core.py The Megahertz
10 snips
Jul 12, 2025 Discover the exciting new built-in sampling profiler introduced in Python 3.15, revolutionizing how developers analyze performance. Dive into the complexities of profiling AsyncIO applications and the challenges of multi-threaded environments. The conversation also highlights the launch of memory.python.org and the advancements in profiling tools, including the benefits of efficient sampling to reduce overhead. Enjoy a light-hearted take on the technical intricacies and common misconceptions surrounding hashing and performance optimization!
AI Snips
Chapters
Transcript
Episode notes
Tracing vs Sampling Profilers
- Tracing profilers introduce high overhead by stopping program execution at every call and return.
- Sampling profilers provide a statistical understanding of program hotspots with significantly less overhead.
Avoid Using Tracing Profilers
- Avoid using Python tracing profilers like profile due to high overhead and complexity.
- Prefer the new sampling profiler for better accuracy and minimal performance impact.
Profiling the Profiler on macOS
- Pablo used Apple's Instruments on macOS to profile the sampler's initial performance.
- He discovered it ran at only 3 Hz initially, which was later optimized to 10,000 Hz and beyond.