From: | Sam Mason <sam(at)samason(dot)me(dot)uk> |
---|---|
To: | pgsql-general(at)postgresql(dot)org |
Subject: | optimisations to aggregates |
Date: | 2010-01-05 15:26:48 |
Message-ID: | 20100105152648.GA5407@samason.me.uk |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-general |
Hi,
I've just realised that I'm performing the same rewrite on lots of my
queries to get performance reasonable. They take the form of something
like:
SELECT a.x, b.y, COUNT(*) AS n
FROM foo a, bar b
WHERE a.z = b.z
GROUP BY a.x, b.y;
And I rewrite them to:
SELECT a.x, b.y, SUM(b.count) AS n
FROM foo a, (
SELECT y, z, COUNT(*)
FROM bar
GROUP BY y, z) b
WHERE a.z = b.z
GROUP BY a.x, b.y;
Obviously this is only a win when "bar" is large enough that doing the
aggregation reduces the number of rows significantly, hence we're also
predicated on there being a small number of distinct (y,z) values.
This seems like a somewhat easy rewrite that the planner could be doing
itself, but the general case seems harder. Extending the aggregate
abstraction as I suggested here:
http://archives.postgresql.org/pgsql-hackers/2009-11/msg00322.php
would make this transform possible in the general case. It still seems
a bit fiddly to detect though.
--
Sam http://samason.me.uk/
From | Date | Subject | |
---|---|---|---|
Next Message | Adrian Klaver | 2010-01-05 15:27:45 | Re: FM format modifier does not remove leading zero from year |
Previous Message | howardnews@selestial.com | 2010-01-05 15:24:11 | Using regex to update a table |