SQL Views for Postgres

Adds first-class support for PostgreSQL Views in the Django ORM. For of the original django-pgviews by mypebble with support for Django 2.2+.

Installation

Install via pip:

pip install django-pgviews-redux

Add to installed applications in settings.py:

INSTALLED_APPS = (
  # ...
  'django_pgviews',
)

Examples

from django.db import models

from django_pgviews import view as pg


class Customer(models.Model):
    name = models.CharField(max_length=100)
    post_code = models.CharField(max_length=20)
    is_preferred = models.BooleanField(default=False)

    class Meta:
        app_label = 'myapp'

class PreferredCustomer(pg.View):
    projection = ['myapp.Customer.*',]
    dependencies = ['myapp.OtherView',]
    sql = """SELECT * FROM myapp_customer WHERE is_preferred = TRUE;"""

    class Meta:
      app_label = 'myapp'
      db_table = 'myapp_preferredcustomer'
      managed = False

NOTE It is important that we include the managed = False in the Meta so Django 1.7 migrations don't attempt to create DB tables for this view.

The SQL produced by this might look like:

CREATE VIEW myapp_preferredcustomer AS
SELECT * FROM myapp_customer WHERE is_preferred = TRUE;

To create all your views, run python manage.py sync_pgviews

You can also specify field names, which will map onto fields in your View:

from django_pgviews import view as pg


VIEW_SQL = """
    SELECT name, post_code FROM myapp_customer WHERE is_preferred = TRUE
"""


class PreferredCustomer(pg.View):
    name = models.CharField(max_length=100)
    post_code = models.CharField(max_length=20)

    sql = VIEW_SQL

Usage

To map onto a View, simply extend pg_views.view.View, assign SQL to the sql argument and define a db_table. You must always set managed = False on the Meta class.

Views can be created in a number of ways:

1. Define fields to map onto the VIEW output
2. Define a projection that describes the VIEW fields

Define Fields

Define the fields as you would with any Django Model:

from django_pgviews import view as pg


VIEW_SQL = """
    SELECT name, post_code FROM myapp_customer WHERE is_preferred = TRUE
"""


class PreferredCustomer(pg.View):
    name = models.CharField(max_length=100)
    post_code = models.CharField(max_length=20)

    sql = VIEW_SQL

    class Meta:
      managed = False
      db_table = 'my_sql_view'

Define Projection

django-pgviews can take a projection to figure out what fields it needs to map onto for a view. To use this, set the projection attribute:

from django_pgviews import view as pg


class PreferredCustomer(pg.View):
    projection = ['myapp.Customer.*',]
    sql = """SELECT * FROM myapp_customer WHERE is_preferred = TRUE;"""

    class Meta:
      db_table = 'my_sql_view'
      managed = False

This will take all fields on myapp.Customer and apply them to PreferredCustomer

Features

Updating Views

Sometimes your models change and you need your Database Views to reflect the new data. Updating the View logic is as simple as modifying the underlying SQL and running:

python manage.py sync_pgviews --force

This will forcibly update any views that conflict with your new SQL.

Dependencies

You can specify other views you depend on. This ensures the other views are installed beforehand. Using dependencies also ensures that your views get refreshed correctly when using sync_pgviews --force.

Note: Views are synced after the Django application has migrated and adding models to the dependency list will cause syncing to fail.

Example:

from django_pgviews import view as pg

class PreferredCustomer(pg.View):
    dependencies = ['myapp.OtherView',]
    sql = """SELECT * FROM myapp_customer WHERE is_preferred = TRUE;"""

    class Meta:
      app_label = 'myapp'
      db_table = 'myapp_preferredcustomer'
      managed = False

Materialized Views

Postgres 9.3 and up supports materialized views which allow you to cache the results of views, potentially allowing them to load faster.

However, you do need to manually refresh the view. To do this automatically, you can attach signals and call the refresh function.

Example:

from django_pgviews import view as pg


VIEW_SQL = """
    SELECT name, post_code FROM myapp_customer WHERE is_preferred = TRUE
"""

class Customer(models.Model):
    name = models.CharField(max_length=100)
    post_code = models.CharField(max_length=20)
    is_preferred = models.BooleanField(default=True)


class PreferredCustomer(pg.MaterializedView):
    name = models.CharField(max_length=100)
    post_code = models.CharField(max_length=20)

    sql = VIEW_SQL


@receiver(post_save, sender=Customer)
def customer_saved(sender, action=None, instance=None, **kwargs):
    PreferredCustomer.refresh()

Concurrent refresh

Postgres 9.4 and up allow materialized views to be refreshed concurrently, without blocking reads, as long as a unique index exists on the materialized view. To enable concurrent refresh, specify the name of a column that can be used as a unique index on the materialized view. Unique index can be defined on more than one column of a materialized view. Once enabled, passing concurrently=True to the model's refresh method will result in postgres performing the refresh concurrently. (Note that the refresh method itself blocks until the refresh is complete; concurrent refresh is most useful when materialized views are updated in another process or thread.)

