justIN           Dashboard       Workflows       Jobs       AWT       Sites       Storages       Docs       Login

Workflow 1613, Stage 1

Priority50
Processors1
Wall seconds80000
RSS bytes2621440000 (2500 MiB)
Max distance for inputs30.0
Enabled input RSEs
Enabled output RSEs
Enabled sites
Scopefardet-vd
Events for this stage

Output patterns

 DestinationPatternLifetimeFor next stage
0Rucio fardet-vd:fardet-vd-reco2_ritm1780305_anu_rhc_skip15000_end_1613*reco2.root2592000False
0Rucio fardet-vd:fardet-vd-reco2ana_ritm1780305_anu_rhc_skip15000_end_1613*reco2_ana.root2592000False
0Rucio fardet-vd:fardet-vd-pandora_ritm1780305_anu_rhc_skip15000_end_1613*.pndr2592000False
0Rucio fardet-vd:fardet-vd-validation_ritm1780305_anu_rhc_skip15000_end_1613*Validation.root2592000False

Environment variables

NameValue
CONVERT_DIR/cvmfs/fifeuser2.opensciencegrid.org/sw/dune/075c0e56c49d2ffca32bc443c9c2fcc3efb66d49
DETPRODVD
HCPRODRHC

File states

Total filesFindingUnallocatedAllocatedOutputtingProcessedNot foundFailed
5870000056350235

Job states

TotalSubmittedStartedProcessingOutputtingFinishedNotusedAbortedStalledJobscript errorOutputting failedNone processed
1529500006784027253212151410600
Files processed0020020040040060060080080010001000120012001400140016001600180018002000200022002200240024002600260028002800Feb-21 06:00Feb-21 10:00Feb-21 14:00Feb-21 18:00Feb-21 22:00Feb-22 02:00Feb-22 06:00Feb-22 10:00Feb-22 14:00Feb-22 18:00Feb-22 22:00Feb-23 02:00Feb-23 06:00Feb-23 10:00Feb-23 14:00Feb-23 18:00Feb-23 22:00Feb-24 02:00Feb-24 06:00Feb-24 10:00Files processedBin start timesNumber per binUK_RAL-PPDCERNFR_CCIN2P3US_UChicagoCZ_FZUNL_SURFsaraUS_FNAL-T1US_FNAL-FermiG…US_FNAL-FermiGridUS_BNLCA_SFUUK_ManchesterUS_ColoradoUK_RAL-Tier1US_SU-ITSUK_ImperialCH_UNIBE-LHEPUK_OxfordUK_EdinburghUK_BristolES_CIEMATUS_WisconsinUS_NotreDameUK_BrunelIT_CNAFCA_VictoriaUS_PuertoRicoES_PICUK_SheffieldUK_LiverpoolUK_LancasterNL_NIKHEFUK_QMUL
Replicas per RSE5870490.025244.55870269.975244.50000000000003Replicas per RSEDUNE_FR_CCIN2P3_DISK (50%)DUNE_US_FNAL_DISK_STAGE (50%)

RSEs used

NameInputsOutputs
DUNE_FR_CCIN2P3_DISK689195
DUNE_US_FNAL_DISK_STAGE56447738
DUNE_US_BNL_SDCC03605
RAL_ECHO03578
RAL-PP02190
PRAGUE01918
NIKHEF01306
DUNE_CERN_EOS01130
SURFSARA0574
DUNE_UK_LANCASTER_CEPH0470
DUNE_ES_PIC0430

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-Tier11611061
US_UChicago155508
US_FNAL-FermiGrid58626
US_BNL56229
US_NotreDame4763
UK_Manchester45533
US_Colorado4267
US_Wisconsin3685
US_SU-ITS3496
CZ_FZU34159
CERN26110
NL_NIKHEF2530
UK_QMUL24132
UK_RAL-PPD20181
US_PuertoRico1914
IT_CNAF1884
UK_Imperial1860
UK_Brunel1039
UK_Oxford926
NL_SURFsara670
CA_SFU516
CA_Victoria547
US_UConn-HPC524
US_FNAL-T1329
CH_UNIBE-LHEP332
UK_Bristol327
UK_Liverpool312
UK_Sheffield221
UK_Edinburgh227
ES_PIC225
ES_CIEMAT220
UK_Lancaster111
FR_CCIN2P319
US_UCSD03

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


##TODO -- write usage



DETECTOR="${DETPROD:-HD}"
HC="${HCPROD:-FHC}"

## Build up the fcls to use
if [ "${DETECTOR}" = "HD" ]; then
  ANAFCL="standard_ana_dune10kt_1x2x6.fcl"
  MDDETTYPE="fardet-hd"

  if [ "$HC" = "FHC" ]; then
    RECOFCL="standard_reco2_dune10kt_nu_1x2x6.fcl"
  elif [ "$HC" = "RHC" ]; then
    RECOFCL="standard_reco2_dune10kt_anu_1x2x6.fcl"
  fi

elif [ "${DETECTOR}" = "VD" ]; then
  ANAFCL="anatree_dunevd10kt_1x8x6_3view_30deg_geov3.fcl"
  MDDETTYPE="fardet-vd"

  if [ "$HC" = "FHC" ]; then
    RECOFCL="reco2_dunevd10kt_nu_1x8x6_3view_30deg_geov3.fcl"
  elif [ "$HC" = "RHC" ]; then
    RECOFCL="reco2_dunevd10kt_anu_1x8x6_3view_30deg_geov3.fcl"
  fi

elif [ "${DETECTOR}" = "TEST" ]; then
  RECOFCL="${INPUT_TAR_DIR_LOCAL}/test.fcl"

elif [ "${DETECTOR}" = "TEST2" ]; then
  RECOFCL="${INPUT_TAR_DIR_LOCAL}/test2.fcl"

else
  echo "NEED TO REQUEST EITHER HD OR VD. USER REQUESTED ${DETECTOR}"
  exit 1
fi

echo "Using Reco fcl: ${RECOFCL}"
echo "and Ana fcl: ${ANAFCL}"


#Setup recent lar software suite
setup dunesw \
   "${DUNE_VERSION:-v09_81_00d02}" \
   -q "${DUNE_QUALIFIER:-e26:prof}"
echo "printing env"
env


##Force certain env vars to force good behavior
#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   

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=`$JUSTIN_PATH/justin-get-file`
#getfileExit=$?

#if [ $getfileExit -ne 0 ] ; then
#  # Success !
#  # Error -- exit immediately 
#  jobscriptExit=1
#  echo "Error in justin-get-file"
#  exit $jobscriptExit
#fi

##Check that any file was returned
if [ "${DID_PFN_RSE}" == "" ] ; then
  echo "Could not get file"
  exit 0
fi

FILE=`echo ${DID_PFN_RSE} | cut -f2 -d' '`
DID=`echo ${DID_PFN_RSE} | cut -f1 -d' '`
echo "DID: ${DID}"

metacat file show -mj ${DID} > old_md.json
mcExit=$?
if [ $mcExit -eq 0 ] ; then
  echo "old metadata:"
  cat old_md.json
else
  echo "Could not retrieve old metadata"
  exit 1 
fi

#
#
now=$(date -u +"%Y%m%dT%H%M%SZ")
#OUTFILE="${OUTPREFIX:-fd_mc_prod_test}.${JUSTIN_REQUEST_ID}"
#OUTFILE="$OUTFILE.$JUSTIN_JOBSUB_ID.${now}"
#OUTFILE=`echo $OUTFILE | sed -e 's/@/./'`
#
####Run reco2
lar -c "${RECOFCL}" \
    -n "${NEVENTS:--1}" \
    "$FILE" >reco2.log 2>&1
    #-o "$OUTFILE.root" \
    #-n "${NEVENTS:--1}" \
    #"$FILE" >$OUTFILE.log 2>&1

larExit=$?
echo "Reco step lar exit code $larExit"

if [ $larExit -eq 0 ] ; then
  # Success !
  # Log the file for justin and move on to the next step
  #echo "$FILE" > justin-processed-pfns.txt
  echo "Moving on to analysis "
else
  # Error -- exit immediately 
  #jobscriptExit=1
  tail -100 reco2.log
  exit $larExit 
fi

if [ "${DETECTOR}" = "TEST" ] || [ "${DETECTOR}" = "TEST2" ]; then
  echo "Done testing. Exiting"
  exit 0
fi
ORIG_OUTFILE=`ls *reco2.root`
OUTFILE=`echo "${ORIG_OUTFILE}_${now}_reco2" | sed -e 's/reco2.root//'`

mv ${ORIG_OUTFILE} ${OUTFILE}.root

### Get the metadata
##TODO -- convert this is the right file extensions
extractor_prod.py --infile ${OUTFILE}.root --campaign ${CAMPAIGN:-fd_mc_2023a_reco2} \
                  --requestid ritm1780305 --no_crc > ${OUTFILE}.root.json
extractorExit=$?
if [ $extractorExit -eq 0 ] ; then
  # Success !
  echo "Extracted metadata"
else
  # Error -- exit immediately 
  jobscriptExit=1
  echo "Failed to extract md"
  exit $extractorExit
fi

