Support cleanup between tests with multiple databases

[ftobia]

pytest-django doesn't clean up between tests when using Django with multiple databases. The problem is related to this StackOverflow question: http://stackoverflow.com/questions/10121485/django-testcase-not-using-transactions-on-secondary-database

The db fixture uses Django's TestCase under the covers.

Here is the code from the original db fixture:

case = TestCase(methodName='__init__')
case._pre_setup()
request.addfinalizer(case._post_teardown)
request.addfinalizer(_django_cursor_wrapper.disable)

Here is how I patched it as a work-around in my code:

case = TestCase(methodName='__init__')
case.multi_db = True
case._pre_setup()
request.addfinalizer(case._post_teardown)
request.addfinalizer(_django_cursor_wrapper.disable)

Obviously that multi_db flag exists and defaults to False for a reason. I'm not sure the right way to incorporate multi_db support into pytest_django, and I'm not sure the right way to test such a change.

If you have a suggestion I can work on a pull request.

[flub]

Could you give an example test using your patched version which uses two databases? I think that would help understand how multiple databases are used and what you need control over other then setting multi_db to true.

[ftobia]

DATABASES = {
    'default': {
        'ENGINE'   : 'django.db.backends.sqlite3',
        'NAME'     : 'default.db',
    },
    'operations': {
        'ENGINE'   : 'django.db.backends.sqlite3',
        'NAME'     : 'operations.db',
    },
}

Most / all of my models are associated with the second database "operations". I believe Django's TestCase will flush transactions correctly on the default db, but not others unless multi_db=True.

Here are some sample tests:

from collection.models import Foo

def test_one(db):
    Foo.objects.create()
    assert Foo.objects.count() == 1

def test_two(db):
    assert Foo.objects.count() == 0

Here is the output with my patched db fixture, described above:

$ py.test multi_db_test/ -vv
======================================= test session starts =======================================
platform linux2 -- Python 2.7.2 -- py-1.4.20 -- pytest-2.5.2 -- bin/python
plugins: django
collected 2 items

multi_db_test/test_foo.py:3: test_one PASSED
multi_db_test/test_foo.py:7: test_two PASSED

==================================== 2 passed in 5.75 seconds =====================================

Here is the output with pytest_django's own db fixture:

$ py.test multi_db_test/ -vv
======================================= test session starts =======================================
platform linux2 -- Python 2.7.2 -- py-1.4.20 -- pytest-2.5.2 -- bin/python
plugins: django
collected 2 items

multi_db_test/test_foo.py:3: test_one PASSED
multi_db_test/test_foo.py:7: test_two FAILED

============================================ FAILURES =============================================
____________________________________________ test_two _____________________________________________

db = None

    def test_two(db):
>       assert Foo.objects.count() == 0
E       assert 1 == 0
E        +  where 1 = <bound method Manager.count of <django.db.models.manager.Manager object at 0x3cd0890>>()
E        +    where <bound method Manager.count of <django.db.models.manager.Manager object at 0x3cd0890>> = <django.db.models.manager.Manager object at 0x3cd0890>.count
E        +      where <django.db.models.manager.Manager object at 0x3cd0890> = Foo.objects

multi_db_test/test_foo.py:8: AssertionError
=============================== 1 failed, 1 passed in 5.85 seconds ===============================

[flub]

So from a quick looks it seems like a multi_db fixture and pytest.mark.django_db(multi=True) could work as an API? Anyone else got any better ideas?

[ftobia]

That sounds reasonable. I certainly don't have other ideas, better or otherwise.

[slafs]

Hello. Is there any discussion about multi_db support besides this issue? The docs say that one should get in touch but whats the current status on this?

[pelme]

@slafs There has not been much discussions/requests for multi database support apart from this.

It would be pretty simple to pass on multi_db=True, but we also needs some kind of tests for this, so we also need a secondary database set up. If anyone is willing to work on this, it would be great :)

[robdennis]

I've been using the work-around @ftobia mentions in the opening post, but after upgrading to Django 1.8.4 from 1.7 it stopped working.

There was Django change to make the code handling setup/tearDown look for a multi_db class attribute instead of an instance attribute.

Changing @ftobia's work-around to the following worked for me:

case = TestCase(methodName='__init__')
case.__class__.multi_db = True
case._pre_setup()
request.addfinalizer(case._post_teardown)
request.addfinalizer(_django_cursor_wrapper.disable)

[adrianandreias]

+1 👍

This is stopping us from switching from Django test runner to pytest.

[eremeevdev]

Why no use django_case.multi_db = True here https://github.com/pytest-dev/pytest-django/blob/master/pytest_django/fixtures.py#L107 by default?

[knabben]

+1

[guilhermemaba]

+1

[pelme]

I think that having a marker on the django_db marker like would by I good API:

@pytest.mark.django_db(transactional=True, multi_db=True)
def test_a():
    pass

The implementation should be relatively similar to reset_sequences (PR #308) or serialized_rollback (PR #353).

The "tricky" part would be to improve pytest-django's internal test suite to contain multiple databases.

If someone wants to work in this I will for sure review it and help in getting it merged. Feel free to start working on a PR if this is something that would be interesting to you!

[blueyed]

Left some comments at #397 (comment), and created a PR, which would allow to change this in a generic way: #431.

[cheniel]

Another option to enable multi_db for those looking for a temporary solution. Put this in conftest.py

def pytest_sessionstart(session):
    from django.test import TransactionTestCase
    TransactionTestCase.multi_db = True