relations.py

Serializer relations

Data structures, not algorithms, are central to programming.

Rob Pike

Relational fields are used to represent model relationships. They can be applied to ForeignKey, ManyToManyField and OneToOneField relationships, as well as to reverse relationships, and custom relationships such as GenericForeignKey.


Note: The relational fields are declared in relations.py, but by convention you should import them from the serializers module, using from rest_framework import serializers and refer to fields as serializers.<FieldName>.



Note: REST Framework does not attempt to automatically optimize querysets passed to serializers in terms of select_related and prefetch_related since it would be too much magic. A serializer with a field spanning an orm relation through its source attribute could require an additional database hit to fetch related objects from the database. It is the programmer's responsibility to optimize queries to avoid additional database hits which could occur while using such a serializer.

For example, the following serializer would lead to a database hit each time evaluating the tracks field if it is not prefetched:

class AlbumSerializer(serializers.ModelSerializer):
    tracks = serializers.SlugRelatedField(
        many=True,
        read_only=True,
        slug_field='title'
    )

    class Meta:
        model = Album
        fields = ['album_name', 'artist', 'tracks']

# For each album object, tracks should be fetched from database
qs = Album.objects.all()
print(AlbumSerializer(qs, many=True).data)

If AlbumSerializer is used to serialize a fairly large queryset with many=True then it could be a serious performance problem. Optimizing the queryset passed to AlbumSerializer with:

qs = Album.objects.prefetch_related('tracks')
# No additional database hits required
print(AlbumSerializer(qs, many=True).data)

would solve the issue.


Inspecting relationships.

When using the ModelSerializer class, serializer fields and relationships will be automatically generated for you. Inspecting these automatically generated fields can be a useful tool for determining how to customize the relationship style.

To do so, open the Django shell, using python manage.py shell, then import the serializer class, instantiate it, and print the object representation…

>>> from myapp.serializers import AccountSerializer
>>> serializer = AccountSerializer()
>>> print(repr(serializer))
AccountSerializer():
    id = IntegerField(label='ID', read_only=True)
    name = CharField(allow_blank=True, max_length=100, required=False)
    owner = PrimaryKeyRelatedField(queryset=User.objects.all())

API Reference

In order to explain the various types of relational fields, we'll use a couple of simple models for our examples. Our models will be for music albums, and the tracks listed on each album.

class Album(models.Model):
    album_name = models.CharField(max_length=100)
    artist = models.CharField(max_length=100)

class Track(models.Model):
    album = models.ForeignKey(Album, related_name='tracks', on_delete=models.CASCADE)
    order = models.IntegerField()
    title = models.CharField(max_length=100)
    duration = models.IntegerField()

    class Meta:
        unique_together = ['album', 'order']
        ordering = ['order']

    def __str__(self):
        return '%d: %s' % (self.order, self.title)

StringRelatedField

StringRelatedField may be used to represent the target of the relationship using its __str__ method.

For example, the following serializer:

class AlbumSerializer(serializers.ModelSerializer):
    tracks = serializers.StringRelatedField(many=True)

    class Meta:
        model = Album
        fields = ['album_name', 'artist', 'tracks']

Would serialize to the following representation:

{
    'album_name': 'Things We Lost In The Fire',
    'artist': 'Low',
    'tracks': [
        '1: Sunflower',
        '2: Whitetail',
        '3: Dinosaur Act',
        ...
    ]
}

This field is read only.

Arguments:

  • many - If applied to a to-many relationship, you should set this argument to True.

PrimaryKeyRelatedField

PrimaryKeyRelatedField may be used to represent the target of the relationship using its primary key.

For example, the following serializer:

class AlbumSerializer(serializers.ModelSerializer):
    tracks = serializers.PrimaryKeyRelatedField(many=True, read_only=True)

    class Meta:
        model = Album
        fields = ['album_name', 'artist', 'tracks']

Would serialize to a representation like this:

{
    'album_name': 'Undun',
    'artist': 'The Roots',
    'tracks': [
        89,
        90,
        91,
        ...
    ]
}

By default this field is read-write, although you can change this behavior using the read_only flag.

