[Dirvish] More than on tree per backup?

[foner-dirvish at media.mit.edu]

    > Date: Fri, 9 Feb 2007 15:10:08 -0800
    > From: Kelsey Cummings <kgc at corp.sonic.net>

    > On Fri, Feb 09, 2007 at 10:11:22AM -0600, Richard Geoffrion wrote:
    > > But then...maybe I'm totally lost...which I think I am.  Good luck though.

    > Who knows?  Perhaps I'm lost.

    > The reason that I want to do it that way (or by exludes if it's the only
    > way) is that I'm talking about scaling dirvish to backup about 150 servers
    > each with at least two local filesystems and some with quite a bit more a
    > good number of which have as many as 48 or more NFS mounts (making managing
    > the exludes painful, and what happens when a new NFS mount show up and
    > someone forgets to update the excludes and dirvish decides to walk down
    > into my netapps and get lost?)

Can't you just use "xdev: 1" in your dirvish config file?  Presumably,
that should prevent rsync from recursing into filesystems other than
the one(s) you tell it to start with, which should prevent it from
walking into any NFS mounts.

(Normally, that's what I'd do on -my- machines, but I want to include
/boot, which isn't the same filesystem as / the way I configure them,
and I have a small & constant set of NFS mounts, so I set "xdev: 0"
and explicitly exclude the mountpoints (and other things like /proc
and /sys and various other special cases like big cache directories).)

If -that- doesn't work for some reason, you could presumably write
some special tool that walks your entire set of machines just before a
backup and tests each close-to-toplevel dir (assuming that NFS mounts
won't show up arbitrarily deep in the filesystem) and stats them to
figure out if it's a mount or not.  That's got timing races and is
generally ugly---but it could autogenerate the exclude list for you.