Thread: subselect query time and loops problem

subselect query time and loops problem

From

pankaj naug

Date:

10 April 2005, 00:22:09

hi,

I am using this query.

SELECT * FROM guild_properties_buy WHERE agent IN (SELECT agent_id FROM guild_agents WHERE address_region = 'midlands' AND active='on' AND status=0) AND country = 'United Kingdom' AND active='on' AND status='0' ORDER BY price DESC LIMIT 10 OFFSET 0

when i run this in my development server i get this.

Limit (cost=1680331.30..1680331.31 rows=1 width=541) (actual time=3631.97..3632.00 rows=10 loops=1)
-> Sort (cost=1680331.30..1680331.31 rows=1 width=541) (actual time=3631.96..3631.98 rows=11 loops=1)
        Sort Key: price
        -> Seq Scan on guild_properties_buy (cost=0.00..1680331.29 rows=1 width=541) (actual time=39.39..3556.80 rows=4747 loops=1)
              Filter: ((country = 'United Kingdom'::character varying) AND (active = 'on'::character varying) AND (status = 0) AND (subplan))
              SubPlan
                -> Materialize (cost=57.15..57.15 rows=1 widt h=4) (actual time=0.00..0.05 rows=88 loops=27235)
                      -> Seq Scan on guild_agents (cost=0.00..57.15 rows=1 width=4) (actual time=0.04..1.76 rows=100 loops=1)
                            Filter: ((address_region = 'midlands'::character varying) AND (active = 'on'::character varying) AND (status = 0))
Total runtime: 3633.46 msec

when i run this in my hosting server i get this.

Limit (cost=847964.41..847964.43 rows=10 width=1036) (actual time=28265.15..28265.19 rows=10 loops=1)
-> Sort (cost=847964.41..847999.30 rows=13957 width=1036) (actual time=28265.15..28265.17 rows=11 loops=1)
        Sort Key: price
        -> Seq Scan on guild_properties_buy (cost=0.00..832943.58 rows=13957 width=1036) (actual time=6.88..28157.11 rows=4790 loops=1)
              Filter: ((country = 'United Kingdom'::character varying) AND (active = 'on'::character varying) AND (status = 0) AND (subplan))
              SubPlan
                -> Seq Scan on guild_agents (cost=0.00..56.15 rows=100 width=4) (actual time=0.01..0.95 rows=87 loops=27173)
                      Filter: ((address_region = 'midlands'::character varying) AND (active = 'on'::character varying) AND (status = 0))
Total runtime: 28269.32 msec

when i use my script to replace sub query then both servers run fine.

SELECT * FROM guild_properties_buy WHERE agent IN (

56259,56397......................

) AND country = 'United Kingdom' AND active='on' AND status='0' ORDER BY price DESC LIMIT 10 OFFSET 0

Both servers have same version with same postgresql.conf. But the time taken by both the servers with subselect are way different. the loops and the query time are completely different. both servers have same indexes.

Any help will be greately appreciated..

Best Regards

Pankaj Naug

Do you Yahoo!?
Yahoo! Small Business - Try our new resources site!

Re: subselect query time and loops problem

From

Tom Lane

Date:

10 April 2005, 03:26:27

pankaj naug <pankajnaug@yahoo.com> writes:
> Both servers have same version with same postgresql.conf. But the time taken by both the servers with subselect are
waydifferent. the loops and the query time are completely different. both servers have same indexes.
 

Evidently one has been analyzed much more recently than the other,
because the estimated row counts are wildly different.

You didn't say which PG version this is, but I gather that it's pre-7.4,
which means that the performance of IN (SELECT ...) is generally going
to be awful.  Either rewrite as a join or update to 7.4 or later.
        regards, tom lane

Re: subselect query time and loops problem

From

pankaj naug

Date:

10 April 2005, 18:54:29

Thanks Tom,

I am using version 7.3.4, having problems updating to postgres 8 because of cpanel problems.

regarding,

>Evidently one has been analyzed much more recently than the other,
because the estimated row counts are wildly different.

Both the explain/analyse queries has been run at the same time.

Best regards

Pankaj

Yahoo! Mail Mobile
Take Yahoo! Mail with you! Check email on your mobile phone.

Re: subselect query time and loops problem

From

Ragnar Hafstað

Date:

10 April 2005, 21:27:25

On Sun, 2005-04-10 at 07:54 -0700, pankaj naug wrote:
> [quoting Tom] 
> >Evidently one has been analyzed much more recently than the other,
> because the estimated row counts are wildly different.
>  
> Both the explain/analyse queries has been run at the same time.

in that case, is the data the same?
if so, what about STATISTICS settings for relevant columns?

just to make things clear, have both databases have been
ANALYZEd or VACUUM ANALYZEd recently ? (in case your
'explain/analyse' only refers to a EXPLAIN ANALYZE)

gnari