Bug#336514: yaird: falls down when new scsi/sata disks are added

[Erik van Konijnenburg <ekonijn@xs4all.nl>]

On Mon, Oct 31, 2005 at 01:20:32PM -0500, Daniel Jacobowitz wrote:
> On Mon, Oct 31, 2005 at 07:18:01PM +0100, Jonas Smedegaard wrote:
> > On Mon, 31 Oct 2005 12:47:58 -0500 Daniel Jacobowitz <dan@debian.org> wrote:
> > > On Mon, Oct 31, 2005 at 06:09:28PM +0100, Jonas Smedegaard wrote:
> > > > Ah, the mknod call. Yes, you are right.
> > > > 
> > > > Erik: Recent versions of mdadm supports a new option "auto" to
> > > > auto-create devices as needed. Perhaps that could be used at least
> > > > on Debian to avoid hardcoded device names?
> > > 
> > > I don't think that will help?  That's for /dev/md0, not /dev/sd*.
> > 
> > Right. So what was the problem again with my proposed hack?
> 
> mdadm presumably needs to open device nodes in order to find RAID
> disks.  If we don't _create_ the device nodes, it can't _find_ them.

Daniel: agreed, the mdadm auto hack won't work for the reason you note.

Doing mknod for all partitions and disks in /sys gets close to coldplugging.

I don't expect it to be trivial.  The reason is yaird currently has a depth-first
recursive walk doing exactly the stuff required for a file system.  With coldplugging,
you work in layers: first all hardware probing, then creating block-special nodes,
then doing scan for raid, lvm or crypt devices.

Coldplugging is more resilient to hardware changes, but harder to get right
if someone has crypt below lvm instead of the other way around for example.

Interesting to experiment with, but not a quick fix for today.

Regards,
Erik