Using mdadm --examine to write mdadm.conf

Question

I'm trying to script the RAID setup for a (growing) number of identical servers. The advice I've seen for using mdadm suggests that once the RAID array has been created you can run mdadm --examine --scan >> /etc/mdadm.conf so that the array is initialised on bootup. But that doesn't seem to work for me - in particular the array is presented as /dev/md/<hostname>:pool, where the mdadm --examine outputs /dev/md/pool.

Am I doing something wrong? Is there something odd about my version of mdadm?

[root@positron /]# mdadm --create /dev/md0 --level=0 --name=pool --raid-devices=6 /dev/sda /dev/sdb /dev/sdc /dev/sdd /dev/sde /dev/sdf
[root@positron /]# mdadm --examine --scan
ARRAY /dev/md/pool metadata=1.2 UUID=567da122:fb8e445e:55b853e0:81bd0a3e name=positron:pool
[root@positron /]# ls /dev/md/*
/dev/md/md-device-map  /dev/md/positron:pool
[root@positron /]# cat /dev/md/md-device-map
md127 1.2 22a17d56:5e448efb:e053b855:3e0abd81 /dev/md/positron:pool
[root@positron /]# mdadm --version
mdadm - v3.1.3 - 6th August 2010

Are you running udev? I've come across a few issues on debian where udev wasn't assigning the right device. — Thomas James, Nov 03 '11 at 12:34
Yes, on RHEL rather than Debian, but running udev. Not sure quite what to do with this tip, but I'll go Googling. — David Waller, Nov 03 '11 at 13:38
Do you get problems when omitting the --name option during creation? — Nils, Nov 07 '11 at 21:19

score 29 · Answer 1 · answered Oct 26 '12 at 00:42

29

And here's the fix, I'm not entirely clear on why it works, but it does!

After updating your mdadm file following the RAID array creation - usually with something like mdadm --detail --scan >> /etc/mdadm/mdadm.conf

Simply run the following command update-initramfs -u

Then reboot, and everything will work as expected.

Voila ;)

answered Oct 26 '12 at 00:42

Matthew O'Riordan

391
3
3

(Very old, but:) the reason this is necessary because without `update-initramfs -u` the array is being assembled early in the boot process via some "autodiscovery"-like mechanism, before `/` is mounted (and so `/etc/mdadm/mdadm.conf` can't be read). Whereas running `update-initramfs -u` copies your `mdadm.conf` file into the initramfs so that it can be read early (`mdadm` installs a hook script which copies the file). – Keeley Hoek Jun 01 '23 at 18:30
When `mdadm` isn't sure that the array it is assembling is supposed to be assembled on your specific host (e.g. when it couldn't find the array in `mdadm.conf`) then it prepends the hostname to the name of the array, as is happening to you. – Keeley Hoek Jun 01 '23 at 18:33

score 1 · Answer 2 · edited Dec 21 '11 at 14:50

1

I've found that I've had to manually edit both the mdadm.conf and udev.conf to get the manual device names to hold.

However, I've not found it to be an issue one LVM is layered on top.

edited Dec 21 '11 at 14:50

Kevin

40,087
16
88
112

answered Nov 21 '11 at 12:20

Thomas James

276
3
5

score 0 · Answer 3 · answered Jun 18 '12 at 17:44

0

I don't see what's the problem. Array will be assembled on boot okay, even if it would be only mentioned with its UUID.

answered Jun 18 '12 at 17:44

poige

6,195
2
30
57

Using mdadm --examine to write mdadm.conf

3 Answers3