django-chronicle 0.2.0

efficient model history using database triggers

Django Chronicle is an implementation of the slowly changing dimensions type 4 which uses database triggers.

How to use?

1. Create a custom revision model. e.g.

   from chronicle.models import AbstractRevision
   
   class Revision(AbstractRevision):
       user = models.ForeignKey(settings.AUTH_USER_MODEL)
       created = models.DateTimeField(auto_now_add=True)
   

2. Set settings.REVISION_MODEL to point to your revision model. e.g.

   REVISION_MODEL = 'revision.Revision'
   

3. Let your models inherit from HistoryMixin e.g.

   from chronicle.models import HistoryMixin
   from django.db import models
   
   class Food(HistoryMixin, models.Model):
       name = models.CharField(max_length=50)
   

4. Create all the migrations and run them:

   $ manage.py makemigrations
   $ manage.py migrate
   

   That should create all the _history tables for your models that inherit from the HistoryMixin.

5. Create the database triggers

   $ manage.py create_history_triggers
   

Now every change to your models should be logged in the _history tables and you can access the model history via the History model which becomes a field of the original class.

Example usage:

# create
food = Food('Carot')
food.save()
assert(Food.History.objects.filter(id=food.id).count() == 1)

# update
food.name = 'Carrot'
food.save()
assert(Food.History.objects.filter(id=food.id).count() == 2)

# delete
food.delete()
assert(Food.History.objects.filter(id=food.id).count() == 3)

Why database triggers?

The obvious choice to implement model history would be to connect a signal handler to the post_save and post_delete signal. This has some rather huge downsides:

1.) QuerySet.update() and a lot of other QuerySet methods do not emit any signals. Having to limit the code to only use save() can be a rather huge performance problem depending on the type of application.

2.) There is a rather large performance impact when creating the history via the Django ORM. A single QuerySet.update() call could result in hundreds or thousands of inserts. While this could mostly be solved using the Manager.bulk_create method a database trigger is a lot faster as there is no extra database roundtrip required.

3.) This works for any kind of raw query - even outside of the Django ORM - as long as the chronicle.revision_id session variable is properly set.

The only real downside is the DB compatibility. Right now this package only supports the PostgreSQL database engine.

How to issue queries without the Django ORM?

Create a revision by inserting a row into the revision table and set the chronicle.revision_id session variable like so:

SET chronicle.revision_id = 42; -- replace 42 by the actual revision id

Once you have made all changes to your models don't forget to reset the session variable. Otherwise you might reuse the same revision by accident in the same DB session:

SET chronicle.revision_id TO DEFAULT;