Observe Tool Use CloudWatch Metrics built-in service metrics + custom metrics (Embedded Metric Format in logs) CloudWatch Logs + Logs Insights central logs with a query language Alarms thresholds or anomaly detection → SNS / auto scaling / incident tools X-Ray / Application Signals distributed traces and service maps (OpenTelemetry supported) CloudTrail who did what via the API — keep an org trail on Set log retention (default is never expire). Control cost Budgets with alerts (actual and forecast), Cost Explorer, Cost Anomaly Detection. Tag everything (project, env, owner) and activate cost-allocation tags. Right-size with Compute Optimizer; buy Savings Plans for steady usage; use Spot for fault-tolerant work. Classic surprises: idle NAT gateways, unattached EBS volumes/EIPs, old…
Key Takeaways
Set log retention explicitly and query logs with Logs Insights.
Alarm on symptoms users feel (errors, latency, queue depth) and route to on-call.
Create budgets with forecast alerts and tag resources for cost allocation.