Questions about coordinator node in a cluster

[José Enrique González]

Hi:

I'am still some newbie in HP clusters, and I must install patches on a node of a production cluster. I've read about the coordinator node in manuals, but it is not explained so far. Thanks in advance if you could explain me this:

1.- How is the coordinator node selected for a cluster? Is it configurable? Does it works the same for OPS and non-OPS clusters?

2.- How could I see which is the current coordinator node of a running cluster? "cmviewcl" command doesn't show anything about this.

3.- How could I transfer this function to another node for mainteinance of a node (if manually transfer needed)? Is it automatic? What could happen if I halt the coordinator node?

Of course, if any of you know about where to find this documented in detail, never better. Thanks a lot (plus one) for all your comments.

José Enrique
Caracas, venezue

[Navin Bhat_2]

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/B3936-90065/B3936-90065_top.html&con=/hpux/onlinedocs/B3936-90065/00/00/17-con.html&toc=/hpux/onlinedocs/B3936-90065/00/00/17-toc.html&searchterms=node%7ccoordinator&queryid=20040512-120358

or go to docs.hp.com and look for "Managing MC/ServiceGuard" and you should find the releveant information there.

[Navin Bhat_2]

with regards to OPS this might be useful.

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/B5158-90044/B5158-90044_top.html&con=/hpux/onlinedocs/B5158-90044/00/00/19-con.html&toc=/hpux/onlinedocs/B5158-90044/00/00/19-toc.html&searchterms=cluster%20coordinator&queryid=20040512-121138

[José Enrique González]

Navin, I had seen both of them before (in fact, they're just the same) but they are useless for me as they don't explain this matter in detail. Anyway, thanks for posting your reply

[Sanjay_6]

Hi,

It is not the co-ordinator node, but the configuration node.

Configuration node is just a concept and not actually a physical reality.

What happens is that in a multinode situation, to keeps things straight, you should do all your changes on one single node and then replicate the changes across other nodes within the cluster. Any nodes that is chosen for such an objective can be called the configuration node.

say at the time of cluster buildup, node1 is the configuration node. Sometime in the future if some changes are required to be done to the cluster / package, you can do all theses changes on node2 and replicate the changes across the other nodes. In this situation the node2 is identified as the configuration node.

Hope this helps.

Regds

[Ashwani Kashyap]

I think I am confused too with this term . I never came across this term so far . But this is what I think it is

1.This is the node where you start the cluster from . This would be the node where you did all your lvm confiuration , cluster confihuration , chekconf ,applyconf etc etc from . No , it is dynamic and could be done from any node in the cluster . Yest it works the same for PS and non OPS clusters .


2. THe current coordiantor will be the node where the clusterwas started from . This could be found out by viewing the syslog files .

3. This is dyanmic , I think . Because you shutdown one node in the cluster , the cluster keep running on the other node and that node is now the only active member , hence the coordinator . THe package manager , the network manager and the cluster manager now runs on the alternate node so that should be it .

I think if you need to do maintenance then just failover your packages to the adoptive nodes and then take the node out of the cluster and do your maintenance . Everything else is transparent .

[Kent Ostby]

Navin --

I don't have all your answers, but I have a few.

1) It does work the same for OPS and non-OPS clusters. I beleive by default, the coordinator will be the first node listed in the ascii cluster file .. changing the order there should change your co-ordinator.

Another idea which is rather manual and which I'm not sure would work either would be to cmruncl -n then cmrunnode . I think this would make node1 your co-ordinator but I'm not sure.

2) Look in syslog.log and whichever node has the lowest node ID is the co-ordinator

3) If you cmhaltnode on the machine you are shutting down (and you have cluster lock disk in place as you should for a two node cluster) then this will happen automatically.

Best regards,

Kent M. Ostby

[Navin Bhat_2]

Hi Kent,
The questions were not mine but anyway, I think Jose is looking for a command to view the clsuter coordinator. I know you can use syslog but then he could also use Clusterview software to get that information. But there is no command to see who the cluster coordinator is.

The co-ordinator node also has responsibility of being a package co-ordinator and also services snmp requests through cmsnmpd and it handles the inbound and outbound requests for all the nodes in the cluster.

I dont think you can do an active switching of the coordinator function when both systems are live.

[Carsten Krege · Accepted solution]

Because there were so many wrong or misledading replies, I try to explain the concept of the cluster coordinator.

First of all each node in the cluster can be the cluster coordinator. It is elected during a cluster reformation by all nodes, e.g. during cmruncl, cmrunnode, if a node fails etc.. The user does not have any influence which node is the coordinator node and doesn't need to care about it at all. This is all handled internally in Serviceguard.

The coordinator is responsible to determine the cluster membership and to distribute this info to all active nodes. It also manages the heartbeat protocol, ie each node has a heartbeat connection open to the coordinator.

The current coordinator can be determined by checking syslog.log for the message:

May 3 14:46:01 grcdg319 cmcld: The new active cluster membership is: grcdg319(i
d=1), hprtdd32(id=2)

The first node in this list is the current cluster coordinator, here grcdg319. After cmruncl it is more likely that the node with the lowest node id (here 1) will become the coordinator. This is only of interest in special error scenarios for support, but the user doesn't need to care about this.

Your questions:

1) The coordinator is elected during cluster reformation by all nodes. The election protocol of SG and SGOPS (now named SGeRAC) is the same, because SG and SGOPS are basically the same in its base functionality.

2) See the syslog message above.

3) You don't need to care. If the coordinator leaves the cluster (cmhaltnode, failure etc), another node will become the coordinator. Serviceguard's election protocol will take care that one node will become the coordinator.

Carsten

[José Enrique González]

Thank you very much, Carsten. That's exactly what I was looking for.