Perfect Match: Correlating Continuous Profiling with Distributed Tracing for Stronger Observability
Talkat Observability Day North America 2024 · on Tuesday, 2024-11-12· at 11:15· Salt Palace | Level 2 | 255 B·
summary
A low‑overhead eBPF profiler (donated to OpenTelemetry) samples CPU stacks (~20 Hz) and writes service/resource and trace context into native thread-local storage via a small JNI extension so traces and profiles correlate exactly; targets under 1% system CPU and <250 MB RAM on amd64/arm64 and handles Java virtual threads. Demo on a microservice cluster finds GC, a bad log loop and a Fibonacci hotspot tied to the "binoculars" span attribute; benchmarks show only low‑nanosecond sync overhead.
AI-written summary · about
chapters (7)
- 0:00Continuous profiling intro
- 3:50Profiling with OpenTelemetry
- 12:20Demo setup overview
- 14:10Flame graphs for CPU
- 18:04Trace-level profiling demo
- 20:59Filter profiling by span
- 22:43Q&A
related talks
40mContinuous Profiling, the missing piece in your observability puzzle!
24mWhy Does Continuous Profiling Matter to Developers?
56mContinuous Profiling the missing piece in your observability puzzle
36mProfiling in the cloud-native era - Matthias Loibl
01h 13mContinuous Profiling: The Missing Piece in Your Observability Puzzle
52mThe missing piece in your observability puzzle!
24mCommunity Driven Breakout Session: Continuous Profiling
34mTrace-Profiling: A New Way About How to Track Application Behavior