justIN           Dashboard       Workflows       Jobs       AWT       Sites       Storages       Docs       Login

Workflow 9885, Stage 1

Workflow9885
Priority50
Processors2
Wall seconds18000
Image/cvmfs/singularity.opensciencegrid.org/fermilab/fnal-wn-sl7:latest
RSS bytes4193255424 (3999 MiB)
Max distance for inputs30.0
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, CERN, CH_UNIBE-LHEP, CZ_FZU, ES_CIEMAT, ES_PIC, FR_CCIN2P3, IT_CNAF, NL_NIKHEF, NL_SURFsara, UK_Bristol, UK_Brunel, UK_Durham, UK_Edinburgh, UK_Lancaster, UK_Liverpool, UK_Manchester, UK_Oxford, UK_QMUL, UK_RAL-PPD, UK_RAL-Tier1, UK_Sheffield, US_Colorado, US_FNAL-FermiGrid, US_FNAL-T1, US_Michigan, US_PuertoRico, US_SU-ITS, US_Swan, US_UChicago, US_UConn-HPC, US_UCSD, US_Wisconsin
Scopevd-protodune-det-reco
Events for this stage

Output patterns

 DestinationPatternLifetimeFor next stageRSE expression
1Rucio vd-protodune-det-reco:fnal-w9885s1p1*_keepup.root630720000False

Environment variables

NameValue
FHICL_TAR/cvmfs/fifeuser4.opensciencegrid.org/sw/dune/0183247c92433907df7c1f561070e5f865254efc
METADATA_DIR/cvmfs/fifeuser3.opensciencegrid.org/sw/dune/ff18e11900bfcc2b467fc81eba35d7e7de647a80

Condor Class Ads

NameValue
HAS_CVMFS_dune_osgstorage_orgtrue

File states

Total filesFindingUnallocatedAllocatedOutputtingProcessedNot foundFailed
29630130700120825423

Job states

TotalSubmittedStartedProcessingOutputtingFinishedNotusedAbortedStalledJobscript errorOutputting failedNone processed
158664000029101266751538901039123
Files processed00100100200200300300400400500500Nov-06 23:00Nov-07 09:00Nov-07 19:00Nov-08 05:00Nov-08 15:00Nov-09 01:00Nov-09 11:00Nov-09 21:00Nov-10 07:00Nov-10 17:00Nov-11 03:00Nov-11 13:00Nov-11 23:00Nov-12 09:00Nov-12 19:00Nov-13 05:00Nov-13 15:00Nov-14 01:00Nov-14 11:00Nov-14 21:00Nov-15 07:00Files processedBin start timesNumber per binUK_OxfordUK_RAL-Tier1FR_CCIN2P3CZ_FZUCERNUK_BristolUK_QMULUK_ManchesterUK_LancasterUK_BrunelUK_RAL-PPDIT_CNAFNL_SURFsaraNL_NIKHEFES_PICUK_SheffieldUK_Durham
Replicas per RSE2938380.00057375369.7499999985656Replicas per RSEDUNE_CERN_EOS (100%)

RSEs used

NameInputsOutputs
DUNE_CERN_EOS95420
DUNE_US_FNAL_DISK_STAGE01202
RAL_ECHO01

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

SiteAllocatedOutputting
UK_RAL-Tier123416
UK_Manchester8670
UK_QMUL7711
UK_Durham6591
NL_SURFsara6270
UK_RAL-PPD5231
NL_NIKHEF3710
ES_PIC2681
IT_CNAF2490
UK_Sheffield2310
UK_Brunel2270
UK_Oxford2101
UK_Lancaster2090
CERN1230
CZ_FZU940
FR_CCIN2P3730
UK_Bristol370
UK_Edinburgh190

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"

DUNE_TAG=v10_11_00d00
#Setup recent lar software suite
DUNE_VERSION=${DUNE_VERSION:-${DUNE_TAG}}
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

export FHICL_FILE_PATH=${FHICL_TAR}:${FHICL_FILE_PATH}
export FW_SEARCH_PATH=${FHICL_TAR}:$FW_SEARCH_PATH

echo "FHICL_FILE_PATH: ${FHICL_FILE_PATH}"

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

# Temporary fix to get the propoer PDS map
export FHICL_FILE_PATH=${METADATA_DIR}:${FHICL_FILE_PATH}
export FW_SEARCH_PATH=${METADATA_DIR}:${FW_SEARCH_PATH}

FCL1=${FCL1:-"standard_reco_stage1_protodunevd_keepup_all.fcl"}
#FCL1=${FCL1:-"standard_reco_stage1_protodunevd_keepup.fcl"}
#FCL1=${FCL1:-"standard_reco_protodunevd_keepup.fcl"}
echo "FCL1 dump:" ${FCL1}
fhicl-dump ${FCL1}
if [ $? -ne 0 ]; then
  echo "fhicl-dump ${FCL1} failed"
  exit 1
fi

# not yet setup
#FCL2=${FCL2:-"protodunevd_data_reco_stage2_calibration.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 ${JUSTIN_JOBSUB_ID:-1.1@1} | cut -f1 -d'@' | sed -e"s/\./_/"`
  extra_line="-T pdvd_${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/reco/reco_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 "vd-protodune-reco-keepup-v0" \
                     --fcl $FCL1 \
                     -f "${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 "vd-protodune-reco-keepup-v0" \
                     --fcl $FCL1 \
                     -f "${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
justIN time: 2026-02-10 09:51:49 UTC       justIN version: 01.06.00