Prefetching indirectly related items using Django ORM - django

Question

I'm trying to optimize the queries for my moderation system, build with Django and DRF.
I'm currently stuck with the duplicates retrieval: currently, I have something like
class AdminSerializer(ModelSerializer):
duplicates = SerializerMethodField()
def get_duplicates(self, item):
if item.allowed:
qs = []
else:
qs = Item.objects.filter(
allowed=True,
related_stuff__language=item.related_stuff.language
).annotate(
similarity=TrigramSimilarity('name', item.name)
).filter(similarity__gt=0.2).order_by('-similarity')[:10]
return AdminMinimalSerializer(qs, many=True).data
which works fine, but does at least one additional query for each item to display. In addition, if there are duplicates, I'll do additional queries to fill the AdminMinimalSerializer, which contains fields and related objects of the duplicated item. I can probably reduce the overhead by using a prefetch_related inside the serializer, but that doesn't prevent me from making several queries per item (assuming I have only one related item to prefetch in AdminMinimalSerializer, I'd still have ~2N + 1 queries: 1 for the items, N for the duplicates, N for the related items of the duplicates).
I've already looked at Subquery, but I can't retrieve an object, only an id, and this is not enough in my case. I tried to use it in both a Prefetch object and a .annotate.
I also tried something like Item.filter(allowed=False).prefetch(Prefetch("related_stuff__language__related_stuff_set__items", queryset=Items.filter..., to_attr="duplicates")), but the duplicates property is added to "related_stuff__language__related_stuff_set", so I can't really use it...
I'll welcome any idea ;)
Edit: the real code lives here. Toy example below:
# models.py
from django.db.models import Model, CharField, ForeignKey, CASCADE, BooleanField
class Book(Model):
title = CharField(max_length=250)
serie = ForeignKey(Serie, on_delete=CASCADE, related_name="books")
allowed = BooleanField(default=False)
class Serie(Model):
title = CharField(max_length=250)
language = ForeignKey(Language, on_delete=CASCADE, related_name="series")
class Language(Model):
name = CharField(max_length=100)
# serializers.py
from django.contrib.postgres.search import TrigramSimilarity
from rest_framework.serializers import ModelSerializer, SerializerMethodField
from .models import Book, Language, Serie
class BookAdminSerializer(ModelSerializer):
class Meta:
model = Book
fields = ("id", "title", "serie", "duplicates", )
serie = SerieAdminAuxSerializer()
duplicates = SerializerMethodField()
def get_duplicates(self, book):
"""Retrieve duplicates for book"""
if book.allowed:
qs = []
else:
qs = (
Book.objects.filter(
allowed=True, serie__language=book.serie.language)
.annotate(similarity=TrigramSimilarity("title", book.title))
.filter(similarity__gt=0.2)
.order_by("-similarity")[:10]
)
return BookAdminMinimalSerializer(qs, many=True).data
class BookAdminMinimalSerializer(ModelSerializer):
class Meta:
model = Book
fields = ("id", "title", "serie")
serie = SerieAdminAuxSerializer()
class SerieAdminAuxSerializer(ModelSerializer):
class Meta:
model = Serie
fields = ("id", "language", "title")
language = LanguageSerializer()
class LanguageSerializer(ModelSerializer):
class Meta:
model = Language
fields = ('id', 'name')
I'm trying to find a way to prefetch related objects and duplicates so that I can get rid of the get_duplicates method in BookSerializer, with the N+1 queries it causes, and have only a duplicates field in my BookSerializer.
Regarding data, here would be an expected output:
[
{
"id": 2,
"title": "test2",
"serie": {
"id": 2,
"language": {
"id": 1,
"name": "English"
},
"title": "series title"
},
"duplicates": [
{
"id": 1,
"title": "test",
"serie": {
"id": 1,
"language": {
"id": 1,
"name": "English"
},
"title": "first series title"
}
}
]
},
{
"id": 3,
"title": "random",
"serie": {
"id": 3,
"language": {
"id": 1,
"name": "English"
},
"title": "random series title"
},
"duplicates": []
}
]