I haven't recently been following d-i development, but had to rebuild a
system over the weekend. I used a daily-built netinst and found what I would
categorize as a regression in the installer from the etch release.

I have two drives on which I had manually created two md-raid devices
(/dev/md0 and /dev/md1). During the (manual) partitioning phase of the
installer, I select to configure software RAID and let it do the modprobe
and starting of the arrays. This appears to work, since both pre-existing
arrays are started, synced and healthy. partman re-inits and displays the
main screen as expected. The problem, however, is that any further action
which involves the md-drives claims that "The partition tables of the
following devices are changed: RAID1 device #1". A quick way to reproduce
this is to run the "Setup software RAID" step twice successively (should
really be idempotent, but isn't).

If I wipe the swraid superblocks and create the md-raid drives from within
the installer, however, things work as expected. This (along with comments
from bug #37791) seems to indicate that a flag is not being set properly
during the initial scan and starting of existing drives, but letting partman
create the devices sets the flag(s) properly.

This becomes a blocker if, as in my case, I put LVM on top of the resulting
md devices and wish to have multiple root installations and a shared data
drive - I want to run d-i multiple times for multiple root installs but am
unable to (cleanly) do so since I need to recreate the underlying md device
each time.

I looked for references relating to this issue, but couldn't find any short
of bug 377391 which appears related, but claims resolved in July of 2006.
Perhaps it is only partially resolved? (I'm not familiar with the partman
internals). Where (presumably under /var/lib/partman/) do I look for and/or
change the relevant flags if that is indeed the/an issue?

As a reference, I tested the same procedure with the following installation
media, all of which failed which indicates that it is a problem that is not
brand new: Debian netinst (19Oct07 and 28Dec07), Ubuntu 7.10 and 8.04
alpha2. The expected behavior *was* obtained with the etch (4.0r1) KDE CD.

Thanks, and I'm more than willing to provide further information to help
resolve the issue. I would assume that partman-md would be the packages
against which to file a bug, but would like to verify that first.

Joel Johnson

P.S. The partition structure is as follows (not exact, but the general idea):

drive 1:
partition1 (raid autodetect)
partition3 (raid autodetect)
partition4 (LVM)

drive 2:
partition1 (raid autodetect)
partition3 (raid autodetect)

partitions3 are combined into a small md0 for /boot, partitions1 are
combined into a large md1 for data and root volumes, and partition4 uses the
extra space of drive1.

To UNSUBSCRIBE, email to debian-boot-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org