[Dirvish] singling out crucial data

[Keith Lofstrom]

On Tue, Jun 23, 2009 at 09:43:31PM -0700, Loren M. Lang wrote:
> 
> My primary suggestion was putting a tree in different vaults which have
> a different backup policy/schedule.  I just it easier to deal per
> partition since some partitions use ACLs or extended attributes which
> need to be enabled in dirvish.

I agree with Loren, although the multiple vault approach does 
necessitate thinking about (and probably implementing) a set of
bare-metal restore scripts.  Dirvish doesn't have very good tools
for that; most backup software doesn't.  Separated vaults facilitate 
restoring different segments from different backups;  for example,
if /usr has been corrupted by a failed upgrade, one can restore
from an earlier /usr and the latest /home.  This can also be done
with one big vault, of course, but it is a little easier if there
are a few smaller vaults.

Dirvish lacks a restore tool.  Restore is typically the most
time-critical and stressful part of the backup/restore cycle,
it is important to build ad hoc scripts for restore and to
test them.  With drives so cheap these days, it is a good idea
to have a spare drive for each drive type in the system, and 
practice doing restores onto the spares.  That way, when a
crisis occurs and a new drive has to be built from bare metal,
the restore can be done with minimum time and effort (leaving
time for all the other simultaneous crises implied by restores).

Keith

-- 
Keith Lofstrom          keithl at keithl.com         Voice (503)-520-1993
KLIC --- Keith Lofstrom Integrated Circuits --- "Your Ideas in Silicon"
Design Contracting in Bipolar and CMOS - Analog, Digital, and Scan ICs