django-baseshift-brancher

Run Django and pytest against a PostgreSQL database that is already migrated. Tests do not CREATE DATABASE and do not run migrations again.

This package applies migrations. Each new snapshot runs BASESHIFT_INIT_SQL before migrate when a migration needs an extension. Seed data and project management commands run on the clones afterwards.

This package does not include the baseshift program that creates those databases. Install that binary separately and put it on PATH. django-baseshift-brancher clone asks that program for one snapshot per git commit that changes migrations, reuses snapshots this checkout shares with other branches, and starts one Postgres per test worker. That program is not the PyPI project named baseshift.

pip install django-baseshift-brancher
# if pytest-django is not already installed:
pip install 'django-baseshift-brancher[pytest]'

Clone the current checkout

From the Django project directory:

export DJANGO_SETTINGS_MODULE=yourproject.settings
unset BASESHIFT_BRANCHES_FILE DATABASE_URL PGHOST PGPORT

django-baseshift-brancher clone --count 4

clone looks at git. Each commit that adds, removes, or replaces migration files is a snapshot. Adding files continues the previous snapshot, so two branches that diverged from the same migration commit share it. Replacing or removing a migration file starts again from the empty snapshot, because that schema cannot be applied on top of the previous one. A rebase writes a new commit, so that checkout gets a new snapshot on the new parent. A branch that left before the rebase keeps its own chain. You do not name snapshots or parents.

The command prints clone JSON and writes it to .brancher/clones.json. stderr prints the BASESHIFT_BRANCHES_FILE export. --count is the number of databases (one per pytest worker).

export BASESHIFT_BRANCHES_FILE="$PWD/.brancher/clones.json"
pytest -n 4
baseshift stop --all

clones.json is not picked up from the working directory. The BASESHIFT_BRANCHES_FILE export is what the tests read.

pytest needs no settings.py change. The plugin loads itself, points each worker at one clone, and forces --reuse-db and --no-migrations. Do not pass --create-db.

Django's runner needs one flag:

python manage.py test --testrunner django_baseshift_brancher.DiscoverRunner --parallel 4

--parallel 4 and pytest -n 4 both need four clones. pytest worker gw0 uses the first clone. Django's worker suffix is 1-based: "1" uses the first clone.

Existing DATABASES USER and PASSWORD are kept. HOST, PORT, and NAME are rewritten onto the clone, and migrations during tests are turned off.

export BASESHIFT_INIT_SQL="CREATE EXTENSION IF NOT EXISTS pg_trgm"   # when a migration needs it

That SQL runs on each new snapshot before migrate. After the clones are up, run project commands yourself — for example ensure_celery_broker_schema — against those DSNs.