Re: Debian etch, backport postgresql 8.3 experiences?

On Wed, 23 Apr 2008 12:04:08 +0200
Peter Eisentraut <peter_e(at)gmx(dot)net> wrote:

> Am Mittwoch, 23. April 2008 schrieb Ivan Sergio Borgonovo:
> > I'd like to know if anyone has experience in using postgresql 8.3
> > for amd64.
>
> There are probably thousands of people with that experience.

I'd like to be in the N*1000+1 happy ones ;)

> > How did you set up your apt config/source.list to just install the
> > minimum required to install 8.3 and php drivers?

> I'd say follow the instructions on backports.org.

Other than adding backport to the source.list and using -t I didn't
see anything else.

> > Considering I'm not concerned of a "short" downtime and the DB is
> > pretty small what were your steps to safely upgrade on a
> > production box? (that mean I can just backup and restore and
> > avoid more complicated configurations involving slony).

> That depends on what your production box currently runs.

plain etch. pg 8.1 client, server and contrib + php5-pgsql

My "dream" upgrade would be:
1 backup
2 stop old server
3 install the new one
4 start the new server
5 restore
without removing the old server so I could revert the process... just
in case...
I see that pg_* stuff in /usr/bin looks for $ENV{'PGCLUSTER'} and
stuff in /etc/init.d is versioned.

I'd say I've to add:
4.1 change PGCLUSTER
4.2 start the new server

and in case of troubles change PGCLUSTER back to the old value and
restart the old server.

> > Is there a way to prepare for a safe "roll back" to the previous
> > version? Does the backported package let you keep installed the
> > previous version?

> That also depends on which package/version your production box
> currently runs.

etch stock stuff for 8.1.
DB:
- main table around 600K rows, 15 fields, very few varchar larger
that 128.
- total size of backup 50M compressed (-Z9) (a restore takes seconds,
not minutes)

I'm going to test the procedure on a test box, but I'd like to avoid
to discover the right method by trial and error... and well it doesn't
mean that the method that succeeded on the test box is the right
one ;)

thanks

--
Ivan Sergio Borgonovo
http://www.webthatworks.it