django-hierarchical-models

This package provides an abstract Django model which supports hierarchical data. The implementation is an adjacency list, which is rather naive, but actually has higher performance in this scenario than other implementations such as path enumeration or nested sets because those implementations store more data with each instance which must be updated before almost every operation, effectively doubling (or more) database queries and killing performance. The performance of this implementation actually holds up pretty well at large numbers of instances.

Usage

from django.db import models
from django_hierarchical_models.models import HierarchicalModel

class MyModel(HierarchicalModel):
    name = models.CharField(max_length=100)

...

child = MyModel.objects.create(name="Betty")
child.parent  # None

parent = MyModel.objects.create(name="Simon")
# checks for pesky cycles
child.set_parent(parent)
# alternative
# child.parent = parent
child.parent  # <MyModel: "Simon">

child.root()  # <MyModel: "Simon">
parent.root()  # <MyModel: "Simon">

parent.direct_children()  # [<MyModel: "Betty">]

child.is_child_of(parent)  # True
parent.is_child_of(child)  # False

parent = vs .set_parent()

parent is a ForeignKeyField which may be directly accessed or set. The .set_parent() method checks to see if the operation would create a cycle, which can be bad for some of the other instance methods. The .set_parent() method is slower because it must determine if a cycle would be formed. .set_parent() makes a call to .save(update_fields=("parent",)), so it is not necessary to call .save() after updating the parent this way.

Refreshing from database

The following is expected behavior:

instance_1 = MyModel.objects.create(name="Betty")
instance_2 = MyModel.objects.create(parent=instance_1, name="Simon")
instance_2.parent  # <MyModel: "Betty">

instance_1.delete()

instance_2.parent  # <MyModel: "Betty">

instance_2.refresh_from_db()

instance_2.parent  # None

instance_1 = MyModel.objects.create(name="Betty")
instance_2 = MyModel.objects.create(parent=instance_1, name="Simon")
instance_3 = MyModel.objects.create(parent=instance_2, name="Finn")
instance_3_copy = MyModel.objects.get(pk=instance_3.pk)

instance_1.root()  # <MyModel: "Betty">
instance_2.root()  # <MyModel: "Betty">
instance_3.root()  # <MyModel: "Betty">
instance_3_copy.root()  # <MyModel: "Betty">

instance_2.set_parent(None)

instance_1.root()  # <MyModel: "Betty">
instance_2.root()  # <MyModel: "Simon">
instance_3.root()  # <MyModel: "Simon">
instance_3_copy.root()  # <MyModel: "Betty">

instance_3_copy.refresh_from_db()

instance_1.root()  # <MyModel: "Betty">
instance_2.root()  # <MyModel: "Simon">
instance_3.root()  # <MyModel: "Simon">
instance_3_copy.root()  # <MyModel: "Simon">

Moral of the story, if your instance's parent might have been edited/deleted, you will want to refresh your instance for that change to be reflected.

Benchmarks

The following benchmarks demonstrate that the query performance of the model stays the same from 10,000 to 1,000,000 models. These tests were done with Postgres. The results are in the form total time (s) / per instance (ms). Eventually the query performance of this model should scale down with the total number of instances in the database, but it appears up to these scales those effects are insignificant compared to other overhead.