Django-rest-framework: Unique constraint prevents nested serializers from updating

[xordoquy]

If you have a nested serializer which model does have a unique constraint on a field you're not able to reference an existing instance.

class Task(serializers.ModelSerializer):
    class Meta:
        model = models.Task
        fields = ('id', 'name')


class Category(serializers.ModelSerializer):
    tasks = Task(many=True)
    class Meta:
        model = models.Category
        fields = ('id', 'name', 'tasks')

Let's say I created a category & task which I'm updating with the very same data they already have:

c = Category(data={"id": 1, "name": "Django", "tasks": [{"id": 1, "name": "Demo", "owner": "admin"}]})
c.is_valid()  # Returns False
c.errors  # Returns {'tasks': [{'name': ['This field must be unique.']}]}

The point is, name remains unique because it matches its own id

[tomchristie]

It may be that we simply have to document this as a limitation in validators.
There's not really going to be any nice automatic ways to handle this.

[xordoquy]

I'm a bit uncomfortable with this.
Let's way our nested resource is either created or fetched from the DB ala get_or_create.
If we remove the validation, it may run in later issues.
If we keep it, it'll be impossible to get an item that already exists without changing it.

[tomchristie]

We don't support (automatic) writable nested serialization, so it's not totally unreasonable to also say 'validators' are not appropraite for this case, here's what you'd need to do. However if there's an alternative that'd def be worth considering.

[arw180]

This sounds very similar to the issue I ran into today: http://stackoverflow.com/questions/32123148/writable-nested-serializer-with-existing-objects-using-django-rest-framework-3-2

[xordoquy]

@arw180 indeed. Thanks for pointing to this issue, totally forgot about it.
Adding a reference to #3312 since it's the same.

[irongomme]

Arrgh, i've search for a day why was it not works!!!

This is a really confusing issue, because it is a common thing to play with nested models, and i don't know how to get around of this problem ...

I've got this :

class Contact(models.Model):
    account = models.OneToOneField(Account, null=True)
    first_name = models.CharField(max_length=40, blank=True)
    last_name = models.CharField(max_length=40, blank=True)
    phone = PhoneNumberField(blank=True, null=True)
    phone_alt = PhoneNumberField(blank=True, null=True)
    birthday = models.DateField(blank=True, null=True)
    comment = models.TextField(blank=True, null=True)
    created = models.DateTimeField(auto_now_add=True)
    modified = models.DateTimeField(auto_now=True)
    is_musician = models.BooleanField(default=False)
    is_volunteer = models.BooleanField(default=False)
    is_active = models.BooleanField(default=True)

class Account(AbstractBaseUser):
    email = models.EmailField(unique=True)
    tagline = models.CharField(max_length=140, blank=True)
    is_admin = models.BooleanField(default=False)
    is_staff = models.BooleanField(default=False)
    created_at = models.DateTimeField(auto_now_add=True)
    updated_at = models.DateTimeField(auto_now=True)

    objects = AccountManager()

    USERNAME_FIELD = 'email'

class ContactSerializer(serializers.ModelSerializer):
    account = AccountContactSerializer()

    class Meta:
        model = Contact
        read_only_fields = ('created', 'modified',)

def create(self, validated_data):
        account_data = validated_data.pop('account')
        account = Account.objects.create_user(**account_data)
        contact = Contact.objects.create(account=account, **validated_data)

        return contact

class AccountContactSerializer(serializers.ModelSerializer):
    password = serializers.CharField(write_only=True, required=False)

    class Meta:
        model = Account
        fields = ('id', 'email', 'password',)

    def create(self, validated_data):
        return Account.objects.create(**validated_data)

And i must save a contact with an account, they are inseparable. How can i do this without this annoying unique validator message ?

[xordoquy]

The discussion group is the best place to take this discussion and other usage questions. Thanks!

[ankitml]

So what is the way out? This to me seems like a major limitation.

1. Writing custom non-model serializer. - Almost all my endpoints would have custom serializers then.
2. Ignoring 'id' from 'to_internal_value' method.* - Might run into trouble later on because of this.
3. Any way to use model serializer but change the validation rule for nested serializr? some sort of mixin.. This would be an ideal way, not sure how to do it.
4. check issue no 2403.

