Choose test database?

I had a similar issue. But I wanted Django to just bypass the creation of a test database for one of my instances (it is not a mirror tough). Following Mark's suggestion, I created a custom test runner, as follows

from django.test.simple import DjangoTestSuiteRunnerclass ByPassableDBDjangoTestSuiteRunner(DjangoTestSuiteRunner):    def setup_databases(self, **kwargs):        from django.db import connections        old_names = []        mirrors = []        for alias in connections:            connection = connections[alias]            # If the database is a test mirror, redirect its connection            # instead of creating a test database.            if connection.settings_dict['TEST_MIRROR']:                mirrors.append((alias, connection))                mirror_alias = connection.settings_dict['TEST_MIRROR']                connections._connections[alias] = connections[mirror_alias]            elif connection.settings_dict.get('BYPASS_CREATION','no') == 'no':                old_names.append((connection, connection.settings_dict['NAME']))                connection.creation.create_test_db(self.verbosity, autoclobber=not self.interactive)        return old_names, mirrors

Then I created an extra dict entry in one of my databases entries inside settings.py, 'BYPASS_CREATION':'yes',

Finally, I configured a new TestRunner with

TEST_RUNNER = 'auth.data.runner.ByPassableDBDjangoTestSuiteRunner'

I would suggest using sqlite3 for testing purposes while keeping on using mysql/postgres/etc for production.

This can be achieved by placing this in your settings file:

if 'test' in sys.argv:    DATABASES['default'] = {'ENGINE': 'django.db.backends.sqlite3'}

see Running django tests with sqlite

a temporary sqlite database file will be created in your django project home which you will have write access to. The other advantage is that sqlite3 is much faster for testing. You may however run in to problems if you are using any mysql/postgres specific raw sql (which you should try to avoid anyway).

I think a better solution might be to define your own test runner.