Re: Worse perfomance on 8.2.0 than on 7.4.14

=?iso-8859-1?q?Rolf=20=D8stvik?= <rolfostvik@yahoo.no> writes:
> First i have some queries to give you a feel of size
> of datasets and plans and times.

You said earlier that essentially all the rows of step_result_subset
have step_parent = 0 ... is that really true?  I can hardly believe
that either 7.4 or 8.2 would use an indexscan for Q-A if so.

I'd be interested to see the results of

prepare foo(int) as select id from step_result_subset sr
    where uut_result = $1 and step_parent = 0;
explain analyze execute foo(42);

(use some representative uut_result value instead of 42).  If it doesn't
want to use an indexscan for this, disable plan types until it does.
This would perhaps shed some light on why 8.2 doesn't want to use a scan
like that as the inside of a nestloop.

            regards, tom lane