benfred/py-spy
Py Spy
py-spy is a sampling profiler and process debugger for Python. It allows for the analysis of running processes to identify performance bottlenecks and diagnose hanging programs without requiring code changes or restarts.
The tool operates by reading the memory of a running process from the outside, which enables non-invasive sampling and state collection without pausing execution. It can resolve binary symbols to capture performance data from native extensions written in compiled languages and generate visual flame graphs for both native extensions and subprocesses.
The project provides capabilities for real-time process monitoring, execution profiling, and thread call stack dumping. These utilities allow for the inspection of local variables and call stacks across all active threads to identify deadlocks or infinite loops.
Features
- Python Profilers - Identifies slow functions and execution bottlenecks in running Python programs without modifying source code.
- Native Stack Unwinders - Traverses call stacks across managed and native code boundaries to resolve function names and line numbers.
- Thread Dumps - Provides snapshots of active threads and their local variables to diagnose deadlocks and infinite loops in running processes.
- Non-Pausing State Sampling - Reads interpreter state while the program continues to run to eliminate execution interruptions during data collection.
- Out-of-Process Sampling - Collects performance data by reading a running process's memory from the outside without interrupting execution.
- User-Space Memory Inspection - Reads a process's user-space memory addresses directly to analyze the call stack without modifying the target code.
- Symbol Resolution Engines - Maps raw memory addresses to human-readable function names by parsing binary debug information from native extensions.
- Process Debuggers - Dumps call stacks and local variables from every thread to identify and diagnose hanging processes.
- Thread State Extractors - Iterates through all active threads to capture individual execution paths and local variable states for debugging.
- Production Program Sampling - Analyzes the behavior of running Python applications in production without pausing execution or restarting services.
- Non-Invasive Program Sampling - Samples the execution of running processes to identify bottlenecks without requiring code changes or restarts.
- Native Extension Profiling - Captures performance data from extensions written in compiled languages by reading symbols from the binary.
- Runtime Offset Bindings - Locates internal runtime structures by calculating version-specific memory offsets for compatibility across different Python interpreters.
- Live Execution Monitoring - Displays a real-time view of functions consuming the most time in a running program.
- Runtime Binding Generators - Generates low-level bindings for multiple runtime versions using offset calculations to ensure environment compatibility.
- Real Time Process Monitors - Provides a real-time view of functions consuming the most CPU time in a live Python process.
- Flame Graph Generators - Converts stack trace data into interactive flame graphs for analyzing the performance of native extensions and subprocesses.
- Execution Profilers - Records execution data to files to generate performance visualizations like flame graphs.
- Debugging And Profiling - Sampling profiler for Python programs.
- Debugging Tools - Listed in the “Debugging Tools” section of the Awesome Python awesome list.
- Performance Profiling - Sampling profiler for Python applications.
- System Monitoring And Profiling - Sampling profiler that visualizes execution without stopping the program.
