Understanding Django transactions

Agenda

This post assumes that you understand transactions at database level. Also this post has been written considering PostgreSQL as database.

To recap, every SQL statement in psql implicitly executes inside a transaction without user/client having to explicitly use transaction constructs like BEGIN and COMMIT.

We will write a view which leads to a race condition and will then fix it using Django’s F expressions and later using Django transaction decorator.

View

Let’s assume we have the following view.

from rest_framework.views import APIView
from rest_framework.responses import Response

class LikeView(APIView):

    def patch(self, request, *args, **kwargs):
        post = get_object_or_404(Post, pk=kwargs['pk'])
        post.likes += 1
        post.save()
        return Response("Voted")

Assume this api is exposed at http://localhost:8000/api/post/<pk>/like.

Let’s assume the correct likes is 0 for post with id 1.

If you make 50 concurrent calls to http://localhost:8000/api/post/1/like, you will notice that the likes might not have increased to 50. It might show 47 or 48 or some other number less than 50.

This happened because there is a race condition in the code. One process/thread might have read a value for likes and before it could save(), another process/thread would have read the same value. So even though save() executed from both processes, still the value would have been incremented only once.

The fix for this is to handle increment of likes using a database call without reading it in memory.

Modify the above code to following:

from django.db.models import F

class LikeView(APIView):

    def patch(self, request, *args, **kwargs):
        Post.objects.filter(pk=kwargs['pk']).select_for_update(votes=F('votes') + 1)

We used ORM select_for_update() which runs underlying database’s select ... for update. As the db update runs in a transaction so each process/thread would correctly increment the value of likes.

Would you like to download 10+ free Django and Python books? Get them here