On Saturday 29 March 2008, Joel Johnson wrote:
> I have previously [1] reported an issue I encountered when using previous
> installation images. I checked the Lenny beta1 installation images
> (netinst) and the same problem still occurs. Namely, newly created md
> devices are find, but scanning and activating existing md devices makes
> partman think that the md device is partitionable.

This is indeed a very annoying issue and one I've seen a few times myself
(e.g. in #470374).

> This is a regression from the Etch installer.

I doubt that as AFAIK the relevant code has not changed.

> I see several related bug reports against partman-md, and am wondering if
> plans are in the works for resolution of this. I'll take a look at it when
> I get some time, but want to see if progress is being made elsewhere.

The problem so far has been that the issue is not 100% reproducible, but
your analysis seems to offer a good starting point.

Nobody is actively working on this, although I may give it a shot again with
your info. Feel free to dive into it if you are motivated to do so.

The main question here is what is different/wrong in the way partman calls
libparted in this situation and how can we avoid it. One possible cause
could be that even though the RAID device(s) are activated, we don't create
the proper partman state files for it/them under /var/lib/partman/devices/
and that the call to parted ends up wrong because of that.
Comparing the state files for a newly created RAID device and a reactivated
one could give a clue.

Cheers,
FJP

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQBH8LtVgm/Kwh6ICoQRAsQtAKDF+xX99GuknJaxs1ZJquHTMNLWlQCdF68L
WTV43FLc9BT1Sf6d54wCPvM=
=7lFy
-----END PGP SIGNATURE-----

On Saturday 29 March 2008, Joel Johnson wrote:
> I have previously [1] reported an issue I encountered when using previous
> installation images. I checked the Lenny beta1 installation images
> (netinst) and the same problem still occurs. Namely, newly created md
> devices are find, but scanning and activating existing md devices makes
> partman think that the md device is partitionable.

I think I've found the cause. The problem was with a combination of
assembled pre-existing RAID devices and LVM.

The cause of the issue was almost certainly that when existing RAID devices
were activated, they were created as /dev/mdX instead of /dev/md/X. The
sync flag exception handling in partman-lvm only allowed for the second and
not for the first, and thus created a "flags change" on the RAID device
which parted would try to write, resulting in the error message.

Both issues have been addressed:
- mdcfg 1.24 will assemble pre-existing RAID devices as /dev/md/X
- partman-lvm 60 excludes both /dev/md/X and /dev/mdX

I'll upload the fixed versions soon (probably later today).

Cheers,
FJP

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQBH8Qqugm/Kwh6ICoQRAkJiAKCD9z24ZIO6/Df+MkZjhIzK4UrncQCdG1z6
Uw/JqcWKu1ym+4w4ZSVNCsM=
=XYEA
-----END PGP SIGNATURE-----