I had many time the need to optimise some performance issues with some ORM queries, then many times the prefetch_related and select_related were my best friends for this topic. But when calling some business logic which reside in some model's methods which try to calculate or to fetch some related model data the optimisation will not have any effect as these method are using filter or some queryset aggregations Sum or Count ...

The solution to this problem is either to redefine the logic to use prefetched data or more clean to make these methods now if there already pre-feteched objects, that's the best solution in my cases:

A basic models example models.py:

    class Teacher(models.Model):
        name = models.CharField(max_length=255)

        def students_count(self):
            return self.student_set.filter(age__gte=20).count()

    class Student(models.Model):
        name = models.CharField(max_length=255)
        teacher = models.ForeignKey(Teacher)
        age = models.IntegerField(default=20)

    for t in Teacher.objects.all():
        print(teacher.students_count())
    # This will results to the number of teachers + 1 SQL queries

Let's use prefetch_related:

    for t in Teacher.objects.all().prefetch_related('student_set'):
        print(teacher.students_count())
    # Our prefetch have no effect as the method students_count is using filter which
    # will ignore the prefetched objects

We have to fix the method students_count to consider checking for prefetched objects:

    class Teacher(models.Model):
        name = models.CharField(max_length=255)

        def students_count(self):
            if hasattr(self, '_prefetched_objects_cache') and 'student' in self._prefetched_objects_cache:
                return len([x for x in self.student_set.all() if x.age >= 20])
            return self.student_set.filter(age__gte=20).count()

This way even calling students_count in a single instance without calling prefetch_related on the queryset will use the filter instead of the prefetched, this can be useful in many cases but I better to use this only when you need to.

Thanks for reading