[OpenAFS] Re: 1.4 (fs) -> 1.5 (dafs) migration

Andrew Deason adeason@sinenomine.net
Fri, 20 Aug 2010 11:13:20 -0500


On Fri, 20 Aug 2010 17:18:38 +0200
Gémes Géza <geza@kzsdabas.hu> wrote:

> Initially I've planed to use the same ip address for the new server as
> the old had. But maybe vos move is a better approach. I have an almost
> empty partition, which I will empty up and reattach to the new server.
> My only remaining concern is about the db migration, as I intend to
> shut down the old servers.

So, are these fileservers also dbservers? (that is, they run ptserver,
vlserver, etc) If so, yes, in general keeping the same IP for your
dbservers will make things easier.

If you have two dbservers, as I think you mentioned, you don't
technically need to really do much to migrate the db data. If you turn
off one old dbserver, and turn on a replacement, the database will be
synchronized from the remaining old dbserver. The synchronization can
take a few minutes; you can tell when it is done by running
"udebug <sync_site> 7003". It will report "Recovery state 1f" when the
database is synchronized and everything is good. After that is done, you
can migrate the other dbserver.

However, to make the database on the new machine available to clients
faster, you can copy the files in /var/lib/openafs/db (or /usr/afs/db)
to the new machine, after you've shut down the OpenAFS daemons on the
old machine. If you copy the database files, you can also migrate both
dbservers at the same time, though I think I prefer doing them one at a
time so I can verify everything's okay at each step before proceeding.

There are certain ways to minimize the downtime of migrating dbservers,
but currently there isn't really a way to eliminate the downtime
entirely (I think it's possible with some code in gerrit, but it's not
in a release yet). So, what I wrote above involves some downtime, while
the dbservers reestablish quorum.

-- 
Andrew Deason
adeason@sinenomine.net