id	summary	reporter	owner	description	type	status	component	version	severity	resolution	keywords	cc	stage	has_patch	needs_docs	needs_tests	needs_better_patch	easy	ui_ux
12579	QuerySet.get_or_create()'s behavior depends on database's uniqueness restrictions	Tim Molendijk	Tim Graham <timograham@…>	"{{{QuerySet.get_or_create()}}} works as expected, IFF the kwargs that are provided correspond to database columns with appropriate uniqueness restrictions. In other words: {{{get_or_create()}}} depends on an IntegrityError being raised under certain circumstances. If you have a database that does not do this, f.e. because a table's uniqueness logic cannot be expressed with the database's {{{UNIQUE}}} construct.

For instance:

Given the following table: {{{CREATE TABLE t1 (i INT, flag TINYINT, parent_id INT, PRIMARY KEY (i)) ENGINE = InnoDB;}}}
An additional restriction, one which cannot be expressed in the table definition, is that only one record can exist with {{{flag=1}}} for every given {{{parent_id}}}.
The following code gives the impression that we enforce said restriction: {{{T1.objects.get_or_create(parent_id=3, flag=1)}}}.
But this is a false impression, as two concurrent transactions executing this same code can (and will on a regular basis) result in two records with {{{parent_id=3}}} and {{{flag=1}}}. This is the result of both transactions trying to {{{get()}}} a record with {{{parent_id=3}}} and {{{flag=1}}} before any of them has inserted this record. So both transactions will conclude they need to {{{create{}}}} the record, which will be allowed by the database and thus an inconsistency will enter the database.

Now the ideal scenario would be that race conditions are prevented regardless of how the underlying database table is defined. But I am not sure if this is a feasible approach. At the very least I think people who use {{{get_or_create()}}} need to be aware that it can give rise to concurrency issues, unless they follow some basic guidelines... something like that."	Bug	closed	Documentation	1.0	Normal	fixed	get_or_create concurrency race unique documentation	tomasz.zielinski@… vijay_shanker	Accepted	1	0	0	0	1	0