[OpenAFS] Further TransArc -> OpenAFS musings/planning

Russ Allbery rra@stanford.edu
Wed, 16 Aug 2006 12:21:28 -0700


Joe Di Lellio <joed@ucsc.edu> writes:

> 1) Bring all the old servers down.  Copy over DB & others files to new
> systems.  Swap in the new systems IP/hostname wise & bring them back up.

> This approach is what John Harris et al suggested, and looks to be
> pretty quick (he estimated 20 minutes - I'll assume twice that since I'm
> hyper- careful/paranoid/take a lot of time on things I haven't done
> before).  It does mean my pack of three DB servers are completely
> offline for this time, but it's our normal service window during the
> mentioned quiet time.  I'll have the old TransArc servers, re-ip'ed but
> kind of as living backups, should I need to get something.  Nothing AFS,
> even as a client, would be running on these old DB servers.

I don't know of any reason why this wouldn't work, but I have to admit
that I'm really partial to transferring the database over protocol rather
than making the new server read the old disk file format.  I know that if
you bring up a new server and let Ubik handle the replication, you don't
have to worry about endianness, word size, etc., and I'm not sure about
just copying the disk file.

In your situation, you probably don't care and I expect this will work
fine, but that's why I generally don't recommend this approach.  Just
always transferring the database over protocol is safer in a wider variety
of circumstances.

-- 
Russ Allbery (rra@stanford.edu)             <http://www.eyrie.org/~eagle/>