Best practices for application observability at scale
summary
Starts from user experience as the primary metric and enforces “one structured request log per unit of work” so each HTTP or queue request emits a request data log that includes trace ID, timings, attributes and dimensions. Uses AWS tools: CloudWatch RUM captures page load and web vitals (example: observed a page taking ~10s to DOM processing), Synthetic canaries run a six‑step random order workflow with screenshots and step durations, and Embedded Metric Format converts log fields into CloudWatch metrics. Alarms use anomaly detection and composite alarms (canary duration AND API latency) to avoid 3:00 AM noise.
AI-written summary · about
chapters (9)
- 0:00The observability challenge
- 1:51Defining observability
- 5:32Customer-centric strategy
- 7:40Dashboarding best practices
- 10:07Designing effective dashboards
- 12:20Capturing rich telemetry
- 17:59Designing actionable alarms
- 20:51Observability tools demo
- 27:55Conclusion and resources
related talks
56mFull-stack observability and application monitoring with
58mObserving Applications in Production
24mObservability with AWS
20mDesigning the Right Observability Stack in AWS Cloud
34mObservability in Serverless Application
31mObservability for Developers: How to Get There
16mBuilding Observability into Cloud-Native Applications
28mOperational visibility in distributed systems