for
- Pipeline shapeYou design stages by failure cost and runtime budget, not by tool feature lists.
- Parallelisation craftYou can shard a real suite and explain the state-isolation choices you made.
- Quality gatesYou know which checks block merge, which warn, and why those defaults are correct.
- Shift-left fluencyYou talk about ownership transfer, not just earlier execution.