Re: Optimizer internals - Mailing list pgsql-performance

From John Vincent
Subject Re: Optimizer internals
Date
Msg-id c841561b0606151623p3ed68f8g3e1df4143549a8d4@mail.gmail.com
Whole thread Raw
In response to Re: Optimizer internals  ("Jim C. Nasby" <jnasby@pervasive.com>)
List pgsql-performance



decibel=# create index test on i ( sum(i) );
ERROR:  cannot use aggregate function in index expression
decibel=#

BTW, there have been a number of proposals to negate the effect of not
having visibility info in indexes. Unfortunately, none of them have come
to fruition yet, mostly because it's a very difficult problem to solve.
But it is something that the community would like to see happen.
--
Jim C. Nasby, Sr. Engineering Consultant       jnasby@pervasive.com
Pervasive Software       http://pervasive.com    work: 512-231-6117
vcard: http://jim.nasby.net/pervasive.vcf       cell: 512-569-9461


Yeah we got the same thing when we tried it.

I thought about the whole thing on the way home and the downside is that we might have to ditch pgsql.

As far as implementing it, it might make sense to translate READ UNCOMMITTED to that new functionality. If the default isolation level stays the current level, the people who need it can use it via WITH UR or somesuch.

I know it's not that easy but it's an idea. I'm also thinking that the table inheritance we're going to be taking advantage of in 8.1 on the new server might make the sequence scan less of an issue. The only reason the sequence scan really blows is that we have a single table with 220M rows and growing.

pgsql-performance by date:

Previous
From: Tom Lane
Date:
Subject: Re: SAN performance mystery
Next
From: Mark Lewis
Date:
Subject: Re: SAN performance mystery