Re: Simple Column reordering

From: Bruce Momjian <bruce(at)momjian(dot)us>
To: Simon Riggs <simon(at)2ndquadrant(dot)com>
Cc: pgsql-hackers(at)postgresql(dot)org
Subject: Re: Simple Column reordering
Date: 2007-02-26 16:20:19
Message-ID: 200702261620.l1QGKJA03363@momjian.us
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers


I realized this proposal has been withdrawn, but the fact the proposal
even illicited comments exploring it requires me to comment.

Folks, how can we entertain ideas that would break SELECT * and
no-column-list INSERTs for a small performance boost? If there was no
other way to get the performance boost, and the features was rarely
used, we might consider such a change, but neither is true in this case.

My point is that this proposal is so far away from our acceptable
criteria that I am worried about how people are analyzing proposals.

---------------------------------------------------------------------------

Simon Riggs wrote:
> Column storage position is the subject of many long threads in recent
> times. Solutions proposed for this have been both fairly complex and
> long enough that nothing seems likely to happen for 8.3. If I'm wrong,
> then of course this proposal would be superceded.
>
> I propose that at CREATE TABLE time, the column ordering is re-ordered
> so that the table columns are packed more efficiently. This would be a
> physical re-ordering, so that SELECT * and COPY without explicit column
> definitions would differ from the original CREATE TABLE statement.
>
> This would be an optional feature, off by default, controlled by a
> USERSET GUC
> optimize_column_order = off (default) | on
>
> When the full column ordering proposal is implemented,
> optimize_column_ordering would be set to default to on. The feature
> would be supported for at least one more release after this to allow bug
> analysis.
>
> The proposed ordering would be:
> 1. All fixed length columns, arranged so that alignment is efficient
> 2. All variable length columns
>
> All column ordering would stay as close as possible to original order
>
> No changes would be made apart from at CREATE TABLE time.
>
> The ordering would be repeatable, so that the order would not change on
> repeated dump/restore of a table with no changes.
>
> --
> Simon Riggs
> EnterpriseDB http://www.enterprisedb.com
>
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 1: if posting/reading through Usenet, please send an appropriate
> subscribe-nomail command to majordomo(at)postgresql(dot)org so that your
> message can get through to the mailing list cleanly

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://www.enterprisedb.com

+ If your life is a hard drive, Christ can be your backup. +

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Enrico 2007-02-26 16:32:57 Sottoscrizione
Previous Message Bruce Momjian 2007-02-26 15:45:22 Re: Proposal for Implenting read-only queries during wal replay (SoC 2007)