Workflow 7106, Stage 1

Priority	50
Processors	1
Wall seconds	80000
Image	/cvmfs/singularity.opensciencegrid.org/fermilab/fnal-wn-sl7:latest
RSS bytes	8387559424 (7999 MiB)
Max distance for inputs	100.0
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, ES_CIEMAT, ES_PIC, FR_CCIN2P3, IN_TIFR, IT_CNAF, NL_SURFsara, UK_Bristol, UK_Brunel, UK_Durham, UK_Edinburgh, UK_Lancaster, 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	Rucio usertests:calcuttj_unet_input-w7106s1p1	*.h5	7776000	False

Environment variables

Name	Value
DUNESW_TAR	/cvmfs/fifeuser3.opensciencegrid.org/sw/dune/c9be5504be824ed976a7e79d7471f1ecffc87d20
DUNESW_VERSION	v10_01_03d00
MCJob	1
NEVENTS	5
pipyaml	1
UNET_DIR	/cvmfs/fifeuser3.opensciencegrid.org/sw/dune/b5765e0d70ac8df928f55219574a442063a5eb63
UTIL_TAR	/cvmfs/fifeuser2.opensciencegrid.org/sw/dune/d0e7b0ea355210af5e3f44d4e21e4b90d8f11883
YAMLFILE	pdhd_unet.yaml

File states

Total files	Finding	Unallocated	Allocated	Outputting	Processed	Not found	Failed
10000	0	3646	3147	29	3178	0	0

Job states

Total	Submitted	Started	Processing	Outputting	Finished	Notused	Aborted	Stalled	Jobscript error	Outputting failed	None processed
10944	2018	0	3147	29	3356	0	0	635	1759	0	0

RSEs used

Name	Inputs	Outputs
MONTECARLO	8746	0
DUNE_US_FNAL_DISK_STAGE	0	1118
RAL_ECHO	0	954
DUNE_UK_MANCHESTER_CEPH	0	621
DUNE_CERN_EOS	0	257
RAL-PP	0	158
QMUL	0	57
DUNE_CA_SFU	0	12
DUNE_FR_CCIN2P3_DISK	0	1

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

File reset events, by site

Site	Allocated	Outputting
UK_RAL-PPD	596	0
US_NotreDame	22	0
US_FNAL-T1	7	0
ES_PIC	6	0
US_FNAL-FermiGrid	2	0

Jobscript

#!/bin/bash
#

#These must be defined
if [ -z $UTIL_TAR ]; then
  echo "Fatal must provide UTIL_TAR env var"
  exit 1
fi
export PYTHONPATH=$UTIL_TAR:$PYTHONPATH

if [ -z $UNET_DIR ]; then
        echo "fatal must provide UNET_DIR env var"
        exit 1
fi

source /cvmfs/dune.opensciencegrid.org/products/dune/setup_dune.sh
setup metacat
setup python v3_9_15
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 [ -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}  

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
export pfn=`echo ${DID_PFN_RSE} | cut -f2 -d' '`
export did=`echo ${DID_PFN_RSE} | cut -f1 -d' '`
 

echo "Justin specific env vars"
env | grep JUSTIN
now=$(date -u +"%Y%m%dT%H%M%SZ")
jobid=`echo "${JUSTIN_JOBSUB_ID:-1}" | cut -f1 -d'@' | sed -e "s/\./_/"`
stageid=${JUSTIN_STAGE_ID:-1}
a=1

###Subshell for lar
#(
 if [ -n "$DUNESW_TAR" ]; then
   stat ${DUNESW_TAR}
   if [ $? -ne 0 ]; then
     echo "Failed to stat $DUNESW_TAR. Exiting safely"
     exit 0
   fi
 
   export PRODUCTS=$DUNESW_TAR:$PRODUCTS 
   echo "Set dunesw to $DUNESW_TAR"
 fi

 if [ -n "$WIRECELL_TAR" ]; then
   stat ${WIRECELL_TAR}
   if [ $? -ne 0 ]; then
     echo "Failed to stat $WIRECELL_TAR. Exiting safely"
     exit 0
   fi
 
   export PRODUCTS=$WIRECELL_TAR:$PRODUCTS 
   echo "Set wirecell to $WIRECELL_TAR"
 fi

