20th March: justIN upgraded to 01.03 - please report any problems in #computing-operations or #user_grid_usage on Slack

Workflow 5777, Stage 1

Priority	50
Processors	1
Wall seconds	80000
Image	/cvmfs/singularity.opensciencegrid.org/fermilab/fnal-wn-sl7:latest
RSS bytes	6815744000 (6500 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, CZ_FZU, ES_CIEMAT, ES_PIC, FR_CCIN2P3, IN_TIFR, 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_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	https://fndcadoor.fnal.gov:2880/dune/scratch/users/ykermaid/05777/1	prod__g4_stage2.
2	https://fndcadoor.fnal.gov:2880/dune/scratch/users/ykermaid/05777/1	prod__reco

Environment variables

Name	Value
DUNESW_TAR	/cvmfs/fifeuser1.opensciencegrid.org/sw/dune/6b323ea5c7cc9b9401e76eefb3bb590ca7ade803
JSONFILE	pdhd_photon_208Tl_base_meta.json
MCJob	1
NEVENTS	10
pipyaml	1
UTIL_TAR	/cvmfs/fifeuser1.opensciencegrid.org/sw/dune/759abd7b624dfc93425879e7a4c1a5f7a98a4657
YAMLFILE	pdhd_scan_0to10MeV_electron_flatrec_gen.yaml

File states

Total files	Finding	Unallocated	Allocated	Outputting	Processed	Not found	Failed
10	0	0	0	0	10	0	0

Job states

Total	Submitted	Started	Processing	Outputting	Finished	Notused	Aborted	Stalled	Jobscript error	Outputting failed	None processed
26	0	0	0	0	18	0	0	4	0	0	4

RSEs used

Name	Inputs	Outputs
MONTECARLO	14	0
None	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)

File reset events, by site

Site	Allocated	Outputting
FR_CCIN2P3	2	0
CZ_FZU	1	0
UK_QMUL	1	0

Jobscript

#!/bin/bash
#

#These must be defined
#if [ -z $FCL_FILE ]; then
#  echo "Fatal Must provide FCL_FILE env var"
#  exit 1
#fi

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

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 [ -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 "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=${JUSTIN_STAGE_ID:-1}
a=1
 if [ -n "$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=${DUNESW_VERSION:-v10_04_07d00}
 setup dunesw \
    "${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
 
export FHICL_FILE_PATH=${DUNESW_TAR}:${FHICL_FILE_PATH}
 
 
 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


 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
 pfn=`echo ${DID_PFN_RSE} | cut -f2 -d' '`
 did=`echo ${DID_PFN_RSE} | cut -f1 -d' '`
 
 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"
 echo "MCJob: $MCJob"
 if [ -z "$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 they'd 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} \
      --overrides ${overrides} \
      --run $run \
      --subrun $subrun \
      --yaml ${UTIL_TAR}/${YAMLFILE} \
      --nevents $nevents \
      --event 0 \
      $inputflag

exitcode=$?
if [ $exitcode -ne 0 ]; then
  echo "Error running. Exiting with ${exitcode}"
  exit $exitcode
fi

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