[xordoquy]

The workaround this is to remove the unique constraint from the serializer and move it to the business logic validation part since the nested serializer can not say - as of today - whether he's updating or creating a new object.

[BurkovBA]

Guys, I've run into this problem, too.

Just wanted to say that (unfortunately) without a generic WritableNestedModelSerializer out-of-the-box, Django REST framework is hardly useful. With hyperlinked APIs only, you often have to create chains of 3-4-5 consecutive POST/PUT ajax requests, which is waste of time. Writing one from scratch is problematic, repetitive and requires good knowledge of DRF internals. Much more importantly, if you want to expose your API to domain area specialists (e.g. bioinformaticians), they won't buy Hyperlinked one. They want a clear API with nested data structures and, to be honest, they are right.

I''ve been studying the details of this for a couple of days now, got stuck on this error and not sure, whether I should continue messing with DRF internals or stop (breaking the deadlines) investing time into this and start learning Loopback.

I'm pretty sure that with advent of update_or_create function in Django 1.7 it's possible to think out a generic way of nesting writable serializers.

[dedsm]

but can't the nested serializer figure out if it's updating an instance by checking parent.instance?

[xordoquy]

@dedsm sure, but that would also break cases where the uniqueness constraint is broken on the "main" serializer and end up in 500 errors with database errors.

[getup8]

Sorry I can't really tell from this thread; was this issue fixed?

I'm trying to update a ManyToMany Field via nested serializer and it gets stuck on run_validators() because the relation object I'm trying to attach to my main object already exists.

Would be nice to at least put the recommended workaround (with a few specifics) for a pretty common use case like this.

[xordoquy]

The workaround this is to remove the unique constraint from the serializer and move it to the business logic validation part since the nested serializer can not say - as of today - whether he's updating or creating a new object.

[ColinBarnwell]

I recently came across this issue (a user object attached to an account object via an intermediary relationship).

Within the context of the account, I wanted to be able to update both the user details and the relationship details as a single endpoint. The problem being that the email field on the user model is unique and fell foul of the issue described here.

I wasn't happy with breaking some very necessary validation, so I took a more direct approach. At the point of serializer initialisation, if the email supplied in the data is _already the same_ as the email on the user, I simply removed it from the data. No change = no validation required*.

def __init__(self, instance, data=empty, **kwargs):
    if (
        data is not empty and
        'details.email' in data and
        instance.user is not None and
        instance.user.email == data['details.email']
    ):
        del data['details.email']
    super(AccountUserSerializer, self).__init__(instance, data=data, **kwargs)

• 
  
  • Okay, technically you could change the email validator and require user's with no-longer-valid emails to change them, but that seems a remote possibility in the extreme.
  
  

[Ostlander]

@xordoquy, would that work?

class MySerializer(serializers.ModelSerializer):
    nested = NestedSerializer()
    ...

    def get_fields(self):
        fields = super().get_fields()

        if self.instance is not None:
            nested_serializer = copy.deepcopy(fields.pop('nested'))
            nested_serializer.instance = self.instance.nested
            fields['nested'] = nested_serializer
        return fields

[xordoquy]

The discussion group is the best place to take further this discussion and other usage questions. Thanks!

[xtornasol512]

I don't kwon if there is a documentation for this now, but i recently fix this problem with a basic zero validators on the field that cause the problem.
I took a lot of time discovering this , so I hope some one else help it
youruniquefield = serializers.CharField(max_length=255, min_length=4, validators=[])
Before this I always raise a already exist error even if I correct the problem with managers to use ger_or_create.

[xordoquy]

I wrote some time ago an article about that if that helps: https://medium.com/django-rest-framework/dealing-with-unique-constraints-in-nested-serializers-dade33b831d9

[m-ali-94]

there is this 3rd party library https://github.com/beda-software/drf-writable-nested, it has a UniqueFieldsMixin which can do the magic to handle unique constraints.

[studioj]

for those ending up here => this SO answer helps
https://stackoverflow.com/a/32452603/2559785