PostgreSQL aggregation and views through unmanaged models

671 views
Skip to first unread message

charettes

unread,
May 21, 2017, 11:05:38 PM5/21/17
to Django developers (Contributions to Django itself)
Hello fellow developers,

As some of you may know PostgreSQL 9.1 added support for GROUP'ing BY
selected table primary keys[0] only. Five years ago it was reported[1] that
Django could rely on this feature to speed up aggregation on models backed
up by tables with either many fields or a few large ones.

Being affected by this slow down myself I decided to dive into the ORM internals
and managed to get a patch that made it in 1.9[2] thanks to Anssi's and Josh's
review[3].

One subtle thing I didn't know back in the time is that PostgreSQL query planner
isn't able to introspect database views columns' functional dependency like it
does with tables and thus prevents the primary key GROUP'ing optimization from
being used.

While Django doesn't support database views officially it documents that
unmanaged models can be used to query them[4] and thereby perform aggregation on
them and generating an invalid query.

This was initially reported as a crashing bug 9 months ago[5] and the consensus
at this time was that it was an esoteric edge case since there was few reports
of breakages and it went off my radar. Fast-forward to a month ago, this is
reported again[6] and it takes the reporter quite a lot of effort to determine
the origin of the issue, pushing me to come up with a solution as I introduced
this behavior.

Before Claude makes me realize this is a duplicate of the former report (which I
completely forgot about in the mean time) I implement a patch and commit it once
it's reviewed [7].

When I closed the initial ticket as "fixed" the reporter brought to my attention
that this was now introducing a performance regression for unmanaged models
relying on aggregation and that we should document how to disable this
optimization by creating a backend subclass as a workaround instead.

In my opinion the current situation is as follow. The optimization introduced a
break in backward compatibility in 1.9 as we've always documented that database
views could be queried against using unmanaged models. If this issue had been
discovered during the 1.9 release cycle it would have been eligible for a
backport because it was a bug in a newly introduced feature. Turning this
optimization off for unmanaged models by assuming they could be views is only
going to degrade performance of queries using unmanaged models to perform
aggregation on tables with either a large number of columns or large columns
using PostgreSQL.

Therefore I'd favor we keep the current adjustment in the master branch as it
restores backward compatibility but I don't have strong feelings about reverting
it either if it's deemed inappropriate.

Another solution I came up while writing this post would be to replace the
feature flag by a callable that takes a model as a single parameter and returns
whether or not the optimization can be performed against it. The default
implementation would return `mode._meta.managed` but it would make it easier for
users affected by this to override in order to opt-in or out based on their
application logic.

Thank you for your time,
Simon


Josh Smeaton

unread,
May 21, 2017, 11:46:02 PM5/21/17
to Django developers (Contributions to Django itself)
> Therefore I'd favor we keep the current adjustment in the master branch as it
> restores backward compatibility but I don't have strong feelings about reverting
> it either if it's deemed inappropriate.

Fixing the crash is the number 1 priority in my opinion, as it broke something that used to work.

Optimising aggregation for unmanaged models is a distant second goal. Make it right. Then make it fast.

It'd be nice to provide said optimisation for unmanaged models provided there was a palatable way of doing so. I'm not sure how users would be able to use your callable approach without subclassing the backend - unless that is the intention? Getting feedback from the reporters would be good of course.

jr...@leukeleu.nl

unread,
May 22, 2017, 2:32:55 PM5/22/17
to Django developers (Contributions to Django itself)
Thanks for that extensive write up!

As the reporter of #27241 it seems I would be arguing against my own self interest when I say that I'm not in favour of the patch, but my reasonings are as follows:

* The current behaviour is preferable in the vast majority of cases, only a couple of projects are affected by the change of behaviour in Django 1.9.
* Anyone using unmanaged models on views in a way that stops working when upgrading Django to >= 1.9 is actively maintaining their code and should be able to implement a workaround.
* The easiest workaround is only ~5 lines of code and pretty much restores the behaviour of Django 1.8 so it's perfectly acceptable for people upgrading from <= 1.8.
* Other workarounds are also possible, especially with the new subquery and other ORM improvements that have been introduced since Django 1.8.
* Upgrading from <= 1.8 doesn't necessarily mean upgrading to 2.0 (i.e. ready to upgrade to the next LTS but not yet ready to migrate to Python 3). These people will still end up looking for a workaround anyway.
* Any project using unmanaged models that works with Django >= 1.9 will suddenly see a performance hit when they upgrade to 2.0.
* The reporter of #28107, #26758 (which is probably the same issue) and I (reporter of #27241) have worked around the issue, so we none of us will benefit from a patch.

So while I think that fixing this bug is noble, in this case I think there's way more downsides than upsides.

Thanks!

Jaap Roes

Dylan Young

unread,
Sep 23, 2017, 1:26:39 PM9/23/17
to Django developers (Contributions to Django itself)
Is there any reason not to query the models on startup to determine which ones support the optimization and which do not (naively, which are proper tables and which are views, but perhaps postgresql can provide the information instead of relying on heuristics which could change in any release)?  Not only would this solve this issue, it would add a general mechanism for auto-configuring optimizations based on underlying model/table characteristics, essential for any ORM IMO. 

Matthew Pava

unread,
Dec 11, 2017, 5:27:19 PM12/11/17
to Django developers (Contributions to Django itself)
You asked me to join in on this conversation as I had an issue using Django 1.11.  My third party apps do not yet all support Django 2.0.

I have an unmanaged model that is connected to a view in PostgreSQL.  I attempted to create an annotated version of the model manager using the .annotate() method.  Unfortunately, I am getting an error that some fields “must appear in the GROUP BY clause or be used in an aggregate function.”  It was a mysterious error message for me.

 

Upon further investigation, it appears that I need a PRIMARY KEY in the backend to avoid having to include columns in a GROUP BY clause, but PostgreSQL does not allow for such constraints to be added to a VIEW.  The “id” column is just a row_number().

vojtec...@avast.com

unread,
Jul 31, 2018, 9:31:42 AM7/31/18
to Django developers (Contributions to Django itself)
Hello,
the fix for https://code.djangoproject.com/ticket/27241 caused significant performance drop for my project, because it pretty much prevents posgresql from using indexes on the particular query. I use unmanaged tables because Django is used to display data from another project's database, so it has just read-only access to the database and I don't want to create any migrations for it.

Another solution I came up while writing this post would be to replace the
feature flag by a callable that takes a model as a single parameter and returns
whether or not the optimization can be performed against it. The default
implementation would return `mode._meta.managed` but it would make it easier for
users affected by this to override in order to opt-in or out based on their
application logic.

This would be perfect for my (arguably niche) use case. Should I try to prepare a patch?

On Monday, May 22, 2017 at 5:05:38 AM UTC+2, charettes wrote:
Reply all
Reply to author
Forward
0 new messages