Profiling in the cloud-native era - Matthias Loibl
summary
Parca open-source continuous profiler scrapes process profiles (originally 100 Hz sampling, practical deployments use 19 Hz sampling (previously 100 Hz)) with ~2–3% CPU overhead, stores stacks in a columnar TSDB built on Arrow and Plonge, and maps addresses via DWARF or JVM perf maps. Agent uses eBPF to collect stacks without code changes and supports C/C++/Rust/Go plus new Ruby and Python work. System claims fleet-wide savings of up to 30% infra spend, runs agents as a daemonset, and currently lacks built-in HA.
AI-written summary · about
chapters (19)
- 0:00Introduction and speaker intro
- 1:05Observability conference mention
- 1:26What is profiling and its importance
- 2:30Tracing vs sampling profiling
- 3:35Why use profiling in engineering
- 4:17Built-in Go profiling tools
- 7:05Using Go’s pprof tool
- 8:11Profile guided optimizations in Go
- 9:35Limitations of snapshot profiling
- 10:18Introducing continuous profiling
- 11:20Why time dimension matters
- 14:12Introducing Parca for profiling
- 16:20What is eBPF and how it works
- 20:36eBPF integration and benefits
- 23:01Flame graph fundamentals
- 25:40Live demo of Parca
- 27:30Parca storage and query engine
- 29:58Parca future and open source
- 30:40Q&A session
related talks
36mContinuous Profiling in the Cloud-Native era
36mContinuous Profiling with Parca- Beneath the Kernel to Beyond...
28mConprof - Profiling in the cloud-native era
40mContinuous Profiling, the missing piece in your observability puzzle!
01h 13mContinuous Profiling: The Missing Piece in Your Observability Puzzle
26mProfiling Mastery: Decode Your Data and Elevate Observability
56mContinuous Profiling the missing piece in your observability puzzle
36mDeveloping Continuous eBPF Profiler to look Beneath the Kernel to Beyond the Clouds