Changeset 1853 for TOOLS/PACK_IPSL/parallelPack.sh

Timestamp:

09/03/12 11:29:13 (12 years ago)

Author:

gpincka

Message:

principalement : enchainement du launcher parallele + gestion reprise du code parallele

File:

• TOOLS/PACK_IPSL/parallelPack.sh (modified) (5 diffs)

TOOLS/PACK_IPSL/parallelPack.sh

@@ -1 +1 @@
 #!/bin/bash
+
+function insertDirectives
+{
+    
+    local batchFile="launch_and_measureTime.sh"
+    local lineBeginMutableDrves=`grep -n "batch directives : begin" $batchFile | awk -F":" '{print $1}' `
+    local lineHeaderMutableDrves=`grep -n "mutable directives" $batchFile | awk -F":" '{print $1}' `
+    local lineEndMutableDrves=`grep -n "batch directives : end" $batchFile | awk -F":" '{print $1}' `
+    if [ "x${lineBeginMutableDrves}" == "x" ] || [ "x${lineHeaderMutableDrves}" == "x" ] || [ "x${lineEndMutableDrves}" == "x" ]
+    then
+        echo "Examen du fichier $batchFile :"
+        echo "Il manque les balises commentees 'mutable directives' et 'batch directives : end' "
+        exit 1
+    fi
+    
+    # echo "lineHeaderMutableDrves=$lineHeaderMutableDrves"
+    # echo "lineEndMutableDrves=$lineEndMutableDrves"
+    
+    if [ $lineHeaderMutableDrves -ge $lineEndMutableDrves ]
+    then
+       echo "Examen du fichier $batchFile :"
+       echo "les lignes 'mutable directives' et 'batch directives : end' n'existent pas ou ne sont"
+       echo "pas bien placees l'une par rapport a l'autre ==> STOP."
+       exit 1
+    fi
+    
+    # Suppression des anciennes directives
+    local firstLineToMove=$(( $lineHeaderMutableDrves + 1 ))
+    local lastLineToMove=$(( $lineEndMutableDrves - 1 ))
+    if [ $firstLineToMove -lt $lineEndMutableDrves ]
+    then
+        sed -i "${firstLineToMove},${lastLineToMove}d" $batchFile
+    fi
+        
+    # insertion des directives chargees a partir de "load_batch_directives.sh"
+    local lineEndMutableDrves=`grep -n "batch directives : end" $batchFile | awk -F":" '{print $1}' `
+    sed -i "${lineEndMutableDrves}i\#MSUB -Qos $QosType" $batchFile
+    sed -i "${lineEndMutableDrves}i\#MSUB -q $queueType" $batchFile
+    sed -i "${lineEndMutableDrves}i\#MSUB -A $projectName" $batchFile
+    sed -i "${lineEndMutableDrves}i\#MSUB -T $computationTime" $batchFile
+    sed -i "${lineEndMutableDrves}i\#MSUB -n $nbProcs" $batchFile
+    
+    # insertion des directives specifiant les sorties des operations de pack
+    sed -i "${lineEndMutableDrves}i\#MSUB -e ${DETAILLED_OUTPUT}/pack_ipsl_%I.e" $batchFile
+    sed -i "${lineEndMutableDrves}i\#MSUB -o ${DETAILLED_OUTPUT}/pack_ipsl_%I.o" $batchFile
+
+}
+
+function check_machine
+{
+    local host_name=`hostname `
+    local isMachineTitane=`echo $host_name | grep "titane" | wc -l `
+    local isMachineCurie=`echo $host_name | grep "curie" | wc -l `
+    
+    if [ "x${isMachineTitane}" == "x0" ] && [ "x${isMachineCurie}" == "x0" ]
+    then
+        echo "Ce programme doit etre lance a partir 'curie' ou 'titane'. STOP."
+        exit 1
+    fi
+
+}
+
+function getNumeroOfPreviousTry
+{
+   local num_try="0"
+   ici=$PWD
+   cd $USER_OUTPUT_PROGRESS
+   
+   tryDirs=`ls | grep -e 'TRY__[[:digit:]]\{1,2\}$' `
+   for dir in $tryDirs
+   do
+      num=`echo $dir | awk -F"__" '{print $NF}' `
+      if [ $num -gt $num_try ]
+      then
+          num_try=$num 
+      fi
+   done  
+   cd $ici
+   echo $num_try
+   
+}
 