Arguments:

  • queryset - The queryset used for model instance lookups when validating the field input. Relationships must either set a queryset explicitly, or set read_only=True.
  • many - If applied to a to-many relationship, you should set this argument to True.
  • allow_null - If set to True, the field will accept values of None or the empty string for nullable relationships. Defaults to False.
  • pk_field - Set to a field to control serialization/deserialization of the primary key's value. For example, pk_field=UUIDField(format='hex') would serialize a UUID primary key into its compact hex representation.

HyperlinkedRelatedField

HyperlinkedRelatedField may be used to represent the target of the relationship using a hyperlink.

For example, the following serializer:

class AlbumSerializer(serializers.ModelSerializer):
    tracks = serializers.HyperlinkedRelatedField(
        many=True,
        read_only=True,
        view_name='track-detail'
    )

    class Meta:
        model = Album
        fields = ['album_name', 'artist', 'tracks']

Would serialize to a representation like this:

{
    'album_name': 'Graceland',
    'artist': 'Paul Simon',
    'tracks': [
        'http://www.example.com/api/tracks/45/',
        'http://www.example.com/api/tracks/46/',
        'http://www.example.com/api/tracks/47/',
        ...
    ]
}

By default this field is read-write, although you can change this behavior using the read_only flag.


Note: This field is designed for objects that map to a URL that accepts a single URL keyword argument, as set using the lookup_field and lookup_url_kwarg arguments.

This is suitable for URLs that contain a single primary key or slug argument as part of the URL.

If you require more complex hyperlinked representation you'll need to customize the field, as described in the custom hyperlinked fields section, below.


Arguments:

  • view_name - The view name that should be used as the target of the relationship. If you're using the standard router classes this will be a string with the format <modelname>-detail. required.
  • queryset - The queryset used for model instance lookups when validating the field input. Relationships must either set a queryset explicitly, or set read_only=True.
  • many - If applied to a to-many relationship, you should set this argument to True.
  • allow_null - If set to True, the field will accept values of None or the empty string for nullable relationships. Defaults to False.
  • lookup_field - The field on the target that should be used for the lookup. Should correspond to a URL keyword argument on the referenced view. Default is 'pk'.
  • lookup_url_kwarg - The name of the keyword argument defined in the URL conf that corresponds to the lookup field. Defaults to using the same value as lookup_field.
  • format - If using format suffixes, hyperlinked fields will use the same format suffix for the target unless overridden by using the format argument.

SlugRelatedField

SlugRelatedField may be used to represent the target of the relationship using a field on the target.

For example, the following serializer:

class AlbumSerializer(serializers.ModelSerializer):
    tracks = serializers.SlugRelatedField(
        many=True,
        read_only=True,
        slug_field='title'
     )

    class Meta:
        model = Album
        fields = ['album_name', 'artist', 'tracks']

Would serialize to a representation like this:

{
    'album_name': 'Dear John',
    'artist': 'Loney Dear',
    'tracks': [
        'Airport Surroundings',
        'Everything Turns to You',
        'I Was Only Going Out',
        ...
    ]
}

By default this field is read-write, although you can change this behavior using the read_only flag.

When using SlugRelatedField as a read-write field, you will normally want to ensure that the slug field corresponds to a model field with unique=True.

Arguments:

  • slug_field - The field on the target that should be used to represent it. This should be a field that uniquely identifies any given instance. For example, username. required
  • queryset - The queryset used for model instance lookups when validating the field input. Relationships must either set a queryset explicitly, or set read_only=True.
  • many - If applied to a to-many relationship, you should set this argument to True.
  • allow_null - If set to True, the field will accept values of None or the empty string for nullable relationships. Defaults to False.

HyperlinkedIdentityField

This field can be applied as an identity relationship, such as the 'url' field on a HyperlinkedModelSerializer. It can also be used for an attribute on the object. For example, the following serializer:

class AlbumSerializer(serializers.HyperlinkedModelSerializer):
    track_listing = serializers.HyperlinkedIdentityField(view_name='track-list')

    class Meta:
        model = Album
        fields = ['album_name', 'artist', 'track_listing']

Would serialize to a representation like this:

{
    'album_name': 'The Eraser',
    'artist': 'Thom Yorke',
    'track_listing': 'http://www.example.com/api/track_list/12/',
}

This field is always read-only.

