[Gllug] best way to update a single production server?

[Khusro Jaleel]

I apologise for coming back to this thread so late, I obviously made a
mistake by stating "MUST be up 24/7" as it seems to have upset quite a
lot of people. I have not been told that this website must be up 24/7,
but it would not be nice if it did go down. I was just trying to find
a way to do it with minimal to no disruption (if possible).
Sorry if I didn't provide much details, but this is a web server on
Debian Etch, and the disks are mirrored.

At the moment I'm just thinking of doing step 2, i.e. setting up the
website on a temp server, changing DNS A records, and when no more
queries can be seen in the Apache logs, doing the updates, then
changing things back. This is just a solution for this moment in time,
though.

It's just that some bad procedures were followed before I got started
here and now I *do* have a chance to put things right, however I admit
I'm a little bit inexperienced when dealing with updating servers and
maintenance windows so in my situation, what would you guys suggest is
the best way forward towards implementing a proper solution so that
servers can be upgraded smoothly? When you guys setup servers for
clients in data centers, do you always get them in pairs and use load
balancers? VMs?

So some questions I have are:

- Do I need to always buy servers in pairs and put load balancers in
front of them?
- create Xen-based servers which allow VMs to be migrated to another
server? (Servers in pairs again)
- just recreate the production server on a test machine or VM, test
then upgrade production machine and hope for the best?
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug