Marcus Engene <mengpg2(at)engene(dot)se> writes:
> ... In a case with some 5000 rows belonging to owner 123, this select really
> takes a long time. Way longer than without the subselect and order by
> filelength. I agree that with the subselect it would take some extra
> juice, but in my mind it would do some hash in memory which wouldn't be
> too slow to lookup in.
8.4 can turn EXISTS subqueries into hash joins, but previous releases
won't...
regards, tom lane