Python Profiling State of the World
summary
Survey of concrete tools and results rather than theory: cProfile and profile are shown with measured overheads — 15% overhead for cProfile and ~106% for the pure‑Python profile; sampling profilers (py-spy, pyinstrument, Austin) run out‑ or in‑process with different tradeoffs and sampling rates, e.g. 9% at 100Hz / ~25% at 1kHz for py‑spy; Flask middleware example to capture per‑request profiles, visualizations via flamegraphs and speedscope, and a production system that aggregates profiles and surfaces automated insights (Sentry in‑process profiler).
AI-written summary · about
chapters (6)
- 0:00Start
- 2:17What is a profiler?
- 3:38A brief history of profiling
- 10:49Deterministic vs. sampling profilers
- 14:42Visualizing profile data
- 47:07Sentry Profiling+
related talks
26mOptimize Performance with Continuous Production Profiling
24mCommunity Driven Breakout Session: Continuous Profiling
21mFrom PyPerf to py-spy - Everything You Need to Know About Python Profilers
51mProduction Profiling: What, Why and How
56mContinuous Profiling the missing piece in your observability puzzle
01h 13mContinuous Profiling: The Missing Piece in Your Observability Puzzle
31mProfiling Formats and Why You Should Care
33mInto the flamegraph: From the primitives through advanced concepts