Обсуждение: BUG #2050: Bad plan by using of LIKE

The following bug has been logged online:

Bug reference:      2050
Logged by:          Johannes
Email address:      postgres@arltus.de
PostgreSQL version: 8.0.3
Operating system:   i686-pc-linux-gnu, compiled by GCC i686-pc-linux-gnu-gcc
(GCC) 3.3.5-20050130 (Gentoo 3.3.5.20050130-r1, ssp-3.3.5.20050130-1,
pie-8.7.7.1
Description:        Bad plan by using of LIKE
Details:

I use this table:
CREATE TABLE content (
    title character(64) NOT NULL,
    content_htm character(128) NOT NULL,
    id serial NOT NULL
);
ALTER TABLE ONLY content ADD CONSTRAINT p_key PRIMARY KEY (id);
CREATE INDEX idx1 ON content USING btree (title);

I have filling this with 1000000 rows by dbmonster and use this statement,
after analyze und reindex:

SELECT title FROM content WHERE title LIKE 'teane%';

It uses to long time, I compare this with sybase
and I was disappointed, but explain shows the reason.


EXPLAIN SELECT title FROM content WHERE title LIKE 'teane%';
Seq Scan on content  (cost=0.00..75647.59 rows=1 width=68)
   Filter: (title ~~ 'teane%'::text)

Now I change this SQL to:
SELECT title FROM content WHERE title >= 'teane' AND title < 'teanez';

I think it means the same but it works very fast by using my index.
(1600 ms up to 2 ms !! sybase uses 4 ms)

Explain shows the reason:

EXPLAIN SELECT title FROM content WHERE title >= 'teane' AND title <
'teanez';
 Index Scan using idx1 on content  (cost=0.00..4.02 rows=1 width=68)
   Index Cond: ((title >= 'teane'::bpchar) AND (title < 'teanez'::bpchar))

I'am not sure is this a bug or not, but without some modifications in the
postgres.conf the difference are more significant ( 22.530 ms ! to 2 ms).

"Johannes" <postgres@arltus.de> writes:
> EXPLAIN SELECT title FROM content WHERE title LIKE 'teane%';
> Seq Scan on content  (cost=0.00..75647.59 rows=1 width=68)
>    Filter: (title ~~ 'teane%'::text)

Apparently you're using a non-C locale.  LIKE can only use an index if
you're in the C locale or you make the index with a special index
operator class.  See
http://www.postgresql.org/docs/8.0/static/indexes-opclass.html

            regards, tom lane

On 11/17/05, Johannes <postgres@arltus.de> wrote:
>
> The following bug has been logged online:
>
> Bug reference:      2050
> Logged by:          Johannes
> Email address:      postgres@arltus.de
> PostgreSQL version: 8.0.3
> Operating system:   i686-pc-linux-gnu, compiled by GCC i686-pc-linux-gnu-=
gcc
> (GCC) 3.3.5-20050130 (Gentoo 3.3.5.20050130-r1, ssp-3.3.5.20050130-1,
> pie-8.7.7.1
> Description:        Bad plan by using of LIKE
> Details:
>
> I use this table:
> CREATE TABLE content (
>    title character(64) NOT NULL,
>    content_htm character(128) NOT NULL,
>    id serial NOT NULL
> );
> ALTER TABLE ONLY content ADD CONSTRAINT p_key PRIMARY KEY (id);
> CREATE INDEX idx1 ON content USING btree (title);
>
> I have filling this with 1000000 rows by dbmonster and use this statement,
> after analyze und reindex:
>
> SELECT title FROM content WHERE title LIKE 'teane%';
>
> It uses to long time, I compare this with sybase
> and I was disappointed, but explain shows the reason.
>
>
> EXPLAIN SELECT title FROM content WHERE title LIKE 'teane%';
> Seq Scan on content  (cost=3D0.00..75647.59 rows=3D1 width=3D68)
>   Filter: (title ~~ 'teane%'::text)
>
> Now I change this SQL to:
> SELECT title FROM content WHERE title >=3D 'teane' AND title < 'teanez';
>
> I think it means the same but it works very fast by using my index.
> (1600 ms up to 2 ms !! sybase uses 4 ms)
>
> Explain shows the reason:
>
> EXPLAIN SELECT title FROM content WHERE title >=3D 'teane' AND title <
> 'teanez';
>  Index Scan using idx1 on content  (cost=3D0.00..4.02 rows=3D1 width=3D68)
>   Index Cond: ((title >=3D 'teane'::bpchar) AND (title < 'teanez'::bpchar=
))
>
> I'am not sure is this a bug or not, but without some modifications in the
> postgres.conf the difference are more significant ( 22.530 ms ! to 2 ms).
>

Maybe you are using a non C-locale? they are known to not use indexes
in LIKE querys... instead, you have to create an index with appropiate
class operator...

--
regards,
Jaime Casanova
(DBA: DataBase Aniquilator ;)

On Thu, Nov 17, 2005 at 06:00:35PM +0000, Johannes <postgres@arltus.de> wrote:

> SELECT title FROM content WHERE title LIKE 'teane%';
>
> It uses to long time, I compare this with sybase
> and I was disappointed, but explain shows the reason.
>
>
> EXPLAIN SELECT title FROM content WHERE title LIKE 'teane%';
> Seq Scan on content  (cost=0.00..75647.59 rows=1 width=68)
>    Filter: (title ~~ 'teane%'::text)
>
> Now I change this SQL to:
> SELECT title FROM content WHERE title >= 'teane' AND title < 'teanez';
>
> I think it means the same but it works very fast by using my index.
> (1600 ms up to 2 ms !! sybase uses 4 ms)


You might want to take a look at

  http://www.postgresql.org/docs/8.1/interactive/indexes-opclass.html

and build your indexes on text fields with *_pattern_ops from now on.
Does the trick.

cheers,
bkw