[Django] #36785: v6 seems to break the behaviour of transaction.atomic

Django

unread,

Dec 8, 2025, 1:58:49 PM12/8/25

to django-...@googlegroups.com

#36785: v6 seems to break the behaviour of transaction.atomic
-------------------------------------+-------------------------------------
Reporter: paulzakin | Type: Bug
Status: new | Component: Database
| layer (models, ORM)
Version: 6.0 | Severity: Normal
Keywords: | Triage Stage:
| Unreviewed
Has patch: 0 | Needs documentation: 0
Needs tests: 0 | Patch needs improvement: 0
Easy pickings: 0 | UI/UX: 0
-------------------------------------+-------------------------------------
Hello!

Just tried upgrading to v6 - and this snippet from a test (django and
pytest) now breaks. Now, the first chunk passses but the second chunk
requires a "current transaction is aborted, commands ignored until end of
transaction block".

Historically, I got around that by wrapping the test in an with
transaction.atomic() to prevent the error from moving to the next part of
the test - but that no longer works.

Not sure if this counts as an error - but I'm not supposed to do this in a
test anyway! But it did break from v5 to v6.

{{{

# Do policies apply to .bulk_create()?
with transaction.atomic(): # Required as any database error, like
RLS, must be in transaction.atomic()
try:
Foo.objects.bulk_create([Foo(name="Foo", study_id=study.pk)])
force_test_to_fail()
except Exception as error:
assert str(error) == 'new row violates row-level security
policy for table "foo"'

with transaction.atomic(): # Required as any database error, like
RLS, must be in transaction.atomic()
try:
# Do policies apply to .bulk_update()?
foo.name = "XXX"
# No rows should have been updated, because the admin does not
have access to the id
Foo.objects.bulk_update([foo], fields=["*"])
force_test_to_fail()
except Exception as error:
assert str(error) == f"id: {foo.pk} in table: foo does not
exist!"
}}}
--
Ticket URL: <https://code.djangoproject.com/ticket/36785>
Django <https://code.djangoproject.com/>
The Web framework for perfectionists with deadlines.

Django

unread,

Dec 8, 2025, 2:14:08 PM12/8/25

to django-...@googlegroups.com

#36785: v6 seems to break the behaviour of transaction.atomic
-------------------------------------+-------------------------------------

Changes (by Simon Charette):

* resolution: => needsinfo
* status: new => closed

Comment:

Hello Paul,

It would help tremendously if you could provide a simplified project that
passes againt Django 5.0 (you didn't mention 5.2 so I assumed you were
still on 5.0) but fails against 6.0 with the exact frozen dependencies you
are using.

Django might be at fault here but since you are using pytest (and I assume
pytest-django) and transaction handling in tests can be quite finicky it
can be quite hard to pinpoint what exactly might cause the problems you
are facing with the little details you've provided (no exact test
definition, no exact package versions, do model definitions, etc).

Please help us help you by providing more details about the problem you
are experiencing so we can reproduce and adequately diagnoze the problem.
In the mean time I'll close the ticket as lacking details but please
reopen when you're able to provide us more details.
--
Ticket URL: <https://code.djangoproject.com/ticket/36785#comment:1>

Django

unread,

Dec 8, 2025, 4:56:59 PM12/8/25

to django-...@googlegroups.com

#36785: v6 seems to break the behaviour of transaction.atomic
-------------------------------------+-------------------------------------

Reporter: Paul Zakin | Owner: (none)

Comment (by Paul Zakin):

Totally fair - our setup is complicated enough that might be pretty
challenging.

However, if I have the time and figure it out - I'll reopen - thanks a
bunch!
--
Ticket URL: <https://code.djangoproject.com/ticket/36785#comment:2>

Django

unread,

Dec 8, 2025, 5:42:57 PM12/8/25

to django-...@googlegroups.com

#36785: v6 seems to break the behaviour of transaction.atomic
-------------------------------------+-------------------------------------
Reporter: Paul Zakin | Owner: (none)
Type: Bug | Status: closed
Component: Database layer | Version: 6.0
(models, ORM) |
Severity: Normal | Resolution: needsinfo
Keywords: | Triage Stage:
| Unreviewed
Has patch: 0 | Needs documentation: 0
Needs tests: 0 | Patch needs improvement: 0
Easy pickings: 0 | UI/UX: 0
-------------------------------------+-------------------------------------
Comment (by Paul Zakin):

Interesting - I bisected and a found the commit

https://github.com/django/django/commit/e1671278e88265e64811657b8b939b5d786295cb

{{{
with transaction.atomic(using=self.db, savepoint=False):
if objs_with_pk and objs_without_pk:
context = transaction.atomic(using=self.db, savepoint=False)
else:
context = nullcontext()
with context:
}}}

I think it is this
--
Ticket URL: <https://code.djangoproject.com/ticket/36785#comment:3>

Django

unread,

Dec 8, 2025, 5:43:59 PM12/8/25

to django-...@googlegroups.com

#36785: v6 seems to break the behaviour of transaction.atomic
-------------------------------------+-------------------------------------
Reporter: Paul Zakin | Owner: (none)

Type: Bug | Status: new

Component: Database layer | Version: 6.0
(models, ORM) |
Severity: Normal | Resolution:

Changes (by Paul Zakin):

* resolution: needsinfo =>
* status: closed => new

--
Ticket URL: <https://code.djangoproject.com/ticket/36785#comment:4>

Django

unread,

Dec 8, 2025, 11:56:09 PM12/8/25

to django-...@googlegroups.com

#36785: v6 seems to break the behaviour of transaction.atomic
-------------------------------------+-------------------------------------
Reporter: Paul Zakin | Owner: (none)

Type: Bug | Status: closed

Component: Database layer | Version: 6.0
(models, ORM) |

Severity: Normal | Resolution: invalid

Changes (by Simon Charette):

* resolution: => invalid

* status: new => closed

Comment:

Well I think you have your answer here.

Prior to #36490 (e1671278e88265e64811657b8b939b5d786295cb) `bulk_create`
was always creating an implicit transaction even if it only needed to
issue a single `INSERT`.

Now that no explicit `transaction.atomic` usage is made the database error
your RLS policy triggers and don't let the error reach
`transaction.Atomic.__exit__` which goes against its documented usage (see
[https://docs.djangoproject.com/en/6.0/topics/db/transactions/#django.db.transaction.atomic
Avoid catching exceptions inside atomic!]) it leaves the current
transaction in an incoherent state.

TL;DR your usage of `transaction.atomic` is incorrect but it was masked by
`bulk_create` creating a nested `atomic` previously (which was
implementation detail) and you should do the following instead

{{{#!python
try:
with transaction.atomic():

Foo.objects.bulk_create([Foo(name="Foo", study_id=study.pk)])
force_test_to_fail()
except Exception as error:
assert str(error) == 'new row violates row-level security policy for
table "foo"'
}}}

--
Ticket URL: <https://code.djangoproject.com/ticket/36785#comment:5>

Reply all

Reply to author

Forward