| From: | Merlin Moncure <mmoncure(at)gmail(dot)com> |
|---|---|
| To: | Linos <info(at)linos(dot)es> |
| Cc: | PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org> |
| Subject: | Re: performance regression in 9.2/9.3 |
| Date: | 2014-06-05 15:51:34 |
| Message-ID: | CAHyXU0xH8SC+KfaNd69vFh=S3=puSvD75KNKJXisknA6JiB17Q@mail.gmail.com |
| Views: | Whole Thread | Raw Message | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
On Thu, Jun 5, 2014 at 9:54 AM, Linos <info(at)linos(dot)es> wrote:
> What I don't understand is why the statistics have this bad information, all my tests are done on a database just restored and analyzed. Can I do something to improve the quality of my database statistics and let the planner do better choices? Maybe increase the statistics target of the columns involved?
By that I meant row count estimates coming out of the joins are way
off. This is pushing the planner into making bad choices. The most
pervasive problem I see is that the row count estimate boils down to
'1' at some juncture causing the server to favor nestloop/index scan
when something like a hash join would likely be more appropriate.
merlin
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Robert Haas | 2014-06-05 16:02:56 | Re: slotname vs slot_name |
| Previous Message | Linos | 2014-06-05 15:28:37 | Re: performance regression in 9.2/9.3 |