Django: Database used for prefetch_related is not the same that the parent query

Question

I'm using a multi-db in my Django app with a master database and a read replica, but to avoid replication lag problems the router always uses default database except few places where I set DB manually.

I'm facing an issue as I don't know how to specify the database used for prefetch_related.

For example I want the next query to use only read_replica DB, but it does 2 queries, the first goes to read_replica as expected and the second goes to default DB.

users = UserProfile.objects.using('read_replica').prefetch_related('usermedia_set').filter(id__in=user_ids)

This are the output of this query:

SELECT @@SQL_AUTO_IS_NULL; args=None

SELECT VERSION(); args=None

SELECT ... FROM ftmanager_userprofile WHERE ( ftmanager_userprofile . id IN (33); args=(33,)

SELECT @@SQL_AUTO_IS_NULL; args=None

SELECT VERSION(); args=None

SELECT ... FROM ftmanager_usermedia WHERE ftmanager_usermedia . user_id IN (33); args=(33,)

I see a related ticket on Django tickets but I do not understand how to apply using() to inner queryset.

Answer 1

I have found the solution thinking about the Django ticket, you need to use the Prefetch class to define the inner queryset used by prefetch_related, it really messes the code but the gain deserves it:

from django.db.models.query import Prefetch
users = UserProfile.objects.using('read_replica').prefetch_related(Prefetch('usermedia_set', queryset=UserMedia.objects.using('read_replica'))).filter(id__in=user_ids)

Answer 2

I've encountered the same problem. Configuring the router (namely, db_for_read and using instance from hints) helped:

def db_for_read(self, model, **hints):
    instance = hints.get('instance')
    if instance is not None:
        if instance._meta.app_label in self.route_app_labels:
            return self.route_db_name 
    elif model._meta.app_label in self.route_app_labels:
        return self.route_db_name