installer fails creating two separate RAID devices and boot fails (regression)

[Daniel Kulesz]

Binary package hint: debian-installer

Using the server install CD, setting up separate RAID devices fails in 10.04, while the same process successfully completes in 9.10.

System used:
Standard x86_64 PC with two 500GB WD Sata Drives (freshly and completely zero'd with dd), standard x86_64 server install cd

Steps to reproduce:
1.) Start the server installer
2.) Continue to the partitioning menu
3.) Select manual partitioning
4.) Create two primary raid partitions on /dev/sda (one 20GB, one 480GB)
5.) Create two primary raid partitions on /dev/sdb (one 20GB, one 480GB)
6.) Create a new RAID 1 array (md0), with members sda1 and sdb1 and no spare drives.
6.) Create a new RAID 1 array (md1), with members sda2 and sdb2 and no spare drives.
7.) Please note, that the installer indicates some "unuseable" space on the RAID devices - the installer in 9.10 does not so.
8.) Create one partition on the first RAID1 array (md0) for / and one on the second RAID array (md1) for /home
9.) Continue with the installation as usual, install grub into /dev/sda and /dev/sdb
10.) Reboot the system

Expected behavior:
The system should boot up normally, and we should have the following partitioning scheme:
/dev/md0p1 - ext4, 20GB
/dev/md1p1 - ext4, 480GB

Actual behavior:
The system fails to boot and has the following scheme (recorded from the rescue CD):
/dev/md0 - no filesystems (!!)
/dev/md1p1 - ext4, 20GB
/dev/md1p2 - ext4, 480GB

The prompt gives a message similar to this one (this message is actually extracted from a forum thread http://swiss.ubuntuforums.org/showthread.php?t=1120840 - i could not capture the real message from screen):

Log of fsck -C3 -R -A -a
Tue Apr 7 22:04:47 2009

fsck 1.41.3 (12-Oct-2008)
/dev/md0: The filesystem size (according to the superblock) is 122096000 blocks
The physical size of the device is 122095984 blocks
Either the superblock or the partition table is likely to be corrupt!

/dev/md0: UNEXPECTED INCONSISTENCY; RUN fsck MANUALLY.
 (i.e., without -a or -p options)
fsck died with exit status 4

Tue Apr 7 22:04:48 2009
----------------

Workarounds:
a.) Do manual partitioning and RAID creating, then boot the 10.04 server install cd
b.) Install 9.10 to get the partition layout done right, then boot the 10.04 server install cd and re-use the partitions, saying "yes, format it".

For me this is a huge regression, since everything worked fine in 9.10 and for a LTS release I would have never expected such a major breakage in the installer!

[Matthew Caron (matt-mattcaron)]

Confirmed.

Also is still broken in 10.04.1.

[David Dombrowsky (davek)]

I had this happen to me last month. My setup is similar, with two drives with RAID1. In addition, I'm using LVM on top of everything else, but I don't think that effects the root issue. My workaround was basically to use cfdisk to make the primary partitions. The exact steps were:

    * zero'd out both drives and removed both partition tables
    * ran the Ubuntu installer to create a normal primary 20 Gb partition on Drive A and installed the OS
    * booted into Linux, ran cfdisk on drive B (cfdisk -z /dev/sdb), wrote the partition table, and rebooted just for good measure
    * then ran cfdisk /dev/sdb again, and created two partitions (one 2 Gb boot partition, and the rest in another primary). I created an ext4 filesystem in the larger partition.
    * next, booted off the USB drive, mounted the new partition and tar-piped everything from the installed partition into the new
    * then zeroed out the partition table on drive A (/mnt/sbin/cfdisk -z /dev/sda), rebooted into the installer, mounted drive B, and ran cfdisk to create the partitions on drive A.
    * once the primary partitions were created, I used the installer to set up RAID1 and LVM.

After I manually created all my primary partitions, I was able to use the installer for everything else. It certainly seems that the installer's disk partitioner is broken in this case.

[Matthew Caron (matt-mattcaron)]

This works better in 10.10, as in it boots and I have access to all the partitions I created. However, fdisk still complains that the partition does not end on a cylinder boundary, so something is still not quite right.

[Matthew Caron (matt-mattcaron)]

This works correctly in 10.04.2. Installs, runs, no warnings from partitioning programs about the partition geometry.

I move to close as fixed.