Profiling Mastery: Decode Your Data and Elevate Observability
Talkat Observability Day North America 2024 · on Tuesday, 2024-11-12· at 09:45· Salt Palace | Level 2 | 255 E·
summary
Sampling CPU profiling using an eBPF agent and an OTLP-compatible wire format produces sampled stack traces, visualized as icicle/flame graphs and cumulative versus self metrics. A demo shows a central storage and query service called Parka that synthesizes metrics and filters by labels or call paths. Concrete fixes include replacing generic ioutil ReadFile patterns in containerd, a microchange that improves global containerd runtime by 5%. Examples cover Go and Rust callers, and distributed-tracing ID attachment for cross-service CPU attribution.
AI-written summary · about
chapters (17)
- 0:00Why profiling matters
- 1:08Transistors and Moore's Law
- 4:59What is profiling
- 6:45Sampling vs tracing
- 8:54How sampling works
- 10:00Flame graphs
- 12:49Cumulative vs self time
- 14:17Flame vs icicle graphs
- 15:21Optimizing code
- 16:46Parka observability tool
- 17:30Parka infrastructure demo
- 18:56Real-world optimization
- 20:00ContainerD case study
- 22:29Rust workloads optimization
- 23:54Custom DB query optimization
- 24:58Distributed tracing
- 25:18Conclusion
related talks
36mContinuous Profiling with Parca- Beneath the Kernel to Beyond...
36mProfiling in the cloud-native era - Matthias Loibl
56mContinuous Profiling the missing piece in your observability puzzle
24mCommunity Driven Breakout Session: Continuous Profiling
36mContinuous Profiling in the Cloud-Native era
01h 13mContinuous Profiling: The Missing Piece in Your Observability Puzzle
33mBuilding a Go Profiler Using Go
31mProfiling Formats and Why You Should Care