[Gllug] Problems on a box running RAID 1

[Nix]

On 14 Sep 2007, Chris Bell verbalised:

> IDE 0 Master  M$, used occasionally during testing
> IDE 0 Slave   Debian etch first of RAID1 pair
> IDE 1 Master  Debian etch second of RAID1 pair
> IDE 1 Slave   DVD R/W drive
>
>    I can successfully boot either M$, Debian, or a CD/DVD, and can access
> the DVD drive from either M$ or Debian, but get consistent CD/DVD read and
> write errors from Debian. I was suspicious that the old drive had a dirty
> lens, but a totally new good quality drive drive which should not show any
> access or DMA problems gives the same errors. I have had to work around
> drive DMA problems on some boxes, but I suspect that on this occasion I may
> have to physically swap the drives around, so that the DVD drive becomes IDE
> 0 Slave and the RAID1 drives share the same cable.

No no no! You'll get *awful* performance if you do that, and on most IDE
and ATAPI systems I've seen, if one drive on a cable dies it takes the
others out with it.

(On the rather aging RAIDed box in my bedroom I benchmarked a speed
reduction from 45Mb/s to *8Mb/s* simply by moving from one drive per
controller to both drives daisy-chained. Well, actually it was the other
way around, and since this machine has arrays containing both those IDE
drives and drives on a SCSI controller that can easily hit well above
50Mb/s, I preferred a bit more speed from my IDE drives... of course
this may have been an extreme case since one of those drives is really
rather aged by modern standards and can only manage 25Mb/s at the bes to
ftimes: it's only in my `archival data, low speed' array, but on the
same cable it was slowing down the far faster disk behind it, which is
in my primary RAID-5 array, the one with / and /usr on it.)

>    Am I likely to hit any unexpected snags when I re-arrange an MD array, or
> do I just have to reconfigure MDADM and GRUB?

Rearranging should work for RAID-1 no matter what: it's a mirror, so it
doesn't matter what order the components are assembled in.

If you ever plan to use anything but RAID-1 and want to avoid destroying
your data on rearrangement, *avoid* using the kernel auto-assembly
feature, use mdadm in an initramfs or initrd to assemble the array, and
use the DEVICE partitions stanza and uuid labelling in /etc/mdadm.conf.
e.g. from one of my RAID installations, with three arrays (one RAID-1,
two RAID-5):

DEVICE partitions
ARRAY /dev/md0 UUID=3a51b74f:8a759fe7:8520304c:3adbceb1
ARRAY /dev/md1 UUID=a5a6cad4:2c7fdc07:88a409b9:192ed3bf
ARRAY /dev/md2 UUID=fe44916d:a1098576:8007fb81:2ee33b5a

MAILADDR postmaster at esperi.org.uk

If you name by uuid like this, the array will always be assembled in the
right order, no matter what happens to the underlying physical devices'
names, device discovery ordering, and so on.

-- 
`Some people don't think performance issues are "real bugs", and I think 
such people shouldn't be allowed to program.' --- Linus Torvalds
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug