On Fri, Jul 1, 2016 at 11:45 AM, Alvaro Herrera
<alvherre@2ndquadrant.com> wrote:
> Merlin Moncure wrote:
>
>> It's pretty easy to craft a query where you're on the winning side,
>> but what's the worst case of doing two pass...is constant folding a
>> non trivial fraction of planning time?
>
> One thing that has been suggested is to re-examine the plan after
> planning is done, and if execution time is estimated to be large (FSVO),
> then run a second planning pass with more expensive optimizations
> enabled to try and find better plans. The guiding principle would be to
> continue to very quickly find good enough plans for
> frequent/small/simple queries, but spend more planning effort on more
> complex ones where execution is likely to take much longer than planning
> time.
>
> So doing constant-folding twice would be enabled for the second planning
> pass.
I like this idea. Maybe a GUC controlling the cost based cutoff (with
0 meaning, "assume the worst and plan the hard way first").
merlin