Tidy a bit and abort as early as possible

Benedikt Volkel · sawenzel · commit 59a57461f3c1 · 2024-03-26T10:56:58.000+01:00
* check for GRID token at the very beginning

* check for jq tool at the very beginning

* set a default seed if nothing is given

* adjust tests
diff --git a/MC/run/ANCHOR/anchorMC.sh b/MC/run/ANCHOR/anchorMC.sh
@@ -15,6 +15,16 @@
 # helper functionality #
 ########################
 
+echo_info()
+{
+  echo "INFO [anchorMC]: ${*}"
+}
+
+echo_error()
+{
+  echo "ERROR [anchorMC]: ${*}"
+}
+
 print_help()
 {
   echo "Usage: ./anchorMC.sh"
@@ -53,7 +63,7 @@ print_help()
 # Prevent the script from being soured to omit unexpected surprises when exit is used
 SCRIPT_NAME="$(basename "$(test -L "$0" && readlink "$0" || echo "$0")")"
 if [ "${SCRIPT_NAME}" != "$(basename ${BASH_SOURCE[0]})" ] ; then
-    echo "ERROR: This script cannot not be sourced" >&2
+    echo_error "This script cannot not be sourced" >&2
     return 1
 fi
 
@@ -70,8 +80,15 @@ while [ "$1" != "" ] ; do
 done
 
 # make sure O2DPG + O2 is loaded
-[ ! "${O2DPG_ROOT}" ] && echo "Error: This needs O2DPG loaded" && exit 1
-[ ! "${O2_ROOT}" ] && echo "Error: This needs O2 loaded" && exit 1
+[ ! "${O2DPG_ROOT}" ] && echo_error "This needs O2DPG loaded" && exit 1
+[ ! "${O2_ROOT}" ] && echo_error "This needs O2 loaded" && exit 1
+
+# check if jq is there
+which jq >/dev/null 2>&1
+[ "${?}" != "0" ] && { echo_error "jq is not found. Install or load via alienv." ; exit 1 ; }
+
+alien-token-info >/dev/null 2>&1
+[ "${?}" != "0" ] && { echo_error "No GRID token found, required to run." ; exit 1 ; }
 
 #################################################################
 # Set all required variables to identify an anchored production #
@@ -87,8 +104,8 @@ export ALIEN_JDL_WORKFLOWDETECTORS=${ALIEN_JDL_WORKFLOWDETECTORS:-ITS,TPC,TOF,FV
 export ALIEN_JDL_ANCHOR_SIM_OPTIONS=${ALIEN_JDL_ANCHOR_SIM_OPTIONS:--gen pythia8}
 # all others MUST be set by the user/on the outside
 export ALIEN_JDL_LPMANCHORPASSNAME=${ALIEN_JDL_LPMANCHORPASSNAME:-${ANCHORPASSNAME}}
-export ALIEN_JDL_MCANCHOR=${ALIEN_JDL_MCANCHOR:-${MCANCHOR}}
-export ALIEN_JDL_LPMPASSNAME=${ALIEN_JDL_LPMPASSNAME:-${PASSNAME}}
+# LPMPASSNAME is used in O2 and O2DPG scripts, however on the other hand, ALIEN_JDL_LPMANCHORPASSNAME is the one that is set in JDL templates; so use ALIEN_JDL_LPMANCHORPASSNAME and set ALIEN_JDL_LPMPASSNAME
+export ALIEN_JDL_LPMPASSNAME=${ALIEN_JDL_LPMANCHORPASSNAME}
 export ALIEN_JDL_LPMRUNNUMBER=${ALIEN_JDL_LPMRUNNUMBER:-${RUNNUMBER}}
 export ALIEN_JDL_LPMPRODUCTIONTYPE=${ALIEN_JDL_LPMPRODUCTIONTYPE:-${PRODUCTIONTYPE}}
 export ALIEN_JDL_LPMINTERACTIONTYPE=${ALIEN_JDL_LPMINTERACTIONTYPE:-${INTERACTIONTYPE}}
@@ -101,29 +118,29 @@ export ALIEN_JDL_ADDTIMESERIESINMC=${ALIEN_JDL_ADDTIMESERIESINMC:-1}
 
