Skip to main content

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

Project description


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


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


* [Installation](#installation)
* [Usage](#usage)
* [Command-line](#command-line)
* [Django](#django)
* [Python](#python)
* [Signal or setprofile mode?](#signal-or-setprofile-mode)
* [Known issues](#known-issues)
* [Further information](#further-information)
* [Call stack profiling?](#call-stack-profiling)


pip install -e git+

pyinstrument supports Python 2.7 and 3.3+.


#### Command-line ####

You can call pyinstrument directly from the command line.

python -m pyinstrument [options] [args...]

-h, --help show this help message and exit
--setprofile run in setprofile mode, instead of signal mode
--html output HTML instead of text
-o OUTFILE, --outfile=OUTFILE
save report to <outfile>
--unicode force unicode text output
--no-unicode force ascii text output
--color force ansi color text output
--no-color force no color text output

This will run `` 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.

##### Per-request profiling #####

Add `?profile` to the end of the request URL to activate the profiler.
Instead of seeing the output of your view, pyinstrument renders an HTML
call tree for the view (as in the screenshot above).


If you're writing an API, it's not easy to change the URL when you want
to profile something. In this case, add
`PYINSTRUMENT_PROFILE_DIR = 'profiles'` to your
pyinstrument will profile every request and save the HTML output to the
folder `profiles` in your working directory.

#### Python ####

from pyinstrument import Profiler

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

# code you want to profile


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.


This overhead is important because code that makes a lot of Python function
calls will appear to take longer than code that does not.

| 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%

To disable signal mode:

* Use flag `--setprofile` if using the command-line interface
* Use setting `PYINSTRUMENT_USE_SIGNAL = False` in Django
* Use argument `use_signal=False` in the constructor for the Python API

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].

- Some system calls can fail with `IOError` when being profiled in signal
mode. If this happens to you, your only option is to run in setprofile
mode, by passing `--setprofile` to the command-line interface or
use_signal=False to the Python API.


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.


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 "./", line 2>)
1 0.001 0.001 1.693 1.693<module>)
1 0.000 0.000 1.586 1.586
1 0.000 0.000 1.586 1.586
1 0.000 0.000 1.142 1.142
43 0.013 0.000 1.124 0.026<module>)
388 0.008 0.000 1.062 0.003
158 0.005 0.000 1.048 0.007
1 0.001 0.001 1.042 1.042
153 0.001 0.000 1.036 0.007
106/102 0.001 0.000 1.030 0.010
1 0.000 0.000 1.029 1.029
1 0.000 0.000 1.021 1.021
2 0.002 0.001 1.011 0.505<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

Project details

Download files

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

Files for pyinstrument, version 0.12
Filename, size File type Python version Upload date Hashes
Filename, size pyinstrument-0.12.tar.gz (45.1 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page