Django Virtual Models
Improve performance and maintainability with a prefetching layer in your Django / Django REST Framework project
Django Virtual Models introduces a new "prefetching layer" to Django codebases that assists developers to express complex read logic without sacrificing maintainability, composability and performance. A Virtual Model allows developers to declare all nesting they need along with all annotations, prefetches, and joins in a single declarative class.
When implementing Django REST Framework serializers, developers need to be careful to avoid causing the N+1 selects problem due to missing
select_related calls on the associated queryset. Additionaly, developers must not miss
annotate calls for fields that are computed at queryset-level.
With Virtual Models integration with DRF, if you change a DRF Serializer, you won't forget to modify the associated queryset with additional annotations, prefetches, and joins. If you do forget to update the queryset, Django Virtual Models will guide you by raising friendly exceptions to assist you to write the correct Virtual Model for the serializer you're changing. This guidance will prevent N+1s and missing annotations in all serializers that use Virtual Models.
For example, imagine if you have following nested serializers starting from
from movies.models import Nomination, Person, Movie class AwardSerializer(serializers.ModelSerializer): class Meta: model = Nomination fields = ["award", "category", "year", "is_winner"] class PersonSerializer(serializers.ModelSerializer): awards = AwardSerializer(many=True) nomination_count = serializers.IntegerField(read_only=True) class Meta: model = Person fields = ["name", "awards", "nomination_count"] class MovieSerializer(serializers.ModelSerializer): directors = PersonSerializer(many=True) class Meta: model = Movie fields = ["name", "directors"]
For good performance and correct functionality, all nested serializers must have a corresponding
prefetch_related on the queryset used by
MovieSerializer. Also, the
nomination_count field should be
annotated on it. Therefore, you'll need to write this complex chain of nested prefetches:
from django.db.models import Prefetch awards_qs = Nomination.objects.filter(is_winner=True) directors_qs = Person.objects.prefetch_related( Prefetch( "nominations", queryset=awards_qs, to_attr="awards" ) ).annotate( nomination_count=Count("nominations") ).distinct() qs = Movie.objects.prefetch_related( Prefetch( "directors", queryset=directors_qs ) )
Conversely, you can declare Virtual Models for this read logic to easily reuse and customize those classes in multiple places of the codebase:
import django_virtual_models as v class VirtualAward(v.VirtualModel): class Meta: model = Nomination def get_prefetch_queryset(self, **kwargs): return Nomination.objects.filter(is_winner=True) class VirtualPerson(v.VirtualModel): awards = VirtualAward(lookup="nominations") nomination_count = v.Annotation( lambda qs, **kwargs: qs.annotate( nomination_count=Count("nominations") ).distinct() ) class Meta: model = Person class VirtualMovie(v.VirtualModel): directors = VirtualPerson() class Meta: model = Movie
To configure your DRF view and serializer to use Virtual Models, inherit from the proper classes:
import django_virtual_models as v class MovieSerializer(v.VirtualModelSerializer): ... class Meta: ... virtual_model = VirtualMovie class MovieList(v.VirtualModelListAPIView): queryset = Movie.objects.all() serializer_class = MovieSerializer ...
Then the library will automatically do the right prefetches and annotations for you!
If, for example, you forget to add the
nomination_count field on
VirtualPerson, the following exception will appear when using
If you aren't using DRF serializers, you hydrate your queryset with virtual fields manually:
qs = VirtualMovie().get_optimized_queryset( Movie.objects.all(), lookup_list=[ "directors__awards", "directors__nomination_count", ] )