CloudWatch

CloudWatch collects metrics, logs, alarms, and events for AWS workloads. Node.js services on Lambda, ECS, or EC2 should emit structured JSON logs to stdout; the agent or platform ships them to log groups with retention policies.

| Signal | Tooling | |--------|---------| | Metrics | Custom namespaces, EMF from Lambda, built-in AWS metrics | | Logs | Log groups, metric filters, Logs Insights queries | | Alarms | SNS/Auto Scaling actions on thresholds | | Dashboards | Cross-service operational view |

Watch cardinality: high-cardinality labels in custom metrics inflate cost. Define SLO-driven alarms (error rate, latency p99, queue age) instead of paging on every CPU blip.

On interviews: Logs Insights query basics, metric filters versus embedded metric format, composite alarms, and how CloudWatch differs from APM tools like X-Ray or third-party vendors.

Common pitfalls: infinite log retention bills; logging secrets or PII; alarms without runbooks; relying only on default AWS metrics while app errors stay invisible.

The trade-off is native integration and low friction versus limited tracing depth and query ergonomics compared with dedicated observability stacks.

Checklist:

• Structured logs with request IDs and service name.
• Set log retention per environment.
• Alarm on user-visible symptoms, not only infrastructure CPU.
• Link dashboards to on-call runbooks and SNS topics.