SATA hotplug + mdadm raid

[Sam Smith]

Hi,

I recently upgraded my home server to an HP ml30 tower server. It came
with a 4 drive hotplug SATA cage. I loaded two old unused drives in it
and installed Debian Stretch on it putting the drives in a software
raid1 via the debian installer. My plan was once I got the new box up
and running that I would shut off the old machine and use the drives
from that, swapping them out one at a time and letting mdadm resync.

Before putting the new box into "production", I pulled the 2nd drive
(sdb) to see what would happen (with the machine still running). Of
course mdadm went into degraded mode with a one disk raid1. I slid the
same drive back in and it came back as /dev/sdb. Later on, after a few
reboots, I pulled the 2nd drive and swapped it with a drive from the old
machine (it was also still up and running with mdadm raid1). However
this time the new drive didn't come up as sdb, but as /dev/sdc. I set up
the partitions the same as sda but I didn't join it to the mdadm array
yet. I wasn't sure if joining as "sdc" and then on reboot having it come
up possibly as "sdb" would mess up anything so I just rebooted. On
reboot it came up as "sdb" and I joined it to the array and all was well.

But I still need to swap the first drive (sda) and I don't really want
to have to reboot this time. So what can I do to ensure that once I pull
the old drive and put in the new one that it comes back up as "sda"? Or
does that even matter? (seems like it would..)


Regards,

Samuel Smith

[Dan Ritter]

It doesn't matter that much. Use this:

mdadm --fail /dev/md0 /dev/sda1
mdadm --remove /dev/md0 /dev/sda1
mdadm --add /dev/md0 /dev/whatever1

then check on progress with cat /proc/mdstat

-dsr-

[Andy Smith]

Hi Sam,

It doesn't matter what your devices are called. In fact you are best
advised to avoid use of the /dev/sd* names where possible as these
names may change for reasons other than drives being hotplugged. For
example if your storage controller needs a module to detect drives,
then order of module loading may affect device naming.

Try to use the paths in /dev/disk/by-id/ or similar.

mdadm itself recognises array component devices by its own metadata
so does not care what they are called (as long as you haven't told
mdadm to ignore those device names).

The only thing you might want to check out is whether your BIOS is
going to see a bootloader on the drive it tries to boot from next
time.

Cheers,
Andy

--
https://bitfolk.com/ -- No-nonsense VPS hosting

[Sam Smith]

On 05/18/2017 12:30 PM, Dan Ritter wrote:
>
> It doesn't matter that much. Use this:
>
> mdadm --fail /dev/md0 /dev/sda1
> mdadm --remove /dev/md0 /dev/sda1
> mdadm --add /dev/md0 /dev/whatever1
>
> then check on progress with cat /proc/mdstat
>
> -dsr-
>

Ok, I just went ahead and yanked the drive and stuck the other one in.
It actually came up as sda, probably because I had previously used dd to
copy the first 1G onto it from the old drive. I partitioned it and added
it to the array, all is well now.

Thanks,

Samuel Smith

[deloptes]

Sam Smith wrote:

> But I still need to swap the first drive (sda) and I don't really want
> to have to reboot this time. So what can I do to ensure that once I pull
> the old drive and put in the new one that it comes back up as "sda"? Or
> does that even matter? (seems like it would..)

As Andy Smith wrote mdadm is using the metadata particularly UUID

blkid /dev/md4
/dev/md4: UUID="b426722d-ec49-4c6b-a638-5941f24debfd" TYPE="swap"


Watch out that you update your mdadm.conf file

man mdadm.conf

and don't forget to update initrd after changing if you boot from raid,
because it is using the conf file to assemble the disks at boot time.

regards