Обсуждение: query optimization

Hi,

By running a relative complex query I get very slow response from
postgresql.

The query is and the explain analyze are:

testsklad19=# explain analyze select
N.IDS,N.IDS_NUM,N.IDS_SLU,N.IDS_SKLAD,SK.MNAME AS
SKNAME,N.ZAPR,N.NOMNUM,NOM.NUM,N.NOMNAME,NOM.MNAME AS

NOMNAME,N.PART,N.SERIENNUM,N.IZV,N.KOL,N.IDS_MED,N.MED,N.IDS_MED_MAIN,N.MED_MAIN,N.OTN_MED,N.CENA,N.VAL,N.KURS,N.CENADDS,N.DDS,N.CENA_LV,N.CENA_LVDDS,N.TOT,N.DTO,N.PTO,N.ORDER_NUM,N.IDS_DOC2,NOM.OSN_MED,NOM.DOP1_MED,NOM.DOP2_MED,NOM.OTN_DOP1_MED,NOM.OTN_DOP2_MED,N.DTON
,N.OFFIC,N.DATE_IZL,N.IZL,N.OTCH_CENA,N.CENAMITALV,N.CENATAKSILV,N.CENATRANSLV,N.CENA,N.CENAZASTRLV
from  A_SKLAD N left outer join A_LOCATION SK ON ( N.IDS_SKLAD=SK.IDS )
left outer join A_NOMEN NOM ON ( N.IDS_NUM = NOM.IDS )  where N.FID = 0
AND  N.IDS_DOC = 'SOF_500'  ORDER BY N.ORDER_NUM  ;
NOTICE:  QUERY PLAN:

Sort  (cost=37.44..37.44 rows=1 width=1118) (actual
time=12140.96..12141.02 rows=48 loops=1)
  ->  Nested Loop  (cost=0.00..37.43 rows=1 width=1118) (actual
time=115.02..12138.57 rows=48 loops=1)
        ->  Nested Loop  (cost=0.00..33.42 rows=1 width=886) (actual
time=0.58..20.77 rows=48 loops=1)
              ->  Index Scan using i_sklad_ids_doc on a_sklad n
(cost=0.00..28.58 rows=1 width=760) (actual time=0.36..8.59 rows=48
loops=1)
              ->  Index Scan using a_location_pkey on a_location sk
(cost=0.00..4.82 rows=1 width=126) (actual time=0.09..0.12 rows=1
loops=48)
        ->  Seq Scan on a_nomen nom  (cost=0.00..3.45 rows=45 width=232)
(actual time=0.03..165.45 rows=6702 loops=48)
Total runtime: 12142.07 msec

EXPLAIN


How can I detect the problem?
Exist any info or docs about how to interpred the analyze results?

Many thanks in advance,
Ivan.

> Sort  (cost=37.44..37.44 rows=1 width=1118) (actual
> time=12140.96..12141.02 rows=48 loops=1)
>   ->  Nested Loop  (cost=0.00..37.43 rows=1 width=1118) (actual
> time=115.02..12138.57 rows=48 loops=1)
>         ->  Nested Loop  (cost=0.00..33.42 rows=1 width=886) (actual
> time=0.58..20.77 rows=48 loops=1)
>               ->  Index Scan using i_sklad_ids_doc on a_sklad n
> (cost=0.00..28.58 rows=1 width=760) (actual time=0.36..8.59 rows=48
> loops=1)
>               ->  Index Scan using a_location_pkey on a_location sk
> (cost=0.00..4.82 rows=1 width=126) (actual time=0.09..0.12 rows=1
> loops=48)
>         ->  Seq Scan on a_nomen nom  (cost=0.00..3.45 rows=45 width=232)
> (actual time=0.03..165.45 rows=6702 loops=48)
> Total runtime: 12142.07 msec

It seems the query planner is completly wrong here, look on the line
Seq Scan on a_nomen nom  (cost=0.00..3.45 rows=45 width=232) (actual time=0.03..165.45 rows=6702 loops=48)

This means the planner expects 45 return rows (guessed from statistics), but actually gets 6702 rows.

Do "VACUUM ANALYZE a_nomen" and try your query again.

If it fails:
Do you have a unique index on a_nomen(ids)?

Regards,
    Mario Weilguni


>
> ---------------------------(end of broadcast)---------------------------
> TIP 5: Have you checked our extensive FAQ?
>
> http://www.postgresql.org/users-lounge/docs/faq.html

>
> The query is and the explain analyze are:
>
> testsklad19=# explain analyze select
> N.IDS,N.IDS_NUM,N.IDS_SLU,N.IDS_SKLAD,SK.MNAME AS
> SKNAME,N.ZAPR,N.NOMNUM,NOM.NUM,N.NOMNAME,NOM.MNAME AS
>
NOMNAME,N.PART,N.SERIENNUM,N.IZV,N.KOL,N.IDS_MED,N.MED,N.IDS_MED_MAIN,N.MED_
MAIN,N.OTN_MED,N.CENA,N.VAL,N.KURS,N.CENADDS,N.DDS,N.CENA_LV,N.CENA_LVDDS,N.
TOT,N.DTO,N.PTO,N.ORDER_NUM,N.IDS_DOC2,NOM.OSN_MED,NOM.DOP1_MED,NOM.DOP2_MED
,NOM.OTN_DOP1_MED,NOM.OTN_DOP2_MED,N.DTON
>
,N.OFFIC,N.DATE_IZL,N.IZL,N.OTCH_CENA,N.CENAMITALV,N.CENATAKSILV,N.CENATRANS
LV,N.CENA,N.CENAZASTRLV
> from  A_SKLAD N left outer join A_LOCATION SK ON ( N.IDS_SKLAD=SK.IDS )
> left outer join A_NOMEN NOM ON ( N.IDS_NUM = NOM.IDS )  where N.FID = 0
> AND  N.IDS_DOC = 'SOF_500'  ORDER BY N.ORDER_NUM  ;
> NOTICE:  QUERY PLAN:
>


Well,
can you tell us the index that are present on table A_SKLAD ?
It seem that there are no index on the field FID, BTW if for you
are always important only the line whit FID = 0 ( or other few values )
you can create a partial index:

CREATE INDEX idx_partial ON a_sklad ( fid ) WHERE FID = 0;

or if the values are more:

CREATE INDEX idx_partial ON a_sklad ( fid ) WHERE FID IN ( val1, val2,
.... );


Dont forget to do:  'vacuum analyze a_skland' after index creation.


Good luck
Gaetano