On 7/4/07, Andrew Sullivan <ajs at crankycanuck.ca> wrote:
> On Mon, Jul 02, 2007 at 11:15:38PM -0400, Jan Wieck wrote:
>
> > against a simple N1->N2 setup bombarded with a -c5 pgbench. That isn't
> > quite the testing you want to have done before committing such a
> > substantial change in the inner core log selection logic of STABLE code,
> > is it?
>
> What, we're not gonna pants-seat fly?  Sigh.  No guts, no glory ;-)
>
> Seriously, I agree with Jan here: let's be _really_ conservative with
> this one.  Indeed, given that it's a small patch, I'd be inclined to
> issue a .11 with a contrib/pgq-apprach.patch file and suggest people
> try it before back patching for real.  The HEAD is a good place for
> architectural changes, but the supposedly STABLE releases aren't.
> I'm not a fan of the Linux-style, "rewrite the PCI subsystem in
> x.x.8" STABLE-style releases.  And I think this project has been
> often enough bitten by such exuberance that we should be cautious.

I think the patch is fine correctness-wise.  Main problem
you can have with new approach is that Postgres gets confused
and turns the whole query into seqscan.

It should not happen in 8.3 but could be a problem with
7.4 or 8.0.

OTOH, my experience was with int8 txid, maybe they are
more intelligent when handling int4.

-- 
marko