Django REST Framework ModelSerializer get_or_create functionality

Question

When I try to deserialize some data into an object, if I include a field that is unique and give it a value that is already assigned to an object in the database, I get a key constraint error. This makes sense, as it is trying to create an object with a unique value that is already in use.

Is there a way to have a get_or_create type of functionality for a ModelSerializer? I want to be able to give the Serializer some data, and if an object exists that has the given unique field, then just return that object.

Answer 1

In my experience nmgeek's solution won't work in DRF 3+ as serializer.is_valid() correctly honors the model's unique_together constraint. You can work around this by removing the UniqueTogetherValidator and overriding your serializer's create method.

class MyModelSerializer(serializers.ModelSerializer):

    def run_validators(self, value):
        for validator in self.validators:
            if isinstance(validator, validators.UniqueTogetherValidator):
                self.validators.remove(validator)
        super(MyModelSerializer, self).run_validators(value)

    def create(self, validated_data):
        instance, _ = models.MyModel.objects.get_or_create(**validated_data)
        return instance

    class Meta:
        model = models.MyModel

Answer 2

The Serializer restore_object method was removed starting with the 3.0 version of REST Framework.

A straightforward way to add get_or_create functionality is as follows:

class MyObjectSerializer(serializers.ModelSerializer):
    class Meta:
        model = MyObject
        fields = (
                  'unique_field',
                  'other_field',
                  )

    def get_or_create(self):
        defaults = self.validated_data.copy()
        identifier = defaults.pop('unique_field')
        return MyObject.objects.get_or_create(unique_field=identifier, defaults=defaults)

def post(self, request, format=None):
    serializer = MyObjectSerializer(data=request.data)
    if serializer.is_valid():
        instance, created = serializer.get_or_create()
        if not created:
            serializer.update(instance, serializer.validated_data)
        return Response(serializer.data, status=status.HTTP_202_ACCEPTED)
    return Response(serializer.errors, status=status.HTTP_400_BAD_REQUEST)

However, it doesn't seem to me that the resulting code is any more compact or easy to understand than if you query if the instance exists then update or save depending upon the result of the query.

Answer 3

@Groady's answer works, but you have now lost your ability to validate the uniqueness when creating new objects (UniqueValidator has been removed from your list of validators regardless the cicumstance). The whole idea of using a serializer is that you have a comprehensive way to create a new object that validates the integrity of the data you want to use to create the object. Removing validation isn't what you want. You DO want this validation to be present when creating new objects, you'd just like to be able to throw data at your serializer and get the right behavior under the hood (get_or_create), validation and all included.

I'd recommend overwriting your is_valid() method on the serializer instead. With the code below you first check to see if the object exists in your database, if not you proceed with full validation as usual. If it does exist you simply attach this object to your serializer and then proceed with validation as usual as if you'd instantiated the serializer with the associated object and data. Then when you hit serializer.save() you'll simply get back your already created object and you can have the same code pattern at a high level: instantiate your serializer with data, call .is_valid(), then call .save() and get returned your model instance (a la get_or_create). No need to overwrite .create() or .update().

The caveat here is that you will get an unnecessary UPDATE transaction on your database when you hit .save(), but the cost of one extra database call to have a clean developer API with full validation still in place seems worthwhile. It also allows you the extensibility of using custom models.Manager and custom models.QuerySet to uniquely identify your model from a few fields only (whatever the primary identifying fields may be) and then using the rest of the data in initial_data on the Serializer as an update to the object in question, thereby allowing you to grab unique objects from a subset of the data fields and treat the remaining fields as updates to the object (in which case the UPDATE call would not be extra).

Note that calls to super() are in Python3 syntax. If using Python 2 you'd want to use the old style: super(MyModelSerializer, self).is_valid(**kwargs)

from django.core.exceptions import ObjectDoesNotExist, MultipleObjectsReturned


class MyModelSerializer(serializers.ModelSerializer):

    def is_valid(self, raise_exception=False):
        if hasattr(self, 'initial_data'):
            # If we are instantiating with data={something}
            try:
                # Try to get the object in question
                obj = Security.objects.get(**self.initial_data)
            except (ObjectDoesNotExist, MultipleObjectsReturned):
                # Except not finding the object or the data being ambiguous
                # for defining it. Then validate the data as usual
                return super().is_valid(raise_exception)
            else:
                # If the object is found add it to the serializer. Then
                # validate the data as usual
                self.instance = obj
                return super().is_valid(raise_exception)
        else:
            # If the Serializer was instantiated with just an object, and no
            # data={something} proceed as usual 
            return super().is_valid(raise_exception)

    class Meta:
        model = models.MyModel

Answer 4

There are a couple of scenarios where a serializer might need to be able to get or create Objects based on data received by a view - where it's not logical for the view to do the lookup / create functionality - I ran into this this week.

Yes it is possible to have get_or_create functionality in a Serializer. There is a hint about this in the documentation here: http://www.django-rest-framework.org/api-guide/serializers#specifying-which-fields-should-be-write-only where:

• restore_object method has been written to instantiate new users.
• The instance attribute is fixed as None to ensure that this method is not used to update Users.

I think you can go further with this to put full get_or_create into the restore_object - in this instance loading Users from their email address which was posted to a view:

class UserFromEmailSerializer(serializers.ModelSerializer):
    class Meta:
        model = get_user_model()
        fields = [
            'email',
        ]

    def restore_object(self, attrs, instance=None):
        assert instance is None, 'Cannot update users with UserFromEmailSerializer'

        (user_object, created) = get_user_model().objects.get_or_create(
            email=attrs.get('email')
        )

        # You can extend here to work on `user_object` as required - update etc.

        return user_object

Now you can use the serializer in a view's post method, for example:

def post(self, request, format=None):

    # Serialize "new" member's email
    serializer = UserFromEmailSerializer(data=request.DATA)

    if not serializer.is_valid():
        return Response(serializer.errors,
                        status=status.HTTP_400_BAD_REQUEST)

    # Loaded or created user is now available in the serializer object:
    person=serializer.object
    # Save / update etc.