 ########  pour option -f #############
@@ -17 +98 @@
 ######################################
 
-export IGCM_DEM=${SCRATCHDIR}/IGCM_DEM
-
-######### Si l'option -f est activee, on detruit tous les fichiers status  ##########
+# On charge les variables d'environnement 
+. load_ipslPack_env.sh
+
+# On charge les directives batch
+. load_batch_directives.sh
+
+# on verifie que la machine est 'curie' ou 'titane'
+check_machine
+
+#########  On insert les directives batch en en-tete du script "launch_and_measureTime.sh"         ##########
+#########  lance plus loin. Seules les directives de nom du job et de sorties restent inchangees   ##########
+
+insertDirectives
+
+# exit 0 # gpdebug : a virer
+
+# export inputCmd="${USER_OUTPUT_PROGRESS}/inputCmd.list"
+# export nextInputCmd="${USER_OUTPUT_PROGRESS}/nextInputCmd.list"
+# export output="${USER_OUTPUT_PROGRESS}/packOutput.log"
+# export reportFile="${USER_OUTPUT_PROGRESS}/report.log"
+
+########## Si l'option -f est activee, on detruit tous les fichiers status  ##########
 if [ "x$restartExecFromZero" == "x1" ]
 then
@@ -41 +141 @@
     echo "Deletion done."
 fi
+
+
+# Par ailleurs, on vide les fichiers de cmds, d'output et de rapport si option -f activee
+if [ "x${restartExecFromZero}" == "x1" ]
+then
+    rm -rf ${USER_OUTPUT_PROGRESS}/TRY__*
+fi
+
+# exit 0 # a virer
+
 #####################################################################################
 
@@ -54 +164 @@
 # fichier pour la transmission des messages d'erreur d'un fils au prog principal
 export errorMsgFile="${PWD}/errorMsg.txt"
-############## gpdebug : fin ############################################################## 
-
-export timeHandlingFile="timeHandlingFile.txt"
+############## gpdebug : fin ##############################################################
+
+check_nco_version # verification de la version de nco #######################
+
+# penser a arreter le present script si le module "libccc_user" n'a pas ete charge
+
+export badFailureFile=${USER_OUTPUT_PROGRESS}/badFailure.txt
+> $badFailureFile # nouvel essai, on vide le fichier d'erreur
+tryNumFile=${USER_OUTPUT_PROGRESS}/numero_current_try.txt
+numPreviousTry=$( getNumeroOfPreviousTry )
+# echo "numPreviousTry=$numPreviousTry"
+numNewTry=$(( $numPreviousTry + 1 ))
+
+# Pas de 4eme essai permis
+# ---------------------------- 
+if [ $numNewTry -ge 3 ]
+then
+    echo "Pas de 4eme essai permis"
+    exit 1
+fi
+
+# echo "numNewTry=$numNewTry"
+mkdir -p ${USER_OUTPUT_PROGRESS}/TRY__${numNewTry}
+echo $numNewTry > $tryNumFile
+
+# exit 0 # a virer
+
+
+
+export timeHandlingFile="${USER_OUTPUT_PROGRESS}/timeHandlingFile.txt"
 > $timeHandlingFile
 startTime=$( getDateMilliSeconds )
@@ -65 +202 @@
 DEM_log -0 "Demarrage de ${SCRIPT_NAME}"
 
-check_nco_version # gpdebug : verification de la version de nco #######################
-
-ccc_msub -A tgcc0013 launch_and_measureTime.sh
-
-
+# ccc_msub -A tgcc0013 launch_and_measureTime.sh
+ccc_msub launch_and_measureTime.sh
 
 DEM_log -0 "Fin de ${SCRIPT_NAME}"
+
+
+elapsedTime=0
+while [ $elapsedTime -lt 500 ]
+do
+    sleep 5
+    elapsedTime=$(( $elapsedTime + 5 ))
+    isFailureFileEmpty=`cat $badFailureFile | wc -l `
+    if [ "x${isFailureFileEmpty}" != "x0" ]
+    then
+       sleep 2 # on attend que le fichier soit correctement rempli
+       cat $badFailureFile
+       exit 0
+    fi
+done
+
+
+
+