dev name of Cluster Lock Disk changes

[mmfpbw]

Hello Experts,

I'm running a cluster of two DL380 Servers. The shared storage consists of two MSA1000 storages which are connected to each server. I'd like to build a RAID1 with these two storages. And I'd like one of the storage to hold the Cluster Lock Lun (CLL). Each storage has its own controller on the server.

The Software I'm using is SLES9 SP3 and MCSGLX 11.16.05.

The problem:

If the servers start the LUNs of the storage are numbered from /dev/sda to /dev/sdo. I've set /dev/sda1 to be the Cluster-Lock-Lun. This worked actually fine. Now I've move one server into another room (I've lengthened the connections by fibre channel). As the servers restarted the partition that appeared to be /dev/sda is now /dev/sdh. I've used scsidev to put an alias on this disk which worked fine. The name of this CLL is now /dev/scsi/locklun and I can verify that the disk which can be accessed as /dev/sda1 and /dev/scsi/locklun on one node is as well /dev/sdh1 and /dev/scsi/locklun on the other node. But if I try to use this name /dev/scsi/locklun an error message appears:

Begin cluster verification...
Error: Cannot get disk partition for device /dev/scsi/locklunError: Invalid data for cluster lock LUN configuration
cmcheckconf : Unable to verify cluster file: /opt/cmcluster/conf/cmclconfig.ascii.


Has anybody expirienced this phanomenon ?
Is there another way to change the way the disks are numbered ?
How can I make SG to use the alias of the CLL ?

gromit:~ # sfdisk -s
/dev/sda: 10485600
/dev/sdb: 10485600
/dev/sdc: 10485600
/dev/sdd: 10485600
/dev/sde: 10485600
/dev/sdf: 10485600
/dev/sdg: 7862160
/dev/sdh: 16320 <-- CLL
/dev/sdi: 10485600
/dev/sdj: 10485600
/dev/sdk: 10485600
/dev/sdl: 10485600
/dev/sdm: 10485600
/dev/sdn: 10485600
/dev/sdo: 7862160

gromit:~ # sfdisk -s /dev/sdh1
16304
gromit:~ # sfdisk -s /dev/scsi/locklun
16304



wallace:~ # sfdisk -s
/dev/sda: 16320 <-- CLL
/dev/sdb: 10485600
/dev/sdc: 10485600
/dev/sdd: 10485600
/dev/sde: 10485600
/dev/sdf: 10485600
/dev/sdg: 10485600
/dev/sdh: 7862160
/dev/sdi: 10485600
/dev/sdj: 10485600
/dev/sdk: 10485600
/dev/sdl: 10485600
/dev/sdm: 10485600
/dev/sdn: 10485600
/dev/sdo: 7862160

wallace:~ # sfdisk -s /dev/sda1
16304
wallace:~ # sfdisk -s /dev/scsi/locklun
16304

[Steven E. Protter]

Shalom,

When the device changes, you need to cmcheckconf and then cmapplyconf

Have you done this since the disk configuration change?

SEP

[Serviceguard for Linux]

I don't have the full write-up ready - but you should be able to use udev to create an alias name. The name still needs to be of the form /dev/sdxxx as the Lock Lun code looks for that name format.

[mmfpbw]

Hello,

the error message that's in the text is the output of the cmcheckconf command.

I'll try the udev method. I'll let you know the results.

Thanks & Regards
Matthias

[mmfpbw]

Hello,

the assignement of a disk by udev solved the problem.

I have installed udev on both servers and setup the boot script for it. The exercise is to modify the file
/etc/udev/udev.rules.

I needed to add the line

BUS=="scsi", SYSFS{vendor}=="COMPAQ ", SYSFS{model}=="MSA1000 VOLUME ", SYSFS{size}=="32608", NAME="sdz%n\
"

to the mentioned file on all cluster nodes. The used CLL is /dev/sdz1.


Thanks & regards
Matthias