RE: [GENERAL] cost difference for 2 similar queries - Mailing list pgsql-general

From Jackson, DeJuan
Subject RE: [GENERAL] cost difference for 2 similar queries
Date
Msg-id F10BB1FAF801D111829B0060971D839F2DCCDD@cpsmail
Whole thread Raw
List pgsql-general
> Hi,
>
> I have 2 queries, which are quite similar:
>
> a) select distinct p1.domain from dom_prot p1, identical p2,
> classification p3
>    where p2.reference = p1.protein
>    and p2.protein ~ 'EPA3_HUMAN'
>    and p3.domain = p1.domain;
>
> b) select distinct p1.domain from dom_prot p1, identical p2,
> classification p3
>    where p2.reference = p1.protein
>    and p2.protein ~ 'EPA3_HUMAN'
>    and p3.domain = p1.domain
>    and p3.class = 'extra';
Not sure (this may take longer):
SELECT DISTINCT p1.domain
  FROM dom_prot AS p1
 WHERE EXISTS (SELECT p2.reference
                 FROM identical AS p2
                WHERE p2.reference = p1.protein AND
                      p2.protein ~ 'EPA3_HUMAN') AND
       EXISTS (SELECT p3.domain
                 FROM classification AS p3
                WHERE p3.domain = p1.domain AND
                      p3.class = 'extra');

I'd select bak the first field in thew subselects from p2 and p3
respectfully.

> As you see, they differ only in the last line. The problem is, that
> the time
> they need differes tremendously. Query a) takes about 5 sec, query b)
> 2 min !!
> I don't understand why this last constraint over an indexed field
> leads to such
> a difference.
>
> Here is what explain says:
>
> a)
> NOTICE:  QUERY PLAN:
>
> Unique  (cost=5193.21 size=0 width=0)
>   ->  Sort  (cost=5193.21 size=0 width=0)
>         ->  Nested Loop  (cost=5193.21 size=1 width=48)
>               ->  Nested Loop  (cost=5191.16 size=1 width=36)
>                     ->  Seq Scan on p2  (cost=5189.11 size=1 width=12)
>                     ->  Index Scan on p1  (cost=2.05 size=62250
> width=24)
>               ->  Index Scan on p3  (cost=2.05 size=189 width=12)
> EXPLAIN
>
> b)
> NOTICE:  QUERY PLAN:
>
> Unique  (cost=680.60 size=0 width=0)
>   ->  Sort  (cost=680.60 size=0 width=0)
>         ->  Nested Loop  (cost=680.60 size=1 width=48)
>               ->  Nested Loop  (cost=4.10 size=330 width=36)
>                     ->  Index Scan on p3  (cost=2.05 size=1 width=12)
>                     ->  Index Scan on p1  (cost=2.05 size=62250
> width=24)
>               ->  Index Scan on p2  (cost=2.05 size=1 width=12)
>
> EXPLAIN
>
>
> Dou you have any idea, how to improve query b) ??
>
> Thanx,
> Joerg
>
Let me know how it performs (or if I made any typoes),
        -DEJ

pgsql-general by date:

Previous
From: Roman Volkoff
Date:
Subject: Apache, php3 & PostgreSQL
Next
From: Byron Nikolaidis
Date:
Subject: NEW ODBC DRIVER v.0246