ServiceGuard Quorum Disk Config Problem

[Jimmy Rogers]

I was trying to configure a quorum disk on a two node cluster and seem to have run into some brick walls. The cluster had been moved to a new disk array previously, and the old quorum disk was on the old array. I apologize for not having the exact error messages, but they scrolled off my terminal after some other issues arose that day.
The package/cluster were stopped, the cmclconfig.ascii file edited with new VG and new disk info. However, when I tried a cmcheckconf, it complained that the VG needed to be activated. I tried to use vgchange -a y, but received the message that that was not the configured mode for that VG. I tried vgchange -a e, and received the message that vgchange with this option wouldn't work since ServiceGuard wasn't running. I even tried running cmcheckconf with ServiceGuard running, and got the expected results, can't run it with SG running. Any guidance here?

[Gordon Morrison]

I'm a bit rusty on ServiceGuard, but I have found this doc useful in the past:

http://docs.hp.com/en/B8467-90001/B8467-90001.pdf

If that's not what you need, go to docs.hp.com & search for "quorum"

[Stephen Doud · Accepted solution]

The cluster lock VG that you chose already has a cluster ID on it, and is marked to activate in exclusive mode.
This means that in order to update the cluster lock structure onto the disk, you have to
1) halt the cluster
2) de-cluster the VG (vgchange -c n vgname)
3) activate it (vgchange -a y vgname)
4) run cmapplyconf on the updated cluster configuration file
5) deactivate the lock VG
6) start the cluster.


-SD

[Jimmy Rogers]

Thanks for the response, it should be very helpful. Is there a step needed after step #5 to "re-cluster" the VG?

[Stephen Doud]

Step 4 - cmapplyconf, should 're-cluster' the vg.

step 5.5 could be:
5.5) Verify the cluster lock VG has been re-clustered:
# vgchange -a e
If it activates, it has been re-clustered.
If it does not activate "activation mode conflicts with current configured mode", then it has not been re-clustered.
If this is so - then do one of the following:
5.5a
- start the cluster
- vgchange -c y
- load and run cminitlock on the disk

5.5b
- with the cluster halted: cmdeleteconf -f
- vgchange -a y vglock
- do a full cmapplyconf

StephenD.