89
votes

With Django REST Framework, a standard ModelSerializer will allow ForeignKey model relationships to be assigned or changed by POSTing an ID as an Integer.

What's the simplest way to get this behavior out of a nested serializer?

Note, I am only talking about assigning existing database objects, not nested creation.

I have hacked away around this in the past with additional 'id' fields in the serializer and with custom create and update methods, but this is such a seemingly simple and frequent issue for me that I'm curious to know the best way.

class Child(models.Model):
    name = CharField(max_length=20)

class Parent(models.Model):
    name = CharField(max_length=20)
    phone_number = models.ForeignKey(PhoneNumber)
    child = models.ForeignKey(Child)

class ChildSerializer(ModelSerializer):
    class Meta:
        model = Child

class ParentSerializer(ModelSerializer):
    # phone_number relation is automatic and will accept ID integers
    children = ChildSerializer() # this one will not

    class Meta:
        model = Parent
12

12 Answers

59
votes

The best solution here is to use two different fields: one for reading and the other for writing. Without doing some heavy lifting, it is difficult to get what you are looking for in a single field.

The read-only field would be your nested serializer (ChildSerializer in this case) and it will allow you to get the same nested representation that you are expecting. Most people define this as just child, because they already have their front-end written by this point and changing it would cause problems.

The write-only field would be a PrimaryKeyRelatedField, which is what you would typically use for assigning objects based on their primary key. This does not have to be write-only, especially if you are trying to go for symmetry between what is received and what is sent, but it sounds like that might suit you best. This field should have a source set to the foreign key field (child in this example) so it assigns it properly on creation and updating.


This has been brought up on the discussion group a few times, and I think this is still the best solution. Thanks to Sven Maurer for pointing it out.

93
votes

Updated on July 05 2020

This post is getting more attention and it indicates more people have a similar situation. So I decided to add a generic way to handle this problem. This generic way is best suitable for you if you have more serializers that need to change to this format

Since DRF doesn't provide this functionality out of the box, we need to create a serializer field first.

from rest_framework import serializers


class RelatedFieldAlternative(serializers.PrimaryKeyRelatedField):
    def __init__(self, **kwargs):
        self.serializer = kwargs.pop('serializer', None)
        if self.serializer is not None and not issubclass(self.serializer, serializers.Serializer):
            raise TypeError('"serializer" is not a valid serializer class')

        super().__init__(**kwargs)

    def use_pk_only_optimization(self):
        return False if self.serializer else True

    def to_representation(self, instance):
        if self.serializer:
            return self.serializer(instance, context=self.context).data
        return super().to_representation(instance)

I am not well impressed with this class name, RelatedFieldAlternative, you can use anything you want. Then use this new serializer field in your parent serializer as,

class ParentSerializer(ModelSerializer):
   child = RelatedFieldAlternative(queryset=Child.objects.all(), serializer=ChildSerializer)

    class Meta:
        model = Parent
        fields = '__all__'

Original Post

Using two different fields would be ok (as @Kevin Brown and @joslarson mentioned), but I think it's not perfect (to me). Because getting data from one key (child) and sending data to another key (child_id) might be a little bit ambiguous for front-end developers. (no offense at all)


So, what I suggest here is, override the to_representation() method of ParentSerializer will do the job.

def to_representation(self, instance):
    response = super().to_representation(instance)
    response['child'] = ChildSerializer(instance.child).data
    return response


Complete representation of Serializer

class ChildSerializer(ModelSerializer):
    class Meta:
        model = Child
        fields = '__all__'


class ParentSerializer(ModelSerializer):
    class Meta:
        model = Parent
        fields = '__all__'

    def to_representation(self, instance):
        response = super().to_representation(instance)
        response['child'] = ChildSerializer(instance.child).data
        return response



Advantage of this method?

By using this method, we don't need two separate fields for creation and reading. Here both creation and reading can be done by using child key.


Sample payload to create parent instance

{
        "name": "TestPOSTMAN_name",
        "phone_number": 1,
        "child": 1
    }



Screenshot
POSTMAN screenshot

55
votes

Here's an example of what Kevin's answer is talking about, if you want to take that approach and use 2 separate fields.

