Trouble assembling multipath md devices between two nodes who see the same LUNs.

Trouble assembling multipath md devices between two nodes who see the same LUNs.

[Anu Matthew]

Hello,

I have two nodes, hostA and hostB, both of them see the same 4 multipath 
LUNs.

md0 to md4 are thus visible to both the hosts, (yeah, they both do not 
write to thoe md devices at the same time, hostB mounts them only when 
hostA is down or has crashed, and viceversa.)

It works for a while.

While testing, the latest we have seen is, after md devices are stopped 
on hostA, hostB cannot start the md devices, as the uuid changes, and
comes back with a message in --verbose mode: 

mdadm: /dev/sdc  has wrong uuid

mdadm: no devices found for /dev/md0

Tried copying over the mdadm.conf with the latest uuids from hostA, 
still hostB cannot assemble them. I have to re-create the md devices on  
hostB eventually, then after a while, hostA starts to act up: it simply 
refuses to assemble.

Is there a way I can circumvent this problem? I am running rhel 3.0 (AS 
and ES), and we're on  mdadm - v1.5.0 - 22 Jan 2004, from Redhat's 
mdadm-1.5.0-9 rpm.

Thank you so much in advance,

--AM

Re: Trouble assembling multipath md devices between two nodes who see the same LUNs.

[Lars Marowsky-Bree]

On 2005-04-18T17:14:53, Anu Matthew <anu.matthew@bms.com> wrote:

> Hello,
> 
> I have two nodes, hostA and hostB, both of them see the same 4 multipath 
> LUNs.
> 
> md0 to md4 are thus visible to both the hosts, (yeah, they both do not 
> write to thoe md devices at the same time, hostB mounts them only when 
> hostA is down or has crashed, and viceversa.)
> 
> It works for a while.

md multipath has on-disk metadata and modifies it. md is NOT
cluster-safe for concurrent activation.


Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX Products GmbH - A Novell Business

-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Trouble assembling multipath md devices between two nodes who see the same LUNs.

[Anu Matthew]

Thanks Larks. It is much appreciated. 

If the original metadata written by hostA changes by "mdadm --assemble" running on hostB, will mdmpd be able to recover the failed links on hostA when they re-surface? I am asking this because, I guess unless hostA goes down, hostA reflects the new metadata written by hostB. Will mdmpd take the new uuid into account? 

Not entirely OT, but devabel gets to the see the same uuid/serial numbers on the shared luns across hosts. It would have been cool if mdadm could too, right?

Thanks,

Anu Matthew 

<snip>
> On 2005-04-18T17:14:53, Anu Matthew 
> 
> md multipath has on-disk metadata and modifies it. md is NOT
> cluster-safe for concurrent activation.
> 
> 
> Sincerely,
>    Lars Marowsky-Brée <lmb@suse.de>
> 
> -- 
> High Availability & Clustering
> SUSE Labs, Research and Development
> SUSE LINUX Products GmbH - A Novell Business
</snip> 

-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Trouble assembling multipath md devices between two nodes who see the same LUNs.

[Luca Berra]

On Mon, Apr 18, 2005 at 08:24:44PM -0400, Anu Matthew wrote:
>Thanks Larks. It is much appreciated. 
>
>If the original metadata written by hostA changes by "mdadm --assemble" running on hostB, will mdmpd be able to recover the failed links on hostA when they re-surface? I am asking this because, I guess unless hostA goes down, hostA reflects the new metadata written by hostB. Will mdmpd take the new uuid into account? 
DONT ACTIVATE THE ARRAY ON MORE THAN _1_ HOST AT THE SAME TIME.
whatever you do in the failover setup make it so that the md is
activated only on the active node.

>Not entirely OT, but devabel gets to the see the same uuid/serial numbers on the shared luns across hosts. It would have been cool if mdadm could too, right?
mdadm can use the UUID stored in the raid superblock, thus avoiding the
need to using devlabel (last time i looked at devlabel it was so buggy,
it could not make head or tails of a non partitioned disk)

-- 
Luca Berra -- bluca@comedia.it
        Communication Media & Services S.r.l.
 /"\
 \ /     ASCII RIBBON CAMPAIGN
  X        AGAINST HTML MAIL
 / \