Skip to main content

A call stack profiler for Python. Inspired by Apple's Instruments.app

Project description

pyinstrument
============

A Python profiler that records the call stack of the executing code, instead
of just the final function in it.

[![Screenshot](screenshot.jpg)](https://raw.githubusercontent.com/joerick/pyinstrument/master/screenshot.jpg)

It uses a **statistical profiler**, meaning the code samples the stack
periodically (every 1 ms). This is lower overhead than event-
based profiling (as done by `profile` and `cProfile`).

This module is still very young, so I'd love any feedback/bug reports/pull
requests!

Installation
------------

pip install -e git+https://github.com/joerick/pyinstrument.git#egg=pyinstrument

pyinstrument supports Python 2.7 and 3.3+.

Usage
-----

- **Command-line**

You can call pyinstrument directly from the command line.

python -m pyinstrument myscript.py [args...]

This will run `myscript.py` to completion or until you interrupt it, and
then output the call tree.

- **Django**

Add `pyinstrument.middleware.ProfilerMiddleware` to `MIDDLEWARE_CLASSES`.
If you want to profile your middleware as well as your view (you probably
do) then put it at the start of the list.

Then add `?profile` to the end of the request URL to activate the
profiler.

- **Python**

```python
from pyinstrument import Profiler

profiler = Profiler() # or Profiler(use_signal=False), see below
profiler.start()

# code you want to profile

profiler.stop()

print(profiler.output_text(unicode=True, color=True))
```

You can omit the `unicode` and `color` flags if your output/terminal does
not support them.

Signal or setprofile mode?
--------------------------

On Mac/Linux/Unix, pyinstrument can run in 'signal' mode. This uses
OS-provided signals to interrupt the process every 1ms and record the stack.
It gives much lower overhead (and thus accurate) readings than the standard
Python [`sys.setprofile`][setprofile] style profilers. **However, this can
only profile the main thread**.

On Windows and on multi-threaded applications, a `setprofile` mode is
available by passing `use_signal=False` to the Profiler constructor. It works
exactly the same as the signal mode, but has higher overhead. See the below
table for an example of the amount of overhead.

[setprofile]: https://docs.python.org/2/library/sys.html#sys.setprofile

| Django template render × 4000 | Overhead
---------------------------|------------------------------:|---------:
Base | 1.46s |
| |
pyinstrument (signal) | 1.84s | 26%
cProfile | 2.18s | 49%
pyinstrument (setprofile) | 5.33s | 365%
profile | 25.39s | 1739%


Known issues
------------

- When profiling Django, I'd recommend disabling django-debug-toolbar,
django-devserver etc., as their instrumentation distort timings.

- In signal mode, any calls to [`time.sleep`][pysleep] will return
immediately. This is because of an implementation detail of `time.sleep`,
but matches the behaviour of the C function [`sleep`][csleep].

[pysleep]: https://docs.python.org/2/library/time.html#time.sleep
[csleep]: http://pubs.opengroup.org/onlinepubs/009695399/functions/sleep.html

Further information
===================

Call stack profiling?
---------------------

The standard Python profilers [`profile`][1] and [`cProfile`][2] produce
output where time is totalled according to the time spent in each function.
This is great, but it falls down when you profile code where most time is
spent in framework code that you're not familiar with.

[1]: http://docs.python.org/2/library/profile.html#module-profile
[2]: http://docs.python.org/2/library/profile.html#module-cProfile

Here's an example of profile output when using Django.

151940 function calls (147672 primitive calls) in 1.696 seconds

Ordered by: cumulative time

ncalls tottime percall cumtime percall filename:lineno(function)
1 0.000 0.000 1.696 1.696 profile:0(<code object <module> at 0x1053d6a30, file "./manage.py", line 2>)
1 0.001 0.001 1.693 1.693 manage.py:2(<module>)
1 0.000 0.000 1.586 1.586 __init__.py:394(execute_from_command_line)
1 0.000 0.000 1.586 1.586 __init__.py:350(execute)
1 0.000 0.000 1.142 1.142 __init__.py:254(fetch_command)
43 0.013 0.000 1.124 0.026 __init__.py:1(<module>)
388 0.008 0.000 1.062 0.003 re.py:226(_compile)
158 0.005 0.000 1.048 0.007 sre_compile.py:496(compile)
1 0.001 0.001 1.042 1.042 __init__.py:78(get_commands)
153 0.001 0.000 1.036 0.007 re.py:188(compile)
106/102 0.001 0.000 1.030 0.010 __init__.py:52(__getattr__)
1 0.000 0.000 1.029 1.029 __init__.py:31(_setup)
1 0.000 0.000 1.021 1.021 __init__.py:57(_configure_logging)
2 0.002 0.001 1.011 0.505 log.py:1(<module>)


When you're using big frameworks like Django, it's very hard to understand how
your own code relates to these traces.

Pyinstrument records the entire stack, so tracking expensive calls is much
easier.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

pyinstrument-0.10.1.tar.gz (43.5 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page