I've struggled with this too. parallel jobs have LIBEXEC stuffed into
the path, but vanilla an other universes are SOL.
I think putting the location of LIBEXEC into an environment variable is
a good idea.
But as I recall, environment that begins _CONDOR_xxx looks to condor
like a PARAM override and can leak from an outer condor to an inner
condor in the batlab.
-tj
On 7/11/2013 4:42 PM, Alan De Smet wrote:
As best I can tell, if I want my job to call condor_chirp, I
either need to search the execute node for it (since it won't be
in the path), or I need to bring it along with me (be sure to get
the architecture of the binary right, and bring along all 16
required HTCondor libraries!). Both of those options seem
crappy.
Am I missing some better solution?
If this really is a good as it gets, I think we should set a
_CONDOR_CHIRP environment for the job that points at the
condor_chirp binary on the execute machine. Is this a good idea?
Are there better ideas?
|
![[Computer Systems Lab]](http://www.cs.wisc.edu/pics/csl_logo.gif){kind=logo}