.NET Rocks Live
summary
Claims that observability must use one wide structured event per request and that tracing should be span-based, not metric-only; recommends OpenTelemetry and intelligent sampling (head vs tail, Honeycomb’s Refinery), plus a Bubble Up diff for root-cause signals. Cites a production footprint: ~700 customers, billions of events per second, 60 days of cheap S3 storage with Lambda-driven query planning. Advises instrumenting the biggest painful outage first, derive SLOs from events, and optimize sampling to control cost.
AI-written summary · about
chapters (22)
- 0:00.NET Rocks Live in Porto
- 2:20Show intro and Better No Framework
- 3:01Pulumi 8 and .NET performance
- 5:11Minimal architecture and ASS
- 7:16Introducing Charity Majors
- 8:18Observability vs telemetry
- 10:24Observability as the bridge between development and operations
- 12:48Modern observability and spans
- 16:39Honeycomb origins and Parse journey
- 19:26Observability at Facebook
- 21:57Cultural shift in debugging
- 24:46Journaling in observability
- 27:37OpenTelemetry pitfalls
- 30:17Observability future and AI
- 34:09Legal accountability in software
- 36:55Debugging in production
- 39:23Introducing observability to teams
- 41:50Data volume and cost in observability
- 45:45Balancing traces metrics SLIs
- 49:04Observability mistakes
- 51:52Getting started with OpenTelemetry
- 53:00Closing remarks and Q&A
related talks
38mThe Murphy's Laws of Observability
39mObservability for Emerging Infra: What Got You Here Won't Get You There
31mObservability for Developers: How to Get There
35mRefining Systems Data without Losing Fidelity
26mDriving O11y x to Nx: Probing Bullet Holes and not Breaking the Bank
42mFounder/CTO Perspectives: The Future of Distributed Tracing
30mSeeing RED
30mPanel: Telemetry That Matters