If you are working on a large Django project, you probably have lots of automated tests running as part of your CI/CD process. As long as tests run fast, everyone is happy. But as your application grows in complexity, your tests start to take more and more time to run and eventually become a real bottleneck. In this post, I will share some ideas that can help you optimize runtime of your test suite. I assume you are using pytest
, but recommendations described in this post should be easily applicable to other runners as well.
Use a simple password hashing algorithm in your tests
Django uses a strong password hashing algorithm (PBFDF2
at the time of writing this post) by default which is a sensible practice to ensure security of your applications. However, such level of security is (usually) not needed in tests. You can use a simple hashing algorithm to significantly reduce hashing operations time. Here is an example configuration block to put in your test settings:
[
"django.contrib.auth.hashers.MD5PasswordHasher",
]
This will tell Django
to use MD5
algorithm for password hashing operations.
Optimize migrations
Every time your test suite starts to run, Django will create an empty database and then run all migrations to create necessary schema. This can take long time, especially if there is a latency between your test runner and test database.
Create test database directly from models
One possible optimization is to create a test database directly from Django models without running any migrations. pytest-django
has a special flag –no-migrations for that. This approach will reduce db creation time if you have lots of migrations in your project. The downside is that you will lose data migrations as well, which some projects may need to rely on in their tests.
Squash existing migrations
Another solution is to squash your existing migrations. Django provides a management command called squashmigrations
. You probably should be squashing your migrations periodically not only to speed tests but also to keep your codebase clean and compact.
Use a pre-made dump to create your database
Sometimes squashing is not an easy option, e.g. when your migrations rely heavily on RunPython
and RunSQL
commands. In that case there is another solution which I applied to a project with immense amount of migrations. The idea is to create a cron
task that will periodically take the recent codebase from your stable branch, create a new database by applying all migrations and upload a resulting dump to a place where it can be downloaded later by your CI runner. In order for this method to work, you need to customize how your test database is being created. pytest-django
has django-db-setup fixture which can be customized to alter database creation process. You need to change it to use a pre-made dump instead of creating database from scratch. After creating the database, only newly added migrations are left to be applied which can save a lot of time.
Use in-memory database if possible
If your data layer is simple enough and does not rely on any database vendor-specific fields or indexes, using in-memory database engine in your tests can be the easiest way to speed things up. Django allows using in-memory version of sqlite
database during tests which will avoid any disk IO altogether and make working with database in your tests blazing fast.
Here is the example configuration for using in-memory database in your test settings:
DATABASES = {
'default': {
'ENGINE': 'django.db.backends.sqlite3',
'NAME': 'file::memory:',
}
}
Despite outstanding speed improvements, this approach is less safe than using the same db engine in your tests as in your prod environment. There is a little chance of difference in behavior between databases which can potentially lead to uncovered scenarios. For that reason it should probably be avoided for large projects with critical functionality.