20th March: justIN upgraded to 01.03 - please report any problems in #computing-operations or #user_grid_usage on Slack

Workflow 5767, Stage 1

Priority	50
Processors	1
Wall seconds	80000
Image	/cvmfs/singularity.opensciencegrid.org/fermilab/fnal-wn-sl7:latest
RSS bytes	2096103424 (1999 MiB)
Max distance for inputs	30.0
Enabled input RSEs	CERN_PDUNE_EOS, DUNE_CA_SFU, DUNE_CERN_EOS, DUNE_ES_PIC, DUNE_FR_CCIN2P3_DISK, DUNE_IN_TIFR, DUNE_IT_INFN_CNAF, DUNE_UK_GLASGOW, DUNE_UK_LANCASTER_CEPH, DUNE_UK_MANCHESTER_CEPH, DUNE_US_BNL_SDCC, DUNE_US_FNAL_DISK_STAGE, FNAL_DCACHE, FNAL_DCACHE_STAGING, FNAL_DCACHE_TEST, MONTECARLO, NIKHEF, PRAGUE, QMUL, RAL-PP, RAL_ECHO, SURFSARA, T3_US_NERSC
Enabled output RSEs	CERN_PDUNE_EOS, DUNE_CA_SFU, DUNE_CERN_EOS, DUNE_ES_PIC, DUNE_FR_CCIN2P3_DISK, DUNE_IN_TIFR, DUNE_IT_INFN_CNAF, DUNE_UK_GLASGOW, DUNE_UK_LANCASTER_CEPH, DUNE_UK_MANCHESTER_CEPH, DUNE_US_BNL_SDCC, DUNE_US_FNAL_DISK_STAGE, FNAL_DCACHE, FNAL_DCACHE_STAGING, FNAL_DCACHE_TEST, NIKHEF, PRAGUE, QMUL, RAL-PP, RAL_ECHO, SURFSARA, T3_US_NERSC
Enabled sites	BR_CBPF, CA_SFU, CA_Victoria, CERN, CH_UNIBE-LHEP, CZ_FZU, ES_CIEMAT, ES_PIC, FR_CCIN2P3, IN_TIFR, IT_CNAF, NL_NIKHEF, NL_SURFsara, UK_Bristol, UK_Brunel, UK_Durham, UK_Edinburgh, UK_Glasgow, UK_Imperial, UK_Lancaster, UK_Liverpool, UK_Manchester, UK_Oxford, UK_QMUL, UK_RAL-PPD, UK_RAL-Tier1, UK_Sheffield, US_Caltech, US_Colorado, US_FNAL-FermiGrid, US_FNAL-T1, US_Michigan, US_MIT, US_Nebraska, US_NotreDame, US_PuertoRico, US_SU-ITS, US_Swan, US_UChicago, US_UConn-HPC, US_UCSD, US_Wisconsin
Scope	usertests
Events for this stage

Output patterns

	Destination	Pattern	Lifetime	For next stage	RSE expression
1	https://fndcadoor.fnal.gov:2880/dune/scratch/users/calcuttj/beamevent/032725/7GeV_14mVfC/05767/1	*.h5

Environment variables

Name	Value
DUNE_VERSION	v10_04_05d00
FCL_FILE	run_pdhd_trigger_timing.fcl
INPUT_DIR	/cvmfs/fifeuser4.opensciencegrid.org/sw/dune/d5f4ec0a3e82cd28de716fb2c92390a8a0dc7a8b
NEVENTS	-1

File states

Total files	Finding	Unallocated	Allocated	Outputting	Processed	Not found	Failed
2501	0	2253	0	0	0	0	248

Job states

Total	Submitted	Started	Processing	Outputting	Finished	Notused	Aborted	Stalled	Jobscript error	Outputting failed	None processed
1536	0	0	0	0	0	0	0	0	1499	0	37

RSEs used

Name	Inputs	Outputs
DUNE_UK_GLASGOW	1499	0

Stats of processed input files as CSV or JSON, and of uploaded output files as CSV or JSON (up to 10000 files included)

