Mixing PostgreSQL and MongoDB (as Django backends)

Since Django 1.2, you can define multiple datbase connections in your settings.py. Then you can use database routers to tell Django which database to go to, transparently for your application.

Disclaimer: this is how I think it should work, I have never used MongoDB in Django, nor have I tested that my code actually works. :)

settings.py

DATABASES = {    'default': {        'ENGINE': 'django_mongodb_engine',        'NAME': 'mydata',        ...    }    'geodata' {        'ENGINE': 'django.db.backends.postgresql_psycopg2',        'NAME': 'geodata',        ...    }}DATABASE_ROUTERS = ['path.to.ModelMetaRouter']

Models

Then add custom Meta variables to your geo-tables, to override their database. Don't add this attribute to models that are supposed to go to the default database.

class SomeGeoModel(models.Model):    ...    class Meta:        using = 'geodata'

Database router

And write a database router to direct all models that have the using meta attribute set, to the appropriate connection:

class ModelMetaRouter(object):    def db_for_read(self, model, **hints):        return getattr(model._meta, 'using', None)    def db_for_write(self, model, **hints):        return getattr(model._meta, 'using', None)    def allow_relation(self, obj1, obj2, **hints):        # only allow relations within a single database        if getattr(obj1._meta, 'using', None) == getattr(obj2._meta, 'using', None):            return True        return None    def allow_syncdb(self, db, model):        if db == getattr(model._meta, 'using', 'default'):            return True        return None

you can't have 'using' in the Meta list.

here is a working solution

add this to models.py:

 import django.db.models.options as options options.DEFAULT_NAMES = options.DEFAULT_NAMES + ('in_db',)

create a router.py in your apps folder:

myapp folder content:   models.py   router.py   ...

Content of router.py:

class ModelMetaRouter(object):    def db_for_read(self, model, **hints):        db = getattr(model._meta, 'in_db', None)   # use default database for models that dont have 'in_db'        if db:            return db        else:            return 'default'    def db_for_write(self, model, **hints):        db = getattr(model._meta, 'in_db', None)        if db:            return db        else:            return 'default'    def allow_relation(self, obj1, obj2, **hints):        # only allow relations within a single database        if getattr(obj1._meta, 'in_db', None) == getattr(obj2._meta, 'in_db', None):            return True        return None    def allow_syncdb(self, db, model):        if db == getattr(model._meta, 'in_db', 'default'):            return True        return None

Reference router in your settings:

   DATABASE_ROUTERS = ['myapp.router.ModelMetaRouter']

I would take a look at the Disqus talk from DjangoCan 2010 about their scaling architecture. They run quite possibly the largest Django website on top of Postgres. They present simple code snippets showing how to start both vertical and horizontal scaling using features built into Django.

My understanding is that they do use MongoDB for some of their analytics thought I don't think it's discussed in that talk.