[SATLUG] multipath issues on SLES 10SP2.

John Pappas j at jvpappas.net
Tue Jul 7 16:32:36 CDT 2009


On Fri, Jul 3, 2009 at 15:19, Enrique Sanchez Vela
<esanchezvela at yahoo.com>wrote:

>
> We're deploying a number of systems using SUSE 10SP2 using EMC storage with
> multipath & LVM, and we have a bit of a problem, we have two groups of
> systems, ones showing /dev/dm-XX physical volume name (groupA) and the ones
> showing /dev/mapper/mpathXX name (groupB) when using  the pvs command, the
> /etc/lvm/lvm.conf and /etc/multipath.conf files are exactly the same on both
> groups.
>

Bit late to this party but:
Is this a FiberChannel setup? iSCSI?  SAS?
What vendor (or driver) are the HBAs/HCAs
Are systems sharing LUNs or are they uniquely assigned?
Is this a DMX or Clariion (Clarions do multipathing differently than DMX)?

here is the /etc/multipath.conf file .. http://pastebin.com/m7bf38584
> and here is the /etc/lvm/lvm.conf one ... http://pastebin.com/m518e5e83
>
> here is the output of pvs on one of the systems of groupA with kernel
> Linux systemA 2.6.16.60-0.39.3-smp #1 SMP Mon May 11 11:46:34 UTC 2009
> x86_64 x86_64 x86_64 GNU/Linux
>
>  http://pastebin.com/m2ebb6f2f
>
>
> and here is the output of one of the servers in groupB with slightly older
> kernel;
> Linux systemB 2.6.16.60-0.21-smp #1 SMP Tue May 6 12:41:02 UTC 2008 x86_64
> x86_64 x86_64 GNU/Linux
>
> http://pastebin.com/m29c88b5d
>

Is there an alias assigned to the command `pvs`?  Also with the older
kernel/LVM-Tools version, the outputs might be different, but the ultimate
device is the same.  I don't have a multipath system availible here with me,
but if you do an `ls -al /dev/dm-* /dev/mapper/*-*` and look at the
major/minor numbers, you can correlate which DM is which mapper entry.

what are the outputs of `vgs` and `lvs` just so I can get a better view of
the environment.

Depending on the setup, you can `cat /proc/scsi/scsi` to check what disks
and/or LUNs are seen.  If you are using the qla-2x drivers, then the files
under the directory /proc/scsi/qla* can help, especially verifying a MPath
setup.  Make sure the target:LUN combos correlate to what you think you have
provisioned.


> I've added a EMC share to the system in group A with a lower serial number
> than the existing ones and restarted the server, the server picks up the new
> lun but renames all physical volumes starting with dm-0 with the newer lun
> and from there up, however, it seems to be able to manage the name change
> and correctly mounts all filesystems, but I still doubt it is working
> correctly.
>

I don't have a good grasp of what the actual problem is.  LVM and MPIO use
the disk sig to ID the disk, not it's HBA:Targ:LUN, thus jockying with the
LUNs is not a huge deal, although certainly not recommended.


> multipath was running just fine on all servers while doing the tests.
>

How do the current results differ from the expected?  Again, not sure what I
am troubleshooting here, so please advise...

Thanks,
jp


More information about the SATLUG mailing list