Migrations fail when changing from OneToOneField to ForeignKey

[Łukasz Harasimowicz]

Hi!

My team and I have encountered a problem in migrations when changing relationship between models from OneToOneField to ForeignKey.

Original thread on django-users can be found here:
​https://groups.google.com/forum/?#!topic/django-users/-NUbqzS4jQs

Environment

• OS: Unix (Ubuntu 14.04)/OSX
• Django: 1.7.3
• DB: MySQL 5.6.22

Steps to reproduce

1. Create two models:

class Model_A(models.Model):
    id_A = models.AutoField(primary_key=True)


class Model_B(models.Model):
    model_a = models.OneToOneField(Model_A)

2. Make migrations and perform migration. So far everything is OK.

3. Now change OneToOneField to ForeignKey:

class Model_A(models.Model):
    id_A = models.AutoField(primary_key=True)


class Model_B(models.Model):
    model_a = models.ForeignKey(Model_A)

4. Run migrations (makemigrations) and apply changes to db (migrate). We get this error:

we get an error:

Cannot drop index 'model_a_id': needed in a foreign key constraint

It seems that Django tries to run SQL commands in following order:

ALTER TABLE 'xxx' DROP INDEX 'yyy' and
ALTER TABLE `xxx` DROP FOREIGN KEY `xxx_yyy_id_{some_hash}_fk_yyy_id`;

but if above commands are run in reverse order:

ALTER TABLE `xxx` DROP FOREIGN KEY `xxx_yyy_id_{some_hash}_fk_yyy_id`;
ALTER TABLE 'xxx' DROP INDEX 'yyy' and

everything is OK.

I have created sample project where it can be reproduced (last three commits are reproducing above steps): ​https://github.com/harnash/django-migration-bug/commits/master

[Markus Holtermann]

Problem only occurs on MySQL (didn't test Oracle though)

[Markus Holtermann]

PR: ​https://github.com/django/django/pull/3951

[Markus Holtermann <info@…>]

In 5792e6a88c1444d4ec84abe62077338ad3765b80:

Fixed #24163 -- Removed unique constraint after index on MySQL

Thanks Łukasz Harasimowicz for the report.

[Markus Holtermann <info@…>]

In e55cb91bd47d054d502c03cdbe74ac98986af3fb:

[1.8.x] Fixed #24163 -- Removed unique constraint after index on MySQL

Thanks Łukasz Harasimowicz for the report.

Backport of 5792e6a88c1444d4ec84abe62077338ad3765b80 from master

[Markus Holtermann <info@…>]

In db2a97870d74bc689428d9c2a942115ef799f2d2:

[1.7.x] Fixed #24163 -- Removed unique constraint after index on MySQL

Thanks Łukasz Harasimowicz for the report.

Backport of 5792e6a88c1444d4ec84abe62077338ad3765b80 from master

[Markus Holtermann]

The recent change breaks tests on Oracle:

​http://djangoci.com/job/django-oracle/lastCompletedBuild/database=oracle11,python=python3.4/testReport/schema.tests/SchemaTests/test_alter_fk_to_o2o/

======================================================================
FAIL [0.001s]: test_alter_fk_to_o2o (schema.tests.SchemaTests)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/jenkins/workspace/django-oracle/database/oracle11/python/python3.4/tests/schema/tests.py", line 629, in test_alter_fk_to_o2o
    self.assertTrue(author_is_unique, "No unique constraint for author_id found")
AssertionError: False is not true : No unique constraint for author_id found

----------------------------------------------------------------------

[Markus Holtermann]

PR with Oracle fix: ​https://github.com/django/django/pull/3960

[Markus Holtermann <info@…>]

In 64ecb3f07db4be5eef4d9eb7687f783ee446c82f:

Refs #24163 -- Fixed failing Oracle test when migrating from ForeignKey to OneToOneField

Thanks Tim Graham for review

[Markus Holtermann <info@…>]

In 20f1aafa14798d8caac74319a335668ffe86c224:

[1.8.x] Refs #24163 -- Fixed failing Oracle test when migrating from ForeignKey to OneToOneField

Thanks Tim Graham for review

Backport of 64ecb3f07db4be5eef4d9eb7687f783ee446c82f from master

[Markus Holtermann <info@…>]

In 70845c6809f53d9d4029d760aa0118a47b92cb9d:

[1.7.x] Refs #24163 -- Fixed failing Oracle test when migrating from ForeignKey to OneToOneField

Thanks Tim Graham for review

Backport of 64ecb3f07db4be5eef4d9eb7687f783ee446c82f from master

[Claude Paroz]

Just to be sure, was the Oracle failure due to an introspection issue with get_constraints on Oracle? If the latter, is there a ticket about it?

[Markus Holtermann]

I haven't figured that out yet. But I'm investing and I'll open an issue if it's another bug.

[Markus Holtermann]

The issue seems to be some kind of query caching on the cx_oracle or Oracle layer. Unless we re-connect to the database, I don't see a way to make that work.

[Shai Berger]

I finally figured out what's going on here, and why it was hard to reproduce. The issue is actually within our control and can be solved once and for all; then we'll be able to remove the connection_persists_old_columns database feature. See #24200.