document that order_by('?') is a huge performance issue

[GomoX <gomo@…>]

order_by('?') generates an SQL query that is horrendous from a performance point of view (the "ORDER BY RAND() LIMIT" type query).

Info on this:
​http://www.titov.net/2005/09/21/do-not-use-order-by-rand-or-how-to-get-random-rows-from-table/

For the current state of affairs, I think at the very least a warning should be added to ​http://www.djangoproject.com/documentation/db-api/#order-by-fields .
That page happily states that you can use the method for obtaining a random row, but in a real scenario that is a very bad idea, and should be avoided at all costs.

On a more useful approach, maybe extra code could be added to a model's Meta class if you plan on grabbing random rows from that particular table. This could set up needed tables/columns/constraints in order to extract a random row without such a big performance hit. If you use order_by('?') on a model with this Meta setting, the enhancement would be transparent. How and if this improvement could be implemented is open for discussion, and is probably database dependent. The page I linked above has some discussion on the topic.

[Simon G. <dev@…>]

I think it's fairly common knowledge that ORDER BY RAND is horrifically inefficient, but it's probably a good idea to place a warning there. Want to write one up?

As for implementing a better random, I think the costs outweigh the benefits, especially if it does mean cracking into weird SQL dialects. This is something to raise on django-developers.

[Matt Boersma]

Added a warning sentence that order_by('?') may be expensive and slow

[Adrian Holovaty]

(In [6293]) Fixed #5267 -- Documented that order_by('?') queries can be slow