[Techtalk] upgrading a file server

[Magni Onsoien]

On 2007-05-14 14:25:31 -0700, Maria McKinley said:
> Hi there,
> 
> I have a drive that is failing. It also has a pretty old linux (debian) 
> distribution, 2.4.27. I have installed linux on a new drive on a test 
> machine, and somehow ended up with 2.4.25. Not sure if the difference 
> means much. I am interested in upgrading the system entirely, though. My 
> current plan is to copy the /etc and /usr/local directories from my old 
> machine to the new machine, and then test my new drive. (Is this a bad 
> idea? Better way to do this?) Once I am sure all services are working, I 
> am thinking the smartest thing to do is copy everything over to another 
> new drive (best way to do this? How do I handle different partitions, is 
> there a hard drive copy tool that paritions the new drive in exactly the 
> same way, or do I need to do this by hand first?), and then upgrade this 
> drive. However, it will be hard to test this, except by shutting down 
> the new working server and trying to run the upgrade as the server, 
> since I can't have two servers running with exactly the same 
> configurations on the same network. But at least, if I can't get the 
> upgrade working in a reasonable amount of time, I can go back to the new 
> working copy.
> 
> Feedback about any of this plan is much appreciated. How do other people 
> handle upgrades on critical servers?

How critical is critical? Can you handle it being unavailable for
seconds, minutes, hours? Many users?

At work we always set up a new server from scratch, with the newest
version of the operating system etc. The server gets a new name - we
have only recycled one name during the 10 years I have been connected to
this place. We use DNS aliases for services - the users always access
their file shares as \\home.staff, not as \\sidney or whatever the
server is called. they don't have to care or think about what we
actually call the physical (or virtual) servers.

All services are then set up, and we can do some testing. Most
testing can be done with substituted names - file services works whether
you use them from fishyserver.example.com or home.staff.example.com, and to
test a webserver or a mailserver you can make testdomains.

When everything is tested, DNS pointers can be moved gradually, and if
something didn't work we can roll back.  Just remember to adjust the TTL
for the DNS data if you need quick rollback.

For a home server a lot of this is overkill, but the concepts can and
should be used. Perhaps you don't need a DNS alias for the file server,
on the other hand it doesn't hurt either (you can use /etc/hosts if you
don't have a name server).

I don't know whether your servers are at home or in a work setting, but
if you have the necessary hardware for a completely new server you
should just set it up, add the services you need (like web, mail, file)
and then move data from the old server. Depending on how customized your
original setup is this may or may not work at first attempt, but since
you will still have a working server with all the data you don't risk
much. If you use DNS aliases you can also move stuff gradually and roll
back only some services if something fails.

Good luck!


Magni :)
-- 
sash is very good for you.