Re: [Condor-users] Fwd: A few basic questions and a modest suggestion

[Todd Tannenbaum <tannenba@xxxxxxxxxxx>]

At 09:34 AM 4/18/2006, Darin Kalisak wrote:

> First, I'm a little bit confused about the roles of the condor_config
> and condor_config.local files.  I have set up a 3-node prototype pool,
> and only the central manager has data in its condor_config_local file.
>  My reading about it in the manual (version 6.6.9) leads me to believe
> that it should be the file with machine-specific settings for all
> machines.  The condor_config file, on the other hand, is different on
> each machine, and seems to be the file that actually contains
> machine-specific info.  In particular, this is the file with the START
> variable, which I understand to hold that machine's classAd.  I'm not
> clear how the two files are supposed to complement each other and work
> together.  In fact, I don't see any use for the .local file at all.
> Any insight would be appreciated.

Many sites keep "condor_config" on a shared filesystem, and have all 
machines in their pool use this same file.  This makes changes to the 
entire pool's configuration easy.  Then they keep a 
condor_config.local, which is "local" to each machine in the 
pool.  This is where changes that are specific to one machine would 
go -- for instance, if the owner of that one machine has a different 
policy than the rest of the machines in the pool.

> Second, I keep seeing comments about pool-wide configuration settings,
> but I can't figure out where these would be stored, or how they would
> be used.  Is there some master config file (presumably on the central
> manager) that I'm unaware of that sets some basic parameters for use
> by all the machines?  I certainly don't see one anywhere.

See comments above.  When people say a pool-wide configuration, they 
usually just mean a condor_config file that is common to all machines 
in the pool (usually accomplished by placing it on a shared filesystem).

> Third, as I understand it, I can declare any variable I like in
> condor_config files, and use them as I see fit in job or machine class
> ads.  So, I should use my own variables to determine which machines
> have third-party software (say Matlab or Fidap).  Any machines without
> the variable explicity defined will return "undefined" as a value for
> it when prompted.  (And I should use the =?= and =!= operators to deal
> with this possibility.)  Are these fair statments?

Yes.  Even better, you could have the variables declare *where* these 
packages are installed, if it varies from machines to machine.  Then 
this information could be used by the job via the $$() macro in your 
submit file.

regards,
Todd


-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Todd Tannenbaum                       University of Wisconsin-Madison
Condor Project Research               Department of Computer Sciences
tannenba@xxxxxxxxxxx                  1210 W. Dayton St. Rm #4257
http://www.cs.wisc.edu/~tannenba      Madison, WI 53706-1685
Phone: (608) 263-7132  FAX: (608) 262-9777