[dm-devel] SLES9 DM-Multipath strange behaviour

[Harald Milz <hm seneca muc de>]

... or maybe it isn't that strange, and it's just a feature. 

Until recently, we using a single 200GB SAN volume with our heartbeat
cluster. The SAN ist a (yuck) Datacor SANsymphony Windows-based crap, but
so be it. Everything was working fine. Multipath just rocked. The volume is
seen twice, and dm-0 was nicely created. 

Later, a second 50GB volume was added from the same SAN controller -
everything worked OK. 

After that, 3 TB sized volumes were added from a Nexsan controller. These
were handled fine but the first 2 Datacor volumes were gone, i.e. no dm-X
was created! Identical behaviour on both machines in the cluster. No
changes whatsoever in any config file compared to the original setup
(single volume). 

Analysis revealed that the /dev/sdX block devices were still present (sdb -
sde, respectively) but all 4 devices had the same UUID! As soon as I
removed one of the volumes, the remaining volume worked just fine. We even
assigned the second volume to a different controller port, thus forcing a
different WWN for the volume, but the UUID (scsi_id -gus <device>) remained
the same. 

What can be wrong here? 

I assume dm-multipath refuses to work with what looks like the same device
(by UUID) but 2 different sizes. 

But why is the same UUID calculated (how?) for 2 apparently different
volumes? 


Strangely, another HB cluster with 3 Datacor SAN volumes has no such
problems. 


Appreciate any help :-) TIA!


-- 
Fortune's Law of the Week (this week, from Kentucky):
	No female shall appear in a bathing suit at any airport in this
State unless she is escorted by two officers or unless she is armed
with a club.  The provisions of this statute shall not apply to females
weighing less than 90 pounds nor exceeding 200 pounds, nor shall it
apply to female horses.