[ENBD] Cleanly rebooting/halting an ENBD machine (RAID, LVM, cryptoloop, etcetera)

[Peter T. Breuer]

"Also sprach Bas van Schaik:"
> I'm just wondering if you guys have consistency problems when rebooting
> or halting an ENBD client. Currently, my configuration is as following:
> 
> enbd devices > RAID5 > LVM > cryptoloop > reiserfs
> 
> When halting or rebooting, services are stopped in default (Debian?) order:
> 1) enbd
> 2) LVM
> 3) RAID
> 
> So, enbd is stopped before RAID has had a chance to stop the array
> cleanly. Correct me if I'm wrong, but I think this can cause consistency

It is indeed a problem when LVM and RAID are interlaced, because one
can either be running LVM over RAID or RAID over LVM and the correct
order of stopping depends on which it is. Obviously the initscripts
don't know, and can't be expected to know.

Throw enbd (or nbd) into the mix, and you have another problem. Anyway,
the initscripts won't know what to do.

> problems when the device is under high load, resulting in an out-of-sync
> RAID-array. Do you guys have any ideas on this?

If you are running in the layered way you state, you need to 

   1) umount reiserfs (all fs's)
   2) stop cryptoloop (that device)
   3) stop lvm        (that device)
   4) stop raid       (that device)
   5) stop enbd.

Note that it's perfectly possible to be running raid over lvm (hic), so
you can't rely on this order as a general solution to all possible
ills.

Peter