On Fri, 5 Nov 2004, Nicolescu, Edward L wrote:
> [testuser@lcg-ui01 bin]$ globus-job-submit lxn1184.cern.ch /bin/pwd
> https://lxn1184.cern.ch:20030/10241/1099671601/
> [testuser@lcg-ui01 bin]$ globus-job-status
> https://lxn1184.cern.ch:20030/10241/1099671601/
> DONE
> [testuser@lcg-ui01 bin]$ globus-job-get-output
> https://lxn1184.cern.ch:20030/10241/1099671601/
> GRAM Job submission failed because the job manager failed to open stderr
> (error code 74)
To use globus-job-run etc. your UI needs *inbound* connectivity, because
the globus-job-manager wants to connect back to the originating host to
send the job's stderr and stdout! :-(
A note in passing: globus-job-run etc. are to be used *only* for testing!
LCG does not support those commands for any real work, because they can
very easily overload the CE, because the globus-job-manager stays around
until the job has finished, continually polling the batch system... :-(
Jobs submitted by an RB do not suffer that problem, because Condor-G tells
the globus-job-manager to exit as soon as the job has been submitted to the
batch system; monitoring of the user's jobs is done by a "grid_monitor"
process running on the CE, communicating with a "condor_gridmanager" process
running on the RB.
|