Beyond OTLP: Unlocking the Potential of OS-native Tracing
summary
Describes writing OpenTelemetry records into kernel-backed ETW and user_events buffers so exports are synchronous, per-CPU, and lock-free — measured as up to 10x faster overall and ~2–6 ns overhead when no listener. Shows working language exporters in .NET, C++, and Rust (contrib) and a collector gap: no ETW/user_events receiver yet. Enables dynamic telemetry (enable checks), a circular flight recorder buffer that can be snapped, and rule-triggered actions like precise call-stack capture or proc dumps on specific event names.
AI-written summary · about
chapters (5)
- 0:00Motivation for dynamic telemetry
- 5:00OS-native tracing architecture
- 11:28Advanced diagnostic scenarios
- 18:52Summary and future directions
- 21:19Q&A and community engagement
related talks
46mThe State of OpenTelemetry
20mObservability and Open Telemetry
30mOpenTelemetry 101
36mObservability For You and Me with OpenTelemetry
58mETW - Monitor Anything, Anytime, Anywhere
18mOpenTelemetry and Observability: What, Why, and Why Now?
39mMonitoring Elixir With OpenTelemetry
24mUnlocking Advanced Processing Capabilities with OpenTelemetry