Arguments:

  • view_name - The view name that should be used as the target of the relationship. If you're using the standard router classes this will be a string with the format <model_name>-detail. required.
  • lookup_field - The field on the target that should be used for the lookup. Should correspond to a URL keyword argument on the referenced view. Default is 'pk'.
  • lookup_url_kwarg - The name of the keyword argument defined in the URL conf that corresponds to the lookup field. Defaults to using the same value as lookup_field.
  • format - If using format suffixes, hyperlinked fields will use the same format suffix for the target unless overridden by using the format argument.

Nested relationships

As opposed to previously discussed references to another entity, the referred entity can instead also be embedded or nested in the representation of the object that refers to it. Such nested relationships can be expressed by using serializers as fields.

If the field is used to represent a to-many relationship, you should add the many=True flag to the serializer field.

Example

For example, the following serializer:

class TrackSerializer(serializers.ModelSerializer):
    class Meta:
        model = Track
        fields = ['order', 'title', 'duration']

class AlbumSerializer(serializers.ModelSerializer):
    tracks = TrackSerializer(many=True, read_only=True)

    class Meta:
        model = Album
        fields = ['album_name', 'artist', 'tracks']

Would serialize to a nested representation like this:

>>> album = Album.objects.create(album_name="The Grey Album", artist='Danger Mouse')
>>> Track.objects.create(album=album, order=1, title='Public Service Announcement', duration=245)
<Track: Track object>
>>> Track.objects.create(album=album, order=2, title='What More Can I Say', duration=264)
<Track: Track object>
>>> Track.objects.create(album=album, order=3, title='Encore', duration=159)
<Track: Track object>
>>> serializer = AlbumSerializer(instance=album)
>>> serializer.data
{
    'album_name': 'The Grey Album',
    'artist': 'Danger Mouse',
    'tracks': [
        {'order': 1, 'title': 'Public Service Announcement', 'duration': 245},
        {'order': 2, 'title': 'What More Can I Say', 'duration': 264},
        {'order': 3, 'title': 'Encore', 'duration': 159},
        ...
    ],
}

Writable nested serializers

By default nested serializers are read-only. If you want to support write-operations to a nested serializer field you'll need to create create() and/or update() methods in order to explicitly specify how the child relationships should be saved:

class TrackSerializer(serializers.ModelSerializer):
    class Meta:
        model = Track
        fields = ['order', 'title', 'duration']

class AlbumSerializer(serializers.ModelSerializer):
    tracks = TrackSerializer(many=True)

    class Meta:
        model = Album
        fields = ['album_name', 'artist', 'tracks']

    def create(self, validated_data):
        tracks_data = validated_data.pop('tracks')
        album = Album.objects.create(**validated_data)
        for track_data in tracks_data:
            Track.objects.create(album=album, **track_data)
        return album

>>> data = {
    'album_name': 'The Grey Album',
    'artist': 'Danger Mouse',
    'tracks': [
        {'order': 1, 'title': 'Public Service Announcement', 'duration': 245},
        {'order': 2, 'title': 'What More Can I Say', 'duration': 264},
        {'order': 3, 'title': 'Encore', 'duration': 159},
    ],
}
>>> serializer = AlbumSerializer(data=data)
>>> serializer.is_valid()
True
>>> serializer.save()
<Album: Album object>

Custom relational fields

In rare cases where none of the existing relational styles fit the representation you need, you can implement a completely custom relational field, that describes exactly how the output representation should be generated from the model instance.

To implement a custom relational field, you should override RelatedField, and implement the .to_representation(self, value) method. This method takes the target of the field as the value argument, and should return the representation that should be used to serialize the target. The value argument will typically be a model instance.

If you want to implement a read-write relational field, you must also implement the .to_internal_value(self, data) method.

To provide a dynamic queryset based on the context, you can also override .get_queryset(self) instead of specifying .queryset on the class or when initializing the field.

