Skip to content

Latest commit

 

History

History
203 lines (151 loc) · 6.19 KB

README.md

File metadata and controls

203 lines (151 loc) · 6.19 KB
Raw

django-spanner

ORM plugin for using Cloud Spanner as a database for Django.

🚨THIS CODE IS STILL UNDER DEVELOPMENT🚨

Table of contents

Installing it

Use the version of django-spanner that corresponds to your version of Django. For example, django-spanner 2.2.x works with Django 2.2.y. (This is the only supported version at this time.)

The minor release number of Django doesn't correspond to the minor release number of django-spanner. Use the latest minor release of each.

pip3 install --user .

Using it

After installing it, you'll need to edit your Django settings.py file:

  • Add django_spanner as the very first entry in the INSTALLED_APPS setting
INSTALLED_APPS = [
    'django_spanner',
    ...
]
  • Edit the DATABASES setting to point to an EXISTING database

Format

DATABASES = {
    'default': {
        'ENGINE': 'django_spanner',
        'PROJECT': '<project_id>',
        'INSTANCE': '<instance_id>',
        'NAME': '<database_name>',
        # Only include this if you need to specify where to retrieve the
        # service account JSON for the credentials to connect to Cloud Spanner.
        'OPTIONS': {
            'credentials_uri': '<credentials_uri>',
        },
    },
}

Example

For example:

DATABASES = {
    'default': {
        'ENGINE': 'django_spanner',
        'PROJECT': 'appdev-soda-spanner-staging', # Or the GCP project-id
        'INSTANCE': 'django-dev1', # Or the Cloud Spanner instance
        'NAME': 'db1', # Or the Cloud Spanner database to use
    }
}

Limitations

Transaction management isn't supported

django-spanner always works in Django's default transaction behavior, autocommit mode. Transactions cannot be controlled manually with calls like django.db.transaction.atomic().

AutoField generates random IDs

Spanner doesn't have support for auto-generating primary key values. Therefore, django-spanner monkey-patches AutoField to generate a random UUID4. It generates a default using Field's default option which means AutoFields will have a value when a model instance is created. For example:

>>> ExampleModel()
>>> ExampleModel.pk
4229421414948291880

To avoid hotspotting, these IDs are not monotonically increasing. This means that sorting models by ID isn't guaranteed to return them in the order in which they were created.

ForeignKey constraints aren't created

Spanner doesn't support ON DELETE CASCADE when creating foreign-key constraints so django-spanner doesn't support foreign key constraints.

Check constraints aren't supported

Spanner doesn't support CHECK constraints so one isn't created for PositiveIntegerField and CheckConstraint can't be used.

DecimalField isn't supported

Spanner doesn't support a NUMERIC data type that allows storing high precision decimal values without the possibility of data loss.

Variance and StdDev database functions aren't supported

Spanner doesn't support these functions.

Meta.order_with_respect_to model option isn't supported

This feature uses a column name that starts with an underscore (_order) which Spanner doesn't allow.

Random QuerySet ordering isn't supported

Spanner doesn't support it. For example:

>>> ExampleModel.objects.order_by('?')
...
django.db.utils.ProgrammingError: 400 Function not found: RANDOM ... FROM
example_model ORDER BY RANDOM() ASC

Schema migrations

Spanner has some limitations on schema changes which you must respect:

  • Renaming tables and columns isn't supported.
  • A column's type can't be changed.
  • A table's primary key can't be altered.
  • Migrations aren't atomic since django-spanner doesn't support transactions.

DurationField arithmetic doesn't work with DateField values (#253)

Spanner requires using different functions for arithmetic depending on the column type:

  • TIMESTAMP columns (DateTimeField) require TIMESTAMP_ADD or TIMESTAMP_SUB
  • DATE columns (DateField) require DATE_ADD or DATE_SUB

Django doesn't provide a way to determine which database function to use. DatabaseOperations.combine_duration_expression() arbitrary uses TIMESTAMP_ADD and TIMESTAMP_SUB. Therefore, if you use a DateField in a DurationField expression, you'll see an error like: "No matching signature for function TIMESTAMP_ADD for argument types: DATE, INTERVAL INT64 DATE_TIME_PART."

Computations that yield FLOAT64 values can't be assigned to INT64 columns

Spanner doesn't support this.

For example, if integer is IntegerField:

>>> ExampleModel.objects.update(integer=F('integer') / 2)
...
django.db.utils.ProgrammingError: 400 Value of type FLOAT64 cannot be
assigned to integer, which has type INT64 [at 1:46]\nUPDATE
example_model SET integer = (example_model.integer /...

Addition with null values crash

For example:

>>> Book.objects.annotate(adjusted_rating=F('rating') + None)
...
google.api_core.exceptions.InvalidArgument: 400 Operands of + cannot be literal
NULL ...

How it works

Overall design

Internals

🚨THIS CODE IS STILL UNDER DEVELOPMENT🚨