Networker

Re: [Networker] Two clients sharing an index?

2004-01-28 02:28:57
Subject: Re: [Networker] Two clients sharing an index?
From: George Scott <George.Scott AT ITS.MONASH.EDU DOT AU>
To: NETWORKER AT LISTMAIL.TEMPLE DOT EDU
Date: Wed, 28 Jan 2004 18:28:16 +1100
Curtis,

> Suppose I had two or three (NetWare cluster) clients that all shared the
> same clustered volume, but the volume was only resident on one client at a
> time -- and it moved around all the time.  By default, if I'm backing up
> each node in the cluster (which I'm forced to do because NetWorker doesn't
> really support NetWare clusters), I'd force a full (or large incremental)
> every time the volume moved.
> What would happen if I:
>
> 1. Create two or three client definitions (one per clustered node), copying
> the same client ID to each one.
> 2. I put just this volume as the saveset entry for each client.
> 3. Symbolicly linked all three client indexes together:
>    cd /nsr/index
>    rm -r node2 node3
>    ln -s node2 node1
>    ln -s node3 node1
> 4. Now whenever I back up this volume on node1, node2, or node3, it shares
> the same index.
> 5. This way, when the clustered volume moves from node1 to node2, it backs
> up the right stuff.
>
> Am I nuts?  Has anyone done this?  Will it flat out not work?  Might it
> work?  Can I have two clients share an index like that?

Yes, you're nuts...

My guess is that you'll be tripped up in the media database somewhere.

I'd have a go at creating just one client (for the service address) and
add all the physical nodes as aliases.

> What about the non-clustered stuff on these nodes, like the system volume?
> Well, I do this:
> 1. Create a virtual IP address on each netware server.
> 2. Create a virtual hostname for each one (e.g. node1system) that points to
> the virtual IP address.

When the client starts the save it contacts the server and tells it
it's hostname.  In this case it will be the name of the physical node
rather than your carefully crafted virtual hostname.  I guess that if
you could somehow persuade it to run save as "save -c node1system" it
would be tricked (maybe by fudging up your own mini savegrp that you
invoke youself and get it to nsrexec the way you want it to - you could
probably just use rsh to do what you want...).

> 3. Create a client definition for that nodename that backs up everything BUT
> the clustered volumes.
>
> Yes, this will cost me a client index per node.  I'll live with that.

I assume you meant license.

If all of the physical nodes are practically identical you might get
away with only backing up one of them.

If the shared disk is big enough you could maintain a copy of each SYS:
etc on it.  Alternatively you could just keep copies of the unique bits
on it and let NetWorker backup SYS: from the currently active node.

You're trading backup complexity for recovery complexity.

George.
--
George Scott           George.Scott AT its.monash DOT edu
Systems Programmer, IT Services, Monash University

--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=