"Ian Harding" <ianh(at)tpchd(dot)org> writes:
> Doh! I set enable_seqscan = off and things are snappy again.
> Same problem though, how can I fix the query so the optimizer comes to the same conclusion?
I notice that in the "before" state, the planner thinks there will be
some thousands of rows coming out of each join, but in the "after"
state only one row. Which of these is more correct? (If you are
running 7.2, it'd be helpful to show EXPLAIN ANALYZE not just
EXPLAIN results for each case.)
regards, tom lane