On 7/18/11 9:39 AM, Brian Cuttler wrote:
On Sat, Jul 16, 2011 at 10:43:25PM -0700, Christ Schlacta wrote:
On 7/16/2011 17:04, Chris Hoogendyk wrote:
That's perfect. I can gradually shuffle tapes over to the new system,
and, eventually, all the tapes and all the clients will be there.
During the transition, I can recover stuff from either server
according to which was covering the period of time in interest for the
client of interest.
while migrating, you should probably have a few servers backed up to
both os and ns at a time, removing them from os only after several
successful runs on ns and a few successful test-recoveries.
That is a little tricker, but only from the timing point of view.
An amanda client can only respond to one amanda server at a time,
so while you can backup a client from more than one server, you can
not do so simultaniously, you have to stagger when during the day
the servers run.
An additional difficulty with that is tracking full and incremental backups. On the Solaris systems,
ufsdump keeps its own records of when a filesystem had backups. Two servers trying to back up the
same client will get each other confused, and you could end up not getting proper incrementals from
either.
It is not enough to backup different DLEs on a single client from
a single server, you the client will only communicate with a single
server at any give time.
It is ALWAYS a good idea to test your restores.
A good argument can be made that you are not so much interested in
backing up your system as in being able to restore it.
You could, if you wanted, call that the first rule of backups -- Test your
recovery.
I like Jon LaBadie's idea. I think the simplest approach is to get ns up and running backing up
itself and drop it off the disklist on os. Then, wean os of one client at a time, moving the
disklist entries from os to ns. And, of course, test recovery of every transitioned client.
The slightly complex part, unless I label a bunch of new tapes for the purpose, is transitioning the
tapes. I have a tape cycle that spans about 6 weeks. I could take the oldest tape on os, use
amrmtape to remove it from os, and add it to ns. Then I'm using two tapes per night, as os would use
the next oldest. The tape cycle for os would shrink as the tape cycle for ns grows. Actually,
that'll work alright, because I would probably get all the clients transitioned before the tape
cycle on os dropped to 4 weeks. Then the tape cycle on ns would just continue to grow. It would be
slightly funky for a while, because I would have to manually transition the tapes on a daily basis
until it was done. Otherwise, I would not have the backups on the os during the period of time when
they might be needed. But, I can smooth the whole thing out by labeling a week or two of new tapes
to get the process started on ns.
[os=old amanda server, ns=new amanda server, as per Jon's post]
--
---------------
Chris Hoogendyk
-
O__ ---- Systems Administrator
c/ /'_ --- Biology& Geology Departments
(*) \(*) -- 140 Morrill Science Center
~~~~~~~~~~ - University of Massachusetts, Amherst
<hoogen...@bio.umass.edu>
---------------
Erdös 4
