Skip to content

zsoldosp/django-performance-testing

Repository files navigation

Django Performance Testing

WARNING:This project is currently frozen and will not longer be maintained. If there is someone interested in continuing to maintain this project please contact: [email protected] .
https://travis-ci.org/PaesslerAG/django-performance-testing.svg?branch=master

Don't wait with performance testing until the end of the project! We have learned already that more frequent feedback on smaller chunks of changes is much better, e.g.: TDD, CI, DevOps, Agile, etc.

This library helps by providing performance testing from the start - integrating it seamlessly into your existing development cycle, without requiring changes to your development workflow.

Unlike regular performance testing tools (ab, tsung, etc.), this libary relies on indirect (proxy) indicators to performance - e.g.: the number of queries executed. It's a good rule of thumb that the more SQL there is, the slower it will be. And this way "performance" testing won't be slower than your normal tests! (Disclaimer: while this tool is useful, classic performance testing is still recommended!)

  • install it via pip install django-performance-testing

  • add it to your settings and it auto-registers itself

    settings.INSTALLED_APPS = [
       ...
       'django_performance_testing',
       ...
    ]
    
  • set your limits (see below for detail)
  • and run your test manage.py test <your app>

For any limit violations, there will be a test failure.

After the test run, you could generate the Worst Items Report by running manage.py djpt_worst_report.

The data is collected into settings.DJPT_DATAFILE_PATH file, or into djpt.results_collected.

Sets the limit in the number of queries executed inside the given scope. Limits can be set for the total number of queries, or more specifically, based on types of queries - read (SELECT), write ( INSERT, UPDATE, DELETE), and other (e.g.: transaction (savepoints)).

When no (or None) value is provided for a given limit type, that is ignored during the check, as if there were no limit rules for. Thus it's possible to only focus on no write queries, while ignoring all the other queries that might be executed.

Sets the limit on the total elapsed seconds.

Following are the keys that are currently supported for settings.PERFORMANCE_LIMITS dictionary

  • django.test.client.Client - every call to its request method is limited, i.e.: GET, POST, etc.
  • Template.render - every render call is checked for limits. Note: it's recursive, i.e.: include and similar tags result in a check
  • for testcase classes, there is
    • test method - the actual various unittest test methods that you write for your app
    • test setUp - the TestCase.setUp methods you write for your test classes
    • test tearDown - the TestCase.tearDown methods you write for your test classes
    • test setUpClass - the TestCase.setUpClass methods you write for your test classes
    • test tearDownClass - the TestCase.tearDownClass methods you write for your test classes

For each of the above keys, there is a dict that holds the actual limits. The keys are the limit types (queries and/or time), and the value is yet another dict, holding the actual limit values. For valid values, see the description of the limits above, or look at the sample settings

PERFORMANCE_LIMITS = {
    'test method': {
        'queries': {'total': 50},  # want to keep the tests focused
        'time': {'total': 0.2},  # want fast integrated tests, so aiming for 1/5 seconds
    },
    'django.test.client.Client': {
        'queries': {
            'read': 30,
            'write': 8,  # do not create complex object structures in the web
                         # process
        },
    },
    'Template.render': {
        'queries': {
            'write': 0,  # rendering a template should never write to the database!
            'read': 0
        }
    }
}

While the built-in measurement points are great, sometimes, when profiling and trying to improve sections of the code, more granular limits are needed.

All limits can be used as context managers, e.g.:

from django_performance_testing.queries import QueryBatchLimit
from django_performance_testing.timing import TimeLimit
...

def my_method_with_too_many_queries(request):
    with QueryBatchLimit(write=0, read=10):  # initialize form
        form = MyForm(request.POST)
    with QueryBatchLimit(write=0, read=3):  # validate it
        is_valid = form.is_valid()
    if is_valid:
        with QueryBatchLimit(read=0, write=8):  # save it
            form.save()
        with QueryBatchLimit(read=0, write=0):  # redirect
            return HttpResponseRedirect(...)
    else:
        with QueryBatchLimit(write=0):  # render form
            with TimeLimit(total=0.01):   # we need superfast templates
                return form_invalid(form)

There is a single template tag that can be used after {% load djpt_limits %}, namely djptlimit. It takes

  • a single string positional argument, the name of the limit - as per settings.DJPT_KNOWN_LIMITS_DOTTED_PATHS, see below
  • keyword arguments that will be passed to the actual limit.

