[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] trouble get started



From: Lee Gramling <leegramling@xxxxxxxxx>
Date: 02/18/2016 08:58 AM
 
> Yes, I used trusty and all installed and is running well now.

> I was originally hoping to use the deb install so my IT guy could install
> them on each machine, which I think is a more manageable solution.

> Now I just have to learn how to write the condor_config files for each machine. :)

One nice thing about the HTCondor configuration mechanisms is that you often don't need to write a separate config file for each system.

My current 8.4.4 systems have the unmodified /etc/condor/condor_config file - the only thing I do on the local machine is add an identical /etc/condor/condor_config.local file which sets CONFIG_DIR to /home/condor/config/config.d and LOCAL_CONFIG to /home/condor/config/host.d/condor_config.$(HOSTNAME)

In the host.d directory, I have two files, one for the central manager role and the other for the exec node role:

In cm_role.conf:
        CENTRAL_MANAGER_ROLE = True

and in ded_exec.conf:
        DEDICATED_EXEC_ROLE = True

Then the condor_config.hostname file is merely a symlink to the appropriate role file.

Next, I have all the customizations in the /home/condor/config/config.d directory. The "001roles" file in there sets the DAEMON_LIST based on the role value:

if CENTRAL_MANAGER_ROLE:
    DAEMON_LIST = MANAGER, NEGOTIATOR, COLLECTOR, SCHEDD
else:
    DAEMON_LIST = MANAGER, STARTD

For a pool where I have opportunistic machines, I'd have a desktop_role.config file, and set the daemon list to include the KBDD in the 001roles config file, as well as modifying its start _expression_ in a different config file pertaining to opportunistic machine configuration details.

With this approach every single machine uses the exact same configuration files with the sole exception of that one-line role-definition file which sets the value used in the configuration's "if" statements elsewhere.

If I want to tinker with one machine, I can drain it and put whatever expressions I want in the /etc/condor/condor_config.local file at the end, or create a new role file and change the symlink for just that host in the /home/condor/config/host.d directory.

The caveat is that NFS has to be up and running for the machine to be able to reach the config files before HTCondor starts up, but if you make sure that it starts in runlevel 3 you should be fine there.

Kudos to the CHTC team - the new config design that enables this approach is a very big improvement over the requirements of 7.8 where I started.

        -Michael Pelletier
_