Re: Oddball data distribution giving me planner headaches

-----Original Message-----
From: Jeff Amiel [mailto:becauseimjeff(at)yahoo(dot)com]
Sent: Friday, December 02, 2011 4:15 PM
To: pgsql-general(at)postgresql(dot)org; David Johnston
Subject: RE: [GENERAL] Oddball data distribution giving me planner headaches

--- On Fri, 12/2/11, David Johnston <polobo(at)yahoo(dot)com> wrote:

> What happens if you disable, say, nested loops and/or index scans?

planner selects different join/indexing techniques (query is slower) but row
estimates (bad) remain identical.

---------------------------------------------------------
My, possibly naïve, observation:

So aside from the fact the estimates seem to be off the planner has still
chosen the most effective plan? In that situation no matter how accurate
you get the statistics you will not gain any performance because the planner
will never choose a different plan.

IF you can find a combination of settings that forces a better performing
plan THEN steps can be taken (by others more skilled than myself) to
configure the statistics/planner system to get that better plan naturally.

David J.