mdadm and raidtools

[Bill Rugolsky Jr.]

On Wed, Jan 14, 2004 at 01:32:30AM -1000, Warren Togami wrote:
> Yes, but you didn't do it remotely with the threat of DEATH should the 
> reboot fail. =)

We currently upgrade our Solaris boxes by failing out one drive from
a RAID1 pair, doing the upgrade on that drive, and then rebooting
from the updated drive and resyncing.  If the update fails, reboot
into the old configuration and try again.  (lilo -R or grub --once.)

With mdadm support in initscripts (and soon initrd, I hope), it would
not be terribly difficult to do this in Fedora.  Add in Eric Biederman's
kexec(), and even the reboot is quick.  Extra credit if one tests it first
with UML or Vserver(*). :-)

Regards,

    Bill Rugolsky

(*) Vserver is useful for other reasons related to package dependencies,
    multiple package installation (stable+testing on one box),
    and configuration management. It remains to be seen whether
    SELinux + CKRM + multi-mount/private-namespaces will meet those needs.