Skip to main content

Keep detailed records of the performance of your Django code.

Project description

https://img.shields.io/github/actions/workflow/status/adamchainz/django-perf-rec/main.yml?branch=main&style=for-the-badge https://img.shields.io/pypi/v/django-perf-rec.svg?style=for-the-badge https://img.shields.io/badge/code%20style-black-000000.svg?style=for-the-badge pre-commit

“Keep detailed records of the performance of your Django code.”

django-perf-rec is like Django’s assertNumQueries on steroids. It lets you track the individual queries and cache operations that occur in your code. Use it in your tests like so:

def test_home(self):
    with django_perf_rec.record():
        self.client.get("/")

It then stores a YAML file alongside the test file that tracks the queries and operations, looking something like:

MyTests.test_home:
- cache|get: home_data.user_id.#
- db: 'SELECT ... FROM myapp_table WHERE (myapp_table.id = #)'
- db: 'SELECT ... FROM myapp_table WHERE (myapp_table.id = #)'

When the test is run again, the new record will be compared with the one in the YAML file. If they are different, an assertion failure will be raised, failing the test. Magic!

The queries and keys are ‘fingerprinted’, replacing information that seems variable with # and . This is done to avoid spurious failures when e.g. primary keys are different, random data is used, new columns are added to tables, etc.

If you check the YAML file in along with your tests, you’ll have unbreakable performance with much better information about any regressions compared to assertNumQueries. If you are fine with the changes from a failing test, just remove the file and rerun the test to regenerate it.

For more information, see our introductory blog post that says a little more about why we made it.

Installation

Use pip:

python -m pip install django-perf-rec

Requirements

Python 3.8 to 3.12 supported.

Django 3.2 to 5.0 supported.


Are your tests slow? Check out my book Speed Up Your Django Tests which covers loads of ways to write faster, more accurate tests.


API

record(record_name: str | None=None, path: str | None=None, capture_traceback: callable[[Operation], bool] | None=None, capture_operation: callable[[Operation], bool] | None=None)

Return a context manager that will be used for a single performance test.

The arguments must be passed as keyword arguments.

path is the path to a directory or file in which to store the record. If it ends with '/', or is left as None, the filename will be automatically determined by looking at the filename the calling code is in and replacing the .py[c] extension with .perf.yml. If it points to a directory that doesn’t exist, that directory will be created.

record_name is the name of the record inside the performance file to use. If left as None, the code assumes you are inside a Django TestCase and uses magic stack inspection to find that test case, and uses a name based upon the test case name + the test method name + an optional counter if you invoke record() multiple times inside the same test method.

Whilst open, the context manager tracks all DB queries on all connections, and all cache operations on all defined caches. It names the connection/cache in the tracked operation it uses, except from for the default one.

When the context manager exits, it will use the list of operations it has gathered. If the relevant file specified using path doesn’t exist, or doesn’t contain data for the specific record_name, it will be created and saved and the test will pass with no assertions. However if the record does exist inside the file, the collected record will be compared with the original one, and if different, an AssertionError will be raised. When running on pytest, this will use its fancy assertion rewriting; in other test runners/uses the full diff will be attached to the message.

Example:

import django_perf_rec

from app.models import Author


class AuthorPerformanceTests(TestCase):
    def test_special_method(self):
        with django_perf_rec.record():
            list(Author.objects.special_method())

capture_traceback, if not None, should be a function that takes one argument, the given DB or cache operation, and returns a bool indicating if a traceback should be captured for the operation (by default, they are not). Capturing tracebacks allows fine-grained debugging of code paths causing the operations. Be aware that records differing only by the presence of tracebacks will not match and cause an AssertionError to be raised, so it’s not normally suitable to permanently record the tracebacks.

For example, if you wanted to know what code paths query the table my_table, you could use a capture_traceback function like so:

def debug_sql_query(operation):
    return "my_tables" in operation.query


def test_special_method(self):
    with django_perf_rec.record(capture_traceback=debug_sql_query):
        list(Author.objects.special_method())

The performance record here would include a standard Python traceback attached to each SQL query containing “my_table”.

capture_operation, if not None, should be a function that takes one argument, the given DB or cache operation, and returns a bool indicating if the operation should be recorded at all (by default, all operations are recorded). Not capturing some operations allows for hiding some code paths to be ignored in your tests, such as for ignoring database queries that would be replaced by an external service in production.

For example, if you knew that in testing all queries to some table would be replaced in production with something else you could use a capture_operation function like so:

def hide_my_tables(operation):
    return "my_tables" in operation.query


def test_special_function(self):
    with django_perf_rec.record(capture_operation=hide_my_tables):
        list(Author.objects.all())

TestCaseMixin

A mixin class to be added to your custom TestCase subclass so you can use django-perf-rec across your codebase without needing to import it in each individual test file. It adds one method, record_performance(), whose signature is the same as record() above.

Example:

# yplan/test.py
from django.test import TestCase as OrigTestCase
from django_perf_rec import TestCaseMixin


class TestCase(TestCaseMixin, OrigTestCase):
    pass


# app/tests/models/test_author.py
from app.models import Author
from yplan.test import TestCase


class AuthorPerformanceTests(TestCase):
    def test_special_method(self):
        with self.record_performance():
            list(Author.objects.special_method())

get_perf_path(file_path)

Encapsulates the logic used in record() to form path from the path of the file containing the currently running test, mostly swapping ‘.py’ or ‘.pyc’ for ‘.perf.yml’. You might want to use this when calling record() from somewhere other than inside a test (which causes the automatic inspection to fail), to match the same filename.

get_record_name(test_name, class_name=None)

Encapsulates the logic used in record() to form a record_name from details of the currently running test. You might want to use this when calling record() from somewhere other than inside a test (which causes the automatic inspection to fail), to match the same record_name.

Settings

Behaviour can be customized with a dictionary called PERF_REC in your Django settings, for example:

PERF_REC = {
    "MODE": "once",
}

The possible keys to this dictionary are explained below.

HIDE_COLUMNS

The HIDE_COLUMNS setting may be used to change the way django-perf-rec simplifies SQL in the recording files it makes. It takes a boolean:

  • True (default) causes column lists in queries to be collapsed, e.g. SELECT a, b, c FROM t becomes SELECT ... FROM t. This is useful because selected columns often don’t affect query time in typical Django applications, it makes the records easier to read, and they then don’t need updating every time model fields are changed.

  • False stops the collapsing behaviour, causing all the columns to be output in the files.

MODE

The MODE setting may be used to change the way django-perf-rec behaves when a performance record does not exist during a test run.

  • 'once' (default) creates missing records silently.

  • 'none' raises AssertionError when a record does not exist. You probably want to use this mode in CI, to ensure new tests fail if their corresponding performance records were not committed.

  • 'all' creates missing records and then raises AssertionError.

  • 'overwrite' creates or updates records silently.

Usage in Pytest

If you’re using Pytest, you might want to call record() from within a Pytest fixture and have it automatically apply to all your tests. We have an example of this, see the file test_pytest_fixture_usage.py in the test suite.

Download files

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

Source Distribution

django_perf_rec-4.25.0.tar.gz (21.7 kB view hashes)

Uploaded Source

Built Distribution

django_perf_rec-4.25.0-py3-none-any.whl (17.5 kB view hashes)

Uploaded Python 3

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