- Use SQLAlchemy to join tables with one-to-one, one-to-many, and many-to-many relationships.
We already know that we can build our SQL tables such that they associate with one another via primary keys and foreign keys. We can also use SQLAlchemy to access data across different tables by establishing relationships in code, without having to write tons of code ourselves, following the idea of convention over configuration.
SQLAlchemy relationships make it easy to establish relationships between our models, without having to write a ton of SQL ourselves. Sounds great, right? Now that we have you totally hooked, let's take a look at how we use these SQLAlchemy relationships.
Before we begin, run pipenv install; pipenv shell
to generate and enter
your virtual environment. This will install sqlalchemy
, alembic
, faker
,
pytest
, and ipdb
.
SQLAlchemy ORM uses a ForeignKey
column to constrain and join data models,
as well as a relationship()
method that provides a property to one model
that can be used to access its related model. SQLAlchemy ORM also uses a
backref()
method to create the property in the related model. This makes the
syntax for accessing related models and creating join tables very simple.
Here is some code that you might use to create a relationship between an order and a customer:
# example
class Customer(Base):
__tablename__ = 'customers'
id = Column(Integer(), primary_key=True)
orders = relationship('Order', backref='customer')
class Order(Base):
__tablename__ = 'orders'
id = Column(Integer(), primary_key=True)
customer_id = Column(Integer(), ForeignKey('customers.id'))
We can see that the Customer
model contains an orders
class attribute
that is set by the relationship()
method. The backref()
method therein
sets the reverse many-to-one relationship that refers back from the
Order
model. Inclusion of a foreign key for a Customer
instance in the
Order
model is all that is needed to complete this relationship.
The backref
goes into the "one" of the "one-to-many" relationship, as the
"one" is a single model object and does not need any additional arguments to
tell SQLAlchemy that it is something more specific like a list or dictionary.
(If you don't think you'll need the many-to-one relationship, you can
leave out the backref
.)
One-to-one relationships aren't tremendously common, but they are simple to build in SQLAlchemy ORM if you need to!
Let's say that orders have a lot of important associated data, but only two or three columns' worth of regularly-queried data. In designing a database to return data quickly, it makes the most sense for us to separate the important attributes into a smaller table and create a one-to-one relationship with the beefier database for metadata.
# example
class Order(Base):
__tablename__ = 'orders'
id = Column(Integer(), primary_key=True)
class OrderMetadata(Base):
__tablename__ = 'orders_metadata'
id = Column(Integer(), primary_key=True)
order_id = Column(Integer(), ForeignKey("orders.id"))
order = relationship('Order',
backref=backref('order_metadata', uselist=False))
The syntax for a one-to-one relationship is identical to that of a one-to-many
relationship, with the exception that the backref()
method takes
uselist=False
as an optional second parameter. This means exactly what it
sounds like: the Order
model refers back to the OrderMetadata
model via a
property order_metadata
that is not a list. (If it were a list, it would be
many!)
In this lesson, we'll be building out a one-to-many relationship between two models: games and reviews. We'll set up our database so that a game has many reviews, and each review belongs to a specific game.
By writing a few migrations and making use of the appropriate SQLAlchemy methods, we'll be able to:
- Ask a game about its reviews.
- Ask a review about its game.
Here's what our Entity Relationship Diagram (ERD) looks like:
A game will have many reviews. Before we worry about the migration that will implement this in our reviews table, let's think about what that table will look like:
id | title | platform | platform | price |
---|---|---|---|---|
1 | Breath of the Wild | Switch | Action-adventure | 60 |
Our games table doesn't need any information about the reviews, so it makes sense to generate this table first: it doesn't have any dependencies on another table. This makes sense even thinking about our domain in the real world: a game can exist without any reviews.
The basic structure for your SQLAlchemy app is already configured. Navigate to
the lib/
directory and you should see the following directory
structure:
.
βββ alembic.ini
βββ debug.py
βββ migrations
βΒ Β βββ README
βΒ Β βββ env.py
βΒ Β βββ script.py.mako
βΒ Β βββ versions
βΒ Β βββ 0d06d41c7860_initialize_database.py
βββ models.py
βββ one_to_many.db
βββ seed.py
βββ testing
βββ conftest.py
βββ game_test.py
βββ review_test.py
alembic.ini
points to a SQLite database called one_to_many.db
. The models
should go into models.py
; it is already configured to create a Base
, and
env.py
is pointing to its metadata.
We made another change to env.py
to support the addition of foreign keys on
line 69:
connection=connection, target_metadata=target_metadata, render_as_batch=True,
SQLite presents a challenge to migration tools as it does not support ALTER
statements, which relational databases frequently rely upon.
