Robert Haas <robertmhaas@gmail.com> writes:
> I had a suspicion we were going to see something like this. You're
> using several NOT EXISTS clauses and 8.4devel is converting those into
> Anti Joins. Aside from the longer planning time, the resulting plan
> appears to have a much higher estimated cost, so I'm suspicious of a
> bug in the new Semi/Anti join stuff,
More likely the other way round --- I think 8.3's rowcounts and hence
costs are flights of fancy, and 8.4 is more likely to be in the right
ballpark. We need to see EXPLAIN ANALYZE to find out though.
regards, tom lane