/sa2django

Convert sqlalchemy database schemas to django database models at runtime.

Primary LanguagePythonMIT LicenseMIT

SqlAlchemy to Django Bridge

Convert sqlalchemy database schemas to django database models at runtime.

Original url: https://github.com/mariushelf/sa2django

The SQLAlchemy to Django Bridge allows to specify your data models in SQLAlchemy and use them in Django, without manually re-specifying all your models and fields for the Django ORM.

Why this package?

Specifying a schema in SQLAlchemy and then using it in Django sounds like... counter- intuitive. There are a lot of Why nots to answer...

Why not specify your models in Django straight away?

Sometimes some or all of the data you serve from Django is maintained or created by sources that are not part of the Django application. If those sources already specify a complete SQLAlchemy model, then the SQLAlchemy to Django bridge is useful.

Why not simply use inspectdb to dynamically generate the Django model specifications?

inspectdb is not that dynamic after all -- it generates a Python file once, which needs to be manually tweaked. And each time the data model changes, you need to adjust that Python file.

Also it is often not possible to automatically derive all relations between models from the database. With third-party datasources, often relations are not manifested as foreign key constraints in the database schema, but just in some documentation that explains the relations in human-, but not machine-readable form.

If the SQLAlchemy models already contain all those relations then it makes sense to convert the SQLAlchemy models to Django ORM at runtime.

Status

The SQLAlchemy to Django Bridge works well for the author's use case.

There are probably a lot of corner cases and advanced (or not so advanced) features of SQLAlchemy that are not (yet) supported.

The tests run in sqlite, and our production system uses PostgreSQL. It may or may not work with other database systems.

Installation

pip install sa2django

Features

  • basic data types (int, float, string, varchar, char, date, datetime etc, bytea)
  • foreign keys and many-to-one relationships
  • many-to-many relationships including through tables
  • automatic inference of all declared models in a sqlalchemy Base
  • alternatively define your Django models as usual, but use the SA2DModel as base class. Then all database fields are added from the corresponding sqlalchemy model, but you can still add properties and functions to the Django model

Usage

Define your SQLAlchemy schema

Say you have the following SQLAlchemy schema:

# sqlalchemy_schema.py
from sqlalchemy import Column, ForeignKey, Integer, String
from sqlalchemy.ext.declarative import declarative_base
from sqlalchemy.orm import relationship

Base = declarative_base()

class Parent(Base):
    __tablename__ = "parent"
    id = Column(Integer, primary_key=True)
    name = Column(String)
    children = relationship("Child", back_populates="parent")


class Child(Base):
    __tablename__ = "child"
    id = Column(Integer, primary_key=True)
    name = Column(String)
    age = Column(Integer)
    parent_id = Column(Integer, ForeignKey("parent.id"))
    parent = relationship(Parent, uselist=False, back_populates="children")

Automatic inference

You can use it in your Django application by simply adding the following code to your models.py:

# models.py
from sa2django.core import generate_sa2d_models, inject_models
from sqlalchemy_models import Base  # this is what we have specified in the snippet above

_models = generate_sa2d_models(Base, __name__)
inject_models(_models, globals())

This will add the classes Child and Parent to your models module. These classes are the Django models that correspond to the SQLAlchemy models.

You can't see it in the file, but Django will recognize them, and you can import them anywhere in the application as if you had declared them manually:

from models import Parent, Child

print(Child.objects.all())
child = Child.objects.get(pk=my_pk)
print(child.parents.all())

Note: your IDE might complain because it thinks that Parent and Child do not exist. It can't know, because the classes are created at runtime. The code will work fine.

Manual specification and custom properties

A strength of Django is that it allows to specify additional properties on a model. sa2django supports this. To do so you can declare your models explicitly and inherit from sa2django.SA2DModel instead of django.db.models.Model, and specify the corresponding SQLAlchemy model in the Meta class under the sa_model attribute.

You also need to register all Django models that are referenced by any model in the SQLAlchemy schema with register_table().

Then you have control over the class name and additional properties and use sa2django only to add all fields and relations to your Django model.

Here's and example:

from sa2django import register_table, SA2DModel
from sa2django.core import extract_tables_from_base
import sqlalchemy_models

# extract all SQLAlchemy tables
_tables = extract_tables_from_base(sqlalchemy_models.Base)

# register the tables
for _tablename, _sa_class in _tables.items():
    register_table(_tablename, _sa_class.__name__)

class Child(SA2DModel):
    class Meta:
        sa_model = sqlalchemy_models.Child
        ordering = ["age"]
    
    @property
    def age_next_year(self):
        return self.age + 1

class Parent(SA2DModel):
    class Meta:
        sa_model = sqlalchemy_models.Parent

At the moment it is not possible to mix manual and automatic model extraction -- if you specify one model manually, you cannot use generate_sa2d_models() anymore, so you need to specify all models manually.

Limitations

SQLAlchemy provides a superset of Django's functionality. For this reason, there's a long list of limitations.

The list is even longer and probably not exhaustive because sa2django is a young project tailored to its author's current needs.

  • at the moment only declarative base definitions are supported, no pure Mapper objects
  • composite foreign keys and primary keys are not supported. Primary keys and foreign keys must contain exactly one column
  • relations that do not use a foreign key are not added to the Django models

Many to many relationships

  • In sqlalchemy, in the mapper of the intermediate table, both foreign keys and relationships linking to both tables must be specified.

    Example:

    class CarParentAssoc(Base):
        __tablename__ = "cartoparent"
        id = Column(Integer, primary_key=True)
        car = relationship("Car")
        parent = relationship("Parent")
        car_id = Column(Integer, ForeignKey("car.car_id"))
        parent_id = Column(Integer, ForeignKey("parent.id"))

    Note that for both links to the car and parent table, both foreign keys and relationship attributes are specified.

Changelog

0.2.1

  • limit to SQLAlchemy <1.4

0.2.0

  • derive foreignkeys from MANYTOONE relationships instead of relying on explicit ForeignKey columns

0.1.3

  • set arbitrary max_length of 2048 on String fields that do not have a defined length in sqlalchemy. Necessary because Django does not support unlimited String fields, even though some backends (e.g., Postgres) do

Contributing

Pull requests are more than welcome! Ideally reach out to us by creating or replying to a Github ticket such that we can align our work and ideas.

License

MIT

Author: Marius Helf (helfsmarius@gmail.com)