Skip to main content

Database abstraction layer for atom objects

Project description

status codecov

atom-db is a database abstraction layer for the atom framework. This package provides api's for seamlessly saving and restoring atom objects from json based document databases and SQL databases supported by sqlalchemy.

The main reason for building this is to make it easier have database integration with enaml applications so a separate framework is not needed to define database models.

This was originally a part of enaml-web but has been pulled out to a separate package.


  • Supports MySQL and Postgres
  • Uses django like queries or raw sqlalchemy queries
  • Works with alembic database migrations
  • Supports MongoDB using motor


The design is based somewhat on django.

There is a "manager" called Model.objects to do queries on the database table created for each subclass.

Serialization and deserialization is done with Model.serializer.

Note: As of 0.3.11 serialization can be customizer per member by tagging the member with a flatten or unflatten which should be a async callable which accepts the value and scope.

Each Model has async save, delete, and restore methods to interact with the database. This can be customized if needed using __restorestate__ and __getstate__.

MySQL and Postgres support

You can use atom-db to save and restore atom subclasses to MySQL and Postgres.

Just define models using atom members, but subclass the SQLModel and atom-db will convert the builtin atom members of your model to sqlalchemy table columns and create a sqlalchemy.Table for your model.

Customizing table creation

To customize how table columns are created you can tag members with information needed for sqlalchemy columns, ex Str().tag(length=40) will make a sa.String(40). See Tagging any member with store=False will make the member be excluded from the db.

atomdb will attempt to determine the proper column type, but if you need more control, you can tag the member to specify the column type with type=sa.<type> or specify the full column definition with column=sa.Column(...).

If you have a custom member, you can define a def get_column(self, model) or def get_column_type(self, model) method to create the table column for the given model.

Primary keys

You can tag a member with primary_key=True to make it the pk. If no member is tagged with primary_key it will create and use _id as the primary key. The_id member will be always alias to the actual primary key. Use the __pk__ attribute of the class to get the name of the primary key member.

Table metadata

Like in Django a nested Meta class can be added to specify the db_name, unique_together, composite indexes and constraints.

If no db_name is specified on a Meta class, the table name defaults the what is set in the __model__ member. This defaults to the qualname of the class, eg myapp.SomeModel.

class SomeModel(SQLModel):
    # ...

    class Meta:
        db_table = 'custom_table_name'

composite indexes must be a list of each composite index description. See sqlachemy's Index for index tuple description.

First element is the index name. If None, the name will be auto-generated according the convention. Followings elements are columns table. Order of columns matters.

class SomeModel(SQLModel):
    # ...

    class Meta:
        composite_indexes = [(None, 'a', 'b'), ('ìndex_b_c', 'b', 'c')]

In the exemple above, two composite indexes are created. The first is on columns 'a' and 'b' with name ix_SomeModel_a_b. The second one on columns 'b' and 'c' with name ìndex_b_c.

Table creation / dropping

Once your tables are defined as atom models, create and drop tables using create_table and drop_table of Model.objects respectively For example:

from atomdb.sql import SQLModel, SQLModelManager

# Call create_tables to create sqlalchemy tables. This does NOT write them to
# the db but ensures that all ForeignKey relations are created
mgr = SQLModelManager.instance()

# Now actually drop/create for each of your models

# Drop the table for this model (will raise sqlalchemy's error if it doesn't exist)
await User.objects.drop_table()

# Create the user table
await User.objects.create_table()

The mgr.create_tables() method will create the sqlalchemy tables for each imported SQLModel subclass (anything in the manager's registry dict). This should be called after all of your models are imported so sqlalchemy can properly setup any foreign key relations.

The manager also has a metadata member which holds the sqlalchemy.MetaData needed for migrations.

Once the tables are created, they are accessible via Model.objects.table.

Note: The sqlachemy table is also assigned to the __table__ attribute of each model class, however this will not be defined until the manager has created it.

Database setup

Before accessing the DB you must assign a "database engine" to the manager's database member.

Note: As of 0.6.2 you can also specify this as a dictionary to use multiple databases.

import os
import re
from import create_engine
from atomdb.sql import SQLModelManager

DATABASE_URL = os.environ.get('MYSQL_URL')

# Parse the DB url
m = re.match(r'mysql://(.+):(.*)@(.+):(\d+)/(.+)', DATABASE_URL)
user, pwd, host, port, db = m.groups()

# Create the engine
engine = await create_engine(
    db=db, user=user, password=pwd, host=host, port=port)

# Assign it to the manager
mgr = SQLModelManager.instance()
mgr.database = engine

This engine will then be used by the manager to execute queries. You can retrieve the database engine from any Model by using Model.objects.engine.

Multiple database

If you need to use more than one database it looks like this.

# Multiple databases
mgr = SQLModelManager.instance()
mgr.database = {
    'default': await create_engine(**default_db_params),
    'other': await create_engine(**other_db_params),

To specify which database is used either using the __database__ class field or specify it as the db_name on the model Meta.

class ExternalData(SQLModel):

    # ... fields
    class Meta:
        db_name = 'other'
        db_table = 'external_data'

Django style queries

Only very basic ORM style queries are implemented for common use cases. These are get, get_or_create, filter, and all. These all accept "django style" queries using <name>=<value> or <name>__<op>=<value>.

For example:

john, created = await User.objects.get_or_create(
        name="John Doe", email="", age=21, active=True)
assert created

jane, created = await User.objects.get_or_create(
        name="Jane Doe", email="", age=48, active=False,
assert created

# Startswith
u = await User.objects.get(name__startswith="John")
assert ==

# In query
users = await User.objects.filter(name__in=[,])
assert len(users) == 2

# Is query
users = await User.objects.filter(active__is=False)
assert len(users) == 1 and users[0].active == False

See sqlachemy's ColumnElement for which queries can be used in this way. Also the tests check that these actually work as intended.