Example

For example, we could define a relational field to serialize a track to a custom string representation, using its ordering, title, and duration:

import time

class TrackListingField(serializers.RelatedField):
    def to_representation(self, value):
        duration = time.strftime('%M:%S', time.gmtime(value.duration))
        return 'Track %d: %s (%s)' % (value.order, value.name, duration)

class AlbumSerializer(serializers.ModelSerializer):
    tracks = TrackListingField(many=True)

    class Meta:
        model = Album
        fields = ['album_name', 'artist', 'tracks']

This custom field would then serialize to the following representation:

{
    'album_name': 'Sometimes I Wish We Were an Eagle',
    'artist': 'Bill Callahan',
    'tracks': [
        'Track 1: Jim Cain (04:39)',
        'Track 2: Eid Ma Clack Shaw (04:19)',
        'Track 3: The Wind and the Dove (04:34)',
        ...
    ]
}

Custom hyperlinked fields

In some cases you may need to customize the behavior of a hyperlinked field, in order to represent URLs that require more than a single lookup field.

You can achieve this by overriding HyperlinkedRelatedField. There are two methods that may be overridden:

get_url(self, obj, view_name, request, format)

The get_url method is used to map the object instance to its URL representation.

May raise a NoReverseMatch if the view_name and lookup_field attributes are not configured to correctly match the URL conf.

get_object(self, view_name, view_args, view_kwargs)

If you want to support a writable hyperlinked field then you'll also want to override get_object, in order to map incoming URLs back to the object they represent. For read-only hyperlinked fields there is no need to override this method.

The return value of this method should the object that corresponds to the matched URL conf arguments.

May raise an ObjectDoesNotExist exception.

Example

Say we have a URL for a customer object that takes two keyword arguments, like so:

/api/<organization_slug>/customers/<customer_pk>/

This cannot be represented with the default implementation, which accepts only a single lookup field.

In this case we'd need to override HyperlinkedRelatedField to get the behavior we want:

from rest_framework import serializers
from rest_framework.reverse import reverse

class CustomerHyperlink(serializers.HyperlinkedRelatedField):
    # We define these as class attributes, so we don't need to pass them as arguments.
    view_name = 'customer-detail'
    queryset = Customer.objects.all()

    def get_url(self, obj, view_name, request, format):
        url_kwargs = {
            'organization_slug': obj.organization.slug,
            'customer_pk': obj.pk
        }
        return reverse(view_name, kwargs=url_kwargs, request=request, format=format)

    def get_object(self, view_name, view_args, view_kwargs):
        lookup_kwargs = {
           'organization__slug': view_kwargs['organization_slug'],
           'pk': view_kwargs['customer_pk']
        }
        return self.get_queryset().get(**lookup_kwargs)

Note that if you wanted to use this style together with the generic views then you'd also need to override .get_object on the view in order to get the correct lookup behavior.

Generally we recommend a flat style for API representations where possible, but the nested URL style can also be reasonable when used in moderation.


Further notes

The queryset argument

The queryset argument is only ever required for writable relationship field, in which case it is used for performing the model instance lookup, that maps from the primitive user input, into a model instance.

In version 2.x a serializer class could sometimes automatically determine the queryset argument if a ModelSerializer class was being used.

This behavior is now replaced with always using an explicit queryset argument for writable relational fields.

Doing so reduces the amount of hidden 'magic' that ModelSerializer provides, makes the behavior of the field more clear, and ensures that it is trivial to move between using the ModelSerializer shortcut, or using fully explicit Serializer classes.

Customizing the HTML display

The built-in __str__ method of the model will be used to generate string representations of the objects used to populate the choices property. These choices are used to populate select HTML inputs in the browsable API.

To provide customized representations for such inputs, override display_value() of a RelatedField subclass. This method will receive a model object, and should return a string suitable for representing it. For example:

class TrackPrimaryKeyRelatedField(serializers.PrimaryKeyRelatedField):
    def display_value(self, instance):
        return 'Track: %s' % (instance.title)

Select field cutoffs

When rendered in the browsable API relational fields will default to only displaying a maximum of 1000 selectable items. If more items are present then a disabled option with "More than 1000 items…" will be displayed.

This behavior is intended to prevent a template from being unable to render in an acceptable timespan due to a very large number of relationships being displayed.

