Horton

Re: wip: functions median and percentile

Tom Lane <tgl@sss.pgh.pa.us>

View thread
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Dean Rasheed <dean.a.rasheed@gmail.com>
Cc: Hitoshi Harada <umi.tanuki@gmail.com>, Robert Haas <robertmhaas@gmail.com>, Greg Stark <gsstark@mit.edu>, Pavel Stehule <pavel.stehule@gmail.com>, David Fetter <david@fetter.org>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2010-10-11T15:44:08Z
Lists: pgsql-hackers 
Dean Rasheed <dean.a.rasheed@gmail.com> writes:
> On 11 October 2010 15:03, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Reflecting on it, I think it'd be best to allow an agg to
>> provide an estimation function that'd be told the input data type and
>> expected number of rows --- even on a per-aggregate basis, a constant
>> estimate just isn't good enough.

> How good will that estimate of the number of rows be though?

It can't possibly be any worse than a hard-wired constant ;-)

> If they're coming from a SRF it could be a huge under-estimate, and you'd
> still risk eating all the memory, if you allowed a hash aggregate.

If, for a particular aggregate, you're too chicken to ever allow hash
aggregation, you could just return a very large number from the
estimation hook function.  I doubt that's a very useful behavior in the
majority of cases though.

			regards, tom lane