It can be used directly in your tempaltes like

{% load djpt_limits %}
{% djptlimit 'TimeLimit' total=3 %}
{{ slow_rendering }}
{% enddjptlimit %}

When debugging more complext template hierarchies, where e.g.: the slow part might not even be our own template, then {{ block.super }} could be helpful

{% extends "base.html" %}
{% block title %}
{% djptlimit 'QueryBatchLimit' read=3 %}
{{ block.super }}
{% enddjptlimit %}
{% endblock %}

This is an array of full class paths, similar to how settings.MIDDLEWARE are defined, e.g.: ['django_performance_testing.timing.TimeLimit'].

The name of the limit is the classname part of the class.

Unless you have written a custom limit, this setting doesn't need to be set explicitly, as the app has defaults that include all limits.

  • 0.7.3 - conform to latest flake8
  • 0.7.1 - bugfix a test
  • 0.7.0 - separate data collection and reporting
    • introduce djpt_worst_report management command
    • backwards incompatibe changes:
      • Collectors are expected to have get_sample_results method to allow easier and more realistic testing
      • Worst Items Report is not printed anymore after the test run.
      • settings.DJPT_PRINT_WORST_REPORT doesn't have much effect anymore, will be dropped in a subsequent release
  • 0.6.1
    • add support for Django 1.11 (and thus for Python 3.6 too)
  • 0.6.0
    • django test runner integration now uses settings.DJPT_KNOWN_LIMITS_DOTTED_PATHS for the collectors/limits it initializes, thus allowing 3rd party collectors/limits
    • new predefined limit points: test setUp, test tearDown, test setUpClass, test tearDownClass
  • 0.5.0
    • backwards incompatible - remove --djpt-no-report and use settings.DJPT_PRINT_WORST_REPORT instead to suppress the printing of the report (to address incompatibilities with third party testrunner extensions)
  • 0.4.0
    • add --djpt-no-report argument to disable output of performance report on shell
  • 0.3.0
    • introduced django_performance_testing.core.limits_registry. This keeps track of all limits, and enforces that across the django project all limits have unique names. This also warranted the introduction of settings.DJPT_KNOWN_LIMITS_DOTTED_PATHS.
    • introduced djptlimit template tag to be used for ad-hoc template debugging
  • 0.2.0
    • add timing measurement that can be limited
    • remove uniqueness check for collector.id_, as the problems it caused for testing outweighed its benefit for developer debugging aid
    • backwards incompatible:
      • change how settings based limits are specified
      • change the worst report data output/data structure
  • 0.1.1 - bugfix release
    • bugfix: attributes set by on test methods (e.g.: @unittest.skip) are now recognizable again and not lost due to the library's patching
  • 0.1.0 - initial release
    • supports Django 1.8, 1.9, 1.10 on python 2.7, 3.3, 3.4, and 3.5
    • query counts are reported and can be limited, by categories - read, write, other, and total
    • support ad-hoc limits by using it as a context manager
    • predefined limits support:
      • django.test.client.Client - all calls to its request method
      • actual unittest test_<foo> methods
      • Template.render

As an open source project, we welcome contributions.

The code lives on github.

Please open an issue on github or provide a pull request whether for code or for the documentation.

For non-trivial changes, we kindly ask you to open an issue, as it might be rejected. However, if the diff of a pull request better illustrates the point, feel free to make it a pull request anyway.

  • for code changes
    • it must have tests covering the change. You might be asked to cover missing scenarios
    • the latest flake8 will be run and shouldn't produce any warning
    • if the change is significant enough, documentation has to be provided
sudo apt-get -y install software-properties-common
sudo add-apt-repository ppa:fkrull/deadsnakes
sudo apt-get update
for version in 3.3 3.5 3.6; do
  py=python$version
  sudo apt-get -y install ${py} ${py}-dev
done

As it is a Django extension, it follows Django's own Code of Conduct. As there is no mailing list yet, please just email one of the main authors (see setup.py file or github contributors)

About

Performance testing for Django through your automated tests!

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published