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

Re: [Condor-users] Args not found (condor-c)



Ryan,

I have set up a test of the case you mention: schedd-to-schedd submission of a job from 6.7.18 --> 6.7.18 --> 6.7.14. I find that the 6.7.18 --> 6.7.14 hop correctly produces a job with "Args" rather than "Arguments", because the 6.7.18 condor_c-gahp knows that 6.7.14 doesn't understand the new argument syntax. I am baffled as to why this is not happening in your environment.

Can you please check that the CGAHPWorkerLog for the middle schedd is reporting that it is truly version 6.7.18? And also that the final schedd log truly reports that it is 6.7.14?

Sorry for the trouble.  We _must_ get to the bottom of this!

--Dan

Ryan Garver wrote:

I am having trouble with backwards compatibility with JobAds being
forwarded through Condor-C.  When the job reaches the destination
startd and a starter is spawned to handle it the starter promptly
fails with the error:

5/4 00:45:56 Starting a VANILLA universe job with ID: 2419.0
5/4 00:45:56 Args not found in JobAd.  Aborting OsProc::StartJob.
5/4 00:45:56 Failed to start job, exiting
5/4 00:45:56 ShutdownFast all jobs.
5/4 00:45:56 **** condor_starter (condor_STARTER) EXITING WITH STATUS 0

The job is then marked back as idle and retried in the negotiation cycle.

The job looks like:
universe = grid
executable = pi-compute
arguments = 5000000
output = out.$(Process)
log = log.$(Process)
grid_resource = $$(Resource)
remote_grid_resource = $$(RemoteResource)
remote_universe = grid
+remote_remote_jobuniverse = 5
queue 1

In the first queue (6.7.18):
Args = "5000000"

To a 6.7.18 queue via condor-c:
Arguments = "5000000"
Environment = ""

Then on to 6.7.14 via condor-c:
Arguments = "5000000"
Environment = ""

Any help is appreciated.