Firstly, apologies for any cross posting. I've run into a couple of issues with the LeSC Globus-SGE adapter, I'm not sure but I think they could be to do with changes made upstream of the original source, possibly in VDT.
I'm running an application on the PRAGMA Grid and am trying to use the recently added Komolongma resource (thanks John). I've noticed that any single type jobs I submit via globus in batch mode end up in the error state in SGE. This is because the sge.pm in the VDT1.10.1a on Komolongma points the job stdout and stderr to /dev/null.real - a file which doesn't exist in a directory I clearly cannot write to. If I use globus-job-run or ask for GASS streaming (-o) from globusrun the job works with the stdout and stderr being directed as I believe they should to files under the globus job directory. Can somebody who claims to read Perl tell whether this is a bug? Also, I've noticed from a couple of VDT Globus servers (on SGE) that the poll method adds output (such as "Job is still queued for execution."). Which, whilst being informative for an interactive user, would probably break any client software that is just expecting the GRAM job state. And lastly, querying the status (globusrun -status) of a batch job which is in the error state in SGE returns GRAM job state 'ACTIVE' - another bug? Regards, -Blair -- In science one tries to tell people, in such a way as to be understood by everyone, something that no one ever knew before. But in poetry, it's the exact opposite. - Paul Dirac
