Workflow 3567, Stage 1

Priority	50
Processors	1
Wall seconds	18000
RSS bytes	4193255424 (3999 MiB)
Max distance for inputs	30.0
Enabled input RSEs	CERN_PDUNE_EOS, 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, MANCHESTER, MONTECARLO, NIKHEF, PRAGUE, QMUL, RAL-PP, RAL_ECHO, SURFSARA, T3_US_NERSC
Enabled output RSEs	CERN_PDUNE_EOS, 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, MANCHESTER, 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, UK_Bristol, UK_Brunel, UK_Durham, UK_Edinburgh, UK_Imperial, UK_Lancaster, UK_Liverpool, UK_Manchester, UK_Oxford, UK_RAL-Tier1, UK_Sheffield, US_BNL, 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	hd-protodune-det-reco
Events for this stage

Output patterns

	Destination	Pattern	Lifetime	For next stage
1	Rucio hd-protodune-det-reco:hd-protodune_reconstruction_keepup_20240930_set0_v09_91_02d01_v0_3567	*keepup.root	7776000	False
2	Rucio hd-protodune-det-reco:hd-protodune_reconstruction_keepup_ntuples_20240930_set0_v09_91_02d01_v0_3567	*hists.root	7776000	False

Environment variables

Name	Value
METADATA_DIR	/cvmfs/fifeuser2.opensciencegrid.org/sw/dune/71f643ddd59465043e3cd3712a1e24b9cd0fa631

Condor Class Ads

Name	Value
HAS_CVMFS_dune_osgstorage_org	true

File states

Total files	Finding	Unallocated	Allocated	Outputting	Processed	Not found	Failed
58	0	0	0	0	45	0	13

Job states

Total	Submitted	Started	Processing	Outputting	Finished	Notused	Aborted	Stalled	Jobscript error	Outputting failed	None processed
172	0	0	0	0	90	0	1	2	78	0	1

RSEs used

Name	Inputs	Outputs
DUNE_US_BNL_SDCC	123	0
DUNE_US_FNAL_DISK_STAGE	0	90

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
#

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 "${DUNESW_DIR}" ]; then
  stat ${DUNESW_DIR}
  if [ $? -ne 0 ]; then
    echo "failed to stat dunesw dir"
    exit 1
  fi

  export PRODUCTS=$DUNESW_DIR:$PRODUCTS
fi

if [ -n "${LARRECO_DIR}" ]; then
  stat ${LARRECO_DIR}
  if [ $? -ne 0 ]; then
    echo "failed to stat larreco dir"
    exit 1
  fi

  export PRODUCTS=$LARRECO_DIR:$PRODUCTS
fi


if [ -n "${DUNEPROTOTYPES_DIR}" ]; then
  stat ${DUNEPROTOTYPES_DIR}
  if [ $? -ne 0 ]; then
    echo "failed to stat dunedetdataformats dir"
    exit 1
  fi

  export PRODUCTS=$DUNEPROTOTYPES_DIR:$PRODUCTS
fi

echo "PRODUCTS $PRODUCTS"

#Setup recent lar software suite
DUNE_VERSION=${DUNE_VERSION:-v09_91_02d01}
setup dunesw \
   "${DUNE_VERSION}" \
   -q "${DUNE_QUALIFIER:-e26:prof}"

if [ $? -ne 0 ]; then
  echo "Failed to setup dunesw $DUNE_VERSION $DUNE_QUALIFIER"
  exit 1
fi

if [ -n "${USE_INPUT_FCL}" ]; then
  
  if [ -z ${INPUT_DIR} ]; then
    echo "Error, INPUT_DIR is undefined but user requested USE_INPUT_FCL"
    exit 1
  fi

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

  FHICL_FILE_PATH=${INPUT_DIR}:${FHICL_FILE_PATH}
  echo "FCL PATH: $FHICL_FILE_PATH"
fi

if [ -n "${METADATA_DIR}" ]; then
  stat ${METADATA_DIR}
  if [ $? -ne 0 ]; then
    echo "failed to stat metadata dir"
  fi

  echo "metadata dir contents:"
  ls $METADATA_DIR
  PYTHONPATH=${METADATA_DIR}:$PYTHONPATH
fi

