Question

I’m in the process of configuring my Django app for multiple databases, and as a stop-gap measure, I’m using the following configuration in my settings.py:

READ_DATABASE = {
    'ENGINE': 'django.db.backends.mysql',
    'NAME': 'thedatabase',
    'USER': 'read_only',
    'PASSWORD': '',
    'OPTIONS': {
        'init_command': 'SET storage_engine=INNODB'
    }
}

WRITE_DATABASE = {
    'ENGINE': 'django.db.backends.mysql',
    'NAME': 'thedatabase',
    'USER': 'root',
    'PASSWORD': '',
    'OPTIONS': {
        'init_command': 'SET storage_engine=INNODB'
    }
}

DATABASES = {'default': WRITE_DATABASE,
             'read': READ_DATABASE}

These point to the same database, in order to emulate a master-slave pair. The read_only user, as you might guess, only has read permissions.

My database router looks like this:

from django.conf import settings 

class MasterSlaveRouter(object):

    def db_for_read(self, model, **hints):
        return 'read' 

    def db_for_write(self, model, **hints):
        return 'default' 

    def allow_relation(self, db1, db2, **hints):
        return True

    def allow_syncdb(self, db, model):
        return db in ('default',)

When running syncdb, it now crashes immediately after setting up the superuser, with the error:

django.db.utils.IntegrityError: (1062, "Duplicate entry 'auth-permission' for key 'app_label'")

The offending SQL that’s failing to execute is:

INSERT INTO `django_content_type` (`name`, `app_label`, `model`) VALUES (permission, auth, permission)

This error only occurs when the second read database is sepcified in my settings.py, when only a default database is present, the syncdb command completes successfully.

Any suggestions what could be causing this?

Edit: Django version is 1.2.3

Answer 1

After a few hours of tinkering, I’ve deduced that the reason this was occurring is because some Django models (in this case django.contrib.contenttype.models.ContentType, but also django.contrib.site.models.Site) have their own inbuilt caching mechanism.

Since I have set up two different connections to the same database, this was causing a problem since the contenttype cache for the read-only connection was not being invalidated by a write to a ContentType model on the write connection.

The solution to this was to change my db_for_read method to something like this:

    def db_for_read(self, model, **hints):
        if model._meta.app_label in ('contenttypes', 'sites'):
            return 'write'
        else:
            return 'read'