Re: broken raid level 5 array caused by user error

[Phil Turmel <philip@turmel.org>]

On 11/10/2015 10:20 AM, Mathias Mueller wrote:
> Hi Phil,
> 
> Is it correct to use --data-offset=1024 when I want to set an offset of
> 2048? At least a mdadm --examine on one of the devices shows the offset
> is set to 2048 when using --data-offset=1024.

This must be a bug in an old mdadm.  Don't use it.

> With --data-offset=1024 and both chunk sizes (64 & 512) I have something
> like a success when using this two orders:
> 
> sde sdd sdb sdc
> sde sdb sdd sdc
> 
> I'm getting:
> 
> e2fsck 1.41.12 (17-May-2010)
> /dev/md0 has unsupported feature(s): FEATURE_C16 FEATURE_C17 FEATURE_C18
> FEATURE_C19 FEATURE_C21 FEATURE_C22 FEATURE_C23 FEATURE_C25 FEATURE_C27
> FEATURE_C28 FEATURE_I29 FEATURE_R29
> e2fsck: Get a newer version of e2fsck!
> 
> So I guess I should try some newer fsck version (I'm back on the old
> machine with centos 6, the failed reshaping was done with the new
> machine with centos 7).

Whoa!  Don't do that.  Upgrade to CentOS 7, or go back to the other
machine.  You didn't mention switching machines, btw.  That would
explain the different drive names vs. roles.  You need to document the
device names vs. drive serial numbers on the current setup before moving
or upgrading.  Then adjust the names worked out above based on the
serial numbers in the new environment.

In general, device names cannot be considered reliable -- that's why
labels and uuids are the preferred way to identify things in today's
systems.  And why mdadm has superblocks to keep track of what's what.

And when you have your system fully running again, document the entire
layout with lsdrv [1].

Phil

[1] https://github.com/pturmel/lsdrv