In your models.py...

class Child(models.Model):
    name = CharField(max_length=20)

class Parent(models.Model):
    name = CharField(max_length=20)
    phone_number = models.ForeignKey(PhoneNumber)
    child = models.ForeignKey(Child)

then serializers.py...

class ChildSerializer(ModelSerializer):
    class Meta:
        model = Child

class ParentSerializer(ModelSerializer):
    # if child is required
    child = ChildSerializer(read_only=True) 
    # if child is a required field and you want write to child properties through parent
    # child = ChildSerializer(required=False)
    # otherwise the following should work (untested)
    # child = ChildSerializer() 

    child_id = serializers.PrimaryKeyRelatedField(
        queryset=Child.objects.all(), source='child', write_only=True)

    class Meta:
        model = Parent

Setting source=child lets child_id act as child would by default had it not be overridden (our desired behavior). write_only=True makes child_id available to write to, but keeps it from showing up in the response since the id already shows up in the ChildSerializer.

4
votes

There is a way to substitute a field on create/update operation:

class ChildSerializer(ModelSerializer):
    class Meta:
        model = Child

class ParentSerializer(ModelSerializer):
    child = ChildSerializer() 

    # called on create/update operations
    def to_internal_value(self, data):
         self.fields['child'] = serializers.PrimaryKeyRelatedField(
             queryset=Child.objects.all())
         return super(ParentSerializer, self).to_internal_value(data)

    class Meta:
        model = Parent
4
votes

A few people here have placed a way to keep one field but still be able to get the details when retrieving the object and create it with only the ID. I made a little more generic implementation if people are interested:

First off the tests:

from rest_framework.relations import PrimaryKeyRelatedField

from django.test import TestCase
from .serializers import ModelRepresentationPrimaryKeyRelatedField, ProductSerializer
from .factories import SomethingElseFactory
from .models import SomethingElse


class TestModelRepresentationPrimaryKeyRelatedField(TestCase):
    def setUp(self):
        self.serializer = ModelRepresentationPrimaryKeyRelatedField(
            model_serializer_class=SomethingElseSerializer,
            queryset=SomethingElse.objects.all(),
        )

    def test_inherits_from_primary_key_related_field(self):
        assert issubclass(ModelRepresentationPrimaryKeyRelatedField, PrimaryKeyRelatedField)

    def test_use_pk_only_optimization_returns_false(self):
        self.assertFalse(self.serializer.use_pk_only_optimization())

    def test_to_representation_returns_serialized_object(self):
        obj = SomethingElseFactory()

        ret = self.serializer.to_representation(obj)

        self.assertEqual(ret, SomethingElseSerializer(instance=obj).data)

Then the class itself:

from rest_framework.relations import PrimaryKeyRelatedField

class ModelRepresentationPrimaryKeyRelatedField(PrimaryKeyRelatedField):
    def __init__(self, **kwargs):
        self.model_serializer_class = kwargs.pop('model_serializer_class')
        super().__init__(**kwargs)

    def use_pk_only_optimization(self):
        return False

    def to_representation(self, value):
        return self.model_serializer_class(instance=value).data

The usage is like so, if you have a serializer somewhere:

class YourSerializer(ModelSerializer):
    something_else = ModelRepresentationPrimaryKeyRelatedField(queryset=SomethingElse.objects.all(), model_serializer_class=SomethingElseSerializer)

This will allow you to create an object with a foreign key still only with the PK, but will return the full serialized nested model when retrieving the object you created (or whenever really).