 # cache the production tag, will be set to a special anchor tag; reset later in fact
 ALIEN_JDL_LPMPRODUCTIONTAG_KEEP=$ALIEN_JDL_LPMPRODUCTIONTAG
-echo "Substituting ALIEN_JDL_LPMPRODUCTIONTAG=$ALIEN_JDL_LPMPRODUCTIONTAG with ALIEN_JDL_LPMANCHORPRODUCTION=$ALIEN_JDL_LPMANCHORPRODUCTION for simulating reco pass..."
+echo_info "Substituting ALIEN_JDL_LPMPRODUCTIONTAG=$ALIEN_JDL_LPMPRODUCTIONTAG with ALIEN_JDL_LPMANCHORPRODUCTION=$ALIEN_JDL_LPMANCHORPRODUCTION for simulating reco pass..."
 ALIEN_JDL_LPMPRODUCTIONTAG=$ALIEN_JDL_LPMANCHORPRODUCTION
 
 # check variables that need to be set
-[ -z "${ALIEN_JDL_LPMANCHORPASSNAME}" ] && { echo "ERROR: Set ALIEN_JDL_LPMANCHORPASSNAME or ANCHORPASSNAME" ; exit 1 ; }
-[ -z "${ALIEN_JDL_MCANCHOR}" ] && { echo "ERROR: Set ALIEN_JDL_MCANCHOR or MCANCHOR" ; exit 1 ; }
-[ -z "${ALIEN_JDL_LPMPASSNAME}" ] && { echo "ERROR: Set ALIEN_JDL_LPMPASSNAME or PASSNAME" ; exit 1 ; }
-[ -z "${ALIEN_JDL_LPMRUNNUMBER}" ] && { echo "ERROR: Set ALIEN_JDL_LPMRUNNUMBER or RUNNUMBER" ; exit 1 ; }
-[ -z "${ALIEN_JDL_LPMPRODUCTIONTYPE}" ] && { echo "ERROR: Set ALIEN_JDL_LPMPRODUCTIONTYPE or PRODUCTIONTYPE" ; exit 1 ; }
-[ -z "${ALIEN_JDL_LPMINTERACTIONTYPE}" ] && { echo "ERROR: Set ALIEN_JDL_LPMINTERACTIONTYPE or INTERACTIONTYPE" ; exit 1 ; }
-[ -z "${ALIEN_JDL_LPMPRODUCTIONTAG}" ] && { echo "ERROR: Set ALIEN_JDL_LPMPRODUCTIONTAG or PRODUCTIONTAG" ; exit 1 ; }
-[ -z "${ALIEN_JDL_LPMANCHORRUN}" ] && { echo "ERROR: Set ALIEN_JDL_LPMANCHORRUN or ANCHORRUN" ; exit 1 ; }
-[ -z "${ALIEN_JDL_LPMANCHORPRODUCTION}" ] && { echo "ERROR: Set ALIEN_JDL_LPMANCHORPRODUCTION or ANCHORPRODUCTION" ; exit 1 ; }
-[ -z "${ALIEN_JDL_LPMANCHORYEAR}" ] && { echo "ERROR: Set ALIEN_JDL_LPMANCHORYEAR or ANCHORYEAR" ; exit 1 ; }
-
-[ -z "${NTIMEFRAMES}" ] && { echo "ERROR: Set NTIMEFRAMES" ; exit 1 ; }
-[ -z "${NSIGEVENTS}" ] && { echo "ERROR: Set NSIGEVENTS" ; exit 1 ; }
-[ -z "${SPLITID}" ] && { echo "ERROR: Set SPLITID" ; exit 1 ; }
-[ -z "${CYCLE}" ] && { echo "ERROR: Set CYCLE" ; exit 1 ; }
-[ -z "${PRODSPLIT}" ] && { echo "ERROR: Set PRODSPLIT" ; exit 1 ; }
+[ -z "${ALIEN_JDL_LPMANCHORPASSNAME}" ] && { echo_error "Set ALIEN_JDL_LPMANCHORPASSNAME or ANCHORPASSNAME" ; exit 1 ; }
+[ -z "${ALIEN_JDL_LPMRUNNUMBER}" ] && { echo_error "Set ALIEN_JDL_LPMRUNNUMBER or RUNNUMBER" ; exit 1 ; }
+[ -z "${ALIEN_JDL_LPMPRODUCTIONTYPE}" ] && { echo_error "Set ALIEN_JDL_LPMPRODUCTIONTYPE or PRODUCTIONTYPE" ; exit 1 ; }
+[ -z "${ALIEN_JDL_LPMINTERACTIONTYPE}" ] && { echo_error "Set ALIEN_JDL_LPMINTERACTIONTYPE or INTERACTIONTYPE" ; exit 1 ; }
+[ -z "${ALIEN_JDL_LPMPRODUCTIONTAG}" ] && { echo_error "Set ALIEN_JDL_LPMPRODUCTIONTAG or PRODUCTIONTAG" ; exit 1 ; }
+[ -z "${ALIEN_JDL_LPMANCHORRUN}" ] && { echo_error "Set ALIEN_JDL_LPMANCHORRUN or ANCHORRUN" ; exit 1 ; }
+[ -z "${ALIEN_JDL_LPMANCHORPRODUCTION}" ] && { echo_error "Set ALIEN_JDL_LPMANCHORPRODUCTION or ANCHORPRODUCTION" ; exit 1 ; }
+[ -z "${ALIEN_JDL_LPMANCHORYEAR}" ] && { echo_error "Set ALIEN_JDL_LPMANCHORYEAR or ANCHORYEAR" ; exit 1 ; }
+
+[ -z "${NTIMEFRAMES}" ] && { echo_error "Set NTIMEFRAMES" ; exit 1 ; }
+[ -z "${NSIGEVENTS}" ] && { echo_error "Set NSIGEVENTS" ; exit 1 ; }
+[ -z "${SPLITID}" ] && { echo_error "Set SPLITID" ; exit 1 ; }
+[ -z "${CYCLE}" ] && { echo_error "Set CYCLE" ; exit 1 ; }
+[ -z "${PRODSPLIT}" ] && { echo_error "Set PRODSPLIT" ; exit 1 ; }
 