To help solve this problem, Alembic provides the batch operations context. With batch operations, a table is named, and then a series of changes to that table are specified. When the batch is complete, a process begins where the existing table structure is copied from the database, a new version of this table is created with the desired changes, data is copied from the old table to the new table, and the old table is dropped and the new one renamed to the original name.
Just remember that alterations to tables require render_as_batch=True
in
env.py
from the start of your first migration.
Run alembic upgrade head
to create your database.
Navigate to lib/models.py
and build a basic model for the games
table:
# models.py
from sqlalchemy import ForeignKey, Column, Integer, String, MetaData
from sqlalchemy.orm import relationship, backref
from sqlalchemy.ext.declarative import declarative_base
convention = {
"fk": "fk_%(table_name)s_%(column_0_name)s_%(referred_table_name)s",
}
metadata = MetaData(naming_convention=convention)
Base = declarative_base(metadata=metadata)
class Game(Base):
__tablename__ = 'games'
id = Column(Integer(), primary_key=True)
title = Column(String())
genre = Column(String())
platform = Column(String())
price = Column(Integer())
Note that we've added a convention
and metadata
to our models. SQLite
requires names for changes to foreign keys (we'll talk about these shortly)
and several other fields in models. Our convention
provides a template for
naming these changes, and metadata
saves them to a sqlalchemy.MetaData
object. Passing this to our declarative_base
object allows Alembic to generate
these names automatically when we autogenerate migrations.
Run alembic revision --autogenerate -m'Create Game Model'
from inside of the
lib/
directory. You should see the following output:
INFO [alembic.runtime.migration] Context impl SQLiteImpl.
INFO [alembic.runtime.migration] Will assume non-transactional DDL.
INFO [alembic.autogenerate.compare] Detected added table 'games'
Generating ...python-p3-sqlalchemy-one-to-many/one-to-many/migrations/versions/62797912f786_create_model.py ... done
Run alembic upgrade head
to generate a database with your Game
model.
A review will belong to a specific game. What does that mean in terms of our database? Think back to what you learned about SQL and joining between multiple tables. How can we connect between a review and its associated game?
That's right, we need a foreign key! Since a review belongs to a specific game, we need some way of indicating on the review which specific game it belongs to.
Let's take a look at what our reviews
table will need to look like:
id | score | comment | id |
---|---|---|---|
1 | 10 | A classic! | 1 |
Ok! Now that we know what we need to create, let's head back to models.py
and write out a new model:
# models.py
class Review(Base):
__tablename__ = 'reviews'
id = Column(Integer(), primary_key=True)
score = Column(Integer())
comment = Column(String())
game_id = Column(Integer(), ForeignKey('games.id'))
Make sure to add the relationship to Game
as well:
# models.py
class Game(Base):
# tablename, columns
reviews = relationship('Review', backref=backref('game'))
Lastly, don't forget to set the __repr__()
for each of your models. You''ll
thank me later!
# models.py
class Game(Base):
# tablename, columns, relationship
def __repr__(self):
return f'Game(id={self.id}, ' + \
f'title={self.title}, ' + \
f'platform={self.platform})'
class Review(Base):
# tablename, columns, relationship
def __repr__(self):
return f'Review(id={self.id}, ' + \
f'score={self.score}, ' + \
f'game_id={self.game_id})'
Great! Now go ahead and run the same commands in your terminal to generate and run our migrations:
$ alembic revision --autogenerate -m'Add Review Model'
# => INFO [alembic.runtime.migration] Context impl SQLiteImpl.
# => INFO [alembic.runtime.migration] Will assume non-transactional DDL.
# => INFO [alembic.autogenerate.compare] Detected added table 'reviews'
# => Generating ...sqlalchemy-relationships/python-p3-sqlalchemy-one-to-many/lib/migrations/versions/1f2ce6b5977d_add_model.py ... done
$ alembic upgrade head
# => INFO [alembic.runtime.migration] Context impl SQLiteImpl.
# => INFO [alembic.runtime.migration] Will assume non-transactional DDL.
# => INFO [alembic.runtime.migration] Running upgrade 62797912f786 -> 1f2ce6b5977d, Add Review Model
There is also some code in the seed.py
file that we'll use to generate
some data for our two models. In the seed file, we first create a game instance,
then use the ID from that game instance to associate it with the corresponding
review. The data itself is generated automatically using faker
.
Run this to fill the database with games and reviews:
$ python seed.py
Let's enter the Python shell and access the first review. Run python debug.py
to set up a session and import the data models:
# Access the first review instance in the database
review = session.query(Review).first()
review
# => Review(id=1, score=7, game_id=1)
# Get the game_id foreign key for the review instance
review.game_id
# => 1
We know that this review has some relationship to data in the games
table. We
could even use the foreign key to access that data directly:
# Find a specific game instance using an ID
session.query(Game).filter_by(id=review.game_id).first()
# => Game(id=1, title=Jacob Floyd, platform=wii u)
But SQLAlchemy provides an even easier route to the related game:
review.game
# => Game(id=1, title=Jacob Floyd, platform=wii u)
Nice!
