Bruno Wolff III <bruno@wolff.to> writes:
> By disabling merge joins and using the updated view, I got the query down
> to about 25% of its original runtime.
> Note the query estimate is off by a factor of more than 10.
This seems to indicate some estimation problems in cost_hashjoin; the
estimated cost for the hashjoin is evidently a lot higher than it should
be.
Are you interested in digging into this; or could you send me a dump of
the tables used in the view and query, so I could look into it?
regards, tom lane