17.8 Monitoring, logging, Kubernetes

Q40. What are the four golden signals?

Latency, traffic, errors, saturation — plain questions about user experience and fullness of the system.

Q41. Prometheus vs CloudWatch (how you answer)?

Both collect metrics and support alerts. This course practised Prometheus + Node Exporter + Grafana as primary lab path and treated CloudWatch as the AWS-native alternate. Pick based on team standard; do not invent product limits in the interview — say you would check current docs.

Q42. What is alert fatigue?

Too many noisy alerts train humans to ignore pages. Prefer fewer, actionable alerts with runbooks.

Q43. Why log to stdout in containers?

The platform can collect streams uniformly (docker logs, agents, cluster log shipping).

Q44. What must you never log?

Passwords, tokens, raw .env, unnecessary PII.

Q45. Pod vs Deployment vs Service?

Pod runs container(s); Deployment keeps replica Pods updated; Service provides stable network access to Pods via selectors.

Q46. Why not rely on latest in production?

Tags move; you lose certainty about which bits run. Prefer immutable SHA/version tags.

Q47. How do you debug CrashLoopBackOff?

kubectl describe Events + kubectl logs; fix image/command/config; check probes.

Q48. What local cluster tool did you practise?

kind (Kubernetes in Docker). Mention k3d/minikube exist but stick to what you used.

Q49. Correlation id — why care?

Ties one user request across multiple log lines/services so you are not matching timestamps by eye.

Q50. Metrics vs logs vs traces?

Metrics detect; logs explain; traces follow a request across services.