On Fri, Jul 13, 2012 at 11:11 AM, Eoghan Murray <eoghan(at)qatano(dot)com> wrote:
> 8.4: 314ms: http://explain.depesz.com/s/GkX
> 9.1: 10,059ms :http://explain.depesz.com/s/txn
> 9.1 with setting `enable_material = off`: 1,635ms
> http://explain.depesz.com/s/gIu
I think the problem is it's using a merge join, with a sort inside
that's producing 600x more rows than expected, while 8.4 does a hash
join with no intermediate big tables instead.
What's your configuration like in both servers? (that could explain
planning differences)