 # also for this keep a real default
 NWORKERS=${NWORKERS:-8}
+# set a default seed if not given
+SEED=${ALIEN_PROC_ID:-${SEED:-1}}
 
 
 # default async_pass.sh script
@@ -143,14 +160,14 @@ fi
 # if there is no setenv_extra.sh in this directory (so no special version is "shipped" with this rpodcution), copy the default one
 if [[ ! -f setenv_extra.sh ]] ; then
     cp ${DPGSETENV} .
-    echo "[INFO alien_setenv_extra.sh] Use default setenv_extra.sh from ${DPGSETENV}."
+    echo_info "Use default setenv_extra.sh from ${DPGSETENV}."
 else
-    echo "[INFO alien_setenv_extra.sh] setenv_extra.sh was found in the current working directory, use it."
+    echo_info "setenv_extra.sh was found in the current working directory, use it."
 fi
 
 chmod u+x setenv_extra.sh
 
-echo "[INFO alien_async_pass.sh] Setting up DPGRECO to ${DPGRECO}"
+echo_info "Setting up DPGRECO to ${DPGRECO}"
 
 # settings that are MC-specific, modify setenv_extra.sh in-place
 sed -i 's/GPU_global.dEdxUseFullGainMap=1;GPU_global.dEdxDisableResidualGainMap=1/GPU_global.dEdxSplineTopologyCorrFile=splines_for_dedx_V1_MC_iter0_PP.root;GPU_global.dEdxDisableTopologyPol=1;GPU_global.dEdxDisableGainMap=1;GPU_global.dEdxDisableResidualGainMap=1;GPU_global.dEdxDisableResidualGain=1/' setenv_extra.sh
@@ -167,30 +184,28 @@ touch list.list
 ./async_pass.sh ${CTF_TEST_FILE:-""} 2&> async_pass_log.log
 RECO_RC=$?
 
-echo "RECO finished with ${RECO_RC}"
+echo_info "async_pass.sh finished with ${RECO_RC}"
 
 if [[ "${RECO_RC}" != "0" ]] ; then
     exit ${RECO_RC}
 fi
 
 ALIEN_JDL_LPMPRODUCTIONTAG=$ALIEN_JDL_LPMPRODUCTIONTAG_KEEP
-echo "Setting back ALIEN_JDL_LPMPRODUCTIONTAG to $ALIEN_JDL_LPMPRODUCTIONTAG"
+echo_info "Setting back ALIEN_JDL_LPMPRODUCTIONTAG to $ALIEN_JDL_LPMPRODUCTIONTAG"
 
 # now create the local MC config file --> config-config.json
 ${O2DPG_ROOT}/UTILS/parse-async-WorkflowConfig.py
 ASYNC_WF_RC=${?}
 
 # check if config reasonably created
 if [[ "${ASYNC_WF_RC}" != "0" || `grep "o2-ctf-reader-workflow-options" config-json.json 2> /dev/null | wc -l` == "0" ]]; then
-  echo "Problem in anchor config creation. Exiting."
+  echo_error "Problem in anchor config creation. Exiting."
   exit 1
 fi
 
