Python Performance at Scale - Making Python Faster at Instagram
summary
Shows concrete changes: a production Cinder fork of CPython with immortalized objects to cut copy‑on‑write, hot-path wins from inline caching and eager async that save allocations, build-time gains from PGO and method-at-a-time JIT plus bolt and huge pages, and experiments like Static Python and Pyro. Claims are a measured 20–30% RPS uplift in production, up to ~7x on some benchmarks, with profiling via linux perf and uwsgi process model driving the work.
AI-written summary · about
chapters (5)
related talks
48mPython Performance: Past, Present and Future
22mPyPy in production
57mPython and PyPy performance (not) for dummies
01h 03mA tour of CPython's runtime
23mMemoirs of a Python Object: Memory Management and Improving Performance
28m"Performant Python"
58mPython performance Past, Present, Future
22mPyPy and the future of the Python ecosystem