Re: split brain mode after reboot

[NeilBrown <neilb@suse.de>]

[-- Attachment #1: Type: text/plain, Size: 1602 bytes --]

On Tue, 20 Nov 2012 15:08:45 +0100 Carsten Aulbert
<Carsten.Aulbert@aei.mpg.de> wrote:

> Hi all
> 
> a colleague of mine created a raid 1 on a fairly recent machine
> 
> Kernel 3.5.0-sabayon
> 
> mdadm - v3.2.3 - 23rd December 2011
> 
> During operation sda seemed to have been disconnected by the
> system/motherboard/whatever but this was not detected before a reboot
> was done, after the reboot, sda re-appeared but of course with a much
> older version of the mirror:
> 
> Personalities : [linear] [raid0] [raid1] [raid10] [raid6] [raid5]
> [raid4] [multipath] [faulty]
> md126 : active raid1 sdb1[1]
>       4194240 blocks [2/1] [_U]
> 
> md127 : active raid1 sdb3[1]
>       235808704 blocks [2/1] [_U]
> 
> md0 : active raid1 sda1[0]
>       4194240 blocks [2/1] [U_]
> 
> md1 : active raid0 sda2[0] sdb2[1]
>       8387584 blocks 512k chunks
> 
> md2 : active raid1 sda3[0]
>       235808704 blocks [2/1] [U_]
> 
> unused devices: <none>
> 
> As no vital information were on these disks, my question for the list is
> just if this is an expected/wanted behavior after such an event and what
> one could do to prevent this (besides monitoring via mdadm).
> 

Expected - probably.
Wanted - no.

I think one half gets assembled by "mdadm --incremental" run from udev, and
the other by a subsequence "mdadm -As" or similar.

mdadm-3.3, which is still under development, has a fix for this so that
--incremental and --assemble don't trip over each other.

http://git.neil.brown.name/?p=mdadm.git;a=commitdiff;h=0431869cec4c673309d9aa

NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]