[MINOR] add opt-in multi-database support for db fixtures

[averagechris]

Adds opt-in multi-database support for database fixtures (db, transactional_db) to cleanup in databases other than default.

This approach uses a setting in settings.DATABASES["db_name"]["TEST"]["PYTEST_DJANGO_ALLOW_TRANSACTIONS"] to determine whether the database fixtures should create a transaction, or cleanup that database after each test. It does this by extending the collection of database names on the relevant django TestCase's databases attribute or by setting multi_db=True for compatibility with django version <=1.8.x.

TODO:

• update documentation
• ask maintainers if they'd like these tests to run against other databases and add the DATABASES config if so
• I found Expose fixtures to change Django's {Transaction,}TestCase #431 and will be adjusting the approach based on the feedback in that PR

[bluetech]

Hi @mistahchris,

I don't have much experience with non-replica mutli-db but I'll try to find time to research the issues and look into your PR. Any more details/considerations you can add would be useful.

In the meantime, I noticed you have some code for django versions <= 1.8.X. Current pytest-django only supports Django>=2.2 so you don't need to worry about <=1.8.

[averagechris]

Hi @bluetech,

Thanks for taking the time to look into this PR! Let me know if I can help answer any questions. I've had to work with multiple databases quite a bit for work, which prompted me to make this PR in the first place.

I'll ditch the django 1.8.x stuff and fix the failing check on Monday or Tuesday this week (in next couple of days).

[aijorgenson]

This PR may also indirectly resolve #297

[averagechris]

Sorry about ghosting on this PR. My computer died and I just got my new one a few days ago. I'm working on setting up my python environment today so I can make the tweaks to this PR this weekend.

[jgb]

@mistahchris any update on this PR? Would love to try it out.

[bluetech]

So I'm going over all of the multi-db issues and PRs, trying to understand the material.

The approach you've taken here is to have a global setting, which affects all of the tests. I think this is too coarse -- one should be able to configure the databases on a per-test way, for efficiency and control. This is what the underlying Django TestCase provides. This means it should be configured on the mark/fixture.

I'll close this PR then, but I'll open an issue discussing multi-db support and will cc this.