[CentOS] enterprise backup solution (probably amanda?)

Paul Bijnens paul.bijnens at xplanation.com
Wed Jul 30 21:46:34 UTC 2008

Joseph L. Casale wrote:
>> I'm an Amanda user since 9+ years now.
>> Done several restores, even bare metal ones.
>> Amanda never failed on me.
> Paul,
> I bought the Enterprise version of Amanda and was blown away to find it cant do a verify of a backup once completed.
> You can verify the contents of a single tape manually, but I thought that was a bit of an oversight!
> Do you guys just trust what you hope was written to tape to be ok, or how do you verify backups?

There is the command "amverify", which you can run just after a backup
or even a few weeks later, when you believe a tape has gone bad.

But, indeed, that command verifies only to some depth.
For gnutar dumptypes, it pipes the backup image into gnutar to verify,
but for "dump" dumptypes, this cannot be generalized, because the server 
OS does not necessarily has a compatible "restore" program (your client 
could be a Solaris with ufsdump, while your server could be a Linux 
without ufsrestore/ufsdump).  In that case the amverify command limits 
itself to verify if the bytes can be read from tape without errors, and 
uncompressed without errors if the image had software compression enabled.

Adding checksum verification is currently work in progress.

Of course, tapes can fall on the ground, and get damaged, after being 
verified, so, if you really really want more certainty to be able to 
restore, there is the RAIT support in Amanda, where you can mirror (2 
tapes), or write a N tapes with one parity (N-1 tapes + 1 parity) in 
parallel. Now, you can loose a tape, and still be able to restore.

I even ran a RAIT-mirror with one virtual-tape-on-disk (for easy and 
fast restores) and one real physical tape (stored offsite for security)
for some time.  Worked very well.

When there are errors DURING a tape write Amanda will notice, unless 
your hardware does not detect those. Tapedrives actually have a 
read-after-write head, that verify the bits that were written 
microseconds before -- if you don't trust those, buy another tapedrive 
from another manufacturer.

One of the frequent errors "competitive" backup programs make is, that 
they close() after each image (or even a partial chunk), and then reopen 
the tapedrive again to write the next image.  The current version of 
Amanda does not even allow appends to tape, just to avoid this, because 
in the small time between the close and open, there is a chance that 
e.g. the scsi subsystem gets reset, and the tape rewound.  Amanda would 
have noticed any error here.  Otherwise you need indeed a verify pass
to notice that kind of errors.

As long as I use Amanda, however, I did indeed not have got any tape
that got zero errors while writing, but was unreadable afterwards.
Maybe I was just lucky.

The feature I miss most in Amanda is the rsync-like stuff.  That's
why I use BackupPC for our MS Windows Clients, mostly laptops (using 
rsyncd), and rdiff-backup (rsync with history) for the remote backups 
over a slow link.


More information about the CentOS mailing list