Re: Multiple postgresql packages proposed
On Fri, 31 Oct 2003 01:20:16 +1000
Anthony Towns <firstname.lastname@example.org> wrote:
> On Wed, Oct 29, 2003 at 12:19:11AM +0000, Oliver Elphick wrote:
> > This is to get round problems with upgrading major versions, and to
> > allow people to have multiple database clusters, possibly at
> > different software versions.
> ] This is a major problem for Red Hat and Debian, because a dump and
> ] reload is not required by every upgrade and by the time the need for
> a] dump is realised, the old software may be deleted. Debian has
> certain] rather unreliable procedures to save the old software and use
> it to do] a dump, but these procedures often go wrong.
> Having multiple versions installed might be necessary for some weird
> sites, but all the extra complexity it involves is a good thing to
> avoid wherever possible.
No, the real reason for having multiple versions is to allow graceful
upgrade when postgresql syntax changes. Some of these changes have been
fairly substantial and fairly nasty. For example, postgresql formerly
accepted casting 0/1 to boolean. It no longer does. The new date
comparison routines broke comparison to timestamp 'infinity'.
It would be very desirable to have the option of running the current
version and moving items over table by table during an upgrade. This
would allow a site to do pre-validation, and gradual movement rather
than big-bang upgrades.
I have, in the past, put off upgrades precisely because I knew that they
were going to cause me three days to a week of agony, with substantial
user inconvenience. Having multiple versions would let me move over
more gradually, and with more testing before users were affected.
> Anthony Towns <email@example.com> <http://azure.humbug.org.au/~aj/>
> I don't speak for anyone save myself. GPG signed mail preferred.
> Australian DMCA (the Digital Agenda Amendments) Under Review!
> -- http://azure.humbug.org.au/~aj/blog/copyright/digitalagenda