| From: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
|---|---|
| To: | Greg Stark <gsstark(at)mit(dot)edu> |
| Cc: | Simon Riggs <simon(at)2ndquadrant(dot)com>, Csaba Nagy <nagy(at)ecircle-ag(dot)com>, "Jonah H(dot) Harris" <jonah(dot)harris(at)gmail(dot)com>, Neil Conway <neilc(at)samurai(dot)com>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org> |
| Subject: | Re: generalizing the planner knobs |
| Date: | 2005-12-05 03:17:27 |
| Message-ID: | 12469.1133752647@sss.pgh.pa.us |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
Greg Stark <gsstark(at)mit(dot)edu> writes:
> Plan stability is also an important feature, especially for OLTP
> systems which have hard real-time requirements. OLTP systems typically
> don't care about getting the "best" plan for a query, only a plan that
> is "good enough".
> "Good enough" means it can keep up with the rate of incoming requests; it
> doesn't matter whether it keeps up with 10% headroom or 20% headroom. But if
> one incoming query even one in a thousand takes 1000% of the time available
> then the entire system risks falling down.
Is it worth pointing out that using the same plan all the time is *no*
recipe for guaranteeing response time? There is no such thing as a plan
that is good for every case --- outlying data values can make a
usually-good plan blow out your performance guarantee anyway. Disabling
the planner is just a recipe for ensuring that that will happen, IMHO.
regards, tom lane
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Bruce Momjian | 2005-12-05 03:18:10 | Re: port/snprintf.c (was Re: Numeric 508 datatype) |
| Previous Message | Tom Lane | 2005-12-05 03:07:38 | Re: Reducing relation locking overhead |