MCSG and Powerpath

[Baz Sollitt]

Morning All,

I have a 2-node cluster and we are changing from edge switches to directors for our SAN connection.

This means the controller number of the lock disks will change.

I appreciate that the cluster will need to be taken down and the cluster rebuilt if the lock disk is changed, but has anyone had any experience with the lock disk under powerpath control ?

I am sure that Powerpath will still handle i/o to the lock disk even though the ascii file will be wrong, the cluster should still function correctly ?

The cluster can then be taken offline in the future to update with the new lock disk name, as long as it is managed properly and not forgotten.

Any advice would be much appreciated.

Regards

[Stephen Doud · Accepted solution]

The cluster binary file (/etc/cmcluster/cmclconfig) contains linkage pointing to the cluster lock disk, from each node in the cluster. To see which disk paths Serviceguard will expect to see the cluster lock disk path on, use the following command:

# cmviewconf | grep -e "Node name" -e "lock pv"

Example output:
Node name: helix
first lock pv name: /dev/dsk/c4t0d0
Node name: torus
first lock pv name: /dev/dsk/c4t0d0

In this above case, the path to the lock disk from each nodes' point of view is the same special file name. It is not always the case.

If changing your SAN switched network causes the disk path to change, the cluster binary file must be updated to allow Serviceguard to continue to test and validate the lock disk on an hourly basis. If the lock disk is not accessible because the cluster binary filename is incorrect, both nodes could TOC (reboot) if the lock disk is needed in order to negotiate a split-brain situation.

Edit the cluster configuration ASCII file with the correct disk paths (after the SAN switch change) to point at the correct path to the lock disk. Then with the cluster down, perform the following:
# cmapplyconf -f -C
Use cmviewconf to confirm the modification.
Use cmruncl to restart the cluster.

[nanan]

Hi! Baz
If your lock device name which defined in cluster config binary is changed. you have to reflect the change devices in cluster binary, as you mantioned recompiling the ascii file during halted cluster.

But i am wondering what are you meaning

"I am sure that Powerpath will still handle i/o to the lock disk even though the ascii file will be wrong, the cluster should still function correctly ?"

Could you explan more detail ?
Do you mean even though you changed san switch and still the lock device name not changed or maintain the same device name with powerpath function?


Regards
nanan

[IT_2007]

Since you are moving from switches to directors, device file will be changed for lock disk. You need to recompile cluster to generate new binaries.

Cluster lock volume group is not active during cluster running, you don't have to worry about powerpath control. It will still show multipaths for cluster lock disk. Make sure that you noted down correct lock disk LUN number.

[Baz Sollitt]

Hi Nanan,

What I am meaning is although the lock disk device file is for example;

c56t5d0

Powerpath knows the alternate path as c57t5d0, so i/o can still travel to the correct physical disk with powerpath control, even though the cluster config says the lock disk is c56t5d0.

Couldn't I also use a symbolic link to the from the old the new device file ?

I am sure it isn't recommended, but it would still work right ?

[nanan]

It seems possible as you say if you link the device name.
But it is not normal case, isn't it?
so most admin might not go through such experiences.

But you can test that without any hurt in your system, as you know the lock device is for quorum, If your plan is not working, you can get mesages in syslog.log file hourly; cluster check the lock device periodically.

Even i agree with your plan, i am not sure
so if you got the result, post that


Regards
nanan

[IT_2007]

Baz,

You mean that powerpath doesn't see both paths for lock disk? If so, then you may be in trouble because I had same experience before when device files got changed. Only way to check them is no body got that device name (if you don't want to change cluster device name in config file).
use command something like this.. don't remember exactly.
powermt display dev=clusterlock every=5

which will keep on showing IO activity for it.