Обсуждение: BUG #5230: Limit operator slows down

The following bug has been logged online:

Bug reference:      5230
Logged by:          aftab
Email address:      akhangd@hotmail.co.uk
PostgreSQL version: 8.3.8
Operating system:   Centos 5
Description:        Limit operator slows down
Details:

S1="SELECT *
FROM position WHERE
position.POSITION_STATE_ID=2 AND
position.TARGET_ID=18
ORDER BY position.ID DESC
 ";

 S2="SELECT *
 FROM position WHERE
 position.POSITION_STATE_ID=2 AND
 position.TARGET_ID=18
 ORDER BY position.ID DESC
 LIMIT 1
 ";

 S1 takes 0.16ms compared to S2 which takes 5 secs. Both S1 and S2 are same
except "LIMIT 1 " is added to S2.

Query Plan for S1:

 QUERY PLAN
----------------------------------------------------------------------------
-------------------------------------------------------------
 Sort  (cost=209341.40..209767.55 rows=170457 width=297) (actual
time=0.200..0.200 rows=1 loops=1)
   Sort Key: id
   Sort Method:  quicksort  Memory: 25kB
   ->  Bitmap Heap Scan on "position"  (cost=5050.22..146754.50 rows=170457
width=297) (actual time=0.188..0.189 rows=1 loops=1)
         Recheck Cond: (target_id = 18)
         Filter: (position_state_id = 2)
         ->  Bitmap Index Scan on position_target_fk  (cost=0.00..5007.61
rows=170912 width=0) (actual time=0.134..0.134 rows=2 loops=1)
               Index Cond: (target_id = 18)
 Total runtime: 0.275 ms
(9 rows)

Query Plan for S2:

 QUERY PLAN
----------------------------------------------------------------------------
----------------------------------------------------------------------------
---
 Limit  (cost=0.00..2.43 rows=1 width=297) (actual time=3053.220..3053.221
rows=1 loops=1)
   ->  Index Scan Backward using position_pkey on "position"
(cost=0.00..413713.28 rows=170459 width=297) (actual time=3053.216..3053.216
rows=1 loops=1)
         Filter: ((position_state_id = 2) AND (target_id = 18))
 Total runtime: 3053.297 ms
(4 rows)

On Thu, Dec 3, 2009 at 7:07 AM, aftab <akhangd@hotmail.co.uk> wrote:
>
> The following bug has been logged online:
>
> Bug reference: =A0 =A0 =A05230
> Logged by: =A0 =A0 =A0 =A0 =A0aftab
> Email address: =A0 =A0 =A0akhangd@hotmail.co.uk
> PostgreSQL version: 8.3.8
> Operating system: =A0 Centos 5
> Description: =A0 =A0 =A0 =A0Limit operator slows down
> Details:
>
> S1=3D"SELECT *
> FROM position WHERE
> position.POSITION_STATE_ID=3D2 AND
> position.TARGET_ID=3D18
> ORDER BY position.ID DESC
> =A0";
>
> =A0S2=3D"SELECT *
> =A0FROM position WHERE
> =A0position.POSITION_STATE_ID=3D2 AND
> =A0position.TARGET_ID=3D18
> =A0ORDER BY position.ID DESC
> =A0LIMIT 1
> =A0";
>
> =A0S1 takes 0.16ms compared to S2 which takes 5 secs. Both S1 and S2 are =
same
> except "LIMIT 1 " is added to S2.
>
> Query Plan for S1:
>
> =A0QUERY PLAN
> -------------------------------------------------------------------------=
---
> -------------------------------------------------------------
> =A0Sort =A0(cost=3D209341.40..209767.55 rows=3D170457 width=3D297) (actual
> time=3D0.200..0.200 rows=3D1 loops=3D1)
> =A0 Sort Key: id
> =A0 Sort Method: =A0quicksort =A0Memory: 25kB
> =A0 -> =A0Bitmap Heap Scan on "position" =A0(cost=3D5050.22..146754.50 ro=
ws=3D170457
> width=3D297) (actual time=3D0.188..0.189 rows=3D1 loops=3D1)
> =A0 =A0 =A0 =A0 Recheck Cond: (target_id =3D 18)
> =A0 =A0 =A0 =A0 Filter: (position_state_id =3D 2)
> =A0 =A0 =A0 =A0 -> =A0Bitmap Index Scan on position_target_fk =A0(cost=3D=
0.00..5007.61
> rows=3D170912 width=3D0) (actual time=3D0.134..0.134 rows=3D2 loops=3D1)
> =A0 =A0 =A0 =A0 =A0 =A0 =A0 Index Cond: (target_id =3D 18)
> =A0Total runtime: 0.275 ms
> (9 rows)
>
> Query Plan for S2:
>
> =A0QUERY PLAN
> -------------------------------------------------------------------------=
---
> -------------------------------------------------------------------------=
---
> ---
> =A0Limit =A0(cost=3D0.00..2.43 rows=3D1 width=3D297) (actual time=3D3053.=
220..3053.221
> rows=3D1 loops=3D1)
> =A0 -> =A0Index Scan Backward using position_pkey on "position"
> (cost=3D0.00..413713.28 rows=3D170459 width=3D297) (actual time=3D3053.21=
6..3053.216
> rows=3D1 loops=3D1)
> =A0 =A0 =A0 =A0 Filter: ((position_state_id =3D 2) AND (target_id =3D 18))
> =A0Total runtime: 3053.297 ms
> (4 rows)

This is the same thing you posted yesterday....

...Robert

On Thu, Dec 3, 2009 at 10:45 AM, aftab khan <akhangd@hotmail.co.uk> wrote:
> This is not exactly the same what I posted yesterday, What I have posted
> today also include EXPLAIN ANALYZE OUTPUT .

[ adding the list back to the CC line ]

It looks like you have a problem very similar to one Laurence Laborde
was complaining about recently on pgsql-performance.  My discussion of
the problem is here:

http://archives.postgresql.org/pgsql-performance/2009-12/msg00022.php

In your case, the reason why the estimates are off is because of the
lack of multi-column statistics, rather than because of the choice of
operator, but the issues are similar.

...Robert