This is something I went through recently - I swapped from a mongo backend to postgres. When I set up the original project I had some models and a DataLayer. The datalayer (dl) had quite a simple interface that I used throughout my app.
# note: this is half python / half pseudocode
class Model1(object):
__collection__ = 'model1'
__tablename__ = 'model1'
# field definitions etc
class MongoDataLayer(object):
def __init__(self, mongo_db_connection):
self.conn = mongo_db_connection
def load(self, model, conditions):
raw = self.conn[model.__collection__].find(...)
return model(**raw)
def persist(self, obj):
self.conn[obj.__collection__].save(obj.as_dict())
class SQLDataLayer(object):
def __init__(self, sa_session_factory):
self.Session = sa_session_factory
self.session = self.Session()
def load(self, model, conditions):
return self.session.query(model).find_by(conditions).one() # ...etc
def persist(self, obj):
self.conn[obj.__collection__].save(obj)
# connections - mongo and postgres (I use SQLAlchemy)
dl_mongo = MongoDataLayer(db...)
dl_sql = SQLDataLayer(Session...)
# using them - you don't care which one you have
m = dl_mongo.load(models.Model1)
dl_mongo.persist(m)
m = dl_sql.load(models.Model1)
dl_sql.persist(m)
In my app I load up the dl in the initial load and then inject it into the app whenever data access needs to happen. The app itself then knows about models but not the details of how to load / save them.
Maybe not the best way to do it but it's worked well for me. Would be interested to hear how other people deal with it.