Ah, and I forgot to mention that jobmanager-fork works just fine. Per job GRAM job-manager logs don't show any errors.
On 11 April 2014 17:35, Blair Bethwaite <[email protected]> wrote: > Hi GT folks, > > I've got a pretty ancient cluster I'm trying to use - it's running GT4.0.7 > with both WS and non-WS GRAM available. > > Funny thing is, jobs submitted via "globusrun-ws -s -submit -F ... -Ft SGE > -c /bin/hostname" work ok but then I use the equivalent non-WS > "globus-job-run ..../jobmanager-sge /bin/hostname" the jobs are run on the > cluster but then finish with error. The SGE job status indicates they could > not find their stdout and stderr files, and indeed if I watch the progress > of the job I can see these get created and deleted quite quickly. > > On the other hand if I submit a GRAM batch job it runs to completion just > fine, but attempting to query its status returns "DONE" immediately. So I > guess for some reason the GRAM SEG thinks the job has completed immediately > whereas the WS-GRAM SEG doesn't make this mistake. Any idea what might > cause this? > > -- > Cheers, > ~Blairo > -- Cheers, ~Blairo
