Into the flamegraph: From the primitives through advanced concepts
summary
Shows how continuous production profiling with an eBPF-based profiler and flamegraphs finds real fixes: a Python fork/exec kernel hotspot led to a Python 310 patch that removes expensive page-table copying, and a SQLAlchemy endpoint change (caching a compiled expression) cuts one function from ~34% to ~28% of samples and drops container CPU from ~55% to under 50%. Covers sampling vs tracing, mixed native/kernel stacks, table views (total vs own time), and an end-to-end profile→fix→verify demo.
AI-written summary · about
chapters (4)
- 0:00Start
- 3:13Today's focus - CPU Profiling
- 5:02Continuous and Production Profiling
- 10:19#1 Mixed stacks - kernel side importance
related talks
26mOptimize Performance with Continuous Production Profiling
56mContinuous Profiling the missing piece in your observability puzzle
17mExtending Flamegraphs for Multi-Dimensional Performance Analysis
31mTracing, Profiling & Debugging in Production (eBPF)
26mProfiling Mastery: Decode Your Data and Elevate Observability
50mVisualizing Performance - The Developers’ Guide to Flame Graphs
33mA Tale of Two Flamegraphs: Continuous Profiling in Ruby
01h 07meBPF: Fueling New Flame Graphs & more