Continuous Improvement for Existing Solutions
Continuous-improvement questions test your ability to observe, diagnose, and optimize running workloads. Know the observability stack: CloudWatch metrics, logs, and alarms; CloudTrail for API audit; AWS Config for configuration drift; and X-Ray for distributed tracing. Understand how to set meaningful alarms with composite alarms and anomaly detection, and how to use CloudWatch Logs Insights for querying.
Optimization levers
For cost, use Cost Explorer, Budgets, Savings Plans, and rightsizing recommendations; know that Compute Optimizer analyzes utilization. For performance, examine caching layers (ElastiCache, CloudFront), read replicas, and provisioned throughput. For reliability, review retry and timeout settings, health checks, and chaos testing with AWS Fault Injection Service. For operational excellence, use Infrastructure as Code, CI/CD pipelines, blue/green and canary deployments, and runbooks.
Traps to watch
- Treating high CPU as the only scaling signal; queue depth and latency often matter more.
- Forgetting that CloudTrail records control-plane API calls, not application behavior.
- Assuming Savings Plans cover all services; they cover compute families specifically.
- Overlooking that Config evaluates resource configuration, not runtime performance.
Worked example
An application shows rising p99 latency while CPU stays low. Check X-Ray traces to find a downstream database call dominating latency, then add a read replica or cache hot queries in ElastiCache. Confirm with CloudWatch metrics before changing capacity. If the issue is spiky traffic, use target-tracking scaling on request count rather than CPU. Rule of thumb: measure first, change one variable, and verify with the same metric that exposed the problem. Pair every optimization with an alarm so regressions are caught.