Fixed get_or_create() to re-raise IntegrityError when object still does not exist. Refs #36489 #19676

Jkhall81 · 2025-07-27T01:21:38Z

Trac ticket number

Branch description

Previously, QuerySet.get_or_create() would suprress all IntegrityError exceptions that occurred during object creation, under the assumption that they were due to race conditions. However, this also masked real issues like missing required fields, unique constraint violations, or invalid foreign keys.

This change modifies the behavior to re-raise the IntegrityError if a subsequent .get() still fails after .create() -- ensuring only true race conditions are handled silently.

Checklist

[ x] This PR targets the main branch.
[ x] The commit message is written in past tense, mentions the ticket number, and ends with a period.
[x ] I have checked the "Has patch" ticket flag in the Trac system.
[x ] I have added or updated relevant tests.
I have added or updated relevant docs, including release notes if applicable.
I have attached screenshots in both light and dark modes for any UI changes.

Notes

Includes a PostgreSQL-only regression test (get_or_create_race) using ThreadPoolExecutor to simulate concurrent access.
The test confirms that reverse OneToOne cache remains correct and only one object is created in a race.

…omatic zoom issue in mobile Safari.

…nt exist

sarahboyce

Thank you for the PR

sarahboyce · 2025-07-30T15:14:29Z

.gitignore

@@ -17,3 +17,4 @@ tests/.coverage*
 build/
 tests/report/
 tests/screenshots/
+test_postgres.py


Suggested change

test_postgres.py

sarahboyce · 2025-07-30T15:15:10Z

django/db/models/query.py

@@ -979,17 +979,17 @@ def get_or_create(self, defaults=None, **kwargs):
            return self.get(**kwargs), False
        except self.model.DoesNotExist:
            params = self._extract_model_params(defaults, **kwargs)
-            # Try to create an object using passed params.


I think we can keep this comment

comment has been put back.

sarahboyce · 2025-07-30T15:20:28Z

tests/get_or_create_race/models.py

@@ -0,0 +1,12 @@
+from django.db import models


We don't need to create a new test folder for this fix and can use get_or_create for example
We should try to reuse existing models where we can

tests moved, removed my postgres test.

sarahboyce · 2025-07-30T15:22:13Z

tests/get_or_create_race/tests.py

+
+
+@unittest.skipUnless(
+    connection.vendor == "postgresql", "This test only runs on PostgreSQL"


Not sure why this is only being run only on postgres? I don't see anything in the ticket discussion to suggest this is a postgres specific issue

I use PostgreSQL because this bug revolves around a race condition during concurrent get_or_create() calls involving a OneToOneField. So to reliably reproduce that the database needs to support concurrency.

SQLite doesn't support concurrent writes. It locks the entire database file during a write.
MySQL was failing with errors like Lock wait timeout exceeded.
PostgreSQL was the third one i tried and it allowed me to reproduce the race condition.

sarahboyce · 2025-07-30T15:22:57Z

django/db/models/query.py

-                raise
+                    # Re-raise if the object still doesn't exist.
+                    # this wasn't a race condition but an integrity error.
+                    raise e


I don't think this is the correct solution. Note the test in the original ticket is still failing for me and based off the ticket discussion, I expect the solution to be more involved

solution updated, and now passes test (modified to use models in get_or_create) from original ticket description.

Have you given any thoughts on how systematically clearing one-to-one relationship cache will break select_related("parent").get_or_create(parent=parent), how this relates to update_or_create and if something similar should be done with defaults and create_defaults as described in the ticket?

I appreciate that the test are passing but it seems like there are still a few open questions about which approach should be taken.

Regarding that pruned comment, I was having git issues. Fixed that.

So instead of systematically clearing the one-to-one relationship cache, it should be done conditionally. Only if the cache looks stale. I can make that change real quick. I'm still getting used to how everything works.

I appreciate the guidance. I'll get that changed pushed shortly.

