Reuse database state on CI. Run migrations on CI tests only for changes.
Migrations are slow, but you have to run it on CI for testing reasons, so avoid to run them when the database state was already tested.
Install the package with pip:
pip install django-migrations-ci
Add django_migrations_ci
to Django settings INSTALLED_APPS
.
INSTALLED_APPS = [
..., # other packages
"django_migrations_ci",
]
The command migrateci
execute all migrations and save dump files migrateci-*
.
If these files already exist on disk, they are used to prepare the database without running all migrations again.
This is how the "run test" CI job should work.
./manage.py migrateci
./manage.py test --keepdb
It works with pytest-django
too as a plugin:
pytest --migrateci --reuse-db
The recommended way to work with it is configuring default pytest addopts
with --migrateci --reuse-db
to run without recreating database. When you want to recreate, run pytest with --create-db
that has precedence over --reuse-db
.
./manage.py migrateci --parallel $(nproc)
./manage.py test --keepdb --parallel $(nproc)
pytest --migrateci --reuse-db --parallel $(nproc)
Also check database names for parallel tests.
File storage class. The django-storages package has many backends implemented.
Saving cache files to an external storage allow the lib to reuse partial migrations. When you write a new migration, it will try to get a cache without this last migration and load from it, running only the new migrations.
An S3 example,, but it works with any custom backend:
from storages.backends.s3boto3 import S3Boto3Storage
class MigrateCIStorage(S3Boto3Storage):
bucket_name = "mybucket-migrateci-cache"
region_name = "us-east-1"
object_parameters = {
"StorageClass": "REDUCED_REDUNDANCY",
}
File storage API has a location arg that all backend use in some way.
If no storage is defined, it defaults to ~/.migrateci
to make it easy to work local.
The pytest-django
package use custom test database names.
If you use it and don´t change their default fixtures, just use MIGRATECI_PYTEST=True
.
Before tests, Django execute all migrations in one database and clone it to be able to run parallel tests.
Use MIGRATECI_PARALLEL="auto"
to create one database per process or define the exact number of processes with MIGRATECI_PARALLEL=4
.
It supports how Django test and how pytest-xdist works.
This is how we decide which migration cache to use.
First, it'll try to find a cache with all migration files, but in some cases it's not possible, like when you just pushed a new migration.
For MIGRATECI_DEPTH=1
, it'll remove one migration a time for each Django app installed and check if some cached migration exists. It support the most common use case and it's reasonably fast.
Bigger values cause a cost operation, it'll remove N migrations a time and check if some cached migration exists. It's a combination of every Django app. E.g. for 10 apps, it'll take at most 10^N checks, with some hashing operations.
All below settings can be defined through command line args.
manage.py migrateci [-h] [-n PARALLEL] [--storage STORAGE_CLASS] [--location LOCATION]
[--pytest] [--depth DEPTH] [-v {0,1,2,3}]
options:
-h, --help show this help message and exit
-n PARALLEL, --parallel PARALLEL
--storage STORAGE_CLASS
--location LOCATION
--pytest
--depth DEPTH
--checksum Prints the current checksum and exits.
-v {0,1,2,3}
As a stretch of this package, it's possible to use the same strategy during local
development. It'll by default cache files at ~/.migrateci
.
./manage.py migrateci --parallel $(nproc)
./manage.py test --keepdb --parallel $(nproc)
Django migrations are slow because of state recreation for every migration and other internal Django magic.
In the past, I tried to optimize that on Django core, but learnt it's a running issue.
- mysql
- postgresql
- sqlite3
Django default run sqlite3 tests as in memory database and does not work because
migrateci
runs in a different process. Add a test database name to settings,
like sqlite test settings.
Django supports oracle, but the dump function is not implemented here.
Django test framework has a --parallel N
flag to test with N parallel processes,
naming databases from 1 to N.
- On sqlite3, a
db.sqlite3
generatedb_N.sqlite3
files. - On other databases, a
db
generatetest_db_N
.
Pytest pytest-django
use pytest-xdist
for parallel support, naming databases
from 0 to N-1.
- On sqlite3, a
db.sqlite3
generatedb.sqlite3_gwN
files. - On other databases, a
db
generatetest_db_gwN
.