 # -- CREATE THE MC JOB DESCRIPTION ANCHORED TO RUN --
 
 MODULES="--skipModules ZDC"
-# introduce variable to make usage clear
-SEED=${ALIEN_PROC_ID}
 # Since this is used, set it explicitly
 ALICEO2_CCDB_LOCALCACHE=${ALICEO2_CCDB_LOCALCACHE:-$(pwd)/ccdb}
 
@@ -205,28 +220,31 @@ remainingargs="${remainingargs} -productionTag ${ALIEN_JDL_LPMPRODUCTIONTAG:-ali
 # since the last passed argument wins, e.g. -productionTag cannot be overwritten by the user
 remainingargs="${ALIEN_JDL_ANCHOR_SIM_OPTIONS} ${remainingargs} --anchor-config config-json.json"
 
-echo "baseargs: ${baseargs}"
-echo "remainingargs: ${remainingargs}"
+echo_info "baseargs passed to o2dpg_sim_workflow_anchored.py: ${baseargs}"
+echo_info "remainingargs forwarded to o2dpg_sim_workflow.py: ${remainingargs}"
 
 # query CCDB has changed, w/o "_"
 ${O2DPG_ROOT}/MC/bin/o2dpg_sim_workflow_anchored.py ${baseargs} -- ${remainingargs} &> timestampsampling_${ALIEN_JDL_LPMRUNNUMBER}.log
-if [ "$?" != "0" ] ; then
-    echo "Problem during anchor timestamp sampling and workflow creation. Exiting."
-    exit 1
+WF_RC="${?}"
+if [ "${WF_RC}" != "0" ] ; then
+    echo_error "Problem during anchor timestamp sampling and workflow creation. Exiting."
+    exit ${WF_RC}
 fi
 
 TIMESTAMP=`grep "Determined timestamp to be" timestampsampling_${ALIEN_JDL_LPMRUNNUMBER}.log | awk '//{print $6}'`
-echo "TIMESTAMP IS ${TIMESTAMP}"
+echo_info "TIMESTAMP IS ${TIMESTAMP}"
 
 # -- Create aligned geometry using ITS ideal alignment to avoid overlaps in geant
 CCDBOBJECTS_IDEAL_MC="ITS/Calib/Align"
 TIMESTAMP_IDEAL_MC=1
 ${O2_ROOT}/bin/o2-ccdb-downloadccdbfile --host http://alice-ccdb.cern.ch/ -p ${CCDBOBJECTS_IDEAL_MC} -d ${ALICEO2_CCDB_LOCALCACHE} --timestamp ${TIMESTAMP_IDEAL_MC}
-if [ ! "$?" == "0" ]; then
-  echo "Problem during CCDB prefetching of ${CCDBOBJECTS_IDEAL_MC}. Exiting."
-  exit 1
+CCDB_RC="${?}"
+if [ ! "${CCDB_RC}" == "0" ]; then
+  echo_error "Problem during CCDB prefetching of ${CCDBOBJECTS_IDEAL_MC}. Exiting."
+  exit ${CCDB_RC}
 fi
 
+# TODO This can potentially be removed or if needed, should be taken over by o2dpg_sim_workflow_anchored.py and O2_dpg_workflow_runner.py
 echo "run with echo in pipe" | ${O2_ROOT}/bin/o2-create-aligned-geometry-workflow --configKeyValues "HBFUtils.startTime=${TIMESTAMP}" --condition-remap=file://${ALICEO2_CCDB_LOCALCACHE}=ITS/Calib/Align -b
 mkdir -p $ALICEO2_CCDB_LOCALCACHE/GLO/Config/GeometryAligned
 ln -s -f $PWD/o2sim_geometry-aligned.root $ALICEO2_CCDB_LOCALCACHE/GLO/Config/GeometryAligned/snapshot.root
@@ -235,22 +253,22 @@ ln -s -f $PWD/o2sim_geometry-aligned.root $ALICEO2_CCDB_LOCALCACHE/GLO/Config/Ge
 
 export FAIRMQ_IPC_PREFIX=./
 
-echo "Ready to start main workflow"
+echo_info "Ready to start main workflow"
 
 ${O2DPG_ROOT}/MC/bin/o2_dpg_workflow_runner.py -f workflow.json -tt ${ALIEN_JDL_O2DPGWORKFLOWTARGET:-aod} --cpu-limit ${ALIEN_JDL_CPULIMIT:-8}
 MCRC=$?  # <--- we'll report back this code
 exit 0
 if [[ "${ALIEN_JDL_ADDTIMESERIESINMC}" != "0" ]]; then
   # Default value is 1 so this is run by default.
-  echo "Running TPC time series"
+  echo_info "Running TPC time series"
   ${O2DPG_ROOT}/MC/bin/o2_dpg_workflow_runner.py -f workflow.json -tt tpctimes
 fi
 
