On Sun 21 Nov 2010, Jeffrey Sheinberg wrote:
>
> and rsync is run (using make) like this,
>
> env time sh -c "set -e ; set -u ; cd / ; rsync -rlptgoDH -x --del
> --inplace ./ /dev/sdb9"
I'm sure you don't run rsync with a block device as destination...
> After the rsync copy completes, I run a find script to compare the
> filesystem structure of the source and destination filesystems, note
> that only the structure is compared, not the data that it contains.
[...]
>
> As one can see, "/sbin/findfs" has been incorrectly hard linked by rsync to
> "/sbin/e2label" and "/sbin/tune2fs". Since the find script does not show the
> inodes, here is a normal "ls" listing, taken after the system was brought to
> runlevel 2,
>
> # ls -il {,/fs/p/sdb9}/sbin/{e2label,findfs,tune2fs}
> 56919 -rwxr-xr-x 3 root root 30652 Oct 12 2008 /fs/p/sdb9/sbin/e2label
> 56919 -rwxr-xr-x 3 root root 30652 Oct 12 2008 /fs/p/sdb9/sbin/findfs
> 56919 -rwxr-xr-x 3 root root 30652 Oct 12 2008 /fs/p/sdb9/sbin/tune2fs
> 170746 -rwxr-xr-x 2 root root 30652 Oct 12 2008 /sbin/e2label
> 170953 -rwxr-xr-x 1 root root 4036 Sep 29 03:53 /sbin/findfs
> 170746 -rwxr-xr-x 2 root root 30652 Oct 12 2008 /sbin/tune2fs
>
> I have been making these partition backup copies using rsync for some time,
> without any discrepancies occurring. Then I put many new files into the
> source "/" partition (the Linux kernel source tree) , and that is when the
> above discrepancy first occurred.
I'm wondering whether at some stage the three files were in fact
hard-linked together.
The problem with using --inplace is that such hardlinks at the
destination will nog be broken when using --inplace. Usually --inplace
is not helpful anyway, so I recommend against it unless you have very
good reasons for using it.
> Now this is important - then I ran the rsync backups again to sdb6 and sda9,
> which cleaned up the pax symlink discrepancies, but rsync did *not* re-create
> the discrepancy that I have described above.
That is because I doubt that rsync is the cause of these files being
linked together. I suspect very strongly that at some point these files
*were* all hardlinked together.
> Then I ran the rsync backup again to sdb9, and the discrepancy as described
> above again occurred just as described above. So, maybe the copying of the
No, it didn't occur again, it just wasn't corrected -- due to the use of
--inplace.
I always recommend having multiple copies (generations) of backups, so
that you can check the previous version when something strange happens.
When using --link-dest you can hardlink files common to successive
generations, so that space is not wasted. Check out dirvish for
automating that.
Paul
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
