Packages can be run on node 1 but quickly switch to node 2

[Andrew Kaplan]

Hi there --

We have a two node cluster, and while all packages can be run on node 1, they quickly switch to node 2. Why is this occurring, and how can I correct it? Thanks.

[Patrick Wallek · Accepted solution]

>>Why is this occurring,
Without a whole lot more information, it could be any one, or more, of 1,000 different things.

>>and how can I correct it?
We can't, without first figuring out what the problem is.

Have you looked at /var/adm/syslog/syslog.log to see if anything is logged there?

What about your package log files in /etc/cmcluster/?

[A. Clay Stephenson]

Could you answer your question with the data provided? We could use "The Force" to diagnose your problem but I suspect that examining the package logs would prove more fruitful. Make sure that you examine the logs on both nodes. You should also look on the syslogs on both nodes. This might be a problem with your package monitoring but that is using "The Force".

[Andrew Kaplan]

I checked the syslog.log file and I came across the following messages:

Package cannot run on this node because switching has been disabled for this node.
...
nfs.monitor terminated due to an exit(1)

Sorry about not including this...my bad.

[Patrick Wallek]

It sounds like you need to enable package switching on that node then. The cmmodpkg command can do this.

# cmmodpkg -n nodename -e pkgname

# man cmmodpkg
for more information and examples.

Check the status of your cluster with:

#cmviewcl -v

[Steven E. Protter]

Shalom,

I would surmise that some error is occurring when you try to run the package on node1.

Check the /var/adm/syslog/syslog.log file on node1 and node2 for clues. This is the default logging location for SG, but you can change that.

Check the package script configuration, they may be set up to go to node2 by default as a preferred node.

This is what the force and my brief conversation with Yoda brought to mind.

SEP

[A. Clay Stephenson]

The first error message can be addressed with a cmmodpkg -e package_name. When run w/o specifying a node, the package is enabled globally.

The second error is almost certainly your problem and it appears that "The Force" was strong with me. It appears that as soon as the monitoring begins the package switches.

Since this is NFS related, you may find that purchasing the NFS Toolkit for SG is money well spent. You can certainly do it yourself but someone else has almost certainly run NFS through more combinations and permutations of NFS failure that you will likely ever encounter.

[Sundar_7]

Could be because nfs.monitor (most probably configured as a SERVICE) is failing.

How are the NFS services started in the system ? Are the NFS daemons started through the package control script ?

May be you want to post the contents of the package control log file.

[Andrew Kaplan]

The nfs daemon, although running, was having problems with RPC. I restarted nfs, and that appears to have solved the problem.