multipath + evms problems

[Bernd Zeimetz <bernd@bzed.de>]

Hi,

we have some trouble with runnign an EMC CX400 with multipath and evms.

If we run the machine with systemtype 0x16 (it identifies itself as FSC FC66 
then), Arraycommpath = 0 and Failovermode = 2, failover seems to work well 
(not much tested), and evms uses the dm/multipath devices. The bad thing on 
this way is a trespass of the LUNs every few seconds, which is not the way it 
should work.

Setting the machine to something useful - systemtype=3  (==DGC), 
arraycommpath=1, failovermode=1 - it is recognized by multipath-tools and 
dm-emc as emc-machine. With

zeus-1:/# cat /etc/multipath.conf

defaults {
        path_grouping_policy                    failover
        no_path_retry                           queue
}

multipaths {
        multipath {
                wwid            3600601f2a20c0000fa13caa656b0d711
                alias           LUN_0
        }
[....]
        multipath {
                wwid            3600601f4a20c000099dc229956b0d711
                alias           LUN_7
        }
}


we have some nice devices:
zeus-1:/# ll /dev/mapper/
total 0
brw-------  1 root root 254,  0 Oct 29 12:09 LUN_0
brw-------  1 root root 254,  4 Oct 29 12:09 LUN_0p1
[....]
brw-------  1 root root 254,  2 Oct 29 12:09 LUN_7
brw-------  1 root root 254, 13 Oct 29 12:09 LUN_7p1
crw-rw----  1 root root  10, 63 Oct 29 12:08 control
zeus-1:/#


but evms does not use them, it directly accesses and uses sd*
zeus-1:/# ll /dev/evms
[...]
drwxr-xr-x  2 root root       60 Oct 29 12:09 dm
brw-r-----  1 root disk   8,   0 Oct 29 12:09 sda
brw-r-----  1 root disk   8,  16 Oct 29 12:09 sdb
brw-r-----  1 root disk   8,  32 Oct 29 12:09 sdc
brw-r-----  1 root disk   8,  48 Oct 29 12:09 sdd
brw-r-----  1 root disk   8,  64 Oct 29 12:09 sde
brw-r-----  1 root disk   8,  80 Oct 29 12:09 sdf
brw-r-----  1 root disk   8,  96 Oct 29 12:09 sdg
brw-r-----  1 root disk   8, 112 Oct 29 12:09 sdh

Of course this results in a not working failover - this should work without 
using multipath at all.
After excluding the sd* devices in evms.conf it doesn't find any evms-devices 
at all.


We also have a ton of
multipathd: uevent trigger error
messages in the daemon's log.

Operating system is Debian sarge, but with kernel 2.6.14, the actual git 
version of the multipath-tools, evms 2.5.3, udev 0.070, libdevmapper/dmsetup 
1.01.04.


Since this behaviour seems to depend on the use of the emc-hwhandler/dm-emc 
I'm posting to this list.
Hope you have any ideas how to fix it, or do we run in an evms-only bug here?



Greetings from Darmstadt, Germany


Bernd Zeimetz
Darmstadt University of Technology