[vox] Help: Mongo smash RAID-10 (long-ish).

[vox@lists.lugod.org]

Okay...there is either something I'm missing, or someone is going to
fine one very headless maintainer of the Linux-RAID-HOWTO.

I'm trying to set up a software-based IDE RAID-10 on a Linux machine
which is destined to be a log server; the RAID is composed of four
disks, each idential (80G Maxtors), and each is a master on a separate
IDE bus (courtesey of an additional Promise IDE controller).  I am
RAIDing the third partition on each drive, and have grouped them as
such:

/dev/hda3,/dev/hdc3 -> /dev/md0 |
				+-> /dev/md2
/dev/hde3,/dev/hdg3 -> /dev/md1	|

So, /dev/md0 and /dev/md1 are both RAID-0 arrays (stripes), and these
are mirrored against each other, the mirror being /dev/md2.

Now, I can set this all up without a problem; performance is good, the
RAID-10 seems to work, birds sing, bees polinate, champagne rains from
the heavens.  If I fail out any of these disks, the RAID does indeed
continue working as advertised -- the disk is marked as failed, and all
is well.

Until I replace the disk and try to get things working again.

First off, 'mdadm' refuses to see /dev/md2 -- it just pretends that it
doesn't exist, even though I've set up the config file with the
appropriate 'DEVICE' entries and such (let me know if you'd like to see
a copy of this file).  I can use raidtools to manage the device, but if
I do, I am unable to bring /dev/md2 back online.  I can easily restore
the 'failed' disk within /dev/md0 or /dev/md1, but /dev/md2 refuses to
integrate the failed disk back into the array -- it just sort of sits
there on the side, doing nothing, with the old disk marked as 'failed'
in the output of 'mdadm -D'.

Any ideas?  Questions?  I'd be a bit more coherent were I not so utterly
pissed off at this point in time...

-- 
Don Werve <donw@examen.com> (Unix System Administrator)

Yorn desh born, der ritt de gitt der gue,
Orn desh, dee born desh, de umn bork! bork! bork!