Different path on two HPUX system for the same volume.

[Romeo_4]

Hello,

I have installed 2 EVA 4000 connected at two HPUX RX2620. The 2 systems must be configured with service guard. The problem:

iosca -fnC disk and autopath display commands show a different path for the same volumes.

Do You know lime assign teh same controller number at two servers?

Example of autopath display output system A

Array WWN : 5000-1FE1-5007-C5A0



==================================================================

Lun WWN : 6005-08B4-0010-3F0B-0000-E000-0011-0000

Load Balancing Policy : No Load Balancing

==================================================================

Device Path Status

==================================================================

/dev/dsk/c16t0d1 Active

/dev/dsk/c17t0d1 Active

/dev/dsk/c18t0d1 Active

/dev/dsk/c19t0d1 Active



Example autopath display of the same volum system B

Array WWN : 5000-1FE1-5007-C5A0



==================================================================

Lun WWN : 6005-08B4-0010-3F0B-0000-E000-0011-0000

Load Balancing Policy : No Load Balancing

==================================================================

Device Path Status

==================================================================

/dev/dsk/c14t0d1 Active

/dev/dsk/c15t0d1 Active

/dev/dsk/c18t0d1 Active

/dev/dsk/c19t0d1 Active

Thanks

[Jeff_Traigle]

There's no requirement that the instance numbers be the same on the nodes. (The instance number you see in ioscan is what shows up as the cX portion of the device file name.) If you want to reorder them so the device files are the same on the nodes, however, this procedure should do the trick for you.

http://www2.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000080025317

[Patrick Wallek]

The paths don't HAVE to be the same. That's the beauty of LVM.

Just create your VG on one of the nodes, then run a 'vgexport -p -m /var/tmp/mapfile -v -s vg??' (where vg?? is the vgname) on the first machine. Copy the /var/tmp/mapfile to the 2nd machine.

Then on the 2nd machine, create your /dev/vg?? directory and the /dev/vg??/group file and then run a 'vgimport -m /var/tmp/mapfile -v -s vg??'. This will import the VG on the 2nd machine from teh appropriate disks.

No need for identical device names.

[Abhijit P.]

Hi...


Thats right you don't need to have same path on both the servers.
Just create the VG and while exporting it export it with shared mode on and with a map file and then import it on the second node of the cluster.

Regards,
Abhijit

[John Bigg]

All the replies you have had so far are correct. The instance numbers and therefore the device file names do not need to be the same across different nodes in the cluster although personally I prefer this for the sake of managability when you have 2 systems which are essentially the same.

In this case the solution is to use ioinit to re-map the instance numbers. You can create a text file where you specify the hardware path class and instance number you reqiure and pass this to ioinit to allow you to remap these so they will be the same on all systems. See ioinit(1M).

However, you do not have to do this. There is no problems in having your volumes using different device files on different nodes in the cluster as long as you are careful and are aware they are different.

[Cesarato]

thank you,

I have used thi procedure to rescan every devices:

disconnected all fibre cables

mv /etc/ioconfig /etc/ioconfig.save
mv /stand/ioconfig /stand/ioconfig.save
shutdown -ry 0 (da console)


At reboot this message:

/sbin/ioinitrc /etc/ioconfig is missing. Restore it frombackup or invoke /sbin/ioinit -c to recreate it from the kernel. When finished type ^d to continue boot.

The follow command has been typed

(in ioinitrc)# ioinit -c ^d


After reboot:

ioscan
insf -e

For future use :

# cp /etc/ioconfig /etc/ioconfig.clean
# cp /stand/ioconfig /stand/ioconfig.clean

Riconnected all fibre cables

ioscan
insf -e
shutdown -ry 0

for existent Vg:


vgimport -p -s -m vgxx.map vgxx

[Kent Ostby]

Actually, since they are on a shared bus, you can't really have the address be the same. In fact, its a common PROBLEM on new SG installations.

If you have two cards with address 7 on the same bus, you'll get SCSI errors and system crashes.

[Cube Over]

What if I need to set instance number of a HBA controller to another value, persistently? How can it be done?

[nanan]

For EX
to make 8/0 SCSI card have an instance of 2, and the 8/8 SCSI card have an instance 0,

create a file called /tmp/mappings which contains;

8/0 ext_bus 2
8/4 ext_bus 1
8/8 ext_bus 0

then excute ioinit -f /tmp/mappings -r and the system reboot with new mapping..