On Sat, Oct 06, 2001 at 12:54:02PM -0400, Mike Sowka wrote:
> Russ Ingram wrote:
>
> >Seth Mos wrote:
> >
> >>
> >>It is a shame that tools Like Symantec Ghost are really slow in taking up
> >>on different filesystems. ReiserFS or XFS are unsupported. ext2 is your
> >>only choice for most commercial tools.
> >>
> >
> >On the contrary, my friends. :-P There was a post to this list
> >probably a month or so ago about a linux Norton Ghost like util
> >that supports just about every major fs available in Linux. I
> >didn't actually get to try it out cuz I had just finished cloning
> >the drives I needed with xfsdump/xfsrestore when the message came
> >across but I remember it because I had needed exactly that not 2
> >days before the message hit the list. The link was
> >http://www.partimage.org. You can also always just do what I did
> >and pipe the output of xfsdump to xfsrestore(or do the same with
> >tar for that matter), too.
> >
> >
> >Russ
> >
> Hello Russ,
> I was hoping to tap into your vast knowledge on XFS... :) I've looked at
> www.partimage.com and it seems they don't support the bootdisk/rootdisk
> method (no CD drives on my cluster nodes) on the latest version... I
> like the principle of KISS (keep it simple stupid), and using partition
> image when I've got xfsdump seems a bit redundant :).
> I have a question or two about xfsdump, any sugestion are MUCH appreciated:
> #1) How is it that xfsdump is able to dump / while it's mounted? And how
> "industrial-strength" is that?
I'm not exactly sure what you're asking here. I don't know the technicalities
of how it does it but if you're just asking if it can, yes, it can. If you
want to know how, someone else will have to step in and answer that.
> #2) Can you elaborate a bit on your method of piping xfsdump to
> xfsrestore :)?
The most basic form of xfsdump and its corresponding xfsrestore can be run
back to back through a pipe to make a copy of a filesystem. Like this:
xfsdump / - | xfsrestore - /mnt/
The above command says run xfsdump on the root filesystem, send the output
to stdout and pipe it to xfsrestore. The dash on the xfsrestore command
says to read stdin as the dump input and /mnt/ is where it is to write to.
I noticed that when I ran it on my / filesytem it didn't do something right
on the /dev/ dir so you might want to use tar instead and that can be done
similarly like so:
cd /; tar lcf - .| (cd /mnt/; tar xvpf -)
> #3) Here is my "plan" for the cluster:
> - after installing the "golden-node" I plan on dumping all of its
> partitions /boot and / and storing the dumps on our head node
> - in order to clone: using etherboot (if I can get it running) I run a
> system off of NFS on my head node and xfs restore the partitions onto
> the rest of the nodes (linux xfs doesn't support simultaneous
> xfsrestores yet does it?)
> NOTE: my nodes are head-less no == no video adapter
> HOW VIABLE IS THIS?
> Thank You,
> Mike
Ok, I'm not entirely sure I'm following you here but here's what I can tell
you in regards to what I understand. If you are saying you will boot new
nodes with etherboot with nothing on the harddrive but bare xfs partitions
and run xfsrestore with a xfsdump file as the input device that sounds like
it should work just fine. I don't know about running simultaneous xfsrestores.
I haven't ever tried it but I can't imagine why it wouldn't work if you are
restoring from a file rather than an actual device. Never hurts to try.
Just one suggestion -- it might be just as easy to just do the filesystem
replication by physically installing the new drive for additional nodes into
the "golden node" and doing the piped method as shown above to copy the
system. It will save you space on your server node.
Russ
--
Russel H. Ingram
Gargoyle Computer Consulting
(307)742-1361 or (307)760-1317
www.gargoylecc.com
|