Blackfire Continuous Profiler
Low-overhead continuous profiling for app performance optimization.
py-spy
Zero‑overhead sampling profiler for live Python applications
A low‑overhead, Rust‑based sampling profiler that attaches to running Python processes without code changes, generating flame graphs, live top views, and stack dumps across Linux, macOS, Windows, and FreeBSD.
Overview
Overview
py-spy is a Rust‑implemented sampling profiler that can attach to any running CPython process without modifying its code. By reading the target's memory directly, it provides accurate timing information while keeping overhead minimal, making it safe for use in production environments.
Capabilities & Deployment
The tool offers three sub‑commands: record for generating flame‑graph SVGs or speedscope files, top for a live, top‑like view of hot functions, and dump for a one‑off stack trace with optional locals. It supports profiling native extensions (--native), tracing subprocesses (--subprocesses), and works on Linux, macOS, Windows, and FreeBSD. Installation is straightforward via pip, Homebrew, Cargo, or distribution packages, and the binary runs out‑of‑process, avoiding any impact on the profiled application.
Developers, SREs, and performance engineers can quickly diagnose bottlenecks, even in multi‑process services, without redeploying or restarting the application.
Highlights
No code instrumentation; attaches to running processes
Extremely low overhead thanks to Rust and out‑of‑process sampling
Generates flame graphs, speedscope files, live top view, and stack dumps
Supports native extension profiling and automatic subprocess tracing
Pros
- Safe for production use
- Cross‑platform support for Linux, macOS, Windows, and FreeBSD
- Simple installation via pip, Homebrew, Cargo, etc.
- Multiple output formats for flexible analysis
Considerations
- Root or sudo may be required to attach to existing processes
- Limited to CPython; does not support PyPy or Jython
- Native profiling needs symbols and is platform‑restricted
- Strict security policies can block memory‑reading operations
Managed products teams compare with
When teams consider py-spy, these hosted platforms usually appear on the same shortlist.
Looking for a hosted option? These are the services engineering teams benchmark against before choosing open source.
Fit guide
Great for
- Developers debugging performance issues in live services
- Ops teams needing quick insight without redeploying
- Projects using C/C++ or Cython extensions that require combined profiling
- Applications employing multiprocessing or Gunicorn workers
Not ideal when
- Environments where root access cannot be granted
- Projects using alternative Python interpreters
- Scenarios requiring deterministic, line‑by‑line tracing
- Systems with hardened security that block process memory reads
How teams use it
Generate flame graph for a production web service
Visualize hot paths without restarting, identify bottlenecks, and reduce latency.
Live monitoring of a data‑processing script
Use `py-spy top` to see real‑time function usage and spot stalls.
Profile a Cython‑accelerated module
Enable `--native` to include native frames, helping optimize compiled extensions.
Trace multiprocessing workers in a Gunicorn deployment
Attach with `--subprocesses` to collect samples from all worker processes.
Tech snapshot
Rust99%
Python1%
Shell1%
Frequently asked questions
Why do we need another Python profiler?
py-spy can profile any running Python program without code changes, making it safe for production where other profilers add noticeable overhead.
How does py-spy work?
It reads the target process's memory via system calls, inspects the interpreter state, and reconstructs call stacks without running inside the Python process.
Can py-spy profile native extensions?
Yes, using the `--native` flag. For best results the extension should be compiled with symbols.
How can I profile subprocesses?
Pass the `--subprocesses` flag; py-spy will automatically attach to child Python processes and include their samples.
When do I need to run py-spy with sudo?
Attaching to an existing process often requires root on Linux and always on macOS; creating the process with py-spy avoids this requirement.
Project at a glance
Active- Stars
- 14,853
- Watchers
- 14,853
- Forks
- 496
LicenseMIT
Repo age7 years old
Last commit2 weeks ago
Primary languageRust
Last synced yesterday