Horton

Re: ECPG FETCH readahead

Noah Misch <noah@leadboat.com>

View thread
From: Noah Misch <noah@leadboat.com>
To: Michael Meskes <meskes@postgresql.org>
Cc: Boszormenyi Zoltan <zb@cybertec.at>, PG Hackers <pgsql-hackers@postgresql.org>, Robert Haas <robertmhaas@gmail.com>, Heikki Linnakangas <heikki.linnakangas@enterprisedb.com>, Bruce Momjian <bruce@momjian.us>
Date: 2012-03-05T18:50:04Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Fix some "translator:" comments mangled by pgindent

  2. Make sure float4in/float8in accept all standard spellings of "infinity".

On Sun, Mar 04, 2012 at 05:16:06PM +0100, Michael Meskes wrote:
> On Fri, Mar 02, 2012 at 11:41:05AM -0500, Noah Misch wrote:
> > I suggest enabling the feature by default but drastically reducing the default
> > readahead chunk size from 256 to, say, 5.  That still reduces the FETCH round
> > trip overhead by 80%, but it's small enough not to attract pathological
> > behavior on a workload where each row is a 10 MiB document.  I would not offer
> > an ecpg-time option to disable the feature per se.  Instead, let the user set
> > the default chunk size at ecpg time.  A setting of 1 effectively disables the
> > feature, though one could later re-enable it with ECPGFETCHSZ.
> 
> Using 1 to effectively disable the feature is fine with me, but I strongly
> object any default enabling this feature. It's farily easy to create cases with
> pathological behaviour and this features is not standard by any means. I figure
> a normal programmer would expect only one row being transfered when fetching
> one.

On further reflection, I agree with you here.  The prospect for queries that
call volatile functions changed my mind; they would exhibit different
functional behavior under readahead.  We mustn't silently give affected
programs different semantics.

Thanks,
nm