Example:

from django_pgviews import view as pg


VIEW_SQL = """
    SELECT id, name, post_code FROM myapp_customer WHERE is_preferred = TRUE
"""

class PreferredCustomer(pg.MaterializedView):
    concurrent_index = 'id, post_code'
    sql = VIEW_SQL

    name = models.CharField(max_length=100)
    post_code = models.CharField(max_length=20)


@receiver(post_save, sender=Customer)
def customer_saved(sender, action=None, instance=None, **kwargs):
    PreferredCustomer.refresh(concurrently=True)

Indexes

As the materialized view isn't defined through the usual Django model fields, any indexes defined there won't be created on the materialized view. Luckily Django provides a Meta option called indexes which can be used to add custom indexes to models. pg_views supports defining indexes on materialized views using this option.

In the following example, one index will be created, on the name column. The db_index=True on the field definition for post_code will get ignored.

from django_pgviews import view as pg


VIEW_SQL = """
    SELECT id, name, post_code FROM myapp_customer WHERE is_preferred = TRUE
"""

class PreferredCustomer(pg.MaterializedView):
    sql = VIEW_SQL

    name = models.CharField(max_length=100)
    post_code = models.CharField(max_length=20, db_index=True)

    class Meta:
        managed = False  # don't forget this, otherwise Django will think it's a regular model
        indexes = [
             models.Index(fields=["name"]),
        ]

WITH NO DATA

Materialized views can be created either with or without data. By default, they are created with data, however pg_views supports creating materialized views without data, by defining with_data = False for the pg.MaterializedView class. Such views then do not support querying until the first refresh (raising django.db.utils.OperationalError).

Example:

from django_pgviews import view as pg

class PreferredCustomer(pg.MaterializedView):
    concurrent_index = 'id, post_code'
    sql = """
        SELECT id, name, post_code FROM myapp_customer WHERE is_preferred = TRUE
    """
    with_data = False

    name = models.CharField(max_length=100)
    post_code = models.CharField(max_length=20)

Conditional materialized views recreate

Since all materialized views are recreated on running migrate, it can lead to obsolete recreations even if there were no changes to the definition of the view. To prevent this, version 0.7.0 and higher contain a feature which checks existing materialized view definition in the database (if the mat. view exists at all) and compares the definition with the one currently defined in your pg.MaterializedView subclass. If the definition matches exactly, the re-create of materialized view is skipped.

This feature is enabled by setting the MATERIALIZED_VIEWS_CHECK_SQL_CHANGED in your Django settings to True, which enables the feature when running migrate. The command sync_pgviews uses this setting as well, however it also has switches --enable-materialized-views-check-sql-changed and --disable-materialized-views-check-sql-changed which override this setting for that command.

This feature also takes into account indexes. When a view is deemed not needing recreating, the process will still check the indexes on the table and delete any extra indexes and create any missing indexes. This reconciliation is done through the index name, so if you use custom names for your indexes, it might happen that it won't get updated on change of the content but not the name.

Custom Schema

You can define any table name you wish for your views. They can even live inside your own custom PostgreSQL schema.

from django_pgviews import view as pg


class PreferredCustomer(pg.View):
    sql = """SELECT * FROM myapp_customer WHERE is_preferred = TRUE;"""

    class Meta:
      db_table = 'my_custom_schema.preferredcustomer'
      managed = False

Dynamic View SQL

If you need a dynamic view SQL (for example if it needs a value from settings in it), you can override the run_sql classmethod on the view to return the SQL. The method should return a namedtuple ViewSQL, which contains the query and potentially the params to cursor.execute call. Params should be either None or a list of parameters for the query.

from django.conf import settings
from django_pgviews import view as pg


class PreferredCustomer(pg.View):
    @classmethod
    def get_sql(cls):
        return pg.ViewSQL(
            """SELECT * FROM myapp_customer WHERE is_preferred = TRUE and created_at >= %s;""",
            [settings.MIN_PREFERRED_CUSTOMER_CREATED_AT]
        )

    class Meta:
      db_table = 'preferredcustomer'
      managed = False

Sync Listeners

django-pgviews 0.5.0 adds the ability to listen to when a post_sync event has occurred.

view_synced

Fired every time a VIEW is synchronised with the database.

Provides args:

• sender - View Class
• update - Whether the view to be updated
• force - Whether force was passed
• status - The result of creating the view e.g. EXISTS, FORCE_REQUIRED
• has_changed - Whether the view had to change

all_views_synced

Sent after all Postgres VIEWs are synchronised.

Provides args:

• sender - Always None

Django Compatibility

Django 1.7 Note

Django 1.7 changed how models are loaded so that it's no longer possible to do sql = str(User.objects.all().query) because the dependent models aren't yet loaded by Django.

Django 1.9 Note

You now have to use the .view module directly.

Django 1.10 Note

When updating to Django 1.10, if you see AttributeError: can't set attribute when you try to migrate or run tests, you need to check your migrations for where _base_manager or _default_manager get set on the model and replace it with objects inside the migration.

This also applies to Django MPTT who have covered this in a bit more detail.

Python 3 Support

Django PGViews Redux only supports Python 3.6+.