On Tue, Dec 22, 2009 at 04:30:20PM +0000, Peter Geoghegan wrote:
> Yeah, me too - sort of begs the question of why all releases from the
> 2.0.* branch up until now haven't been release candidates or betas.

Surely it's more likely to raise (not beg) the question of whether the
wider community, beyond the tiny developer community, is pulling its
weight in performing testing.  This is a free software project, after
all, and that means that the quality control you get is approximately
as good as that which you put in.  It would appear that significant
issues are showing up in the software, and that suggests to me that
maybe people aren't testing very thoroughly, or haven't worked out a
good exercise of all the features, especially under load, in order to
demonstrate that it really works as intended.

Slony is unfortunately very complicated software.  This may be a
design flaw, but it's the way it is now.  It would be better if we
could demonstrate that the code has obviously no bugs, but I suspect
the best we can get to is no obvious bugs.  The problem is that our
current tests for "obvious" are way too shallow.  If people would like
that to be better, I heartily suggest that they get to work on
improving the test cases rather than complaining that the developers
aren't labelling the software correctly.  AFAICT, each release goes
through long periods of being RCd, and the 2.0.0 release was coming
RSN for a long time.  But if nobody exercises the code in
realish-but-safe circumstances, we'll never find the big nasty bugs
until someone trips over them in production.

A


-- 
Andrew Sullivan
ajs at crankycanuck.ca