How change HP Serviceguard Secondary node to primary

[Larry Leone]

I learned much from a session initiated by Asif on Apr 8, 2010 12:16:46 GMT.

I have a variation on that theme.

My primary node is down and unable to boot.

When I try to make the other node the primary I get the following error:

ads02> cmcheckconf -v -P node.conf
Checking existing configuration ... Done
Gathering configuration information ... Done
Parsing package file: node.conf.
cmcheckconf : Error found in package file: node.conf.
Package irqads already exists. It will be modified.
Error: Unable to communicate with node ads01

I am changing the primary to ads02 because ads01 is down.

Is there a way around this?

[melvyn burnard]

remove the entry for the down node, or comment it out, from the package configuration file.
then try to apply the change

[Larry Leone]

Taking the node name out of the conf file didn't work. cmcheckconf still complains about ads01.

I did cmqueyrcl and see the node is returned from the query. From there I noticed that the node name is in the binary cmclustert file in /etc/cmcluster. I moved that file to see what would happen and when I tried to cmcheckconf I get Checking existing configuration ... Done
Warning: Unable to get local cluster configuration.
cmcheckconf : Cluster is not currently configured. -C option is required.

What does "-C option required" mean? What is the software asking me to do?

[melvyn burnard]

Ok, so it still sees the "failed" node in the binary.
Now you need to edit your cluster ascii file, and comment out the entires that are for that node, an dredo the cmapplyconf this time using the -C option and specifying the cluster ascii file

[Larry Leone]

More specifically, is the command:
cmapplyconf -v -C ads.conf


Will that create a new binary?

[melvyn burnard]

yes it will, or if you put the origionsl binsry bsck, it will simply modify it (the preferred way)

[Larry Leone]

"cmapplyconf -v -C irqads.conf"? command not found

[Dennis Handly]

>I learned much from a session initiated by Asif

That previous thread:
http://forums.itrc.hp.com/service/forums/questionanswer.do?threadId=1420406

[Larry Leone]

Thank you for this latest response. I went back to the link pasted into message, but I didn't see at the link mention of the problem I have when I use the -C option. When I use the -C option, I get a "command not found" error. Can someone tell me if I can correct that error and how?

[Larry Leone]

I am curious, what if I put things back the way they were and change the secondary server IP address and node name to that of the primary server? In essence ADS2 becomes ADS1. Would the cluster start because the primary node is available, or would the software complain it can't find the secondary server?

It this sounds like a viable solution, can you tell me how to change the IP address and node name on an HP-UX box?

Thanks for your help so far.

[melvyn burnard]

This would NOT work.
What are you ACTUALLY trying to achieve here? Would it not be simpler just to run the package on the current running node, rather than try to fiddle with the configuration?
I assume the second node is running as a cluster?
If not, restore your cmclconfig binary file to /etc/cmcluster, then start the cluster with just the one node using:
cmruncl -v -n ads02

[Larry Leone]

Thank you. Our intention is to bring up the second of two nodes on a cluster where the primary node is out of commission. We will give the command you provided a try.

[Stephen Doud]

No need to cmcheckconf/cmapplyconf (which will fail as long as the down-node is a member of the cluster). If you simply want to start the cluster and package on a one node cluster, do this:

# cmruncl -n ads02 <--- will start a one-node cluster on ads02

If the package does not start, use
# cmmodpkg -e -n ads02
# cmmodpkg -e

These should start the package on the ads02 node if that node was ever an adoptive node of the package.

[Larry Leone]

New problem. The cmcluster binary file is 0 bytes. How do we create a new cmcluster binary file? We've tried the -C switch before, but get a "command not found" error. Can I recreate the cmcluster binary using the existing cluster configurtion files?

[Stephen Doud]

/etc/cmcluster/cmclconfig is created and distributed to all nodes listed in the cluster configuration ASCII file when cmapplyconf is performed (and succeeds) on the file. If the node does not have a cmclconfig file, either it was never a member of the cluster or someone removed the file. If it was never a member of the cluster and you want to add it to a cluster where it's partner node is unreachable, you will not be able to do so. In this scenario, you might consider creating a one-node cluster for now and add packages to the cluster as needed. When the disconnected node returns to runnable status, dissolve this one-node cluster and add the node to the other cluster.

If the node was a member of the cluster but someone deleted cmclconfig, copy if from the other node or an archive. You can use strings on the binary file to ascertain node membership.

[Larry Leone]

OK... trying to create the 1node cluster.

cmapplyconf -v -C adscluster.conf

Error: node ads02 is refusing Serviceguard communication.....
cmapplyconf: Failed to gather configuration information

If I try to start cmclconfd by hand, I get an error that something is already bound to port5302

[Stephen Doud]

You cannot cmapplyconf a cluster configuration file that contains nodes that are not reachable.

To create a one-node cluster configuration file, use
# cmquerycl -C -n
Edit the output file to customize it as needed, then cmapplyconf it.

This assumes that no other system on the network contains a reference to the one-node cluster nodename in it's cluster binary file.

[Larry Leone]

Will cmapplyconf succeed if we are unable to have cmconfd (I think that was it -- it is the cluster management configuration daemon) run?

[Stephen Doud]

Do you mean cmclconfd? This daemon must be able to run in order to service any Serviceguard commands. The daemon is spawned through inetd (see 'hacl' lines in /etc/services and /etc/inetd.conf, and also enable identd a.k.a 'auth' in these files) when a Serviceguard command is executed and when hostname resolution needs of Serviceguard are met (see the Managing Serviceguard manual for details).

What is the current error message that you get, and any related syslog.log messages?

[melvyn burnard]

From your response from 17th September:
I did cmqueyrcl and see the node is returned from the query. From there I noticed that the node name is in the binary cmclustert file in /etc/cmcluster. I moved that file to see what would happen

Have you restored the cmclconfig binary you moved back into /etc/cmcluster?
If so, the cmruncl -v -n ads02 should run

[Larry Leone]

It turns out that the binary cluster file is corrupt [0 bytes]. I think that the commands would have worked if we had a good data file.

[Larry Leone]

I would like to assign points and understand the point system. How do I assign points?

[Dennis Handly]

>I would like to assign points and understand the point system. How do I assign points?

The FAQ has more info about it:
http://forums.itrc.hp.com/service/forums/helptips.do?#33
http://forums.itrc.hp.com/service/forums/helptips.do?#28