2
votes

I think the approach outlined by Kevin probably would be the best solution, but I couldn't ever get it to work. DRF kept throwing errors when I had both a nested serializer and a primary key field set. Removing one or the other would function, but obviously didn't give me the result I needed. The best I could come up with is creating two different serializers for reading and writing, Like so...

serializers.py:

class ChildSerializer(serializers.ModelSerializer):
    class Meta:
        model = Child

class ParentSerializer(serializers.ModelSerializer):
    class Meta:
        abstract = True
        model = Parent
        fields = ('id', 'child', 'foo', 'bar', 'etc')

class ParentReadSerializer(ParentSerializer):
    child = ChildSerializer()

views.py

class ParentViewSet(viewsets.ModelViewSet):
    serializer_class = ParentSerializer
    queryset = Parent.objects.all()
    def get_serializer_class(self):
        if self.request.method == 'GET':
            return ParentReadSerializer
        else:
            return self.serializer_class
2
votes

Here's how I've solved this problem.

serializers.py

class ChildSerializer(ModelSerializer):

  def to_internal_value(self, data):
      if data.get('id'):
          return get_object_or_404(Child.objects.all(), pk=data.get('id'))
      return super(ChildSerializer, self).to_internal_value(data)

You'll just pass your nested child serializer just as you get it from the serializer ie child as a json/dictionary. in to_internal_value we instantiate the child object if it has a valid ID so that DRF can further work with the object.

2
votes

There is a package for that! Check out PresentablePrimaryKeyRelatedField in Drf Extra Fields package.

https://github.com/Hipo/drf-extra-fields

1
votes

Based on the answers of both JPG and Bono, I came up with a solution that handles the OpenAPI Schema generator of DRF as well.

The actual field class is:

from rest_framework import serializers


class ModelRepresentationPrimaryKeyRelatedField(serializers.PrimaryKeyRelatedField):
    def __init__(self, **kwargs):
        self.response_serializer_class = kwargs.pop('response_serializer_class', None)
        if self.response_serializer_class is not None \
                and not issubclass(self.response_serializer_class, serializers.Serializer):
            raise TypeError('"serializer" is not a valid serializer class')

        super(ModelRepresentationPrimaryKeyRelatedField, self).__init__(**kwargs)

    def use_pk_only_optimization(self):
        return False if self.response_serializer_class else True

    def to_representation(self, instance):
        if self.response_serializer_class is not None:
            return self.response_serializer_class(instance, context=self.context).data
        return super(ModelRepresentationPrimaryKeyRelatedField, self).to_representation(instance)

The extended AutoSchema class is:

import inspect
from rest_framework.schemas.openapi import AutoSchema

from .fields import ModelRepresentationPrimaryKeyRelatedField


class CustomSchema(AutoSchema):
    def _map_field(self, field):
        if isinstance(field, ModelRepresentationPrimaryKeyRelatedField) \
                and hasattr(field, 'response_serializer_class'):
            frame = inspect.currentframe().f_back
            while frame is not None:
                method_name = frame.f_code.co_name
                if method_name == '_get_request_body':
                    break
                elif method_name == '_get_responses':
                    field = field.response_serializer_class()
                    return super(CustomSchema, self)._map_field(field)

                frame = frame.f_back

        return super(CustomSchema, self)._map_field(field)

Then on your Dganjo's project settings you can define this new Schema class to be used globally like:

REST_FRAMEWORK = {
    'DEFAULT_SCHEMA_CLASS': '<path_to_custom_schema>.CustomSchema',
}

Lastly from within your models you can use the new field type like:

class ExampleSerializer(serializers.ModelSerializer):
    test_field = ModelRepresentationPrimaryKeyRelatedField(queryset=Test.objects.all(), response_serializer_class=TestListSerializer)
1
votes

I started by implementing something similar to JPG's solution before I found this answer, and noticed that it breaks the built-in Django Rest Framework's templates. Now, that isn't such a big deal (as their solution works wonderfully via requests/postman/AJAX/curl/etc.), but if someone's new (like me) and wants the built-in DRF form to help them along the way, here's my solution (after cleaning it up and integrating some of JPG's ideas):

class NestedKeyField(serializers.PrimaryKeyRelatedField):
    def __init__(self, **kwargs):
        self.serializer = kwargs.pop('serializer', None)
        if self.serializer is not None and not issubclass(self.serializer, serializers.Serializer):
            raise TypeError('You need to pass a instance of serialzers.Serializer or atleast something that inherits from it.')
        super().__init__(**kwargs)

    def use_pk_only_optimization(self):
        return not self.serializer

    def to_representation(self, value):
        if self.serializer:
            return dict(self.serializer(value, context=self.context).data)
        else:
            return super().to_representation(value)

