Django get_or_create race condition when creating related objects

Question

I have an ExtendedUser model like this, that just points to Django's User model:

class ExtendedUser(models.Model):
    user = models.OneToOneField(User)

When a request comes in, I want to get or create an User and its related ExtendedUser if the user that I'm looking for doesn't exist. I have this code :

def get_or_create_user(username):
    with transaction.atomic():
        user, created = User.objects.get_or_create(
            username=username,
            defaults={
                'first_name': "BLABLA",
            },
        )

        if created:
             extended_user = ExtendedUser()
             extended_user.user = user
             extended_user.save()

    return user

I wrap it inside a transaction, to be sure I don't create an User but not its associated ExtendedUser.

But, when two requests come in simultaneously that will create the same user, this fails from time to time with an IntegrityError, for both requests. So I end up with no user being created..

IntegrityError: (1062, "Duplicate entry 'BLABLABLA' for key 'username'")

Note that I use MySQL, but I force READ COMMITTED isolation level.

What do I do wrong ? How should I handle this ? The models need to stay as they are.

create only if the record doesn't exists .. either by handling ObjectDoesNotExist exception or by checking exists() . — Saji Xavier
– Saji Xavier, Commented Feb 2, 2018 at 14:17
@SajiXavier this naive approach is the best way to have more race conditions - a record can be inserted between when you check if it exists and when you try to create it. The whole point of Queryset.get_or_create() is actually to try and handle this situation, but it's alas not failsafe either. — bruno desthuilliers
– bruno desthuilliers, Commented Feb 2, 2018 at 14:21
As hard as the Django maintainers try to make get_or_create failsafe, there are indeed occasional issues. You could of course wrap the call in a try/except block, but for this concrete use case using the post_save signal as explained in Dalvtor's answer is the simplest and safest solution. — bruno desthuilliers
– bruno desthuilliers, Commented Feb 2, 2018 at 14:29
I don't see how post_save signal will help this problem at all. Wrapping the get_or_create in a try/except is exactly what you need to do here. — Anentropic
– Anentropic, Commented Feb 2, 2018 at 14:37
@brunodesthuilliers according to the question both requests are failing with IntegrityError, so even signal handling will not work becuse User itself is not getting created. Maybe need to use table LOCK feature (i know LOCK works with postgreSQL) — Saji Xavier
– Saji Xavier, Commented Feb 2, 2018 at 14:41

Dalvtor · Accepted Answer · 2018-02-02 14:27:55Z

1

Maybe you should solve this problem using signals.

After a User is saved, a signal will be fired. It is in the signal handler where you should create your ExtendedUser.

from django.contrib.auth.models import User
from django.db.models.signals import post_save
from django.dispatch import receiver

@receiver(post_save, sender=User)
def create_user_profile(sender, instance, created, **kwargs):
    if created:
        ExtendedUser.objects.create(user=instance)

edited Feb 2, 2018 at 14:27

answered Feb 2, 2018 at 14:24

Dalvtor

3,2963 gold badges25 silver badges40 bronze badges

Sign up to request clarification or add additional context in comments.

5 Comments

bruno desthuilliers Over a year ago

That's indeed the canonical solution here. Been using it in a dozen projects without any issue.

Anentropic Over a year ago

This doesn't solve the problem at all, which is IntegrityError: (1062, "Duplicate entry 'BLABLABLA' for key 'username'") on the get_or_create call. This is a wrong use of signals, which are for decoupling separate apps rather than chaining a series of always-required actions together.

fougerejo Over a year ago

This will not work, as just after calling my user = get_or_create_user(username) function, then I try to access the user.extendeduser. So the ExtendedUser should be created at the same time as the User, so the slower query can retrieve it along with the user.

bruno desthuilliers Over a year ago

@fougerejo this WILL work. Django's signals are NOT async, so the the ExtendedUser is garanteed to have been created when User.objects.create() returns. I've been using this for years on dozen projects without a single issue. This wont of course solve the occasional race condition, though - but I'm afraid there's actually no 101% failsafe solution with MySQL here.

fougerejo Over a year ago

Ok I see. But then at the end, what's the difference between using signals and creating the ExtendedUser right away?

Collectives™ on Stack Overflow

Django get_or_create race condition when creating related objects

1 Answer 1

5 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

5 Comments

Your Answer

Sign up or log in

Post as a guest

Related