SQLAlchemy Declarative Extensions#

Actions Status Coverage Status Documentation Status

See the full documentation here.

Adds extensions to SQLAlchemy (and/or Alembic) which allows declaratively stating the existence of additional kinds of objects about your database not natively supported by SQLAlchemy/Alembic.

This includes:

  • Schemas

  • Views

  • Roles

  • Privileges (Grants/Default Grants)

  • Functions

  • Triggers

  • Rows (i.e. data)

  • “audit tables” (i.e. triggers which record data changes to some source table)

The primary function(s) of this library include:

  • Registering onto the SQLAlchemy event system such that metadata.create_all creates these objects.

  • (Optionally) Registers into Alembic such that alembic revision --autogenerate automatically creates/updates/deletes declared objects.

Kitchen Sink Example (using all available features)#

from sqlalchemy import Column, types, select
from sqlalchemy.orm import as_declarative
from sqlalchemy_declarative_extensions import (
    declarative_database, Schemas, Roles, Row, View, view,
)
from sqlalchemy_declarative_extensions.dialects.postgresql import (
    DefaultGrant, Function, Trigger, Role
)
from sqlalchemy_declarative_extensions.audit import audit


@declarative_database
@as_declarative
class Base:
    # Note: each object type also has a plural version (i.e. Schemas/Roles/etc) where you can specify
    # collection-level options like `ignore_unspecified`).
    #
    # If you dont set any collection-level options, you can instead use raw list/iterable as the collection.
    schemas = Schemas().are("example")
    roles = Roles(ignore_unspecified=True).are(
        Role("read", login=False),
        Role(
            "app",
            in_roles=['read']
        ),
    )
    grants = [
        DefaultGrant.on_tables_in_schema("public", 'example').grant("select", to="read"),
        DefaultGrant.on_sequences_in_schema("public").grant("usage", to="read"),
        Grant.new("usage", to="read").on_schemas("example")
    ]
    rows = [
        Row('foo', id=1),
    ]
    views = [
        View("low_foo", "select * from foo where i < 10"),
    ]
    functions = [
        Function(
            "fancy_function",
            """
            BEGIN
            INSERT INTO foo (id) select NEW.id + 1;
            RETURN NULL;
            END
            """,
            language="plpgsql",
            returns="trigger",
        ),
    ]
    triggers = [
        Trigger.after("insert", on="foo", execute="fancy_function")
        .named("on_insert_foo")
        .when("pg_trigger_depth() < 1")
        .for_each_row(),
    ]


@audit()
class Foo(Base):
    __tablename__ = 'foo'

    id = Column(types.Integer(), primary_key=True)


audit_table = Foo.__audit_table__


@view(Base)
class HighFoo:
    __tablename__ = "high_foo"
    __view__ = select(Foo.__table__).where(Foo.__table__.c.id >= 10)

Note, there is also support for declaring objects directly through the MetaData for users not using sqlalchemy’s declarative API.

Event Registration#

By default, the above example does not automatically do anything. Depending on the context, you can use one of two registration hooks: register_sqlalchemy_events or register_alembic_events.

register_sqlalchemy_events#

This registers events in SQLAlchemy’s event system such that a metadata.create_all(...) call will create the declared database objects.

from sqlalchemy_declarative_extensions import register_sqlalchemy_events

metadata = Base.metadata  # Given the example above.
register_sqlalchemy_events(metadata)
# Which is equivalent to...
register_sqlalchemy_events(metadata, schemas=False, roles=False, grants=False, rows=False)

All object types are opt in, and should be explicitly included in order to get registered.

Practically, this is to avoid issues with testing. In most cases the metadata.create_all call will be run in tests, a context where it’s more likely that you dont need grants or grants, and where parallel test execution could lead to issues with role or schema creation, depending on your setup.

register_alembic_events#

This registers comparators in Alembic’s registration system such that an alembic revision --autogenerate command will diff the existing database state against the declared database objects, and issue statements to create/update/delete objects in order to match the declared state.

# alembic's `env.py`
from sqlalchemy_declarative_extensions import register_alembic_events

register_alembic_events()
# Which is equivalent to...
register_sqlalchemy_events(schemas=True, roles=True, grants=True, rows=True)

All object types are opt out but can be excluded.

