[hobbit] Monitoring processes in a Service Guard environment

[Henrik Stoerner]

On Fri, Sep 16, 2005 at 12:17:17PM -0400, Terry Barnes wrote:
> We recently changed several of our systems that are managed by Service Guard to automatic failover. They were previously a manual failover only. Now - it is impossible to predict which system in the cluster will have the critical processes active.  The Service Guard managed pseudo hostname and IP address, as well as disk, processes, etc. may be on any system in the cluster at any time. I run a client on each system in the cluster, but cannot monitor the processes effectively. 
> 
> 1. Is there a way to run a client against the Service Guard owned hostname and IP address?
> 2. How well would adding the client to the cluster work? Inother words, if the package switches to another host system in the cluster, so would the client. The actual hostname of the new system would be different and I am guessing this would cause problems.

I think 2) answers your 1). Let the client migrate along with the other
critical processes, and it shouldn't be a problem to check if those
processes are running.

The Hobbit client picks up the hostname from the "uname -n" command
command, but you can start it with a "--hostname=MY.PSEUDO.HOSTNAME"
(at least in the current snapshot) and it will use that instead.

Another solution that I am looking at - it was suggested some time ago
here on the list - is to have a process check that isn't related to any
one host, but rather goes like "process foo must run on N of the nodes
X, Y and Z". (E.g. "httpd" must be running on at least 5 out of our 10 
web servers). That is on the to-do list.


Henrik