Обсуждение: performance question

Поиск
Список
Период
Сортировка

performance question

От
"Reinoud van Leeuwen"
Дата:
Can somebody explain to me:

> radius=# explain select count (radiuspk) from radius ;
> NOTICE:  QUERY PLAN:
>
> Aggregate  (cost=12839.79..12839.79 rows=1 width=8)
>   ->  Seq Scan on radius  (cost=0.00..11843.43 rows=398543 width=8)
>
> EXPLAIN


This query answers me *instantly* after hitting return

> radius=# select count (radiuspk) from radius ;
>  count
>  --------
>   398543
>  (1 row)

This query takes about 3 seconds. But the query plan *already* knows the 
number of rows ("rows=398543"). So why does it take 3 seconds. Is my 
assumption correct that the optimiser still can be optimized a little? :-)

Reinoud (not that this is a real problem, just wondering)




Re: performance question

От
Peter Eisentraut
Дата:
Reinoud van Leeuwen writes:

> > radius=# select count (radiuspk) from radius ;
> >  count
> >  --------
> >   398543
> >  (1 row)
>
> This query takes about 3 seconds. But the query plan *already* knows the
> number of rows ("rows=398543").

This is only an estimate which is only updated by VACUUM.  Presumably you
didn't add or remove any rows since your last VACUUM.

-- 
Peter Eisentraut   peter_e@gmx.net   http://funkturm.homeip.net/~peter



Re: performance question

От
Stephan Szabo
Дата:
On Tue, 28 Aug 2001, Reinoud van Leeuwen wrote:

> Can somebody explain to me:
> 
> > radius=# explain select count (radiuspk) from radius ;
> > NOTICE:  QUERY PLAN:
> >
> > Aggregate  (cost=12839.79..12839.79 rows=1 width=8)
> >   ->  Seq Scan on radius  (cost=0.00..11843.43 rows=398543 width=8)
> >
> > EXPLAIN
> 
> 
> This query answers me *instantly* after hitting return
> 
> > radius=# select count (radiuspk) from radius ;
> >  count
> >  --------
> >   398543
> >  (1 row)
> 
> This query takes about 3 seconds. But the query plan *already* knows the 
> number of rows ("rows=398543"). So why does it take 3 seconds. Is my 
> assumption correct that the optimiser still can be optimized a little? :-)

Not in this case.  The row numbers from explain are just estimates 
from the last vacuum.  As you modify the table, the estimated rows
will be off.

For example:
sszabo=> create table a (a int);
CREATE
sszabo=> insert into a values (100);
INSERT 808899 1
sszabo=> insert into a values (101);
INSERT 808900 1
sszabo=> explain select count(a) from a;
NOTICE:  QUERY PLAN:

Aggregate  (cost=22.50..22.50 rows=1 width=4) ->  Seq Scan on a  (cost=0.00..20.00 rows=1000 width=4)

EXPLAIN
sszabo=> vacuum analyze a;
VACUUM
sszabo=> explain select count(a) from a;
NOTICE:  QUERY PLAN:

Aggregate (cost=1.02..1.02 rows=1 width=4) ->  Seq Scan on a  (cost=0.00..1.02 rows=2 width=4)

EXPLAIN
sszabo=> insert into a values (102);
INSERT 808902 1
sszabo=> explain select count(a) from a;
NOTICE:  QUERY PLAN:

Aggregate  (cost=1.02..1.02 rows=1 width=4) ->  Seq Scan on a  (cost=0.00..1.02 rows=2 width=4)

EXPLAIN
sszabo=> vacuum analyze a;
VACUUM
sszabo=> explain select count(a) from a;
NOTICE:  QUERY PLAN:

Aggregate  (cost=1.04..1.04 rows=1 width=4) ->  Seq Scan on a  (cost=0.00..1.03 rows=3 width=4)

EXPLAIN




Re: performance question

От
"Reinoud van Leeuwen"
Дата:
> On Tue, 28 Aug 2001, Reinoud van Leeuwen wrote:
> 
>> Can somebody explain to me:
>> 
>> > radius=# explain select count (radiuspk) from radius ;
>> > NOTICE:  QUERY PLAN:
>> >
>> > Aggregate  (cost=12839.79..12839.79 rows=1 width=8)
>> >   ->  Seq Scan on radius  (cost=0.00..11843.43 rows=398543 width=8)
>> >
>> > EXPLAIN
>> 
>> 
>> This query answers me *instantly* after hitting return
>> 
>> > radius=# select count (radiuspk) from radius ;
>> >  count
>> >  --------
>> >   398543
>> >  (1 row)
>> 
>> This query takes about 3 seconds. But the query plan *already* knows
>> the  number of rows ("rows=398543"). So why does it take 3 seconds. Is
>> my  assumption correct that the optimiser still can be optimized a
>> little? :-)
> 
> Not in this case.  The row numbers from explain are just estimates 
> from the last vacuum.  As you modify the table, the estimated rows will
> be off.

Yes, I just found out that somebody else is running a script on our test 
server that vacuums all databases each night. That explains a lot.

Thanx for thinking with me

Reinoud