Richard Rauch wrote:

> Thanks for your perspective, especially on the "build.sh
> distribution" time.
> 
> I recently got around to setting up an NFS server with home
> directories on it, so only /etc, maybe some /root, and a couple of
> tweaked files in /usr/src have anything of real importance in them.
> 
> But if disaster strikes, I wasn't sure how best to recover.  (E.g.,
> if one day trying to build -current kills the system.)  I know that
> it is a more likely danger (as compared with a release), and still
> recall the recent debacle about the filesystem changes causing (or
> failing to fix?) corruption.  If that kind of thing is rare enough, I
> can always just wipe the disk and go back to a snapshot + CVS +
> build.sh to get back to where I was.  But if the need to restore from
> backup while running -current is frequent enough, maybe some way of
> building my own snapshots (and stashing them on the NFS server) is a
> good idea to allow quicker recovery.  How hard is that?

If the break happens to be your NIC driver, it could be pretty bad. :)

The snapshot issue is that you'd have multiple snapshots. Just do a base 
load, have your configs in a tar file, and consider all the /usr/src 
stuff based in anoncvs.

> What did you mean by a "dual hard disk machine" being broke under
> 1.6? Do you really mean that two hard disks do not work (at all?)?  I
> have an old 20GB drive in my main system, but the 20GB drive is
> normally not mounted.  (It's a Maxtor that has problems with seek
> performances.  I intended to do something like a nighly, or weekly,
> mount of the disk and copy backup archives to the Maxtor.  It should
> be fast enough for that, though it is miserably slow for certain
> operations.  That's another story though (see netbsd-users or -help,
> I think, going a ways back in the archives).  The replacement is a
> Western Digital 40GB.  I haven't had them both active at the same
> time for very long, but am *pretty* sure that I've had them both up,
> under 1.6.

Sparcs can boot from either disk. So I have both laid out exactly the
same and a script that uses (ufs)dump to copy everything off the other
disk. It even sets the bootblock and edits fstab. Before I do major work 
I run the script. If the -current breaks anything I can't fix, I just 
boot from the other drive and run the script back the other way to wipe 
the mess and start over.

However, dump was broken a while back, and I started playing with tar.
tar is probably better, but my skill with it is less.

My idea for you may be to copy the partition as a tar file. If you then
boot from cd and rebuild the machine, you can then just put back the
files you need and update from CVS for new sources.

I think, but can't give the exact syntax, that you can build a tarfile 
include list that would save just the files you need. You could then 
untar it in / and be set.

If I were on a single disk machine I'd make sure I had a 1.6.1 cd and  a 
disk layout page handy. That way if it dies I could just rebuild with a 
known good setup, then ftp in my tar files.

> Thanks again.

You don't know how thankful *I* am for the people on this list who've
helped me.

> P.S.: Please include me in the To: line of replies.  Otherwise I only
>  see the message when and if I click on the link on
> mail-index.netbsd.org.

Sorry. No problem.

ciao!

leam