Re: [HTCondor-users] Conditional output file transfer

[Brian Candler <B.Candler@xxxxxxxxx>]

On Thu, May 09, 2013 at 12:17:16PM -0500, R. Kent Wenger wrote:
> For some reason when answering this last week, I forgot about "starter
> scripts".  These are scripts that are run on the execute machine.  I
> think you should be able to run a starter post script that would delete
> zero-size error files before they ever get transferred back.  (I
> haven't tried this, so I can say with total certainty that it will
> work.)

Thank you. This is potentially a cleaner solution, by putting a single
+PostCmd entry in the .sub file instead of explicitly setting a SCRIPT POST
on every individual DAG node.

> As far as I know, though, the starter post script can't find out the
> return code of the job (hmm -- we should probably add that
> feature!).

That would be useful.

Also, the starter post script needs to know the filename of the error file,
which is different for each job.  If it gets $_CONDOR_JOB_AD in its
environment then it could read and parse that file.  But it might be simpler
if you could pass macros, e.g.

+PostCmd = cleanup.sh $RETURN $(error)

Cheers,

Brian.