Yet Another Python Profiler (Yappi)

1.7.6 · active · verified Thu Apr 09

Yappi (Yet Another Python Profiler) is a fast, C-implemented tracing profiler for Python applications, notably supporting multithreaded, asyncio, and gevent code. It overcomes limitations of standard profilers like `cProfile` by offering per-thread CPU time analysis, on-the-fly profiling capabilities, and flexible mechanisms for filtering and sorting profiling results. The library is currently at version 1.7.6 and maintains an active release schedule.

Warnings

breaking Older versions of Yappi (pre-1.x) supported Python 2.6.x up to 3.4. Recent versions (from 1.7.0 onwards) officially support Python 3.6 and higher, including Python 3.14. Users on older Python 2 or early Python 3 versions will need to use an older Yappi release or upgrade their Python interpreter.
Fix: Upgrade Python to 3.6+ or downgrade Yappi to a compatible older version if strict Python version requirements exist.
gotcha Yappi operates as a per-interpreter resource. Attempting to run multiple Yappi profiler instances within the same Python interpreter concurrently will raise an exception.
Fix: Ensure only one `yappi.start()` is active at any given time per interpreter. Use `yappi.stop()` and `yappi.clear_stats()` between profiling sessions if necessary.
gotcha When interpreting profiling results, distinguish between 'tottime' and 'tsub'. 'tottime' (total time) includes time spent in sub-functions called from the profiled function, while 'tsub' (subtime) represents time spent directly within the function itself, excluding its sub-calls.
Fix: Understand the definitions: `tottime` is cumulative, `tsub` is self-time. Choose the appropriate metric based on whether you want to identify bottlenecks within a function's own execution or its entire call graph.
gotcha By default, Yappi profiles using 'CPU time'. For I/O-bound or concurrent applications (e.g., asyncio, gevent), 'Wall time' might be more relevant to capture actual elapsed time, including waits.
Fix: Explicitly set the clock type using `yappi.set_clock_type("wall")` before starting the profiler if Wall time is desired. Default is `"cpu"`.

Install

pip install yappi Install stable version

Imports

yappi
```
import yappi
```
The `_yappi` module is an internal C implementation detail and should not be imported directly. Attempting to do so or encountering an error related to `_yappi` often indicates an installation issue.

Quickstart

This quickstart demonstrates how to profile a simple Python function using `yappi.start()`, executing the target code, `yappi.stop()`, and then printing function statistics. It also shows how to set the clock type (CPU or Wall time) and clear the collected stats.

import yappi
import time

def calculate_sum(n):
    total = 0
    for i in range(n):
        total += i
    return total

def main_task():
    time.sleep(0.1) # Simulate some I/O or other work
    result = calculate_sum(1_000_000)
    time.sleep(0.05)
    return result

# Start profiling
yappi.set_clock_type("cpu") # Can be "wall" for wall time
yappi.start()

# Run the code to be profiled
main_task()

# Stop profiling and print statistics
yappi.stop()

# Get function statistics and print them
func_stats = yappi.get_func_stats()
print("Function Statistics:")
func_stats.print_all(
    columns={0:('ncall', 5), 1:('tottime', 8), 2:('avgtime', 8), 3:('file', 30)},
    sort_type='tottime'
)

# Clear collected statistics
yappi.clear_stats()

view raw JSON →