FCL1=${FCL1:-"standard_reco_stage1_protodunehd_keepup.fcl"}
echo "FCL1 dump:" ${FCL1}
fhicl-dump ${FCL1}
if [ $? -ne 0 ]; then
  echo "fhicl-dump ${FCL1} failed"
  exit 1
fi

FCL2=${FCL2:-"standard_reco_stage2_calibration_protodunehd_keepup.fcl"}
echo "FCL2 dump:" ${FCL2}
fhicl-dump ${FCL2}
if [ $? -ne 0 ]; then
  echo "fhicl-dump ${FCL2} failed"
  exit 1
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}  

echo "printing env"
env

echo "Will use justin-get-file"
#
DID_PFN_RSE=`$JUSTIN_PATH/justin-get-file`
##Check that any file was returned
if [ "${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' '`
echo "pfn: ${pfn}"
echo "did: ${did}"
now=$(date -u +"%Y%m%dT%H%M%SZ")

nevents=${NEVENTS:--1}

extra_line=""
if [ -n "${SKIPFCL2}" ]; then
  jobsub_id=`echo ${JUSTIN_JOBSUB_ID:-1.1@1} | cut -f1 -d'@' | sed -e"s/\./_/"`
  extra_line="-T pdhd_${jobsub_id}_${JUSTIN_WORKFLOW_ID}_${now}_decoder.root"
fi
echo "Running reco stage1"
touch reco.log
starttime=`date +"%s"`.0
LD_PRELOAD=$XROOTD_LIB/libXrdPosixPreload.so lar \
    -c ${FCL1} \
    -n ${nevents} \
    ${extra_line} ${pfn} #>reco.log 2>&1
larExit=$?
endtime=`date +"%s"`.0

if [ $larExit -ne 0 ]; then
  echo "Error in reco1"
  cat reco.log
  exit $larExit
fi

if [ -n "${SKIPFCL2}" ]; then
  echo "$pfn" > justin-processed-pfns.txt
  exit 0
fi

output_stage1_file=`ls *stage1.root`

starttime=`date +"%s"`.0
lar -c ${FCL2} \
    $output_stage1_file #>reco.log 2>&1
larExit=$?
endtime=`date +"%s"`.0

if [ $larExit -ne 0 ]; then
  echo "Error in reco2"
  cat reco.log
  exit $larExit
fi



output_reco_file=`ls *keepup.root`
output_mr_file=`ls *keepup_hists.root`

new_mr_file=`echo $output_reco_file | sed -e "s/keepup/keepup_hists/"`
mv $output_mr_file $new_mr_file
output_mr_file=$new_mr_file

echo "Output files:"
echo "\tReco: ${output_reco_file}"
echo "\tHists: ${output_mr_file}"

echo "Forming reco metadata"
python -m meta_maker --start_time $starttime --end_time $endtime --file_format "artroot" \
                     --app_family "dunesw" --app_name "reco" --app_version ${DUNE_VERSION} \
                     --data_tier "full-reconstructed" --get_events -p "$did" \
                     --campaign "hd-protodune-reco-keepup-v0" \
                     --fcl $FCL2 \
                     --past_fcls $FCL1 --past_apps "reco1" \
                     -f "${JUSTIN_SCOPE}:$output_reco_file" -j "${output_reco_file}.json"
if [ $? -ne 0 ]; then
  echo "Error in reco metadata"
  exit 1
fi
echo "Ran successfully"
## TODO -- CHECK
cat ${output_reco_file}.json

echo "Forming hist metadata"
python -m meta_maker --start_time $starttime --end_time $endtime --file_format "root" \
                     --app_family "dunesw" --app_name "reco" --app_version ${DUNE_VERSION} \
                     --data_tier "root-tuple-virtual" -p "$did" \
                     --campaign "hd-protodune-reco-keepup-v0" \
                     --fcl $FCL2 \
                     --past_fcls $FCL1 --past_apps "reco1" \
                     -f "${JUSTIN_SCOPE}:$output_mr_file" -j "${output_mr_file}.json"
 #--parent_as_json \
if [ $? -ne 0 ]; then
  echo "Error in hist metadata"
  exit 1
fi
echo "formed"
cat ${output_mr_file}.json



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