Обсуждение: group by query plan on already clustered index

Поиск
Список
Период
Сортировка

group by query plan on already clustered index

От
Sandeep Gupta
Дата:
On a table T with two fields, f1 and f2, the sql command 

select count(f2), f1 
from  T
group by f1

result is seq scan followed by a sort on f1 (see the query plan below):

   GroupAggregate  (cost=21566127.88..22326004.09 rows=987621 width=8)
   ->  Sort  (cost=21566127.88..21816127.88 rows=100000000 width=8)
         Sort Key: pid
         ->  Seq Scan on tc  (cost=0.00..1442478.00 rows=100000000 width=8)


However, the table is already has clustered index on f1.  My question is why doesn't 
postgres simply scan the table and compute the group? Why it needs to sort on f1 again?
How can I force postgres to use the clustered index for group by?


Thanks.
Sandeep


Re: group by query plan on already clustered index

От
Pavel Stehule
Дата:
Hello

2013/7/25 Sandeep Gupta <gupta.sandeep@gmail.com>:
> On a table T with two fields, f1 and f2, the sql command
>
> select count(f2), f1
> from  T
> group by f1
>
> result is seq scan followed by a sort on f1 (see the query plan below):
>
>    GroupAggregate  (cost=21566127.88..22326004.09 rows=987621 width=8)
>    ->  Sort  (cost=21566127.88..21816127.88 rows=100000000 width=8)
>          Sort Key: pid
>          ->  Seq Scan on tc  (cost=0.00..1442478.00 rows=100000000 width=8)
>
>
> However, the table is already has clustered index on f1.  My question is why
> doesn't
> postgres simply scan the table and compute the group? Why it needs to sort
> on f1 again?
> How can I force postgres to use the clustered index for group by?
>

PostgreSQL doesn't support index organized tables. Statement CLUSTER
doesn't ensure order of heap for ever. After any INSERT, UPDATE you
can lost a order.

So every time, you have to check order.

PostgreSQL 9.2 support a index only scan - but it is not used in your plan.

Regards

Pavel Stehule




>
> Thanks.
> Sandeep
>
>