I have run into this in my administration of HP-UX and Tru64 where just like you, my HP-UX servers use the short names, and the Tru64 are fully qualified. What I have done is in the bb-hosts file put the entry like this:
IP server.fqdn.com # CLIENT:shortname [tests]
Ex:
1.2.3.4 foo.bar.com # CLIENT:foo !ftp ssh !telnet
Because then when the servers send their messages to hobbit, Hobbit is able to translate the messages no matter what the name is...
Hope this helps,
Adam
From: Christian, Ron [mailto:rchristian at columbia.com] Sent: Wednesday, October 26, 2005 1:11 PM To: hobbit at hswn.dk Subject: [hobbit] short vs. canonical name
We're bringing up 4.1.2 on a QA machine, and have noticed an anomaly.
Our environment is mostly Solaris 9 with a few machines running Red Hat Enterprise AS. We have Hobbit 4.0.4 running on our production administrative server and Hobbit 4.1.2 server running on our QA box. The intention was to qualify 4.1.2 and push it into the production environment if it seemed reasonable to do so.
I've run into a problem with the QA instance on Red Hat machines. I can see a couple ways to fix it but want to find out what is considered "best practices"
The Solaris node names are the short (non-canonical) form, but the node names of RHEE tend to be fully qualified. It looks like we will have more Red Hat and fewer Solaris machines in the near future, so this issue will become significant.
It looks like different parts of 4.1.2 arrive at the machine name in different ways. The name of the server is set explicitly as BBSERVERHOSTNAME in hobbitserver.cfg. It appears that this is used both for the built-in tests and server extensions.
The client, on the other hand, appears to get it's hostname in a different fashion depending on whether it's a built-in test or an extension. With the client short name in bbhosts, but the client uname defaulting to the fully qualified name, the built-in tests work but extensions do not. It appears that the built-in tests somehow figure out the client's short name, and the server recognizes it and displays the data. However, when extensions are invoked, the client name is set to the fully qualified name, the name is munged to hostname,my,domain (note commas) and the server doesn't recognize the client.
I see a few ways around this. We could set the uname on all the Linux
boxes to the short name, and remember that we have to do this for any
new installs. (I'm not sure if this will break anything else.) Or, we
could set up a hobbit alias for each Linux machine (ick...). Or, we
could change runclient.sh to use "hostname -s" instead of "uname -n"
(But only on Linux boxes.) or change the "MACHINE="echo $MACHINEDOTS | sed [...]" expression to dump the domain name instead of changing the
dots to commas.
Have others run into this, and is there an accepted solution?
RonSilence is golden. Duct tape is silver.