Andree's Debian & General Musings

Whilst my contributions are totally dwarfed by those of Joey Hess, I wholeheartedly agree with what he says - Debian should be more than just a large collection of readily available packages. Ubuntu is great, but Debian should be able to innovate in different ways because of its different structure and longer release cycles (which can be both a burden and a blessing). I really would like to see Debian staying relevant to end-users, otherwise many people may eventually just stop contributing. That would be a big blow for all the derivatives as well, not just Debian itself.

[Update: Fixed 'what he says' URL.]

I spent some time the past week looking into virtualisation to make installing and restoring installations with Mondo Rescue a bit less disruptive and more efficient. So, I ended up doing a fair few Debian installations. I figured that with all these virtual machines all of a sudden, it would make sense to upgrade my caching-only name server to handle the local network to save me the effort of having to maintain lots of hosts files.

To my pleasant surprise, I found that this also resulted in the hostname and domain being automatically filled in by the Debian Installer (even without DHCP)! Good stuff! Also, the graphical installer in the i386 daily builds is just awesome!

Conclusion: The d-i people do positively rock big time!

CeBIT Australia ate substantial amounts of my time for the last two weeks (and almost killed both my feet and my voice but was still interesting and fun), so things took a little longer in Mondo Rescue land. But at last, here it is: mindi-1.07-3 and mondo-2.07-2 are in incoming and add the much overdue RAID support for Debian!

What it does (from the shiny new NEWS.Debian file):

This is the first release supporting RAID via mdadm. It is supposed to work with all RAID levels/varieties supported by mdadm and the kernel.

The following limitations and oddities apply:

• All information about existing RAID arrays is drawn from /proc/mdstat, the contents of an mdadm.conf file is ignored.

• Using RAID for the boot device is currently untested. The system will likely be unbootable after a restore.

• Using LVM in conjunction with RAID is currently untested. Restore will likely fail.

• Building a RAID5 array may appear to hang the system because of extended screen inactivity. However, disk activity should be high and eventually the restore will continue.

• The synchronisation progress bar and the formatting progress bar may overwrite each other leaving the screen somewhat garbled. Not nice but harmless.

I have intensely tested this with various RAID levels, chunk sizes, parity algorithms, with and without spare disks and so forth. In particular, I've done successful full archive and restore runs using NFS as storage media for the following:

• amd64, RAID1, kernel 2.6.16-1-amd64-k8 (2.6.16-12)
• amd64, RAID5, kernel 2.6.16-1-amd64-k8 (2.6.16-12)
• amd64, RAID10, kernel 2.6.16-1-amd64-k8 (2.6.16-12)
• i386, no RAID, kernel 2.6.16-1-k7-smp (2.6.16-10)

I'll post the patch to #325877 for reference purposes and will also commit to upstream SVN shortly. Bruno was kind enough to review that patch and likes it, so there shouldn't be a problem (other than some fiddling because I have worked off the Debian 2.07-1 package and not tracked upstream).

Apart from fixing issues with the above, I plan the next steps to be:

• support for RAID boot partitions
• getting LVM to work

However, I think I'll do a fresh install of Sid AMD64 before that, to ensure that d-i created systems work with Mondo Rescue.

Also, I am experiencing a strange behaviour with busybox on AMD64 in that I get the following message on virtual terminals (and no prompt):

./sh: Cannot set tty process group (Operation not permitted)

when I run:

./openvt 8 ./sh

(where ./openvt and ./sh are both links to ./busybox)

This happens with 1.00 as well but in either case only on amd64, i386 is fine. Also, it only happens if NFS support is enabled for busybox mount.

I'm currently working on a better routine to parse the output of /proc/mdstat in Mondo Rescue to get RAID to work on Debian.

Fiddling around with pointers and doing nasty things in general finally got me to think that there must be a better way.

I'm sure everyone else is in the know, but I was quite happy to find the fine GNU C Library Manual where in turn I found strtok() which allows for much cleaner and safer code. Cool.

Also found www.snippets.org. Cool, too.