On Thu, Jul 2, 2009 at 12:48 PM, Euler Taveira de Oliveira<eu...@timbira.com> wrote: > I know that it didn't solve the estimation problem but ... IMHO the > [under|over]estimation should be treated by an external tool (autoexplain?). > So when we enable the query progress and some node reports a difference > between estimated and real more than x%, log the plan. Doing it, we will be > helping DBAs to investigate the bad plans.
Keep in mind that it is frequently the case that the estimates are substantially off but the plan still works OK. I just put a dirty hack into one of my apps to improve the selectivity estimates by a factor of 200, but they're still off by a factor of 5. Even when they were off by 1000x the bad plan happened only intermittently. You notice the cases where the estimates are off and it makes for a bad plan, but there are lots of other cases where the estimates are off but the plan is still OK. ...Robert -- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers
