http://bugzilla.novell.com/show_bug.cgi?id=586837
http://bugzilla.novell.com/show_bug.cgi?id=586837#c2
--- Comment #2 from Dennis Olsson 2010-03-17 00:01:49 CET ---
After having done some more debugging, it seems that there is also a problem
with "mkinitrd".
After having installed the packages, YaST issues a pop-up window with a count
down to a reboot.
During this reboot MD device does *not* get (re)activated, which then causes
problems with the next stage of the installation process, while the disk setup
is now screwed-up (in my case the LVM gets activated using "/dev/sdb3" instead
of "/dev/md/raid1p3"!).
But, instead of using the YaST "reboot", it does not help either to do a real
reboot, while the MD device does *not* get activated during the running of the
"initrd" either!
The reason why the MD device does not get initialized during the running of the
"initrd" is that the "md_dev" variable tested for in "boot/51-md.sh" has *not*
been set, which means that "mdadm -A" does not get called! And thus does not
assemble and starts the partitioned MD device.
The "md_dev" variable is *only* getting set in either
"boot/03-storage.sh"
and here _only_ when the *root* device is a MD device (/dev/md*), i.e. when
root is a MD device itself (which is not the case here -- see below),
or
"boot/51-md.sh"
and here _only_ when the root device name consists of hexadecimal digits,
i.e. (as far as I can deduce) when root is kept in a (MD) container.
But, in this case, we have:
/dev/sda + /dev/sdb => /dev/md/raid1
where /dev/md/raid1 has 3 (three) partitions:
/dev/md/raid1p1 containing /boot file system
/dev/md/raid1p2 containing swap device (used as resume device)
/dev/md/raid1p3 containing LVM partition with volume group named "raid1"
where the LVM volume group named "raid1" contains:
/dev/raid1/root113 containing / file system
In the used configuration (see comment #1) the "rootdev" is defined (in
"/config/storage.sh") as "/dev/block/253:1" (device mapper minor 1), which has
the consequence that the "md_dev" variable is being left empty, and thus not
calling "mdadm -A".
As I have not been able to figure out how "mkinitrd" determines how to setup
these things, I cannot say what should be done for configurations like the one
I am using.
Instead, I stopped the YaST2 installation at the pop-up "reboot" countdown by
clicking "Stop", switch to a text console, and made the following changes to
the generated "initrd":
config/md.sh (adding new line; only works, when doing "slow_boot=on"):
md_dev="/dev/md/raid1"
run_all.sh (added after line with "root_md"; only works, when doing "fast
boot"):
md_dev="/dev/md/raid1"
etc/mdadm.conf (changed existing line to these two lines instead):
ARRAY /dev/md/raid1 metadata=1.00 name=raid1 UUID=<value from existing file>
HOMEHOST Loke
Returning to the YaST console and clicking "OK" will *not* work, while for
reasons I do not understand the MD device does _not_ get activated, resulting
in LVM getting very confused about finding the volume group "raid1" twice on
"/dev/sdb3" as well as "/dev/sda3", choicing "/dev/sdb3" over "/dev/sda3".
It should not have - of course - seen the "/dev/sd*" partitions at all, but
instead the "/dev/md/raid1p3" alias "/dev/md127p3" and have used this one
instead.
So, instead, after having replaced the "initrd" with the changes from above
incorporated, I *rebooted* the system by issueing:
vgchange -a n
mdadm --stop /dev/md/raid1
reboot
But, also here LVM gets confused -- this time, though, the MD device *does* get
started, but the LVM activation still sees the "/dev/sdb3" and "/dev/md/raid1"
and prefers "/dev/sdb3" over "/dev/md/raid1" (which, of course, should really
have been "/dev/md/raid1p3"!).
The reason for this confusion is that the MD activation only activates one
device (here "/dev/sda"), but not the other device (here "/dev/sdb"), which is
(also) part of the MD device!
But, why not?? I have not been able to figure that one out...
In any instances, the installation cannot be continued, while the root file
system has a completely wrong setup, eventual corrupting it beyond repair.
It seems that the changes made by Arvin Schnell to implement feature #305883
still needs a few fixes in YaST as well as "mkinitrd", before it works
correctly. ;-)
--
Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.