Offline optimistic locking for Django
Python Makefile HTML
Switch branches/tags
Nothing to show
Fetching latest commit…
Cannot retrieve the latest commit at this time.



Implements an offline optimistic lock [1] for Django models.


Add a VersionField and inherit from VersionedMixin.

from ool import VersionField, VersionedMixin

class MyModel(VersionedMixin, models.Model):
    version = VersionField()

Whenever MyModel is saved, the version will be checked to ensure the instance has not changed since it was last fetched. If there is a conflict, a ConcurrentUpdate exception will be raised.


A VersionField is just an integer that increments itself every time its model is saved. VersionedMixin overrides _do_update (which is called by save to actually do the update) to add an extra condition to the update query -- that the version in the database is the same as the model's version. If they match, there have been no concurrent modifications. If they don't match, the UPDATE statement will not update any rows, and we know that someone else saved first.

This produces SQL that looks something like:

UPDATE mymodel SET version = version + 1, ... WHERE id = %s AND version = %s

When no rows were updated, we know someone else won and we need to raise a ConcurrentUpdate.

Comparison to django-concurrency

django-concurrency before version 0.7 used SELECT FOR UPDATE to implement the version checking. I wanted to avoid database-level locking, so django-optimistic-lock adds a version filter to the update statement, as described by Martin Fowler [1].

Additionally, ool takes a more minimalistic approach than django-concurrency by only doing one thing -- optimistic locking -- without any monkey-patching, middleware, settings variables, admin classes, or form fields. django-concurrency would probably make more sense if you're looking for something that will attempt to accommodate every situation out of the box. Use ool if you just want a straightforward model implementation and need to handle the UI and surrounding architecture yourself.

Running the tests

make test
[1](1, 2)