[linux-lvm] LVM2 snapshot weirdness

[Nick Morrison]

Hello list,

I subscribed a few minutes ago, and I've got a question already :-)  I've done
some searching and research on my problem, but haven't had any success yet. So
I thought I'd pose it to this list, in the hope of finding some tips for
solving it.

I have an HP server running RHEL 5.7. It contains four hard disks, in two
hardware RAID-1 groups. The first RAID-1 contains the operating system and
utilities; the second is slated for compressed snapshot dumps of the first.
There's nothing particularly freaky about the configuration of the machine, as
far as I can tell.

I wrote a script which does the following:

# lvcreate -pr -L 2G -s -n LogVol00-snapshot /dev/VolGroup00/LogVol00
# dd bs=8k if=/dev/VolGroup00/LogVol00-snapshot | gzip -3 -c > /backup/LogVol00.img.gz

After this completes, I verify the dump by comparing the output of:

# md5sum /dev/VolGroup00/LogVol00-snapshot

with

# gunzip -c /backup/LogVol00.img.gz | md5sum

This method seems to have been repeatedly successful on four other servers of
similar hardware configuration. I am running the same script on all of them.
The other machines have a larger logical volume (500Gb vs this machine, 260G)
but the PVs are all the same (whole disk except the /boot partition).  However,
on this machine, I got an md5sum mismatch.

Seeking the source of the weirdness, I ran a second dd from the same snapshot:

# dd bs=8k if=/dev/VolGroup00/LogVol00-snapshot | gzip -3 -c > /backup/LogVol00-again.img.gz

.. and md5sum gave me a third, different, result.

It seems as if the data read from the snapshot is changing.

Do I have a conceptual misunderstanding of what I'm doing? Can anyone suggest
things for me to look at, verify, or test? Am I missing something blindingly
obvious here? :-)

Any and all suggestions and comments welcomed!

Cheers and beers,
Nick