Django REST framework (DRF) serializer speedup

drf, cache, performance
pip install drf-serializer-cache==0.3.4


Django REST framework serializers speedup

PyPI version Build Status Coverage Status

Simple short-living cache implementation for DRF serializers. It solves two issues with DRF serializers:

  • Serializer.fields property can be extremely slow in some cases. Such cases include complex ModelSerializer, serializer hierarchies with repeated serializers and recursive serializers. fields is cached and computed once per class for each serialization run. This leads to limitation - fields should not be dynamically adjusted inside of serializer, which inherits from SerializerCacheMixin;
  • By default serializers will re-compute object representation event if they encounter same object with same serializer twice. This can turn into an issue, if multiple object include same dependent hard-to-serialize objects. Cache avoid re-computing instance representation if it encounters same instances multiple times.

Performance results

Performance results

Cache life-time and invalidation

Cache is designed to be simple and non-error-prone. Cache is created when top-level serializer to_representation starts and is cleaned up once this method is finished. Thus there is no need for timeouts or complex cache invalidation. No data is shared between requests. Thus, following cases will work fine:

user = User.objects.get(name='Lee')
data_1 = UserSerializer(user).data
# OrderedDict([('name', 'Lee')])
user.name = 'Bob'
data_2 = UserSerializer(user).data
# OrderedDict([('name', 'Bob')])


Usage should be pretty simple - inject SerializerCacheMixin before inheriting from Serializer or ModelSerializer:

from drf_serializer_cache import SerializerCacheMixin
from rest_framework import serializers

class UserSerializer(SerializerCacheMixin, serializers.ModelSerializer):
    class Meta:
        model = User
        fields = ('id', 'name')

Common pitfalls

Too often cache cleanup

Cache lives in serializer hierarchy root, but it's life-time is defined by nearest tree node, which inherits from SerializerCacheMixin. Ideal situation is when root inherits from SerializerCacheMixin. That's why SerializerCacheMixin uses custom list_serializer_class, which also inherits from SerializerCacheMixin. If you use custom list as root of serializer hierarchy - it's recommended to use SerializerCacheMixin as it's base class.

Too many hierarchies

By default serializers build nice hierarchy by calling bind on their child serializers. This case can be broken for serializers.SerializerMethodField if it uses some serializer without calling bind on it:

from drf_serializer_cache import SerializerCacheMixin
from rest_framework import serializers

class ResultSerializer(SerializerCacheMixin, serializers.Serializer):
    results = serializers.SerializerMethodField()

    def get_results(self, instance):
        # recursive serializer
        serializer = self.__class__(instance.results, many=True)
        serializer.bind('*', self)  # bind call is essential for efficient cache !
        return serializer.data