How do I integrate a custom input argument into a Graphene DjangoFilterConnectionField queryset?

Question

I'm wondering what the cleanest way to integrate a Graphene DjangoFilterConnectionField custom input argument into the default queryset is?

I'm iterating on the following Graphene Query class whose custom input attribute attempts to integrate the PgVector CosineDistance function:

class Query(graphene.ObjectType):

    all_the_small_things = DjangoFilterConnectionField(
        VectorNode,
        embedding=graphene.List(graphene.Float),
    )

    def resolve_all_the_small_things(
        root,
        info,
        **kwargs,
    ):
        from pgvector.django import CosineDistance

        annotations = {}

        if kwargs.get("embedding"):
            annotations["embedding"] = CosineDistance(
                "embedding", kwargs.pop("embedding")
            )

        if annotations:
            # How best to integrate this into the existing behavior?
            # VectorNode.objects.annotate(**annotations)
            # now use filter functionality provided by
            # the default resolver.

I would like to augment the default query behavior (i.e. filtering) if the custom input is provided and use the default behavior as-is if it isn't. Do I need to implement my own queryset and/or filterset? Is there a way to reflect/modify the default behavior using the function arguments?

Note: I think the long term answer to this is to add support for PG Vector columns to Graphene and/or creating a django-filters filter subclass but I really need to understand how this all comes together before seriously thinking about how to approach any of that upstream.

score -1 · Answer 1 · edited Jun 08 '23 at 14:45

To integrate a custom input argument into the default queryset of a Graphene DjangoFilterConnectionField:

Create a Custom FilterSet

Override the get_queryset method:

from django_filters import FilterSet, filters
from graphene_django.filter import DjangoFilterConnectionField

class VectorFilterSet(FilterSet):
    cosine_distance = filters.Float()  # Define your custom input argument here

    class Meta:
        model = VectorNode
        fields = []  # Specify other fields for filtering if needed

     class Query(graphene.ObjectType):

    all_the_small_things = DjangoFilterConnectionField(
        VectorNode,
        filterset_class=VectorFilterSet,  # Use your custom FilterSet
        cosine_distance=graphene.Float(),  # Add the custom input argument to the Query field
    )

    def resolve_all_the_small_things(
        root,
        info,
        **kwargs,
    ):
        from pgvector.django import CosineDistance

        queryset = VectorNode.objects.all()

        cosine_distance = kwargs.pop("cosine_distance", None)
        if cosine_distance is not None:
            queryset = queryset.annotate(
                cosine_distance=CosineDistance("embedding", cosine_distance)
            )

        return queryset

If your requirements become more complex or you need more fine-grained control over the filtering behavior, you may consider implementing a custom resolver or creating a custom connection Field that extends DjangoFilterConnectionField.

That code's indentation is totally out of whack (I only added (equal) leading space to make it format properly (and deleted trailing space)). This is Python code. — Peter Mortensen, Jun 08 '23 at 14:48
Are you posting ChatGPT-generated answers to a lot of bounty questions? — Peter Mortensen, Jun 08 '23 at 14:49
I have worked on seniors before that's why I just copy my code just edit them and pasted. — Hasan Raza, Jun 09 '23 at 06:36

How do I integrate a custom input argument into a Graphene DjangoFilterConnectionField queryset?

1 Answers1