DynamicField

MariaDB has a feature called Dynamic Columns that allows you to store different sets of columns for each row in a table. It works by storing the data in a blob and having a small set of functions to manipulate this blob. (Docs).

Django-MySQL supports the named Dynamic Columns of MariaDB 10.0+, as opposed to the numbered format of 5.5+. It uses the mariadb-dyncol python package to pack and unpack Dynamic Columns blobs in Python rather than in MariaDB (mostly due to limitations in the Django ORM).

class DynamicField(spec=None, **kwargs)

A field for storing Dynamic Columns. The Python data type is dict. Keys must be strs (unicode on Python 2) and values must be one of the supported value types in mariadb-dyncol:

  • str (unicode on Python 2)
  • int (and also long on Python 2)
  • float
  • datetime.date
  • datetime.datetime
  • datetime.datetime
  • A nested dict conforming to this spec too

Note that there are restrictions on the range of values supported for some of these types, and that decimal.Decimal objects are not yet supported though they are valid in MariaDB. For more information consult the mariadb-dyncol documentation.

Values may also be None, though they will then not be stored, since dynamic columns do not store NULL, so you should use .get() to retrieve values that may be None.

To use this field, you’ll need to:

  1. Use MariaDB 10.0.2+
  2. Install mariadb-dyncol (pip install mariadb-dyncol)
  3. Use either the utf8mb4 or utf8 character set for your database connection.

These are all checked by the field and you will see sensible errors for them when Django’s checks run if you have a DynamicField on a model.

spec

This is an optional type specification that checks that the named columns, if present, have the given types. It is validated against on save() to ensure type safety (unlike normal Django validation which is only used in forms). It is also used for type information for lookups (below).

spec should be a dict with string keys and values that are the type classes you expect. You can also nest another such dictionary as a value for validating nested dynamic columns.

For example:

import datetime

class SpecModel(Model):
    attrs = DynamicField(spec={
        'an_integer_key': int,
        'created_at': datetime.datetime,
        'nested_columns': {
            'lat': int,
            'lon': int,
        }
    })

This will enforce the following rules:

  • instance.attrs['an_integer_key'], if present, is an int
  • instance.attrs['created_at'], if present, is an datetime.datetime
  • instance.attrs['nested_columns'], if present, is a dict
  • instance.attrs['nested_columns']['lat'], if present, is an int
  • instance.attrs['nested_columns']['lon'], if present, is an int

Trying to save a DynamicField with data that does not match the rules of its spec will raise TypeError. There is no automatic casting, e.g. between int and float. Note that columns not in spec will still be allowed and have no type enforced.

For example:

>>> SpecModel.objects.create(attrs={'an_integer_key': 1})  # Fine
>>> SpecModel.objects.create(attrs={'an_integer_key': 2.0})
Traceback (most recent call last):
...
TypeError: Key 'an_integer_key' should be of type 'int'
>>> SpecModel.objects.create(attrs={'non_spec_key': 'anytype'})  # Fine

DynamicFields in Forms

By default a DynamicField has no form field, because there isn’t really a practical way to edit its contents. If required, is possible to add extra form fields to a ModelForm that then update specific dynamic column names on the instance in the form’s save().

Querying DynamicField

You can query by names, including nested names. In cases where names collide with existing lookups (e.g. you have a column named 'exact'), you might want to use the ColumnGet database function. You can also use the ColumnAdd and ColumnDelete functions for atomically modifying the contents of dynamic columns at the database layer.

We’ll use the following example model:

from django_mysql.models import DynamicField, Model

class ShopItem(Model):
    name = models.CharField(max_length=200)
    attrs = DynamicField(spec={
        'size': str,
    })

    def __str__(self):  # __unicode__ on Python 2
        return self.name

Exact Lookups

To query based on an exact match, just use a dictionary.

For example:

>>> ShopItem.objects.create(name='Camembert', attrs={'smelliness': 15})
>>> ShopItem.objects.create(name='Cheddar', attrs={'smelliness': 15, 'hardness': 5})

>>> ShopItem.objects.filter(attrs={'smelliness': 15})
[<ShopItem: Camembert>]
>>> ShopItem.objects.filter(attrs={'smelliness': 15, 'hardness': 5})
[<ShopItem: Cheddar>]

Name Lookups

To query based on a column name, use that name as a lookup with one of the below SQL types added after an underscore. If the column name is in your field’s spec, you can omit the SQL type and it will be extracted automatically - this includes keys in nested dicts.

The list of SQL types is:

  • BINARY - dict (a nested DynamicField)
  • CHAR - str (unicode on Python 2)
  • DATE - datetime.date
  • DATETIME - datetime.datetime
  • DOUBLE - float
  • INTEGER - int (and also long on Python 2)
  • TIME - datetime.time

These will also use the correct Django ORM field so chained lookups based on that type are possible, e.g. dynamicfield__age_INTEGER__gte=20.

Beware that getting a named column can always return NULL if the column is not defined for a row.

For example:

>>> ShopItem.objects.create(name='T-Shirt', attrs={'size': 'Large'})
>>> ShopItem.objects.create(name='Rocketship', attrs={
...     'speed_mph': 300,
...     'dimensions': {'width_m': 10, 'height_m': 50}
... })

# Basic template: DynamicField + '__' + column name + '_' + SQL type
>>> ShopItem.objects.filter(attrs__size_CHAR='Large')
[<ShopItem: T-Shirt>]

# As 'size' is in the field's spec, there is no need to give the SQL type
>>> ShopItem.objects.filter(attrs__size='Large')
[<ShopItem: T-Shirt>]

# Chained lookups are possible based on the data type
>>> ShopItem.objects.filter(attrs__speed_mph_INTEGER__gte=100)
[<ShopItem: Rocketship>]

# Nested keys can be looked up
>>> ShopItem.objects.filter(attrs__dimensions_BINARY__width_m_INTEGER=10)
[<ShopItem: Rocketship>]

# Nested DynamicFields can be queried as ``dict``s, as per the ``exact`` lookup
>>> ShopItem.objects.filter(attrs__dimensions_BINARY={'width_m': 10, 'height_m': 50})
[<ShopItem: Rocketship>]

# Missing keys are always NULL
>>> ShopItem.objects.filter(attrs__blablabla_INTEGER__isnull=True)
[<ShopItem: T-Shirt>, <ShopItem: Rocketship>]