-[[ ! -z "${DISABLE_QC}" ]] && echo "INFO: QC is disabled, skip it."
+[[ ! -z "${DISABLE_QC}" ]] && echo_info "QC is disabled, skip it."
 
 if [[ -z "${DISABLE_QC}" && "${MCRC}" = "0" && "${remainingargs}" == *"--include-local-qc"* ]] ; then
   # do QC tasks
-  echo "Doing QC"
+  echo_info "Doing QC"
   ${O2DPG_ROOT}/MC/bin/o2_dpg_workflow_runner.py -f workflow.json --target-labels QC --cpu-limit ${ALIEN_JDL_CPULIMIT:-8} -k
   # NOTE that with the -k|--keep-going option, the runner will try to keep on executing even if some tasks fail.
   # That means, even if there is a failing QC task, the return code will be 0
diff --git a/MC/run/ANCHOR/tests/test_anchor_2023_apass2_PbPb.sh b/MC/run/ANCHOR/tests/test_anchor_2023_apass2_PbPb.sh
@@ -8,9 +8,7 @@
 # taken from https://its.cern.ch/jira/browse/O2-4586
 export ALIEN_JDL_LPMANCHORPASSNAME=apass2
 export ALIEN_JDL_MCANCHOR=apass2
-export ALIEN_JDL_COLLISIONSYSTEM=Pb-Pb
 export ALIEN_JDL_CPULIMIT=8
-export ALIEN_JDL_LPMPASSNAME=apass2
 export ALIEN_JDL_LPMRUNNUMBER=544121
 export ALIEN_JDL_LPMPRODUCTIONTYPE=MC
 export ALIEN_JDL_LPMINTERACTIONTYPE=PbPb
@@ -25,12 +23,15 @@ export SPLITID=100
 export PRODSPLIT=153
 export CYCLE=0
 
-# on the GRID, this is set, for our use case, we can mimic any job ID
-export ALIEN_PROC_ID=2963436952
+# on the GRID, this is set and used as seed; when set, it takes precedence over SEED
+#export ALIEN_PROC_ID=2963436952
+export SEED=5
 
 # run the central anchor steering script; this includes
 # * derive timestamp
 # * derive interaction rate
 # * extract and prepare configurations (which detectors are contained in the run etc.)
 # * run the simulation (and QC)
+# To disable QC, uncomment the following line
+#export DISABLE_QC=1
 ${O2DPG_ROOT}/MC/run/ANCHOR/anchorMC.sh
diff --git a/MC/run/ANCHOR/tests/test_anchor_2023_apass2_pp.sh b/MC/run/ANCHOR/tests/test_anchor_2023_apass2_pp.sh
@@ -8,9 +8,7 @@
 # taken from https://its.cern.ch/jira/browse/O2-4586
 export ALIEN_JDL_LPMANCHORPASSNAME=apass2
 export ALIEN_JDL_MCANCHOR=apass2
-export ALIEN_JDL_COLLISIONSYSTEM=p-p
 export ALIEN_JDL_CPULIMIT=8
-export ALIEN_JDL_LPMPASSNAME=apass2
 export ALIEN_JDL_LPMRUNNUMBER=535069
 export ALIEN_JDL_LPMPRODUCTIONTYPE=MC
 export ALIEN_JDL_LPMINTERACTIONTYPE=pp
@@ -25,15 +23,18 @@ export SPLITID=100
 export PRODSPLIT=153
 export CYCLE=0
 
-# on the GRID, this is set, for our use case, we can mimic any job ID
-export ALIEN_PROC_ID=2963436952
+# on the GRID, this is set and used as seed; when set, it takes precedence over SEED
+#export ALIEN_PROC_ID=2963436952
+export SEED=5
 
 # for pp and 50 events per TF, we launch only 4 workers.
-export NWORKERS=4
+export NWORKERS=2
 
 # run the central anchor steering script; this includes
 # * derive timestamp
 # * derive interaction rate
 # * extract and prepare configurations (which detectors are contained in the run etc.)
 # * run the simulation (and QC)
+# To disable QC, uncomment the following line
+#export DISABLE_QC=1
 ${O2DPG_ROOT}/MC/run/ANCHOR/anchorMC.sh
