Decoupling the ORM class from the data model class
I'm working on a rather large project where we would want to replace a WSGI ORM with an ASGI ORM, but it's tangled up into everything and ORM queries are executed from all over the business logic. If the ORM would've been decoupled from the objects tossed around in the business logic, it would've been much easier to replace the ORM.
This blog post outlines an example of how this can be done with Pydantic. I'm also including a "bonus section" on decoupling the data store communication from the business logic with some inspiration of the "Repository pattern".
This post was written with Python 3.11 in mind. You'll need to install SQLAlchemy and Pydantic:
Defining the ORM models
Let's start with defining ORM models and related functions in
from sqlalchemy import Column, Engine, Integer, String, create_engine from sqlalchemy.orm import declarative_base Base = declarative_base() class UserOrm(Base): __tablename__ = "users" id = Column(Integer, primary_key=True, nullable=False) name = Column(String(50), nullable=False) password = Column(String(50), nullable=False) email = Column(String(50), nullable=False, unique=True)
Defining the business logic entity models
Let's now implement the internal "entity" model we'll use when passing around a user object in our business logic, in
entities.py. Even if we replace the ORM or the database, this object likely won't change as it should still carry the same attributes and should not require refactorings to our business logic.
What's nice about using Pydantic models for such entities is we'll get the awesome validation Pydantic is known for, for "free".
Please note the
from_attributes configuration. This is where the magic happens, as this enables creating the entity from the ORM object:
Note how the ORM model contains a
password attribute, mapping to the database column of the same name. But in our business logic, I don't want this to be available. Therefore my entity model doesn't have it.
Defining the repositories
With the "Repository pattern", you want to define a tight responsibility scope for the code which communicates with an external data store, such as a database. This code would be known as a "repository" and lives outside of your (likely domain-driven) business logic. The idea is to have your business logic call into this repository whenever it needs to interact with the external data source and allow the repository to be switched for another repository.
For example, you might want to use SQLAlchemy with Postgres in prod, but for tests maybe you want to use SQLAlchemy with an in-memory SQLite database for faster execution and less setup. Or maybe you want your app to gradually move over onto a different database, database driver, ORM or similar but without refactoring your business logic.
Let's define a couple of classes in
repositories.py. First off, we define the abstract class
UserRepositoryAbc that explains which required methods all user repositories must include. In this case it's the
get_all_users methods. Then we implement the
UserRepository class, which implements logic on how to communicate with our SQLite database using SQLAlchemy.
I'm just going to call this repository
UserRepository for now, but imagine we could've had
UserFakeRepository, all inheriting from
UserRepositoryAbc. That last one,
UserFakeRepository, could be used in tests and not even communicate with a real database.
import abc from typing import Self from sqlalchemy.orm import Session from entities import User from orm import Base, UserOrm class UserRepositoryAbc(abc.ABC): @abc.abstractmethod def create_user(self, name: str, email: str, hashed_password: str) -> User: raise NotImplementedError @abc.abstractmethod def get_all_users(self) -> list[User]: raise NotImplementedError class UserRepository(UserRepositoryAbc): @property def engine(self: Self) -> Engine: return create_engine("sqlite:///mydatabase.db", echo=True) def create_tables(self: Self, base) -> None: base.metadata.create_all(self.engine) def create_user(self: Self, name: str, email: str, hashed_password: str) -> User: with Session(self.engine) as session: user_orm = UserOrm(name=name, email=email, password=hashed_password) session.add(user_orm) session.commit() user = User.model_validate(user_orm) return user def get_all_users(self: Self) -> list[User]: with Session(self.engine) as session: users_orm: list[UserOrm] = session.query(UserOrm).all() users: list[User] = [ User.model_validate(user) for user in users_orm ] return users
A note on the table creation and inclusion of engine
As you can see, I also added a methods
engine. These doesn't really belong on a users repository, and you might want to implement this on some general SQLAlchemy repository class or abstract the choice of database away from the ORM. But to avoid making this blog post too long and complicated, I just slapped them on there.
Note that all ORM models inherting from
orm.py will have all their respective tables created when executing
Limit ORM queries to the repository
The repository methods can take business logic entities (such as
User) as input, or it can take strings, integers, booleans etc - or no arguments at all. It is likely desirable that it returns business logic entities but that is no strict rule about this. Just have them return what makes the most sense. Just don't return the ORM objects!
The idea is to limit all occurrences of ORM queries to the repositories and not implementing them in the business logic.
Let's run some commands!
