post installation handling of raid devices on boot

Oh, I'm also thinking, since it is such a minor update, perhaps it could be put in for 10.04 as a patch level update, rather than waiting till the next release of ubuntu?

I think it's obviously the right thing not to require people to do this kind of work by hand if at all possible, but I'm cautious about introducing a new interface for it. In particular, Debian already has code which automatically generates the configuration file if it's absent using '/usr/share/mdadm/mkconf generate /etc/mdadm/mdadm.conf', without requiring anything in the configuration file; this code was removed in Ubuntu as part of a general simplification when we moved to udev for RAID assembly. Maybe we just need to restore that little piece of complexity?

Surbhi, could you have a look at this as part of (or following) your mdadm merge in progress?

Colin, yeah, okay, that makes more sense. Perhaps just keeping mkconf in the distro, and only calling it during initramfs is appropriate then?

@Trenton D. Adams,

Just to make sure, did you add the raid through the following commands:
1) mdadm --create /dev/md0 --level=<level> --raid-devices=2 /dev/sda3 /dev/sdb3
2) mdadm --create /dev/md1 --level=<level> --raid-devices=2 /dev/sda5 /dev/sdb5
?
If you do this, then you should not have to do anything more that this. Is you /etc/mdadm/mdadm.conf not getting updated inspite of this? Or else, what command did you use to get your raid under md?

Seems that this bug is found in cases where old superblocks are found on the disks during mdadm install. I could not see this bug on a fresh disk under fresh Lucid install. Linking this bug to another bug 158918.

Please do unmark this as not duplicate, if your disks are fresh. Also, note that I tried to reproduce this on KVM and not on a real machine. Wonder if that made the difference. Thanks!

Yes, I did the commands in comment #6, while the system was running, and my mdadm.conf was not updated. I then did a live pvmove over to the new md device, update-grub, update-initramfs, and rebooted. It wasn't until I did the command to force writing the config to mdadm.conf that it got there.

I'm not sure that the other bug is the same. Mine only occurred when I actually switch to using raid arrays. When I update my mdadm.conf with the appropriate configuration, it just works. But, I'm not an mdadm expert, so it's possibly this is a duplicate of the other one.

Requesting sponsorship: Attaching a tested patch for mdadm-2.6.7.1. This consists of the code cherry picked from the equivalent file from debian. Please consider merging for Maverick.

P.S: Shall create another patch for mdadm-3.2.1(different changelog). This version is yet not merged in Maverick and so created a separate patch for 3.2.1

What is sponsorship?

Trenton D. Adams, someone who has rights for merging the patches in the Ubuntus mdadm package should after reviewing this patch accept it (if it looks ok) and merge it in. So requesting sponsorship means basically requesting someone to review the patch and to merge it in the package.

I sponsored it now, sorry that it took so long :/

The diff looks fine, note that I did not test it, I don't have a mdadm configuration.

This bug was fixed in the package mdadm - 2.6.7.1-1ubuntu16

---------------
mdadm (2.6.7.1-1ubuntu16) maverick; urgency=low

  * debian/initramfs/hook: Added following code (invoked on update-initramfs)
    (LP: #617725):
    - create a mdadm.conf if it is not found in /etc and copy it in initramfs
    - update an existing mdadm.conf in the initramfs if it does'nt include
      a definition of any array
    - warn the user if the definition of an active array is not found in the
      initramfs/etc/mdadm.conf
 -- Surbhi Palande <email address hidden> Mon, 13 Sep 2010 18:59:03 +0300