python - How might I join two unrelated Django models via distance?

Question

I'm working on a Django project that requires me to report, via web service, for each item in the School table all Hospitals that just so happen to be a set distance away. Here is a dummy implementation of the two models:

class School(models.Model):
    location = models.PointField(srid=4326, geography=True)
    ...some other fields...
    objects = models.GeoManager()

class Hospital
    location = models.PointField(srid=4326, geography=True)
    ...some other fields...
    objects = models.GeoManager()

I will state for the record that this is a contrived example, but the point remains that:

1. Neither model has a foreign key relationship with the other (nor should they)
2. The only relationship is via distance

Now, I am using Django Rest Framework (DRF) to render my requests, and I am using the Serializer classes that it provides to do something along these lines:

class SchoolSerializer(serializers.ModelSerializer):

    nearby_hospitals = serializers.SerializerMethodField('get_nearby_hospitals')

    class Meta:
        model = School
        fields = ('location', 'nearby_hospitals',)

    def get_nearby_hospitals(self, obj):

        geom = obj.location

        try:

            locations = Hospitals.objects.filter(loc__dwithin=(geom, 10000))
            return HospitalSerializer(locations, many=True).data
        except:
            return

This works but it is not efficient. In essence what happens is that DRF loads all Schools, and then from there loops over each school and runs the query in get_nearby_hospitals. The number of database queries is equal to the number of items in Schools plus one (to get the Schools).

Ideally I'd like a Django solution that performs something along these lines (obviously with the fields listed out, aliases for clashing ID column names, etc):

SELECT * FROM schools JOIN 
hospitals ON ST_DWithin(schools.location, hospitals.location, 10000)

The above query produces the intersection of all Schools and Hospitals within the proper distance. I could merge these results in manually with the School.objects.all() queryset or write a better query and call the raw QuerySet method to get what I'm after in one shot.

Is there a better or more 'Django Way' solution to this problem?

Answer 1

我建议使用 GraphQL/DataLoaders 来解决这个 N+1 问题，因为 DRF 不太适合它。但是，这可以在 DRF 中通过使用自定义 ListSerializer（django 在幕后自动为您创建一个）并重载两个 List/Child 序列化器的 to_representation 函数来完成：

class SchoolListSerializer(serializers.ListSerializer):
    def to_representation(self, data):
        iterable = data.all() if isinstance(data, models.Manager) else data
        keys = [(item.id, item.location) for item in iterable]
        # perform your bulk nearby query here using the id/locations
        # the result should be a dict with school id keys and list of serialized hospital values
        # {1: [hospital.to_dict(), hospital.to_dict(), ...], ...}
        nearby_by_id = ...
        extra = {
            'nearby_hospitals_by_id': nearby_by_id
        }
        return [
            self.child.to_representation(item, **extra) for item in iterable
        ]

class SchoolSerializer(serializers.ModelSerializer):

    class Meta:
        model = School
        fields = ('location', )
        list_serializer_class = SchoolListSerializer

    def to_representation(self, instance, **extra):
        response = super().to_representation(instance)
        response['nearby_hospitals'] = extra['nearby_hospitals_by_id'][instance.id]
        return response

我省略了实际的数据库查询，但是无论数据位于何处，该示例都适用于任何查询。

Answer 2

我不确定它会更好，但我认为你正在反方向解决问题。我认为根本没有任何加入的理由。

class SchoolSerializer(serializers.ModelSerializer):

    nearby_hospitals = serializers.SerializerMethodField('get_nearby_hospitals')

    class Meta:
        model = School
        fields = ('location', 'nearby_hospitals',)

    def get_nearby_hospitals(self, obj):

        geom = obj.location

        try:
            # why do you query School model if you want nearby hospitals?
            locations = Hospital.objects.filter(location__distance_lte=(geom ,10000))
            return locations # not sure it's the good type to return though
        except:
            return