thoughts on mdadm and dev node creation

[Neil Brown]

On Friday August 18, madduck at debian.org wrote:
> also sprach martin f krafft <madduck at debian.org> [2006.08.18.0914 +0100]:
> > > >   (b) udev should thus create /dev nodes, and nothing else
> > > 
> > > I don't see that.
> > 
> > The theory is simply that mdadm should not know anything about
> > major/minor device numbers anymore.

I don't think that is rational.  mdadm should know what it needs to
know to get the job done.  The "job" is to provide a comprehensible
interface to the user.  In my mind that means that when you create (or
assemble) an mdadm array, you should use a name for the array which
the user will continue to use to manage/mount the array.  This
strongly suggests you should use a /dev/whatever name.  This suggests
that mdadm should (optionally) create the device nodes..

> 
> This on IRC this morning:
> 
> 10:46:52 < Md> madduck: using (and then deleting) /dev/.tmp-md-xx
> may be a workaround until a proper sysfs interface will be
> available, but the final goal should be to remove knowledge of
> major/minor numbers from mdadm (since sooner or later they will
> become dynamic)

partitioned md numbers are already dynamic. mdadm copes just fine.

> 10:47:34 < Md> madduck: mdadm can easily check if udev is being used
> (test -e /dev/.udev) and switch to the "don't create devices" mode
> 

Still don't like the implied user experience...

> 
> 
> Also:
> 
> 10:48:18 < Md> madduck: I do not know much about mdadm.conf (I use
> kernel autostart) so I cannot comment on the merits of requiring it.
> obviously I'd rather do without more configuration files...
> 
> 
> 
> Isn't kernel autostart deprecated? Do you know, Neil?

Not exactly deprecated - it will continue to work as well as it does
today.
I just don't recommend it because I believe it has fatal (though not
enormous) flaws.
But if people want to use it, I'm not going to try to stop them.

NeilBrown