Note: As of 0.4.0 you can pass sqlalchemy filters as non-keyword arguments directly to the filter method.

Caching, select related, and prefetch related

Foreign key relations can automatically be loaded using select_related and prefetch_related. Select related will perform a join while prefetch related does a separate query.

Each Model has a cache available at Model.objects.cache which uses weakrefs to ensure the same object is returned each time. You can manually prefetch objects and atom-db will pull them from it's internal cache when restoring objects.

For example with a simple many to one relationship like this:

class Category(SQLModel):
    name = Str()
    products = Relation(lambda: Product)

class Product(SQLModel):
    title = Str()
    category = Typed(Category)

category = await Category.objects.create(name="PCB")
await Product.objects.create(title="Stepper driver", category=category)

Use select related to load the product's category foreign key automatically.

# In this case the category of each product will automatically be loaded
products = await Product.objects.select_related('category').filter(title__icontains="driver")
# The __restored__ flag can be used check if the model has been loaded
assert products[0] == "PCB"

If a foreign key relation is NOT in the cache or in the state from a joined row it will create an "unloaded" model with only the primary key populated. In this case the __restored__ flag will be set to False.

From the other direction use prefetch related.

category = await Category.objects.prefetch_related('products').get(name="PCB")
assert category.products[0].title == "Stepper driver"

Note: prefetch_related does not apply a limit. If the query has a lot of rows this may be a problem.

Alternatively you can prefetch the related objects and they will be automatically pulled from the internal cache (eg TheModel.objects.cache).

all_categories = await Category.objects.all()
products = await Product.objects.filter(title__icontains="driver")
assert products[0].category in all_categories

Advanced / raw sqlalchemy queries

For more advanced queries using joins, etc.. you must build the query with sqlalchemy then execute it. The sa.Table for an atom model can be retrieved using Model.objects.table on which you can use select, where, etc... to build up whatever query you need.

Then use fetchall, fetchone, fetchmany, or execute to do these queries.

These methods do NOT return an object but the row from the database so they must manually be restored.

When joining you'll usually want to pass use_labels=True. For example:

q = Job.objects.table.join(JobRole.objects.table).select(use_labels=True)

for row in await Job.objects.fetchall(q):
    # Restore each manually, it handles pulling out the fields that are it's own
    job = await Job.restore(row)
    role = await JobRole.restore(row)

Depending on the relationships, you may need to then post-process these so they can be accessed in a more pythonic way. This is trade off between complexity and ease of use.

Connections and Transactions

A connection can be retrieved using Model.objects.connection() and used like normal aiomysql / aiopg connection. A transaction is done in the same way as defined in the docs for those libraries eg.

async with Job.objects.connection() as conn:
    trans = await conn.begin()
        # Do your queries here and pass the `connection` to each
        job, created = await Job.objects.get_or_create(connection=conn, **state)
        await trans.rollback()
        await trans.commit()

When using a transaction you need to pass the active connection to each call or it will use a different connection outside of the transaction!

The connection argument is removed from the filters/state. If your model happens to have a member named connection you can rename the connection argument by with Model.object.connection_kwarg = 'connection_' or whatever name you like.


Migrations work using alembic. The metadata needed to autogenerate migrations can be retrieved from SQLModelManager.instance().metadata so add the following in your alembic's

# Import your db models first
from myapp.models import *

from atomdb.sql import SQLModelManager
manager = SQLModelManager.instance()
manager.create_tables()  # Create sa tables
target_metadata = manager.metadata

The rest is handled by alembic.

Note: As of 0.4.1 the constraint naming conventions can be set using manager.constraints, this must be done before any tables are imported.

NoSQL support

You can also use atom-db to save and restore atom subclasses to MongoDB.

The NoSQL version is very basic as mongo is much more relaxed. No restriction is imposed on what type of manager is used, leaving that to whichever database library is preferred but it's tested (and currently used) with motor and tornado.

Just define models using atom members, but subclass the NoSQLModel.

from atom.api import Unicode, Int, Instance, List
from atomdb.nosql import NoSQLModel, NoSQLModelManager
from motor.motor_asyncio import AsyncIOMotorClient

# Set DB
client = AsyncIOMotorClient()
mgr = NoSQLModelManager.instance()
mgr.database = client.test_db

class Group(NoSQLModel):
    name = Unicode()

class User(NoSQLModel):
    name = Unicode()
    age = Int()
    groups = List(Group)

Then we can create an instance and save it. It will perform an upsert or replace the existing entry.

admins = Group(name="Admins")

# It will save admins using it's ObjectID
bob = User(name="Bob", age=32, groups=[admins])

tom = User(name="Tom", age=34, groups=[admins])

To fetch from the DB each model has a ModelManager called objects that will simply return the collection for the model type. For example.

# Fetch from db, you can use any MongoDB queries here
state = await User.objects.find_one({'name': "James"})
if state:
    james = await User.restore(state)

# etc...

Restoring is async because it will automatically fetch any related objects (ex the groups in this case). It saves objects using the ObjectID when present.

And finally you can either delete using queries on the manager directly or call on the object.

await tom.delete()
assert not await User.objects.find_one({'name': "Tom"})

You can exclude members from being saved to the DB by tagging them with .tag(store=False).


This is currently used in a few projects but not considered mature by any means.

Pull requests and feature requests are welcome!

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

atom_db-0.8.0-py3-none-any.whl (36.4 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page