There are two keyword arguments you can use to control this behavior:

  • html_cutoff - If set this will be the maximum number of choices that will be displayed by a HTML select drop down. Set to None to disable any limiting. Defaults to 1000.
  • html_cutoff_text - If set this will display a textual indicator if the maximum number of items have been cutoff in an HTML select drop down. Defaults to "More than {count} items…"

You can also control these globally using the settings HTML_SELECT_CUTOFF and HTML_SELECT_CUTOFF_TEXT.

In cases where the cutoff is being enforced you may want to instead use a plain input field in the HTML form. You can do so using the style keyword argument. For example:

assigned_to = serializers.SlugRelatedField(
   queryset=User.objects.all(),
   slug_field='username',
   style={'base_template': 'input.html'}
)

Reverse relations

Note that reverse relationships are not automatically included by the ModelSerializer and HyperlinkedModelSerializer classes. To include a reverse relationship, you must explicitly add it to the fields list. For example:

class AlbumSerializer(serializers.ModelSerializer):
    class Meta:
        fields = ['tracks', ...]

You'll normally want to ensure that you've set an appropriate related_name argument on the relationship, that you can use as the field name. For example:

class Track(models.Model):
    album = models.ForeignKey(Album, related_name='tracks', on_delete=models.CASCADE)
    ...

If you have not set a related name for the reverse relationship, you'll need to use the automatically generated related name in the fields argument. For example:

class AlbumSerializer(serializers.ModelSerializer):
    class Meta:
        fields = ['track_set', ...]

See the Django documentation on reverse relationships for more details.

Generic relationships

If you want to serialize a generic foreign key, you need to define a custom field, to determine explicitly how you want to serialize the targets of the relationship.

For example, given the following model for a tag, which has a generic relationship with other arbitrary models:

class TaggedItem(models.Model):
    """
    Tags arbitrary model instances using a generic relation.

    See: https://docs.djangoproject.com/en/stable/ref/contrib/contenttypes/
    """
    tag_name = models.SlugField()
    content_type = models.ForeignKey(ContentType, on_delete=models.CASCADE)
    object_id = models.PositiveIntegerField()
    tagged_object = GenericForeignKey('content_type', 'object_id')

    def __str__(self):
        return self.tag_name

And the following two models, which may have associated tags:

class Bookmark(models.Model):
    """
    A bookmark consists of a URL, and 0 or more descriptive tags.
    """
    url = models.URLField()
    tags = GenericRelation(TaggedItem)


class Note(models.Model):
    """
    A note consists of some text, and 0 or more descriptive tags.
    """
    text = models.CharField(max_length=1000)
    tags = GenericRelation(TaggedItem)

We could define a custom field that could be used to serialize tagged instances, using the type of each instance to determine how it should be serialized:

class TaggedObjectRelatedField(serializers.RelatedField):
    """
    A custom field to use for the `tagged_object` generic relationship.
    """

    def to_representation(self, value):
        """
        Serialize tagged objects to a simple textual representation.
        """
        if isinstance(value, Bookmark):
            return 'Bookmark: ' + value.url
        elif isinstance(value, Note):
            return 'Note: ' + value.text
        raise Exception('Unexpected type of tagged object')

If you need the target of the relationship to have a nested representation, you can use the required serializers inside the .to_representation() method:

    def to_representation(self, value):
        """
        Serialize bookmark instances using a bookmark serializer,
        and note instances using a note serializer.
        """
        if isinstance(value, Bookmark):
            serializer = BookmarkSerializer(value)
        elif isinstance(value, Note):
            serializer = NoteSerializer(value)
        else:
            raise Exception('Unexpected type of tagged object')

        return serializer.data

Note that reverse generic keys, expressed using the GenericRelation field, can be serialized using the regular relational field types, since the type of the target in the relationship is always known.

For more information see the Django documentation on generic relations.

ManyToManyFields with a Through Model

By default, relational fields that target a ManyToManyField with a through model specified are set to read-only.

If you explicitly specify a relational field pointing to a ManyToManyField with a through model, be sure to set read_only to True.

If you wish to represent extra fields on a through model then you may serialize the through model as a nested object.


Third Party Packages

The following third party packages are also available.

DRF Nested Routers

The drf-nested-routers package provides routers and relationship fields for working with nested resources.

Rest Framework Generic Relations

The rest-framework-generic-relations library provides read/write serialization for generic foreign keys.