 |
 |
Hi Graeme, they point both at the same space. ta cheers alessandra Graeme Stewart wrote: > On Wed, Jul 22, 2009 at 14:05, Duncan Rand<[log in to unmask]> wrote: > >> Alessandra Forti wrote: >> >>> Hi Graeme, >>> >>> >>>> ANALY_MANC1: Bad. Build job problems, which seem to be on stage-out >>>> after the code compiles, e.g., >>>> http://panda.cern.ch:25980/server/pandamon/query?job=1016502474. >>>> >>>> >>>> >>> all data servers links are saturated. >>> >>> >>>> ANALY_MANC2: Fair. Running out of local stage space? "Error details: >>>> pilot: Too little space left on local disk to run job: 2050048 kB >>>> (need > 2097152 kB)" - maybe need to clean up disks or run less jobs >>>> on these nodes? Otherwise ok. >>>> >>>> >>>> >>> indeed this is a recurring problem that we will solve when we upgrade to >>> SL5. We have only two cpus per node but only 20GB in scratch. It used to be >>> enough but these analisys seem to copy far more than 10GB per job. >>> >> We now get torque to create a per-job temporary directory which gets cleared >> up at the end of the job: >> >> http://www.clusterresources.com/torquedocs21/users/2.2files.shtml#tmpdir >> > > For historical reasons (remember the EDG RB?) the pre-pilot wrapper > script looks first for EDG_WL_SCRATCH, then TMPDIR: > > if [ -n "$EDG_WL_SCRATCH" ]; then > cd $EDG_WL_SCRATCH > elif [ -n "$TMPDIR" ]; then > cd $TMPDIR > fi > > I think this is now an anachronism and I will reverse the order, but > be kind to me if in the next few hours the pilot wrapper does the > wrong thing... > > Graeme > > -- No man ever steps in the same river twice, for it's not the same river and he's not the same man. (Heraclitus) Northgrid Tier2 Technical Coordinator http://www.hep.manchester.ac.uk/computing/tier2