Installation

pip install pyuwsgimemhog

Requirements

In order to have the needed information you have to run uWSGI with the memory-report enabled.

Please note that it’s possible to have meaningful results only if you are using one thread per process.

Use from the command line

You need to pass a single uWSGI log file to pyuwsgimemhog:

pyuwsgimemhog --logfile /path/to/log
/api 975 200 4.9
/another-api 502 2 251

That means that /api contributed to increase the memory usage by 975 MB, it has been accounted 200 times and it contributed 4.9 MB per call. /another-api contributed 502 MB in two occurences so 251MB per call.

Use as a library

In addition to using pyuwsgimemhog as a command line utility, it can also be used as a library. This allows passing a custom path normalization function into the log analyzer in order to group requests based on that URL path.

A trivial example of using the Django URL resolver in order to normalize URLs would be the following:

from django.urls import resolve, Resolver404
from pyuwsgimemhog.pyuwsgimemhog import normalize_path, uwsgimemhog
from urllib.parse import urlparse

def normalize(url):
    try:
        return resolve(urlparse(url).path).view_name
    except Resolver404:
        # This view was not handled by Django so fall back to the default
        # normalization. Use normalize_path_with_nums to normalize numbers
        # in path to 0.
        return normalize_path(url)

with open(logfile, 'r') as f:
    for view, memory, count in uwsgimemhog(
            f, threshold * 1_000_000, normalize):
        print('{} {} {} {:.1f}'.format(
            view, memory // 1_000_000, count, memory / count / 1_000_000
        ))

License

MIT license

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.

0.3.0 (2021-04-24)

• Make it possible to pass a custom normalization function (Terence D. Honles)

• Fix regular expression to require RSS (Terence D. Honles)

• Use Github actions instead of Travis (Terence D. Honles)

0.2.0 (2020-01-10)

• Count and show occurrences and per call leak count.

0.1.0 (2019-04-12)

• First release on PyPI.