Skip site navigation (1) Skip section navigation (2)

Efficiency again...

From: Michael Richards <miker(at)scifair(dot)acadiau(dot)ca>
To: pgsql-hackers(at)postgresql(dot)org
Subject: Efficiency again...
Date: 1998-07-22 23:17:33
Message-ID: (view raw, whole thread or download thread mbox)
Lists: pgsql-hackers
I just noticed something interesting. I don't know if my idea is better or
if it wasn't implemented because it violates some SQL rule...

searchengine=> create table test ( test1 int4, test2 int4);
searchengine=> create index test_itest1 on test (test1);
<insert a pile of data so it looks like so>
searchengine=> select * from test;
    1|    3
    1|    5
    1|    9
    2|    1
    2|    3
    2|    6
    2|    9
    3|    9
    4|    5
(9 rows)

Now here is the plan I expect for a single test1 value
searchengine=> explain select * from test where test1=1;
Index Scan on test  (cost=0.00 size=0 width=8)

But look:
searchengine=> explain select * from test where test1=1 or test1=2;
Seq Scan on test  (cost=0.00 size=0 width=8)

ugh! Sequential. This may be OK for a small database, but in my
application I have many rows:
searchengine=> explain select * from word_detail where word_id=23423 or

Seq Scan on word_detail  (cost=205938.73 size=510342 width=10)

That costs a _LOT_.

Wouldn't it be better to do n sequential scans where n is the number of
or'd together values? Using IN doesn't help out either...

searchengine=> explain select * from test where test1 IN (5,9);
Seq Scan on test  (cost=0.00 size=0 width=8)

Sometimes I wish I had the power to tell the DBMS how I wanted a query



pgsql-hackers by date

Next:From: Bruce MomjianDate: 1998-07-23 00:40:58
Subject: Re: [HACKERS] Efficiency again...
Previous:From: Ken McGlothlenDate: 1998-07-22 22:34:16
Subject: Re: [GENERAL] Re: [HACKERS] [Fwd: SGVLLUG Oracle and Informix on Linux]

Privacy Policy | About PostgreSQL
Copyright © 1996-2017 The PostgreSQL Global Development Group