hi all

I used mdadm to create two raid devices, md0 which is a 10G stripped (2x5G) and md1 which is 5G mirrored (2x5G). I mounted them using the mount command and tested them out and it seems to work just find.

Then I rebooted the VM and I found that the under /dev, they are no longer named md0 and md1, but md126 and md127.

Why is that?

Thanks

EDIT: using ubuntu 15.04

What is in your mdadm.conf?

Try running
Seems Ubuntu needs this in order to create the device correctly.

My system did that too. But I noticed that it only happens for those md structures given a name. Those are recorded in /dev/md/<name>. When I created a second named device it was given a new
number - and after reboot, the first created became /dev/md126, and the second /dev/md127. It didn't happen when I created some test md devices (or if I forgot to put a name I wanted).

It makes it look like the small numbered md device names are saved for unnamed devices.

I also saw that a /etc/mdadm.conf was not necessary. The devices are created via UUIDs, and organized by the kernel scanning for md devices. The /etc/mdadm.conf can be used to try to rebuild after a disaster, but is not normally used for anything. I didn't create mine until well after both md devices were created (and after several successful reboots).

I used the /dev/md/<name> to remind me which device I was dealing with (fsck/mkfs/partitioning/...) rather than try to remember which /dev/md[0123...] or which ever I was working on.

I find it harder to have a typo with /dev/md/medialib. Much better than accidentally wiping out a home filesystem or something else important.

thanks for the reply guys. It does seem to be that md0 through 125 is already used for something else and it automatically renames them to md126 or higher. I just wish it would remember to automatically edit the /etc/fstab too. or a dynamically created softlink for the new directory.

EDIT:

the underlined statement is wrong. I just rebooted it again and it changed it back to md0 and md1. F**DFKJEIJSF#(!0R(9e8tq0938euf89adsvhjdio newlkrjntgwoeiuv9oaducv iosdjch ck bdslkjvw.

To make it permenant I had to do this:

http://www.ducea.com/2009/03/08/mdadm-cheat-sheet/

The kernel will do it on its own at boot time, and that will use md127, md126,...

In either case, it doesn't matter to me. I mount using volume labels (or UUID) as the only stable names available during the device scan in the initrd.

yeah, suppose I should start using UUIDs too. This is just a play setup so not too important. Honestly UUID is too long. 727cac18-044b-4504-87f1-a5aefa774bda all that just to say /dev/sdb1.

or saying /dev/disk/by-id/ata-SAMSUNG_HD250HJ_S0URJADPC01495-part1 (though knowing the device model/serial is a bit long)

Which is why I like volume labels. There is also a PARTUUID which is much shorter (11 characters); and you can always specify a short UUID as well (which is why I prefer the volume label).

I just find UUIDs hard to remember; but something like "mount LABEL=archive /mnt" is easy (or using /dev/disk/by-label/archive), doesn't depend on the device name (which can change), and reliable.

for fsck (and md devices) I like being able to use /dev/md/mmedia and know I'm referring to the right device (whether it is /dev/md0..1 or md126 or md127).