In article <199801131829.MAA18731(a)vortex.more.net>,
David Drum <david(a)more.net> wrote:
>I was wondering if anyone uses amanda to backup their NetApp(s)?
Yup, we do. At least, we used to, until yesterday :) The load on the
netapp during phase I of the dump is prohibitively high, and consequently
makes our webserver very slow... we're currently rethinking things.
There are a few problems to backing up a toaster via amanda, and here
are the solutions, roughly:
- the toaster doesn't run amandad :) And you can't easily make it do so.
Our solution is to have amanda backup /dev/netapp1/path on another machine.
To keep the network happy, that "other machine" is also the one that
the tape is connected to, and the one that amanda (the master program,
or the client, whatever you want to call it) runs on.
"dump" on this machine is actually a script that intercepts requests to
dump devices of the form /dev/netapp*. If it isn't, it calls the regular
dump. If it is, it calls a separate program, which translates some options,
and then starts dump on the toaster, via rsh. (Note that the devices
that amanda thinks it's backing up need to exist. Just create them and
make them symlinks or hardlinks to /dev/null or /dev/zero. Doesn't really
matter).
Next problem:
- amanda does not grok toaster "dump" output.
Perl to the rescue again (the dump wrapper is perl, too :).
Since we had to start a separate script anyway, we're piping STDERR
through the script, and have it do on-the-fly translations so the output
of "dump" as produced by the toaster is massaged into something that
amanda understands.
Additionally, we trap "dump W" and read the /etc/dumpdates file on the
toaster too, and generate extra output for it.
A few considerations you have to make:
- amanda starts by getting estimates for all filesystems it backs up,
that is, it makes at least 2 dumps for each filesystem, and terminates
the dump once the "estimated" line is in. If the "dump" command on
the toaster creates its own snapshot each time you do this, you
waste a lot of CPU. We settled on generating a snapshot once each
day, and running all backups from that. You have to take extra
special care though not to refresh the snapshot while a backup is
still running.
- amanda doesn't know anything about "oh don't try to backup this so
often". At least I can't configure it to do so (without extreme
hacks). It knows about priorities, but it will always at least try to
make a backup, and if it fits on tape, it will. This generally means
that amanda does make a backup of all data on the toaster, at least
incrementally. Not strictly necessary with the RAID and all that.
If there is interest, I can post the sources we use to the list, or make
them available on the web. Note that the sources are very alpha, and
most definately not directly usable on other sites, some tweaking will
be necessary.
Best regards,
--
#! ##### Jan-Pieter Cornet ##### <johnpc(a)xs4all.net> ##### perl
++$_;$!=$_+++$_;($:,$,,$/,$*)=$!=~/.(.)...(.)(.).(.)/;$!=$_+$_;
($(a),$\,$~)=$!=~/(.)(.).(.)/; $_="$,$/$:"; $@++; $~="$~$_";($_)=
\$$=~/\((.)/;$|=++$_;$_++;$|++;$~="$~ $@$:";`$~$/$\$*$, $|>&$_`