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

Re: [Condor-users] condor_submit -remote

Kewley, J (John) wrote:

I am still having some troubles with condor_submit -remote to submit
to a remote scheduler.

I have to turn on CLAIM_TO_BE (rather than the default no security) since FS_REMOTE
isn't supported without a shared filestore and/or UID domain (at least that is
my understanding).

BTW Note that I had to use LINUX.bat for my job to run rather than
$$(OpSys}$${ARCH}.bat since the target OS isn't known at this
stage. This could do with documenting.
I have Windows and Linux in the pool, central submit node being Linux.

Jobs now run, but then get stuck in "C" state, COMPLETED.

This is normal behavior. When you submit with condor_submit -remote, you have to use condor_transfer_data to fetch the output from the job once it ends in "C" state. You also have to remove the job, either with condor_rm or by using a periodic_remove expression that removes the job after its output has been fetched.

You may ask, "can't I have condor fetch the data and remove the remote job for me automatically?" Yes, you can. It's called Condor-C. Rather than submitting the job directly to the remote schedd, you submit it to your local schedd with universe=grid and GridResource set appropriately for the remote schedd. Condor then does the remote submission for you and fetches back the output when the job is finished.