Workflow 12633, Stage 1
| Workflow | 12633 |
| Campaign | 971 |
| Priority | 50 |
| Processors | 1 |
| Wall seconds | 80000 |
| Image | /cvmfs/singularity.opensciencegrid.org/fermilab/fnal-wn-sl7:latest |
| RSS bytes | 4194304000 (4000 MiB) |
| Max distance for inputs | 0.0 |
| 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, 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_BNL, US_Colorado, US_FNAL-FermiGrid, US_FNAL-T1, US_Michigan, 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/jqin1/fnal/12633/1 | *_reco2.root | | | |
Environment variables
| Name | Value |
|---|
| INPUT_TAR_DIR_LOCAL | /cvmfs/fifeuser4.opensciencegrid.org/sw/dune/cfefef5fb58e1aa63679ae6194314a8d95f9b9f9 |
| MODE | gt |
| NUM_EVENTS | 10 |
File states
| Total files | Finding | Unallocated | Allocated | Outputting | Processed | Not found | Failed |
|---|
|
| 3 | 0 | 3 | 0 | 0 | 0 | 0 | 0 |
Job states
| Total | Submitted | Started | Processing | Outputting | Finished | Notused | Aborted | Stalled | Jobscript error | Outputting failed | None processed |
|---|
| 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 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
# --- CONFIGURATION ---
# 1. Standard DUNE FHiCLs for the downstream chain
FCL_G4="supernova_g4_dune10kt_1x2x6.fcl"
FCL_DS="standard_detsim_dune10kt_1x2x6.fcl"
FCL_R1="reco1_supernova_dune10kt_1x2x6.fcl"
FCL_R2="reco2_supernova_dune10kt_1x2x6.fcl"
# Unified Version for sim+reco
DUNE_VERSION=${DUNE_VERSION:-v09_91_04d01}
DUNE_QUALIFIER=${DUNE_QUALIFIER:-e26:prof}
# 3. Job Setup
MODE=${MODE:-fermi} # Default to 'fermi' if not set via --env
echo "Processing $NUM_EVENTS events per job in MODE: $MODE"
# 4. Trigger Handling (Dummy File)
did_pfn_rse=`$JUSTIN_PATH/justin-get-file`
if [ "$did_pfn_rse" = "" ] ; then
echo "Nothing to process - exit jobscript"
exit 0
fi
echo "$did_pfn_rse" | cut -f1 -d' ' >>all-input-dids.txt
pfn=`echo $did_pfn_rse | cut -f2 -d' '`
# --- MAIN WORKFLOW ---
# A. Setup Environment (Once for the whole chain)
source /cvmfs/dune.opensciencegrid.org/products/dune/setup_dune.sh
setup dunesw "$DUNE_VERSION" -q "$DUNE_QUALIFIER"
echo "Initialized DUNE software $DUNE_VERSION"
# B. Retrieve Custom Files
# Assumes INPUT_TAR_DIR_LOCAL is passed by justIN via --input-tar-local
if [ -d "$INPUT_TAR_DIR_LOCAL" ]; then
echo "Getting custom files from $INPUT_TAR_DIR_LOCAL"
cp $INPUT_TAR_DIR_LOCAL/*.react . 2>/dev/null
cp $INPUT_TAR_DIR_LOCAL/*.fcl . 2>/dev/null
# Validation
if [ ! -f "run_${MODE}.fcl" ]; then
echo "ERROR: run_${MODE}.fcl not found inside the tarball!"
exit 1
fi
echo "" >> run_${MODE}.fcl
echo "source.maxEvents: $NUM_EVENTS" >> run_${MODE}.fcl
echo "DEBUG: Forced source.maxEvents to $NUM_EVENTS in run_${MODE}.fcl"
else
echo "ERROR: INPUT_TAR_DIR_LOCAL not found. Did you use --input-tar-local?"
exit 1
fi
# C. Construct Unique Filenames
now=$(date -u +"%Y-%m-%dT_%H%M%SZ")
# Unique prefix using Workflow+Stage+Job ID
fname="${MODE}_${JUSTIN_WORKFLOW_ID}_${JUSTIN_STAGE_ID}_${JUSTIN_JOB_ID}"
outFile_gen=${fname}_${now}_gen.root
outFile_g4=${fname}_${now}_gen_g4.root
outFile_ds=${fname}_${now}_gen_g4_detsim.root
outFile_r1=${fname}_${now}_gen_g4_detsim_reco1.root
outFile_r2=${fname}_${now}_reco2.root
# --- EXECUTION LOOP ---
(
export LD_PRELOAD=${XROOTD_LIB}/libXrdPosixPreload.so
# STEP 1: GENERATION (Custom)
echo "Running Generation ($MODE)..."
lar -c run_${MODE}.fcl -n ${NUM_EVENTS} -o $outFile_gen > ${fname}_sim_${now}.log 2>&1
if [ $? -ne 0 ]; then echo "GEN failed"; exit 1; fi
# STEP 2: GEANT4
echo "Running G4..."
lar -c $FCL_G4 -n ${NUM_EVENTS} -s $outFile_gen -o $outFile_g4 >> ${fname}_sim_${now}.log 2>&1
if [ $? -ne 0 ]; then echo "G4 failed"; exit 2; fi
# STEP 3: DETSIM
echo "Running Detsim..."
lar -c $FCL_DS -n ${NUM_EVENTS} -s $outFile_g4 -o $outFile_ds >> ${fname}_sim_${now}.log 2>&1
if [ $? -ne 0 ]; then echo "DETSIM failed"; exit 3; fi
# STEP 4: RECO1
echo "Running Reco1..."
lar -c $FCL_R1 -s $outFile_ds -o $outFile_r1 >> ${fname}_sim_${now}.log 2>&1
if [ $? -ne 0 ]; then echo "RECO1 failed"; exit 4; fi
# STEP 5: RECO2
echo "Running Reco2..."
lar -c $FCL_R2 -n ${NUM_EVENTS} -s $outFile_r1 -o $outFile_r2 >> ${fname}_sim_${now}.log 2>&1
if [ $? -ne 0 ]; then echo "RECO2 failed"; exit 5; fi
)
# Check logic for the whole subshell
if [ $? -ne 0 ]; then
echo "Simulation chain failed. See logs."
exit 1
fi
# --- CLEANUP & FINISH ---
# Remove intermediates
rm $outFile_gen $outFile_g4 $outFile_ds $outFile_r1
# Final Check
if [ -f "$outFile_r2" ]; then
echo "$pfn" > justin-processed-pfns.txt
exit_code=0
else
echo "Error: Final reco2 output not found."
exit_code=1
fi
# Tar logs
tar zcf `echo "$JUSTIN_JOBSUB_ID.logs.tgz" | sed 's/@/_/g'` *.log
exit $exit_code