Those other two issues, update_or_create() and defaults / create_defaults, should those be dealt with here? I can take a closer look at both.

It looks like update_or_create uses get_or_create internally:

obj, created = self.select_for_update().get_or_create( create_defaults, **kwargs )

The cache fix will apply here also.

charettes · 2025-07-30T17:58:44Z

django/db/models/query.py

-        # The get() needs to be targeted at the write database in order
-        # to avoid potential transaction consistency problems.


Why did you prune this comment?

Jkhall81 and others added 3 commits March 24, 2024 11:06

Added CSS rule for label element making font-size 1rem to address aut…

51badee

…omatic zoom issue in mobile Safari.

Merge branch 'django:main' into main

bcbfd96

Fix get_or_create() to re-raise IntegrityError when object still does…

1bc7124

…nt exist

Jkhall81 force-pushed the 36489-fix-get-or-create-cache-bug branch 3 times, most recently from 20ed5e1 to f522c80 Compare July 27, 2025 01:48

added doc string back to get_or_create.

00ee71d

Jkhall81 force-pushed the 36489-fix-get-or-create-cache-bug branch 3 times, most recently from a3c5faf to 615c54c Compare July 27, 2025 02:02

sarahboyce reviewed Jul 30, 2025

View reviewed changes

Jkhall81 force-pushed the 36489-fix-get-or-create-cache-bug branch 2 times, most recently from 9ad93a5 to 59b52ab Compare July 30, 2025 16:21

Jkhall81 requested a review from sarahboyce July 30, 2025 16:27

charettes reviewed Jul 30, 2025

View reviewed changes

Jkhall81 force-pushed the 36489-fix-get-or-create-cache-bug branch from 59b52ab to 7c81ccc Compare July 30, 2025 18:11

--amend

7279535

Jkhall81 force-pushed the 36489-fix-get-or-create-cache-bug branch from 7c81ccc to 7279535 Compare July 30, 2025 18:36

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Fixed get_or_create() to re-raise IntegrityError when object still does not exist. Refs #36489 #19676

Fixed get_or_create() to re-raise IntegrityError when object still does not exist. Refs #36489 #19676

Jkhall81 commented Jul 27, 2025

Uh oh!

sarahboyce left a comment

Uh oh!

sarahboyce Jul 30, 2025

Uh oh!

Jkhall81 Jul 30, 2025

Uh oh!

sarahboyce Jul 30, 2025

Uh oh!

Jkhall81 Jul 30, 2025

Uh oh!

sarahboyce Jul 30, 2025

Uh oh!

Jkhall81 Jul 30, 2025

Uh oh!

sarahboyce Jul 30, 2025

Uh oh!

Jkhall81 Jul 30, 2025

Uh oh!

sarahboyce Jul 30, 2025

Uh oh!

Jkhall81 Jul 30, 2025

Uh oh!

charettes Jul 30, 2025 •

edited

Loading

Uh oh!

Jkhall81 Jul 30, 2025 •

edited

Loading

Uh oh!

Jkhall81 Jul 30, 2025

Uh oh!

charettes Jul 30, 2025

Uh oh!

Uh oh!



		@unittest.skipUnless(
		connection.vendor == "postgresql", "This test only runs on PostgreSQL"

		# The get() needs to be targeted at the write database in order
		# to avoid potential transaction consistency problems.

Uh oh!

Fixed get_or_create() to re-raise IntegrityError when object still does not exist. Refs #36489 #19676

Are you sure you want to change the base?

Fixed get_or_create() to re-raise IntegrityError when object still does not exist. Refs #36489 #19676

Conversation

Jkhall81 commented Jul 27, 2025

Trac ticket number

Branch description

Checklist

Notes

Uh oh!

sarahboyce left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

charettes Jul 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Jkhall81 Jul 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

charettes Jul 30, 2025 •

edited

Loading

Jkhall81 Jul 30, 2025 •

edited

Loading