Serviceguard 11.19 on 2 Itanium 11iv3 nodes - quorum

[Tom Haddad]

I have 2 nodes right now in my cluster. Another host is the quorum server. The quorum server was taken down to resolve another issue on it (its still down). The cluster and nodes are up and running!
Only node 1 has a package running. Im want to add a configure a new package on second node.
With the cluster up, etc, when i run checkconf the new package, will it complain about the quorum server being down?? Will it allow me to proceed and 'apply' my package?

[melvyn burnard]

well if you are only adding a package, it should work. It is when you go to modify thye cluster itself that there may be an issue.
If all else fails, why not quickly install the QS software on another server, set it up, and add the origional QS ip address using ifconfig lanx:1 and verify the cluster can see this is now the qs?

[Rita C Workman]

Yes you can do this. You can make any online changes without the quorum server running.
The only time you will have a problem is if the cluster goes 'down', then it will require the Quorum server to come backup. Or as Melvyn suggests, you could set up another device as a quorum server.

Rgrds,
Rita

[Emil Velez]

IF you have LVM volume groups being used by packages why dont you configure a lock disk or a locklun if you have a unused lun.

A lock disk costs nothing if you have LVM volume groups.

A 2 node cluster is risky if your quorum server is down for a long time.

[John Bigg]

Rita, a quorum server is not required to start up a cluster. Cluster lock is only required when arbitration is required after a failure. It is never required when starting things up.

[Rita C Workman]

John,

The Quorum server is what provides arbitration when bringing a cluster up, be it as a result of first turning it on or bringing it back up after a failure. At least that is how I read the following from the QS ver 4 document page 9:

"If the Quorum Server is not available or reachable, it will not adversely affect any clusters using
it, unless a cluster needs to reform and requires the Quorum Serverâ s arbitration to do so.
As of Serviceguard A.11.19, you can change from one quorum server to another, or to or from
another quorum method, while the cluster is running."

Since Tom is at version 11.19, he could make the change to a different QS, as was mentioned, and he can make changes to his running cluster. But if he goes down, then he needs QS to get back up.

Hope this helps,
Rita

[Tom Haddad]

ok..
1. So i can shutdown (not a failure) one of the nodes and the cluster will remain up? (should!!)

2. I read the 11.19 guide but didnt see where you can change to another quorum server online??!!

3. We are looking to implement Quorum Server on a Linux host. I can halt everything and reconfigure (cmquerycl -q quorumserver....) using a new quorum name.

[Rita C Workman]

You can put it on Linux, or even on a simple workstation. Check and see if QS is already loaded on something close by already.

Rgrds,
Rita

[melvyn burnard]

Rita
"The Quorum server is what provides arbitration when bringing a cluster up, be it as a result of first turning it on or bringing it back up after a failure. At least that is how I read the following from the QS ver 4 document page 9: "

when a cluster is first started by doing a cmruncl, NO arbitration "device" (be it Cluster Lock Disk, LockLun or Quorum Server)is required.
The ONLY time an arbitration device is required is when a FAILURE is experienced, and this results in there being EXACTLY 50% left to try to re-form as a cluster.

This is a good document to refer to:
http://bizsupport1.austin.hp.com/bc/docs/support/SupportManual/c02056095/c02056095.pdf

[Ismail Azad]

Hi,

Hmmmm, yes the cluster lock and quorum server are the two methodologies that help in "cluster reformation" when one of the nodes fail. The cmcld always checks for 100% node attendance and if heartbeat messages are not being received within node_timeout then communication is lost between the nodes which divides the system into 2 "seperate subclusters" (in case of a 2 node cluster). And as of my knowledge this is the most momentous requirement for a "cluster lock" or a "quorum server" as it helps in "cluster reformation". Only difference being the quorum server is outside the serviceguard environment for the cluster we are concerned with. "Tie breaker" is one of the terms used when you talk about this concept but few HPUX administrators don't like that term. :)

Regards
Ismail Azad