Jobscript

#!/bin/bash
#

#These must be defined
if [ -z $FCL_FILE ]; then
  echo "Fatal Must provide FCL_FILE env var"
  exit 1
fi

if [ -z $INPUT_DIR ]; then
  echo "Fatal Must provide INPUT_DIR env var"
  exit 1
fi

stat ${INPUT_DIR}
if [ $? -ne 0 ]; then
  echo "Failed to $INPUT_DIR. Exiting safely"
  exit 0
fi



source /cvmfs/dune.opensciencegrid.org/products/dune/setup_dune.sh
setup metacat
export METACAT_SERVER_URL=https://metacat.fnal.gov:9443/dune_meta_prod/app
export METACAT_AUTH_SERVER_URL=https://metacat.fnal.gov:8143/auth/dune

if [ -n "$PRODUCTS_TAR" ]; then
  stat ${PRODUCTS_TAR}
  if [ $? -ne 0 ]; then
    echo "Failed to stat $PRODUCTS_TAR. Exiting safely"
    exit 0
  fi

  export PRODUCTS=$PRODUCTS_TAR:$PRODUCTS 
  echo "Set protoduneana to $PRODUCTS_TAR"
fi


#Setup recent lar software suite
DUNE_VERSION=${DUNE_VERSION:-v10_01_01d00}
setup dunesw \
   "${DUNE_VERSION}" \
   -q "${DUNE_QUALIFIER:-e26:prof}"

setup_exit=$?
if [ $? -ne 0 ]; then
  echo "Failed to setup dunesw $DUNE_VERSION $DUNE_QUALIFIER"
  exit $setup_exit
fi


echo "DUNESW loc:"
ups active | grep dunesw

if [ -z ${JUSTIN_PROCESSORS} ]; then
  JUSTIN_PROCESSORS=1
fi

echo "Justin processors: ${JUSTIN_PROCESSORS}"

export TF_NUM_THREADS=${JUSTIN_PROCESSORS}   
export OPENBLAS_NUM_THREADS=${JUSTIN_PROCESSORS} 
export JULIA_NUM_THREADS=${JUSTIN_PROCESSORS} 
export MKL_NUM_THREADS=${JUSTIN_PROCESSORS} 
export NUMEXPR_NUM_THREADS=${JUSTIN_PROCESSORS} 
export OMP_NUM_THREADS=${JUSTIN_PROCESSORS}  

pip install --user h5py
pipexit=$?
if [ $pipexit -ne 0 ]; then
  echo "Failed to install h5py"
  exit $pipexit
fi


export PYTHONPATH=$INPUT_DIR/gallery/python:$PYTHONPATH

echo "printing env"
env

echo "Justin specific env vars"
env | grep JUSTIN

echo "Will use justin-get-file"
#


echo "Will use justin-get-file"
DID_PFN_RSE=`$JUSTIN_PATH/justin-get-file`
if [ "${DID_PFN_RSE}" == "" ] ; then
  echo "Could not get file"
  exit 0
fi
pfn=`echo ${DID_PFN_RSE} | cut -f2 -d' '`
did=`echo ${DID_PFN_RSE} | cut -f1 -d' '`

now=$(date -u +"%Y%m%dT%H%M%SZ")

##TODO -- edit this
#jobid=`echo "${JUSTIN_JOBSUB_ID:-1}" | awk -F '.' '{print $1}'`
jobid=`echo "${JUSTIN_JOBSUB_ID:-1}" | cut -f1 -d'@' | sed -e "s/\./_/"`
PREFIX=${PREFIX:-"decoded"}
OUTFILE="${PREFIX}_${jobid}_${JUSTIN_STAGE_ID}_${JUSTIN_WORKFLOW_ID}_${now}.root"


nevents=${NEVENTS:--1}

echo "Running lar on $pfn"
touch ana.log
starttime=`date +"%s"`.0
LD_PRELOAD=$XROOTD_LIB/libXrdPosixPreload.so lar \
    -c ${FCL_FILE} \
    -n ${nevents} \
    -o ${OUTFILE} \
    $pfn  >ana.log 2>&1
larExit=$?
endtime=`date +"%s"`.0

if [ $larExit -ne 0 ]; then
  echo "Error in lar"
  cat ana.log
  exit $larExit
fi

python -m trigger_info -i $OUTFILE
trig_exit=$?
if [ $trig_exit -ne 0 ]; then
  echo "Error in trigger_info"
  exit $trig_exit
fi
echo "Ran successfully"

echo "formed"
echo $pfn > justin-processed-pfns.txt