django_orm_views

The package for managing database views using django, without the migrations. It can be used for generating database views and reading from them using django orm system.

What does this support?

• This package adds support to Django for writing Postgres views using:
  • Raw SQL
  • Django Querysets

They look something like this:

class MySQLView(PostgresViewFromSQL):
    sql = """
       SELECT col_a, col_b FROM table_1;
    """

class MyQuerysetView(PostgresViewFromQueryset):
    @classmethod
    def get_queryset(cls):
        return (
            Table1
            .objects
            .values('col_a', 'col_b')
        )

• The package allows reading from a view using django ORM. For a view to be readable, it has to be inherited from one of the readable view abstractions - ReadableViewFromQueryset or ReadableViewFromSQL. It would look like this:

class MyReadableView(ReadableViewFromQueryset):
    @classmethod
    def get_queryset(cls):
        return (
            Table1
            .objects
            .values('col_a', 'col_b')
        )
    
result = MyReadableView.objects.all()
result = MyReadableView.objects.filter(name="test")
result = MyReadableView.objects.get(id=1)

The name for the view in the database is generated automatically in the base class - BasePostgresView. It's not possible at the moment to define a custom name in the database for a readable view. Though it's possible if the view is not readable (not inherited from the readable view abstraction).

Since a readable view is basically a not managed django model, you might want to set some custom meta data for it by defining a Meta class inside the readable view ( For example changing the data source - db_table).

! Note that defining a custom Meta class is not allowed in the current version of the package and will lead to an error!

See some more examples in the tests here

This also supports the construction of materialised views via PostgresMaterialisedViewMixin. Note that the function refresh_materialized_view will need to be managed by the user in order to keep these up to date where required.

What does this not support?

• Any database engines aside from Postgres (unless syntax happens to be the same!)
• Defining custom name for a readable view in the database. Although it's possible for a view that is not readable.
• Defining custom Meta class in a readable view as it can be done for a django model to define additional settings E.g. db_table

When should I use this?

Our use-case is for a database which is managed by Django in which we would like to provide an analytics-friendly representation of some of our data. This involves giving analytics direct access to our database (whilst using a permissions framework), but using views to expose the data in a more simple way, as well as obscuring data which we consider personally identifiable/sensitive.

There are other frameworks existing which do similar things, usually including reads via the ORM. We found that these packages all generate migrations (despite being unmanaged) and we wanted to remove this from the django migrations process altogether - there seemed to be no value add by including migrations and they would just muddy our migration states.

Cool! But how do I use this?

• pip install django-orm-views
• Add 'django_orm_views' to your INSTALLED_APPS
• Create a postgres_views.py (file or package) inside any app
• Add a PostgresViewFromQueryset or PostgresViewFromSQL to your postgres_views.py (as above)
• run ./manage.py sync_views

A postgres_views.py file might look something like the following:

class ComplexViewFromQueryset(PostgresViewFromQueryset):

    prefix = 'test'  # This is optional, but allows you to prefix the table names for views
    
    @classmethod
    def get_queryset(cls):  # Return a `.values` with the query you desire
        return (
            TestModelWithForeignKey
            .objects
            .all()
            .annotate(
                double_integer_col=F('foreign_key__integer_col') * 2
            )
            .values(
                'id',
                'foreign_key__id',
                'foreign_key__integer_col',
                'foreign_key__character_col',
                'foreign_key__date_col',
                'foreign_key__datetime_col',
                'double_integer_col',
            )
        )

When we run the ./manage.py sync_views, we'll create a view called test_complexviewfromqueryset under the views schema.

Note, you can put ./manage.py sync_views into your CI/CD. It works by:

• Opening a transaction
• Dropping the views schema
• Recreating the views schema
• Recreating all views under that schema
• Committing the transaction

What's still to come?

• Support for more database engines. This currently only supports Postgres, but should be a reasonably light shift to support other database engines.
• Making the package more configurable using settings.
• Consideration of 0 downtime deployments with views.
  • Note, this can still be achieved with the current implementation, but a bad migration (with a view depending) could cascade a view and create downtime. Ideally migrations + view creation should happen in a single transaction.

Contributing

Feel free to fork the package and propose changes. The repo comes with a test django project which can be used to effectively test changes. It also demonstrates the functionality pretty well.