Kai Hessing <kai.hessing@xxxxxxxxxx> writes: > PostgreSQL 8.1.4 (The one which is taking... a very long time...) > Filter: (NOT (subplan)) > SubPlan > -> Materialize (cost=7751.81..8497.94 rows=53613 width=4) > PostgreSQL 8.0.8 (The one which takes only a few seconds... on a much > slower system...) > Filter: (NOT (hashed subplan)) > SubPlan > -> Seq Scan on stud_vera (cost=0.00..7424.24 rows=1461 > width=4) > Filter: (veraid = 2) I bet the problem is that you're not getting a hashed subplan in 8.1. What do you have work_mem set to on the two systems? The fact that the rowcount estimate for the subplan is much larger may be causing the change, too. Why is that --- perhaps you haven't ANALYZEd stud_vera recently on one system or the other? How many rows will really come out of the sub-select (ie, what's "select count(*) from stud_vera where veraid = 2")? regards, tom lane
