[Lustre-discuss] Kernel panic on mounting an OST

[Ludovic Francois]

On Dec 13, 2007 12:53 PM, Wojciech Turek <wjt27 at cam.ac.uk> wrote:
>  Hi,

Hello Wojciech

> I just would like add that you could do very simple test to see if mpath is
> working correctly.

By the way we are reworking on it to integrate nicely mulitpath with
the DDN disk array. And generate the multipath.conf file according at
LUNs wwid and s2a labels.

> On your server oss1 run tunefs.lustre --print
> /dev/<all_mpath_devices> then write down target name for each mpath device.
> Reboot the server and do the same and compare if the mpath -> target map is
> the same as it was before reboot.

I just did it, and it's exactly the same, between 2 different reboots.
But I accord you it's possible the device name moved in the time
according our actual configuration.

How do you explain the UUID can move in the time, the uuid of mpath
device and uuid of sd device corresponding should be always the same.
Am I wrong?

I turned off the multipathing, and I got exactly the same issue.

,----
| LustreError: 134-6: Trying to start OBD lustre-OST0030_UUID using the
| wrong disk lustre-OST0000_UUID. Were the /dev/ assignments rearranged
| ?
`----

I think the multipath is not the problem right now, but the fact there
is some weakness, maybe the multipathing helped someone to do a wrong
command.

Do you think it's possible  someone overwrote the "label" with a tunefs command?

I already saw it with some other file system.

Ludo,

-- 
Ludovic Francois               +33 (0)6 20 67 05 42