Now let's access the first game:
game = session.query(Game).first()
game
# => Game(id=1, title=Jacob Floyd, platform=wii u)
As expected, this is the same game that we found through the first review. We can search for reviews using this game's ID as a filter:
reviews = session.query(Review).filter_by(game_id=game.id)
[review for review in reviews]
# => [Review(id=1, score=7, id=1), Review(id=2, score=7, id=1), Review(id=3, score=8, id=1)]
But just as with finding the game for a review, we can access a game's reviews directly using its relationship:
game.reviews
# => [Review(id=1, score=7, game_id=1), Review(id=2, score=7, game_id=1), Review(id=3, score=8, game_id=1)]
In most belongs to relationships like we see here with reviews and games, we want to make sure that when the parent disappears, the child does as well. SQLAlchemy handles this logic with cascades.
A cascade is a behavior of a SQLAlchemy relationship that carries from parents
to children. All SQLAlchemy relationships have cascades. By default, a
relationship's cascade behavior is set to 'save-update, merge'
. This can be
changed to any combination of a set of behaviors:
save-update
: when an object is placed into a session withSession.add()
, all objects associated with it should also be added to that same session.- If a game is added to a session, all of its reviews will be as well.
merge
: if the session contains duplicate objects,merge
eliminates those duplicates.- If a game is merged to a session, its reviews that have already been added to the session will not be added again.
delete
: when a parent is deleted, its children are deleted as well.- If a game is deleted, its reviews will be deleted as well.
all
: a combination ofsave-update
,merge
, anddelete
.delete-orphan
: when a child is disassociated from its parent, it is deleted.- If a review is removed from
game.reviews
, the review will be deleted.
- If a review is removed from
As we begin configuring our cascade, let's ask ourselves a few questions:
- Do we need reviews to be updated when their parent games are updated?
- Do we need to avoid adding duplicate reviews to our session?
- Do we need to delete reviews when their parent games are deleted?
- Do we need to delete reviews when they are no longer associated with games (i.e. orphaned)?
We know that we want reviews to be associated with their parent games, so that's a "yes" to number 1. We also want to avoid adding duplicates to the session, so that's a "yes" to 2 as well.
Reviews are inherently linked to games, so we should delete them if the games no longer exist. That's a "yes" to number 3.
Since we are going to be using save-update
, merge
, and delete
, we can
start our cascade with all
.
reviews = relationship('Review', backref=backref('game'), cascade='all')
Lastly, for the same reason we included delete
, we do want to delete reviews
that are no longer associated with games. What good is a review of nothing?
Let's say "yes" to number 4 as well and finish our cascade:
reviews = relationship('Review', backref=backref('game'), cascade='all, delete-orphan')
Our database is now configured to delete reviews when their parent games are deleted and when they are removed from their parent game, or orphaned.
In this lesson, we explored the most common kind of relationship between two models: the one-to-many or "has-many"/"belongs-to" relationship. With a solid understanding of how to connect databases using primary and foreign keys, we can take advantage of some helpful SQLAlchemy methods that make it much easier to build comprehensive database schemas and integrate them into our Python applications.
Run pytest -x
to make sure all of the tests are passing. In the next lesson,
we'll explore many-to-many relationships in SQLAlchemy
from sqlalchemy import ForeignKey, Column, Integer, String, MetaData
from sqlalchemy.orm import relationship, backref
from sqlalchemy.ext.declarative import declarative_base
convention = {
"fk": "fk_%(table_name)s_%(column_0_name)s_%(referred_table_name)s",
}
metadata = MetaData(naming_convention=convention)
Base = declarative_base(metadata=metadata)
class Game(Base):
__tablename__ = 'games'
id = Column(Integer(), primary_key=True)
title = Column(String())
genre = Column(String())
platform = Column(String())
price = Column(Integer())
reviews = relationship('Review', backref=backref('game'), cascade='all, delete-orphan')
def __repr__(self):
return f'Game(id={self.id}, ' + \
f'title={self.title}, ' + \
f'platform={self.platform})'
class Review(Base):
__tablename__ = 'reviews'
id = Column(Integer(), primary_key=True)
score = Column(Integer())
comment = Column(String())
game_id = Column(Integer(), ForeignKey('games.id'))
def __repr__(self):
return f'Review(id={self.id}, ' + \
f'score={self.score}, ' + \
f'game_id={self.game_id})'