#29016 closed Bug (fixed)
Reuse of UpdateQuery breaks some delete updates
Reported by: | Étienne Loks | Owned by: | Étienne Loks |
---|---|---|---|
Component: | Database layer (models, ORM) | Version: | 1.11 |
Severity: | Release blocker | Keywords: | |
Cc: | Triage Stage: | Ready for checkin | |
Has patch: | yes | Needs documentation: | no |
Needs tests: | no | Patch needs improvement: | no |
Easy pickings: | no | UI/UX: | no |
Description (last modified by )
On a model A, when deleting a foreign key pointing to a model B, some other foreign key of the model A pointing to the same model B may be nullified.
I have isolated this behaviour on a simple project:
models.py:
from django.db import models class ChildModel(models.Model): name = models.CharField(max_length=200) class ParentModel(models.Model): name = models.CharField(max_length=200) child_1 = models.ForeignKey(ChildModel, on_delete=models.SET_NULL, related_name='parents_1', null=True) child_2 = models.ForeignKey(ChildModel, on_delete=models.SET_NULL, related_name='parents_2', null=True)
Django shell session:
from testapp.models import ParentModel, ChildModel child_1 = ChildModel.objects.create(name="child_1") child_2 = ChildModel.objects.create(name="child_2") parent_1 = ParentModel.objects.create(name="parent 1", child_1=child_1, child_2=child_2) parent_2 = ParentModel.objects.create(name="parent 2", child_1=child_2, child_2=child_1) child_1.delete() parent_1 = ParentModel.objects.get(pk=parent_1.pk) parent_2 = ParentModel.objects.get(pk=parent_2.pk) # parent_1.child_2 and parent_2.child_1 should be normaly equal to child_2 but... parent_1.child_2 is not None and parent_2.child_1 is not None # False is returned
This simple project has been tested on an SQLite database. The same behaviour has been first discovered on a PostgreSQL database.
A mis-reuse of an UpdateQuery seems to be the cause of this bug.
After search on the django bug tracker I have found another issue with the same patch attached #28099.
I have opened this new ticket because the issue seems to be more severe (I have experienced large data loss) and more general.
This issue has been found on version 1.11 and 2.0 of Django.
I have created a new branch on my github account with patch and test associated: https://github.com/Nimn/django/tree/ticket_29016
Attachments (1)
Change History (9)
by , 7 years ago
Attachment: | fix_delete_on_update.patch added |
---|
comment:1 by , 7 years ago
Summary: | Reuse of UpdateQuery breaks delete some updates → Reuse of UpdateQuery breaks some delete updates |
---|
comment:2 by , 7 years ago
Description: | modified (diff) |
---|---|
Has patch: | set |
comment:3 by , 7 years ago
Severity: | Normal → Release blocker |
---|---|
Triage Stage: | Unreviewed → Accepted |
I think that qualifies for a backport based on the "data loss" criteria. Rather than adding entirely new models, can you reuse an existing model (perhaps another ForeignKey
will need to be added on an existing model) either in delete
or delete_regress
?
comment:4 by , 7 years ago
Owner: | changed from | to
---|---|
Status: | new → assigned |
I have changed a model in delete_regress
to add two foreign keys (it seems clearer to me). I have sent a pull request for the master branch.
If it really qualifies to a backport (I agree with this qualification), how can I proceed to propose a patch? I miss a proper documentation.
comment:5 by , 7 years ago
Triage Stage: | Accepted → Ready for checkin |
---|
Looks good, I made some cosmetic edits and added release notes.
Simple patch version (no regression test yet)