In view of the forthcoming 8.1.1 release on December 6th, it would
seem a good idea to see about getting the 1.1.3^H5 release of Slony-I
ready on or pretty near that date.  There have been some rumblings on
IRC to this effect, and there is certainly some keen desire to have a
release that supports PostgreSQL 8.1.

After the pain of the 1.1.1/1.1.2 releases, we now have something more
nearly resembling a release checklist:

http://slony-wiki.dbitech.ca/index.php/Release_Checklist#Release_Checklist

In memory of the 1.0.3, oops, 1.0.4, oops, 1.0.5 release, the
tentative plan is to follow 1.1.2 by skipping straight to 1.1.5.  (I
personally would have favored adding sub releases to 1.1.1, so that
we'd have 1.1.1, then 1.1.1.1, and now, 1.1.1.1.1.1, but that's
evidently not popular with others ;-).)

A couple of things that should get pointed at as they have caused some
trouble in past:

1.  RPM packaging.  Devrim Gunduz generally needs a bit of time to get
this prepared; if that was ready by the 5th, that would be pretty
ideal.

2.  Clarity of other "tarball stuff."  We'll see about having a
separate tarball containing pre-built HTML pages for the admin guide;
that should be of assistance for people who have trouble building
them, as is traditionally a problem on Red Hat's Linux
distributions...

I have added in the "new generation" tests from CVS HEAD; I'd expect
them to be compatible even with Slony-I version 1.0; they don't
particularly torture the FAILOVER capabilities, but they are probably,
even at this fairly early point, a bit more functional than the
"ducttape" tests put together, and they should be easier to run
through.

There are three pretty notably significant changes in this release that
mandates running some tests on a reasonably diverse set of systems, on
various releases of PostgreSQL:

1.  A change has been made to "hide" indexes while tables load (e.g. -
when running SUBSCRIBE SET), followed by returning the indexes and
reindexing.  This can be expected to have a pretty substantial effect
on performance.

2.  The change to the "node interlock" where we no longer use
pg_listener to help manage keeping just one slon managing each node.
(This was mandatory to support PG 8.1, as the implementation of
pg_listener / NOTIFY / LISTEN changed a bit.)

3.  SUBSCRIBE SET attempts to use TRUNCATE rather than DELETE FROM to
more quickly empty out tables.

It is probably not necessary to go through any "combinatorial
explosion" of multiplying the number of platforms by the number of
possible PostgreSQL versions, but we definitely need multiple
PostgreSQL versions covered.

Comments welcome, particularly if you have suggestions as to needful
revisions to the Release Checklist.
-- 
(reverse (concatenate 'string "ofni.sailifa.ac" "@" "enworbbc"))
<http://dba2.int.libertyrms.com/>
Christopher Browne
(416) 673-4124 (land)