In contrast to register_sqlalchemy_events, it’s much more likely that you’re declaring most of these object types in order to have alembic track them

Database support#

In principle, this library can absolutely support any database supported by SQLAlchemy, and capable of being introspected enough to support detection of different kinds of objects.

As you can see below, in reality the existence of implementations are going to be purely driven by actual usage. The current maintainer(s) primarily use PostgreSQL and as such individual features for other databases will either suffer or lack implementation.

Postgres

MySQL

SQLite

Schema

View

Role

N/A

Grant

N/A

Default Grant

N/A

Function

*

Trigger

*

Row (data)

“Audit Table”

The astrisks above note pending or provisional support. The level of expertise in each dialects’ particular behaviors is not uniform, and deciding on the correct behavior for those dialects will require users to submit issues/fixes!

Supporting a new dialect can be as simple as providing the dialect-dispatched implementations for detecting existing objects of the given type. Very much the intent is that once a given object type is supported at all, the comparison infrastructure for that type should make it straightforward to support other dialects. At the end of the day, this library is primarily producing SQL statements, so in theory any dialect supporting a given object type should be able to be supported.

In such cases (like Grants/Roles) that different dialects support wildly different options/syntax, there are also patterns for defining dialect-specific objects, to mediate any additional differences.

Alembic-utils#

Alembic Utils is the primary library against which this library can be compared. At time of writing, most (but not all) object types supported by alembic-utils are supported by this library. One might begin to question when to use which library.

Below is a list of points on which the two libraries diverge. But note that you can certainly use both in tandem! It doesn’t need to be one or the other, and certainly for any object types which do not overlap, you might need to use both.

  • Database Support

    • Alembic Utils seems to explicitly only support PostgreSQL.

    • This library is designed to support any dialect (in theory). Certainly PostgreSQL is best supported, but there does exist support for specific kinds of objects to varying levels of support for SQLite and MySQL, so far.

  • Architecture

    • Alembic Utils is directly tied to Alembic and does not support SQLAlchemy’s MetaData.create_all. It’s also the responsibility of the user to discover/register objects in alembic.

    • This library depends only on SQLAlchemy, although it also supports alembic. Support for MetaData.create_all can be important for creating all object types in tests. It also is designed such that objects are registered on the MetaData itself, so there is no need for any specific discovery phase.

  • Scope

    • Alembic Utils declares specific, individual objects. I.e. you instantiate one specific PGGrantTable or PGView instance and Alembic know knows you want that object to be created. It cannot drop objects it is not already aware of.

    • This library declares ths objects the system as a whole expects to exist. Similar to Alembic’s behavior on tables, it will (by default) detect any undeclared objects that should not exist and drop them. That means, you can rely on this object to ensure the state of your migrations matches the state of your database exactly.

  • Migration history

    • Alembic Utils imports and references its own objects in your migrations history. This can be unfortunate, in that it deeply ties your migrations history to alembic-utils.

      (In fact, this can be a sticking point, trying to convert away from alembic_utils, because it will attempt to drop all the (e.g PGView) instances previously created when we replaced it with this library.)

    • This library, by contrast, prefers to emit the raw SQL of the operation into your migration. That means you know the exact commands that will execute in your migration, which can be helpful in debugging failure. It also means, if at any point you decide to stop use of the library (or pause a given type of object, due to a bug), you can! This library’s behaviors are primarily very much --autogenerate-time only.

  • Abstraction Level

    • Alembic Utils appears to define a very “literal” interface (for example, PGView accepts the whole view definition as a raw literal string).

    • This library tries to, as much as possible, provide a more abstracted interface that enables more compatibility with SQLAlchemy (For example View accepts SQLAlchemy objects which can be coerced into a SELECT). It also tends towards “builder” interfaces which progressively produce a object (Take a look at the DefaultGrant above, for an example of where that’s helpful).

A separate note on conversion/compatibility. Where possible, this library tries to support alembic-utils native objects as stand-ins for the objects defined in this library. For example, alembic_utils.pg_view.PGView can be declared instead of a sqlalchemy_declarative_extensions.View, and we will internally coerce it into the appropriate type. Hopefully this eases any transitional costs, or issues using one or the other library.