Dave Dykstra wrote:

> On Fri, Apr 06, 2001 at 03:20:10PM -0700, Jeff Mandel wrote:
> > I'm trying to create a nearline archive.
> >
> > I don't have another volume big enough to hold a full backup of the
> > master volume, so I can't compare the change set in the usual way.
> > I would just like to get whatever was modified in the last day - like
> > the results of a find -mtime 1 (or 0) would give. Something like:
> > rsync -mtime=1 /vol1/ /vol2
> >
> > This would copy the files modified one day ago, but most files would not
> > be copied even though they don't exist on the destination, as the source
> > files would be older than a day.
> >
> > I don't think there's a option to do this with rsync. Did I miss it?
> >
> > All suggestions welcome.
> >
> > Jeff
>
> Would some of the files exist on /vol2?  If not, why bother with using
> rsync?  Why not just use find and cpio?
>
> Alternatively, you could create an include file with find and pass the
> include file to rsync along with --exclude '*' to exclude everything else.
> Currently you also need to explicitly include all parent directories of
> files you want copied.
>
> - Dave Dykstra

Thanks for your suggestions. Do you think a list built with find and passed
to rsync would outperform an incremental ufsdump to file and restore in a
large file system?

I hope my approach here is not too hokey. I'm trying to put together a set of
rotating, and accumulating "snapshots" - sort of like a NetApp filer that
way. My goal is to allow someone to get back a recent version of a file, or
one that was recently deleted without resorting to tape. This is spreadsheet,
graphics and run of the mill business stuff that is not in some kind of cvs
system.

So far I've come up with the following:
1) ufsdump the previous days changes to a file on the snapshot mirror using
an incremental on the dump level
2) restore the files to a daily snapshots directory - that will give me a
near line snapshot of only the last days changes (when using a file this is
quite fast).
3) rsync the daily directory to a "current" directory which would accumulate
the latest revs of each file. If a specific version was needed, you could
browse through the last two week's worth of dailies (or whatever I have space
to keep) to find the version you wanted.

Does anyone note any particular folly or pitfall in this approach?
Since rsync doesn't have anything to compare against to see what's changed in
the last day if the set doesn't exist in it's entirety on the other mirror, I
need to use ufsdump to "initialize" it.

As Dave suggest, maybe a list generated by find would accomplish this faster
than the ufsdump/restore -> rsync.

More comments?

Jeff


Reply via email to