PVlinks + cluster lock disk - service guard

[nishith]

I have clarification in serviceguard configuration. i have LUN presented from EVA 4000 (active /active controllers) to my hp-ux servers. i used PV links (4 paths - 1 pri path & 3 alternative paths) to create a VG with that LUN.
Now during creating serviceguard config, for my FIRST_CLUSTER_LOCK_PV i have mentioned the primary path. suppose this path fails after configuration it will cause problem with cluster.
should i specify the alternative links of the same PVs or is not necessary. if necessary how can i do it?

[paolo barila · Accepted solution]

If you have secure path, you only specify the device ------>

# spmgr display

TGT/LUN Device
0/ 0 c16t0d0 <---------

Controller Path_Instance
xxxxxxxxxxxx
c10t0d1
c13t0d1

Controller Path_Instance
xxxxxxxxxxxx
c11t0d1
c12t0d1

[melvyn burnard]

You cannot use the alternate link to a device for a cluser lock disc, it does not work and is unsuported.

Unless your cluster is architected in one of two specific ways, then a dual (or second) cluster lock disc is also unsupported

[Devender Khatana]

Hi,

As we have to mention the device file of the disk and not the VG, on failure of the path the lock disk will not be accessible. In the event of any requirement to accessing the lock disk during this time will do everything what will happen if the disk is not available.

The best way will be to use secure path and it will make only one device file available from multiple paths. The second cluster disk is also not supported normally.

HTH,
Devender

[nishith]

Thanks all for your quick reply.

but in EVA 4000 usual secure path is not supported. but only autopath (now included in secure path as active/active component). thats why i had to go for PV links.

now setting FIRST_CLUSTER_LOCK_PV is becoming problem.if u have any solution please let me know.

[nishith]

Adding to my above reply:

Autopath display : ---> gives only 4 active paths. it does create a single file like in usual secure path software.

[Steven E. Protter]

Shalom nishith,

I think you'd be better off with a cluster lock disk that is not accessible through multiple pathes.

I think actually even if its a two node cluster a quorom server is a better idea.

SEP

[nishith]

Thanks steven!

but i can't go with quorum server. do u have any other solution.

[Bharat Katkar]

Hi,
Agree with SEP. That is the only wait out.

Regards,

[Mohanasundaram_1]

Hi Nishith,

The Active/Active EVAs will provide 4 paths for each LUN in a dual Fabric configuration, as you have mentioned.

So your query is indeed valid. I did a VCS 4.001 upgrade on a EVA5000 which put me in a similar situation. I have used the primary device file as the PV lock device. The host upgrade manual was mentioning about a migration tool which contained a script called "inquiry".This script is used to identify the device file for the cluster lock PV. But I could not get this migration tool for my upgrade and hence I did it manually. A response from HP guys here would be helpful.

But you need not worry about the cluster if the cluster lock disk fails. The cluster will keep functioning. the lock disk comes into picture only when a cluster arbitration is required. And then for cluster lock to fail there should be a LUN failure in EVA, which will happen only in case of 2 concurrent failures. This chance is less.

with regards,
Mohan.