Re: Regression: 8.3 2 seconds -> 8.4 100+ seconds

Francisco Reyes <lists@stringsutils.com> writes:
> CentOS 5.4 and 5.5
> Query
> SELECT sum(usramt) as givensum,
>               sum (case when usedid > 0 then usramt else 0 end) as usedsum
>               FROM argrades
>               WHERE userstatus in (5) and
>               membid in (select distinct members.membid from members, cards
> where members.membid = cards.membid and members.membershipid = 40 and useraccount in
> ( select useraccount from cards where membid in
> (select membid from members where commonid = 3594)))

Please provide a self-contained test case for this.  In testing here,
8.4 can still figure out that it ought to use an indexscan against the
large table ... so there is some aspect of your tables that you're not
showing us.

            regards, tom lane