mdadm, superblocks, spooky behavior...

[kentborg-KwkGvOEf1og at public.gmane.org (Kent Borg)]

Some mdadm guru, please help me!

I am confused about mdadm.  I have a raid0 array that mdadm won't
assemble, but if I make loopbacks of the components I *can* assemble my
array from that.


I have an entry in /etc/mdadm/mdadm.conf for my array and when I do:

  # mdadm --assemble --scan
  mdadm: no devices found for /dev/md25

Let's risk blowing away my data (after a backup):

  # mdadm --create /dev/md25 --level=0 --raid-devices=4 /dev/md16
/dev/md17 /dev/md18 /dev/md19
  mdadm: Cannot open /dev/md16: Device or resource busy
  mdadm: Cannot open /dev/md17: Device or resource busy
  mdadm: Cannot open /dev/md18: Device or resource busy
  mdadm: Cannot open /dev/md19: Device or resource busy
  mdadm: create aborted

Look at one of the components with "mdadm --misc --examine /dev/md19"
and it knows it is part of md25.  Each of the four components claims the
same UUID for md25.


Some background:  I have a pair of disks that I chopped up into a bunch
of matching partitions, and I have each pair in a raid1 array.  Most of
those are used directly, but I wanted a bigger partition; that is why I
am gluing them into a raid0 array.  And because there is raid1
underneath the raid0, that striping isn't as dangerous as it looks.


By doing this loopback I am up and running again, but it is wrong to run
this way, just wrong.  Can someone explain what mdadm is doing?


Thanks,

-kb, the Kent who newly nervous that his raid1 redundancy won't save him
from operator or software error.