 #Setup recent lar software suite
 DUNESW_VERSION=${DUNESW_VERSION:-v10_02_02d00}
 setup dunesw \
    "${DUNESW_VERSION}" \
    -q "${DUNE_QUALIFIER:-e26:prof}"
 
 setup_exit=$?
 if [ $? -ne 0 ]; then
   echo "Failed to setup dunesw $DUNESW_VERSION $DUNE_QUALIFIER"
   exit $setup_exit
 fi
 
 echo "DUNESW loc:"
 ups active | grep dunesw

 echo "WIRECELL loc:"
 ups active | grep wirecell
 
 
 echo "printing env"
 env
 echo "DUNE specific env vars"
 env | grep DUNE

 # User can specify to pip install pyyaml
 if [ -n "$pipyaml" ]; then
   echo "Installing yaml"
   pip install --user pyyaml
   echo "Done"
 # elif [ -n "${YAML_TAR}" ]; then
 #   echo "Using venv at ${YAML_TAR}"
 #   source ${YAML_TAR}/yaml_venv/bin/activate
 fi


 input_filename=`echo $did | cut -f2 -d':'`
 echo "input file: $input_filename"

 echo "jobsub_id: ${JUSTIN_JOBSUB_ID:-1}"
 
 #Get run/subrun from parent if using input -- else from justin job ID/process
 if [ -z "$MCJob" ]; then
   echo "Getting run subrun from $did"
   run_subrun=`python -m beam_job_utils get_run_subrun -i $did`
   get_run_ret=$?
   if [ $get_run_ret -ne 0 ]; then
     echo "error in get_run_subrun"
   fi
 
   echo $run_subrun
   run=`echo $run_subrun | cut -f1 -d' '`
   subrun=`echo $run_subrun | cut -f2 -d' '`
 else
   run=`echo $jobid | cut -f1 -d'_'`  ##TODO -- check order here
   subrun=`echo $jobid | cut -f2 -d'_'`
 fi
 nevents=${NEVENTS:--1} #Set from env override or -1 
 NSKIP=${NSKIP:-0} 
 echo "nevents: $nevents"
 
 overrides="core.data_tier=full-reconstructed \
  core.application.version=${DUNESW_VERSION} \
  core.application.name=reco \
  core.application.family=dunesw \
  core.application=dunesw.reco \
 "

 #User specifies if theyd like to use an input file or not 
 if [ -n "$MCJob" ]; then
   inputflag=""
 else
   inputflag="-i $pfn --parent $did"
 fi
 
 echo "input flag: $inputflag"
 
 # TODO -- metadata file configurable
 python -m beam_job_utils run_job \
       --json ${UTIL_TAR}/${JSONFILE:-pdsp_beam_base_meta.json} \
       --overrides ${overrides} \
       --run $run \
       --subrun $subrun \
       --yaml ${UTIL_TAR}/${YAMLFILE:-pdsp_beam_gen.yaml} \
       --nevents $nevents \
       --set_esr \
       $inputflag
 
 exitcode=$?
 if [ $exitcode -ne 0 ]; then
   echo "Error running. Exiting with ${exitcode}"
   exit $exitcode
 fi
#)

#if  [ -n $TESTFILE ]; then
#  echo $TESTFILE
#  ln -s $TESTFILE extracted.h5
#  ls -lht
#fi

h5files=extracted*h5
for i in $h5files; do
  echo $i
  mv $i `echo $i | sed -e "s/\.h5/_${now}_${jobid}_${stageid}.h5/"`
done


###subshell for convert
#(
#  echo "installing torch"
#  pip install --user torch h5py
#  
#  exitcode=$?
#  if [ $exitcode -ne 0 ]; then
#    echo "Error installing torch. Exiting with ${exitcode}"
#    exit $exitcode
#  fi
#  
#  
#  ### Run convert on the output
#  #python $UNET_DIR/convert_to_pt.py extracted*h5
#  #exitcode=$?
#  #if [ $exitcode -ne 0 ]; then
#  #  echo "Error converting to torch. Exiting with ${exitcode}"
#  #  exit $exitcode
#  #fi
#  ### Run convert on the output
#  for f in `ls extracted*h5`; do
#    echo $f
#    python $UNET_DIR/convert_to_pt.py $f #$files #$pfn #extracted*h5
#    exitcode=$?
#    if [ $exitcode -ne 0 ]; then
#      echo "Error converting to torch. Exiting with ${exitcode}"
#      exit $exitcode
#    fi
#  done
#)

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