On 03/11/2010 14:13, Nat Makarevitch wrote:
Hi,
After a spindle (physical hard disk, a "drive") failure in a "md" RAID array,
how can we know which spindle must be replaced?
We want to avoid extracting a working spindle by mistakenly thinking it is the
faulty one...
To solve this problem we put on each spindle a physical label (a tag, not a
partition/disk label) showing its device name (sda, sdb...).
To do so: on an otherwise unused system and sane RAID array we ran "dd" for each
each spindle (individual device) in order to read on it, lighting up its LED.
Then we simulated a crash by physically removing two spindles. Upon reboot the
devices names changed (?!) and our labels weren't right anymore albeit we are
pretty sure they were.
Context: raid10, 10 spindles (8 active + 2 spare), layout : near=1, offset=3. On
the integrated controller + a LSI MPT on-board controller.
That's right, drive letters sda sdb etc are allocated in the order
they're discovered at boot time, so if you remove what used to be sdc
and reboot, sdc will now refer to what used to be sdd, sdd to the old
sde etc.
Have a look at /dev/disk/by-path, in there you'll find symlinks which
will always have the same names, or indeed be missing, which point to
sda, sdb etc. In my case I have SATA discs on an ICH10R controller, the
controller is pci-0000:00:1f.2, it appears as up to 6 SCSI controllers
numbered 0-5, and the discs appear as HBA 0, device 0, lun 0, along with
their partitions, as follows:
pci-0000:00:1f.2-scsi-0:0:0:0 -> ../../sda
pci-0000:00:1f.2-scsi-0:0:0:0-part1 -> ../../sda1
pci-0000:00:1f.2-scsi-0:0:0:0-part2 -> ../../sda2
pci-0000:00:1f.2-scsi-1:0:0:0 -> ../../sdb
pci-0000:00:1f.2-scsi-1:0:0:0-part1 -> ../../sdb1
pci-0000:00:1f.2-scsi-1:0:0:0-part2 -> ../../sdb2
pci-0000:00:1f.2-scsi-2:0:0:0 -> ../../sdc
pci-0000:00:1f.2-scsi-2:0:0:0-part1 -> ../../sdc1
pci-0000:00:1f.2-scsi-2:0:0:0-part2 -> ../../sdc2
pci-0000:03:00.0-scsi-0:0:0:0 -> ../../scd0
Oh and the last one is a CD ROM drive on a pata_marvell IDE controller.
You might want to relabel your drives according to what you find in
/dev/disk/by-path.
Cheers,
John.
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