### Convert the metadata to metacat
##TODO -- make sure the RCDS behavior is correct
python $CONVERT_DIR/convert_metadata.py -i ${OUTFILE}.root.json \
                                                -c ${RECOFCL} -j old_md.json \
                                                --app "art.reco2" \
                                                --app_ver "${DUNE_VERSION:-v09_81_00d02}" \
                                                --det "${MDDETTYPE}" \
                                                --parent ${DID}
converterExit=$?
if [ $converterExit -eq 0 ] ; then
  # Success !
  echo "Converted metadata"
else
  # Error -- exit immediately 
  jobscriptExit=1
  echo "Failed to convert md"
  exit $converterExit
fi

##If running VD, there will also be pandora files. Rename these
if [ "${DETECTOR}" = "VD" ]; then
  all_good=true
  mv Validation.root "${OUTFILE}_Validation.root" || all_good=false
  mv Pandora_Events.pndr "${OUTFILE}_Pandora_Events.pndr" || all_good=false

  ##Copy over the metadata for pndr
  ##Edit the fields corresepondingly
  sed ${OUTFILE}.root.json -e 's/full-reconstructed/pandora-info/' > ${OUTFILE}_Pandora_Events.pndr.json || all_good=false
  sed -i ${OUTFILE}_Pandora_Events.pndr.json -e 's/artroot/binary/' || all_good=false
  sed -i ${OUTFILE}_Pandora_Events.pndr.json -e '/art.file_format/d' || all_good=false
  
  ##Copy over the metadata for validation.root 
  ##Edit the fields corresepondingly
  sed ${OUTFILE}.root.json -e 's/full-reconstructed/pandora-info/' > ${OUTFILE}_Validation.root.json || all_good=false
  sed -i ${OUTFILE}_Validation.root.json -e 's/artroot/root/' || all_good=false
  sed -i ${OUTFILE}_Validation.root.json -e '/art.file_format/d' || all_good=false

  if [ $all_good = false ]; then
    echo "Something failed when editing pandora metadata"
    jobscriptExit=1
    exit $jobscriptExit
  fi

fi

ANAOUTFILE="${OUTFILE}_ana"
echo "Will output ana to $ANAOUTFILE"

###Run ana 
lar -c "${ANAFCL}" \
    -n "${NEVENTS:--1}" \
    -T "$ANAOUTFILE.root" \
    "$OUTFILE.root" >ana.log 2>&1


larExit=$?
echo "Ana step lar exit code $larExit"

### TODO: Handle this correctly
if [ $larExit -eq 0 ] ; then
  # Success !
  # Log the file for justin and move on to the next step
  echo "$FILE" > justin-processed-pfns.txt
else
  # Error -- exit immediately 
  jobscriptExit=1
  tail -100 ana.log
  exit $larExit
fi

sed_good=true
sed ${OUTFILE}.root.json -e 's/full-reconstructed/root-tuple-virtual/' > ${ANAOUTFILE}.root.json || sed_good=false
sed -i ${ANAOUTFILE}.root.json -e 's/artroot/root/' || sed_good=false
sed -i ${ANAOUTFILE}.root.json -e '/art.file_format/d' || sed_good=false
sed -i ${ANAOUTFILE}.root.json -e '/art.process/d' || sed_good=false
sed -i ${ANAOUTFILE}.root.json -e "s/${RECOFCL}/${ANAFCL}/" || sed_good=false
sed -i ${ANAOUTFILE}.root.json -e '/campaign/!s/reco2/anatree/' || sed_good=false
sed -i ${ANAOUTFILE}.root.json -e "s/${DID}/${OUTFILE}.root/" || sed_good=false

if [ $sed_good = false ]; then
  echo "Something failed when editing ana metadata"
  jobscriptExit=1
  exit $jobscriptExit
fi

ls
exit 0
justIN time: 2024-09-29 09:22:43 UTC       justIN version: 01.01.08