    def get_choices(self, cutoff=None):
        queryset = self.get_queryset()
        if queryset is None:
            return {}

        if cutoff is not None:
            queryset = queryset[:cutoff]

        return OrderedDict([
            (
                self.to_representation(item)['id'] if self.serializer else self.to_representation(item), # If you end up using another column-name for your primary key, you'll have to change this extraction-key here so it maps the select-element properly.
                self.display_value(item)
            )
            for item in queryset
        ])

and an example below, Child Serializer class:

class ChildSerializer(serializers.ModelSerializer):
    class Meta:
        model = ChildModel
        fields = '__all__'

Parent Serializer Class:

class ParentSerializer(serializers.ModelSerializer):
    same_field_name_as_model_foreign_key = NestedKeyField(queryset=ChildModel.objects.all(), serializer=ChildSerializer)
    class Meta:
        model = ParentModel
        fields = '__all__'
0
votes

I have been also stuck in the same situation. But what i have done that i have created two serializers for the following models as follow:

class Base_Location(models.Model):
    Base_Location_id = models.AutoField(primary_key = True)
    Base_Location_Name = models.CharField(max_length=50, db_column="Base_Location_Name")

class Location(models.Model):
    Location_id = models.AutoField(primary_key = True)
    Location_Name = models.CharField(max_length=50, db_column="Location_Name")
    Base_Location_id = models.ForeignKey(Base_Location, db_column="Base_Location_id", related_name="Location_Base_Location", on_delete=models.CASCADE)

This is my parent serializer

class BaseLocationSerializer(serializers.ModelSerializer):
    class Meta:
        model = Base_Location
        fields = "__all__"

I'm using this serializer only for get request so in response i got data with foreign key also because of nested serializer

class LocationSerializerList(serializers.ModelSerializer): <-- using for get request 
    Base_Location_id = BaseLocationSerializer() 

    class Meta:
        model = Location
        fields = "__all__"

Screenshot of get method request and response in postman

I'm using this serializer only for post request so while sending post request i do not need to include any additional information rather than primary key field value

class LocationSerializerInsert(serializers.ModelSerializer): <-- using for post request
    class Meta:
        model = Location
        fields = "__all__"

Screenshot of post method request and response in postman

0
votes

Here's what I'm using all over. This may be the simplest, most straight forward method which needs no hacks etc, and is directly using DRF without jumping thru hoops. Happy to hear disagreements with this approach.

In the view's perform_create (or equivalent), fetch the FK model database object corresponding to the field sent in the POST request, and then send that into the Serializer. The field in the POST request can be anything that can be used to filter and locate the DB object, need not be an ID.

This is documented here: https://www.django-rest-framework.org/api-guide/generic-views/#genericapiview

These hooks are particularly useful for setting attributes that are implicit in the request, but are not part of the request data. For instance, you might set an attribute on the object based on the request user, or based on a URL keyword argument.

def perform_create(self, serializer): serializer.save(user=self.request.user)

This method also has the advantage of maintaining parity between the read and write side, by not sending a nested representation for child in the response to the GET or POST.

Given the example posted by the OP:

class Child(models.Model):
    name = CharField(max_length=20)

class Parent(models.Model):
    name = CharField(max_length=20)
    phone_number = models.ForeignKey(PhoneNumber)
    child = models.ForeignKey(Child)

class ChildSerializer(ModelSerializer):
    class Meta:
        model = Child

class ParentSerializer(ModelSerializer):
    # Note this is different from the OP's example. This will send the
    # child name in the response
    child = serializers.ReadOnlyField(source='child.name')

    class Meta:
        model = Parent
        fields = ('name', 'phone_number', 'child')

In the View's perform_create:

class SomethingView(generics.ListCreateAPIView):
    serializer_class = ParentSerializer
    
    def perform_create(self, serializer):
        child_name = self.request.data.get('child_name', None)
        child_obj = get_object_or_404(Child.objects, name=child_name)
        serializer.save(child=child_obj)

PS: Please note that I've not tested this above snippet, however its based on a pattern I'm using in many places so it should work as is.