Обсуждение: make query faster??

WAWI-# select a_nr,a_bez1,a_bez2,a_bez3,a_bez4,a_gewicht,a_bruttog,kl_ean,a_vol from artikel order by a_nr;
NOTICE:  QUERY PLAN:

 

Sort  (cost=341.63..341.63 rows=3809 width=96)
  ->  Seq Scan on artikel  (cost=0.00..115.09 rows=3809 width=96)

EXPLAIN

WAWI=# explain
WAWI-# select a_nr,a_bez1,a_bez2,a_bez3,a_bez4,a_gewicht,a_bruttog,kl_ean,a_vol from artikel;
NOTICE:  QUERY PLAN:

 

Seq Scan on artikel  (cost=0.00..115.09 rows=3809 width=96)

 

EXPLAIN
WAWI=#    

 

On Tue, 4 Dec 2001, [iso-8859-1] DI Hasen�hrl wrote:

> I've a simple query, but when I use *ORDER BY* it takes a long time
> till I can see the result.
>
> a_nr is varchar(20) and the primary key of my table artikel.
>
> Please, can someone give me a hint, how to improve my query.

If you have not yet done so, you may want to increase sort_mem in your
postgresql.conf assuming you have sufficient memory in your machine
(and shared_buffers while you're at it). The defaults are fairly low and
you may see a performance gain in general if you raise them to amounts
more reasonable for your machine.

> In Bruce Momjian's book I read, that indexes are usually not used for
> ORDER BY operations: a sequential scan followed by an explicit sort is
> fatser than an indexscan.
>
> What's an explicit sort????????

Actually sorting the results of the sequence scan rather than using the
fact that the index keys can be gotten in sorted order to get the rows
in order.

----- Original Message -----

From: Stephan Szabo

To: DI Hasenöhrl

Cc: pgsql-sql@postgresql.org

Sent: Tuesday, December 04, 2001 10:26 PM

Subject: Re: [SQL] make query faster??

On Tue, 4 Dec 2001, [iso-8859-1] DI Hasenöhrl wrote:

> I've a simple query, but when I use *ORDER BY* it takes a long time
> till I can see the result.
>
> a_nr is varchar(20) and the primary key of my table artikel.
>
> Please, can someone give me a hint, how to improve my query.

If you have not yet done so, you may want to increase sort_mem in your
postgresql.conf assuming you have sufficient memory in your machine
(and shared_buffers while you're at it). The defaults are fairly low and
you may see a performance gain in general if you raise them to amounts
more reasonable for your machine.

> In Bruce Momjian's book I read, that indexes are usually not used for
> ORDER BY operations: a sequential scan followed by an explicit sort is
> fatser than an indexscan.
>
> What's an explicit sort????????

Actually sorting the results of the sequence scan rather than using the
fact that the index keys can be gotten in sorted order to get the rows
in order.

On Wed, 5 Dec 2001, [iso-8859-1] DI Hasen�hrl wrote:

> Thank you for your response, but I'm not quite clear.
>
> Is *sort_mem* the same as postgres' starting option -S *Sortsize*,
> which is by default 512Kb
> At this time I increased sortsize to 1024Kb, but the costs from
> explain are the same as before and I can't see, that the query is
> faster

IIRC the explain costs may not change.  If your table really only
has 4000 rows, I don't see why it would need to drop off to disk,
but while the query is running, you can see if pg_sorttemp* files
are being created under your data directory (which would show
that it is going out to disk and that you may need to raise -S
again).  Otherwise, is the data involved something you can share?
It might help for us to be able to try it elsewhere to see how
it performs.

> Please, can you explain, what *cost=341.63..341.63* means

It's an estimate of cost.  I believe the first is startup cost
and the last total cost.  You probably want to read the performance
tips section of the user's guide.