Do I need Arbitrator??

[Edgar Canaan]

Hi,
I have a 2 node cluster plus an arbitrator running MC/Service Guard. Sometimes after they reboot the cmcld on the arbitrator hangs waiting. The arbitrator node is a K220 and the other two nodes are N Classes so we set the crontab to reboot them with a 3 minutes difference. Someone suggested me not to reboot the arbitrator or just reboot it at different time from the others.

I have worked with VCS (Veritas Cluster Server) and it does not use arbitrator for a two nodes cluster...?

Comments are welcome...Thanks.

[Sanjay_6]

Hi Edgar,

I have not used an arbitrator. I use the lock disk in a two node cluster.

In such a case i don't need another system to arbitrate between the cluster nodes.

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/B3936-90053/B3936-90053_top.html&con=/hpux/onlinedocs/B3936-90053/00/00/17-con.html&toc=/hpux/onlinedocs/B3936-90053/00/00/17-toc.html&searchterms=lock%20disk&queryid=20020523-125151

Look at the section called "Cluster Lock ".

Hope this helps.

Regds

[Duncan Edmonstone]

Edgar,

The standard method for arbitrating in a two node cluster in MC/ServiceGuard is not with an arbitrator node at all, but with a cluster lock disk. The cluster lock disk is a shared disk (both nodes in the cluster can see it) which is used when either node cannot determine the condition of the other node as a 'tie-break'.

1. How old is this cluster?
2. It's not actually Metrocluster is it?

HTH

Duncan

[Edgar Canaan]

The problem is that I cannot use common disks because the active and standby node are not at the same location...different buildings.

[Sanjay_6]

Hi Edgar,

you can look into the dual cluster lock disk configuration if you are still thinking about a cluster lock disk as tie breaker.

Hope this helps.

Regds

[Edgar Canaan]

I would like to keep my quorum server. I only need to know the reboot order of the nodes...and would like to know (if anyone knows) why you don't see this scenario with VCS..???

[melvyn burnard · Accepted solution]

The cluster lock disc or quorum server has no effect on your scenario.
ServiceGuard has a criteria that for a cluster to form automatically, ALL nodes i.e. 100% MUST be available and ready to form the cluster at the same time.
There should be no reason why you are rebooting all 3 nodes at the same time, so just leave one node up and running the cluster, and when you reboot either or both of the other nodes, they will join the runing cluster.

[Mark van Hassel]

Hi,

Since you are using an arbitrator I assume that you have a metro cluster set up that is spread over three sites usingb EMC SRDF or HP's XP disk arrays. Only in cases like that, you'll need an arbitrator. For a 'normal' two-node cluster you'll use the lock disk as a form of arbitration.
Melvyn is correct, when there is a noticeble difference in boot time, the first system to boot will time out, I belive the default time out is 600 seconds. Check the AUTO_START_TIMEOUT parameter in the Cluster configuration file.

HtH,

Mark