From: | Heikki Linnakangas <hlinnakangas(at)vmware(dot)com> |
---|---|
To: | Peter Geoghegan <pg(at)heroku(dot)com> |
Cc: | Robert Haas <robertmhaas(at)gmail(dot)com>, PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org>, Peter Eisentraut <peter_e(at)gmx(dot)net>, "Magnus Hagander" <magnus(at)hagander(dot)net>, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com>, Gregory Smith <gregsmithpgsql(at)gmail(dot)com> |
Subject: | Re: Last Commitfest patches waiting review |
Date: | 2014-10-09 19:51:36 |
Message-ID: | 5436E748.9010401@vmware.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
On 10/09/2014 10:25 PM, Peter Geoghegan wrote:
> On Thu, Oct 9, 2014 at 12:14 PM, Heikki Linnakangas
> <hlinnakangas(at)vmware(dot)com> wrote:
>> B-tree index build uses tuplesort.c. What's missing?
>
> I don't think that all that much is missing. Tuplesort expects to work
> with an index scankey when sorting B-Tree tuples. There needs to be
> something like a reverse lookup of the sortsupport function. It looks
> like a historical oversight, that would take time to fix, but wouldn't
> be particularly challenging. You'd need to pick out the operators from
> the scankey, so you'd have something like what tuplesort_begin_heap()
> starts off with with tuplesort_begin_index_btree().
Oh, I didn't realize we don't do that already! I'm surprised, I would've
expected index build to have been the first thing we'd use the
SortSupport stuff in.
Yeah, that seems worth doing, independently of the this patch. Can you
write a separate patch to use SortSupport for B-tree index builds,
please? Eliminating the FunctionCallInfoData overhead should shave off
some some cycles from every index build.
- Heikki
From | Date | Subject | |
---|---|---|---|
Next Message | Andres Freund | 2014-10-09 19:53:02 | Re: Deferring some AtStart* allocations? |
Previous Message | Fabien COELHO | 2014-10-09 19:39:59 | Re: pgbench throttling latency limit |