Create the db tables
Let's begin by creating the db tables:
>>> from orm import Base >>> from repositories import UserRepository >>> UserRepository().create_tables(base=Base) 2023-07-02 15:53:47,602 INFO sqlalchemy.engine.Engine BEGIN (implicit) 2023-07-02 15:53:47,602 INFO sqlalchemy.engine.Engine PRAGMA main.table_info("users") 2023-07-02 15:53:47,602 INFO sqlalchemy.engine.Engine [raw sql] () 2023-07-02 15:53:47,603 INFO sqlalchemy.engine.Engine PRAGMA temp.table_info("users") 2023-07-02 15:53:47,603 INFO sqlalchemy.engine.Engine [raw sql] () 2023-07-02 15:53:47,603 INFO sqlalchemy.engine.Engine CREATE TABLE users ( id INTEGER NOT NULL, name VARCHAR(50) NOT NULL, password VARCHAR(50) NOT NULL, email VARCHAR(50) NOT NULL, PRIMARY KEY (id), UNIQUE (email) ) 2023-07-02 15:53:47,603 INFO sqlalchemy.engine.Engine [no key 0.00006s] () 2023-07-02 15:53:47,604 INFO sqlalchemy.engine.Engine COMMIT
Communicate with the db
Finally, we can now communicate with our database using the ORM but always return our entity objects rather than returning ORM objects directly. This is what we'd also want our business logic to do, rather than manage the ORM directly.
>>> from repositories import UserRepository >>> user = UserRepository().create_user(name="John Doe", email="firstname.lastname@example.org", hashed_password="hashed_password") 2023-07-02 15:59:16,700 INFO sqlalchemy.engine.Engine BEGIN (implicit) 2023-07-02 15:59:16,702 INFO sqlalchemy.engine.Engine INSERT INTO users (name, password, email) VALUES (?, ?, ?) 2023-07-02 15:59:16,702 INFO sqlalchemy.engine.Engine [generated in 0.00024s] ('John Doe', 'hashed_password', 'email@example.com') 2023-07-02 15:59:16,703 INFO sqlalchemy.engine.Engine COMMIT 2023-07-02 15:59:16,706 INFO sqlalchemy.engine.Engine BEGIN (implicit) 2023-07-02 15:59:16,708 INFO sqlalchemy.engine.Engine SELECT users.id AS users_id, users.name AS users_name, users.password AS users_password, u sers.email AS users_email FROM users WHERE users.id = ? 2023-07-02 15:59:16,708 INFO sqlalchemy.engine.Engine [generated in 0.00015s] (1,) 2023-07-02 15:59:16,709 INFO sqlalchemy.engine.Engine ROLLBACK >>> print(user) id=1 name='John Doe' firstname.lastname@example.org'
>>> from repositories import UserRepository >>> users = UserRepository().get_all_users() users = UserRepository().get_all_users() 2023-07-02 21:11:27,858 INFO sqlalchemy.engine.Engine BEGIN (implicit) 2023-07-02 21:11:27,859 INFO sqlalchemy.engine.Engine SELECT users.id AS users_id, users.name AS users_name, users.password AS users_password , users.email AS users_email FROM users 2023-07-02 21:11:27,859 INFO sqlalchemy.engine.Engine [generated in 0.00013s] () 2023-07-02 21:11:27,860 INFO sqlalchemy.engine.Engine ROLLBACK >>> print(users) [User(id=1, name='John Doe', email@example.com')]
A final note on switching out the repositories
In the above commands, I called the repository directly, just to show what the output would be like, and how it returns the
User object rather than the ORM object.
But a more desirable pattern is to allow injection of the desired repository into the business logic. Imagine having business logic like below:
def create_user( name: str, email: str, hashed_password: str, repository: UserRepositoryAbc = UserRepository(), ) -> User: return repository.create_user( name=name, email=email, hashed_password=hashed_password, ) def get_all_users(repository: UserRepositoryAbc = UserRepository()) -> list[User]: return repository.get_all_users()
Here you can see how the functions default to using our
UserRepository, but but they can technically accept any other repository that abides by the abstract class of
The above code snippets exhibits "Dependency injection" by allowing the repository to be provided externally, which promotes loose coupling and flexibility. It also aligns with the "Dependency inversion principle", where high-level modules (business logic) should not depend on low-level modules (repositories) directly but should instead depend on abstractions.
To instantiate or not instantiate
You might have noticed how I've instantiated the classes in the signatures. This is yet another example of just trying to avoid this blog post from becoming ginormous. Instantiating like this is not desirable but depending on what your needs are like, there are several solutions to this.
For example, you might want to consider using static methods in your repository classes instead, so you don't have to instantiate:
Or you can use other mechanisms to determine which repository to be used and instantiate the repositories only once.
Caching of instance methods
If you want to implement caching of instance methods, you need to be careful as Python is known for leaking memory when using the
@functools.cache. Especially for long-running processes. It all boils down to that the garbage collection doesn't happen.
This is a rabbit hole of itself and I won't dive into it in this blog post 😄.
I would like to end on a note about over-engineering and the tradeoffs of "clean code".
Depending on what you're building, the solutions outlined in this blog post might resonate with you or it might make no sense to you what so ever. Perhaps you prefer functional programming and are put off by the object oriented approach here. Perhaps you see the abstraction overhead and loss of performance as a problem.
Regardless, having designed the particular project I'm dealing with this way would've substantially helped replacing the ORM with another one. So I figured I'd share this here. Would be super happy for any input or comments on if you see any other ways of dealing with this problem!
Thanks for reading.