1 | #!/bin/bash |
---|
2 | |
---|
3 | function insertDirectives |
---|
4 | { |
---|
5 | |
---|
6 | local batchFile="launch_and_measureTime.sh" |
---|
7 | local lineBeginMutableDrves=`grep -n "batch directives : begin" $batchFile | awk -F":" '{print $1}' ` |
---|
8 | local lineHeaderMutableDrves=`grep -n "mutable directives" $batchFile | awk -F":" '{print $1}' ` |
---|
9 | local lineEndMutableDrves=`grep -n "batch directives : end" $batchFile | awk -F":" '{print $1}' ` |
---|
10 | if [ "x${lineBeginMutableDrves}" == "x" ] || [ "x${lineHeaderMutableDrves}" == "x" ] || [ "x${lineEndMutableDrves}" == "x" ] |
---|
11 | then |
---|
12 | echo "Examen du fichier $batchFile :" |
---|
13 | echo "Il manque les balises commentees 'mutable directives' et 'batch directives : end' " |
---|
14 | exit 1 |
---|
15 | fi |
---|
16 | |
---|
17 | # echo "lineHeaderMutableDrves=$lineHeaderMutableDrves" |
---|
18 | # echo "lineEndMutableDrves=$lineEndMutableDrves" |
---|
19 | |
---|
20 | if [ $lineHeaderMutableDrves -ge $lineEndMutableDrves ] |
---|
21 | then |
---|
22 | echo "Examen du fichier $batchFile :" |
---|
23 | echo "les lignes 'mutable directives' et 'batch directives : end' n'existent pas ou ne sont" |
---|
24 | echo "pas bien placees l'une par rapport a l'autre ==> STOP." |
---|
25 | exit 1 |
---|
26 | fi |
---|
27 | |
---|
28 | # Suppression des anciennes directives |
---|
29 | local firstLineToMove=$(( $lineHeaderMutableDrves + 1 )) |
---|
30 | local lastLineToMove=$(( $lineEndMutableDrves - 1 )) |
---|
31 | if [ $firstLineToMove -lt $lineEndMutableDrves ] |
---|
32 | then |
---|
33 | sed -i "${firstLineToMove},${lastLineToMove}d" $batchFile |
---|
34 | fi |
---|
35 | |
---|
36 | # insertion des directives chargees a partir de "load_batch_directives.sh" |
---|
37 | local lineEndMutableDrves=`grep -n "batch directives : end" $batchFile | awk -F":" '{print $1}' ` |
---|
38 | sed -i "${lineEndMutableDrves}i\#MSUB -Qos $QosType" $batchFile |
---|
39 | sed -i "${lineEndMutableDrves}i\#MSUB -q $queueType" $batchFile |
---|
40 | sed -i "${lineEndMutableDrves}i\#MSUB -A $projectName" $batchFile |
---|
41 | sed -i "${lineEndMutableDrves}i\#MSUB -T $computationTime" $batchFile |
---|
42 | sed -i "${lineEndMutableDrves}i\#MSUB -n $nbProcs" $batchFile |
---|
43 | |
---|
44 | # insertion des directives specifiant les sorties des operations de pack |
---|
45 | sed -i "${lineEndMutableDrves}i\#MSUB -e ${DETAILLED_OUTPUT}/pack_ipsl_%I.e" $batchFile |
---|
46 | sed -i "${lineEndMutableDrves}i\#MSUB -o ${DETAILLED_OUTPUT}/pack_ipsl_%I.o" $batchFile |
---|
47 | |
---|
48 | } |
---|
49 | |
---|
50 | function check_machine |
---|
51 | { |
---|
52 | local host_name=`hostname ` |
---|
53 | local isMachineTitane=`echo $host_name | grep "titane" | wc -l ` |
---|
54 | local isMachineCurie=`echo $host_name | grep "curie" | wc -l ` |
---|
55 | |
---|
56 | if [ "x${isMachineTitane}" == "x0" ] && [ "x${isMachineCurie}" == "x0" ] |
---|
57 | then |
---|
58 | echo "Ce programme doit etre lance a partir 'curie' ou 'titane'. STOP." |
---|
59 | exit 1 |
---|
60 | fi |
---|
61 | |
---|
62 | } |
---|
63 | |
---|
64 | function getNumeroOfPreviousTry |
---|
65 | { |
---|
66 | local num_try="0" |
---|
67 | ici=$PWD |
---|
68 | cd $USER_OUTPUT_PROGRESS |
---|
69 | |
---|
70 | tryDirs=`ls | grep -e 'TRY__[[:digit:]]\{1,2\}$' ` |
---|
71 | for dir in $tryDirs |
---|
72 | do |
---|
73 | num=`echo $dir | awk -F"__" '{print $NF}' ` |
---|
74 | if [ $num -gt $num_try ] |
---|
75 | then |
---|
76 | num_try=$num |
---|
77 | fi |
---|
78 | done |
---|
79 | cd $ici |
---|
80 | echo $num_try |
---|
81 | |
---|
82 | } |
---|
83 | |
---|
84 | ######## pour option -f ############# |
---|
85 | restartExecFromZero=0 |
---|
86 | while [ $# -gt 0 ] |
---|
87 | do |
---|
88 | echo "boucle sur les arguments du script ..." |
---|
89 | echo "@=$@" |
---|
90 | case $1 in |
---|
91 | -f) restartExecFromZero=1 |
---|
92 | echo "force restart from the beginning ..." |
---|
93 | ;; |
---|
94 | esac |
---|
95 | shift |
---|
96 | done |
---|
97 | export restartExecFromZero=${restartExecFromZero} |
---|
98 | ###################################### |
---|
99 | |
---|
100 | # On charge les variables d'environnement |
---|
101 | . load_ipslPack_env.sh |
---|
102 | |
---|
103 | # On charge les directives batch |
---|
104 | . load_batch_directives.sh |
---|
105 | |
---|
106 | # on verifie que la machine est 'curie' ou 'titane' |
---|
107 | check_machine |
---|
108 | |
---|
109 | ######### On insert les directives batch en en-tete du script "launch_and_measureTime.sh" ########## |
---|
110 | ######### lance plus loin. Seules les directives de nom du job et de sorties restent inchangees ########## |
---|
111 | |
---|
112 | insertDirectives |
---|
113 | |
---|
114 | # exit 0 # gpdebug : a virer |
---|
115 | |
---|
116 | # export inputCmd="${USER_OUTPUT_PROGRESS}/inputCmd.list" |
---|
117 | # export nextInputCmd="${USER_OUTPUT_PROGRESS}/nextInputCmd.list" |
---|
118 | # export output="${USER_OUTPUT_PROGRESS}/packOutput.log" |
---|
119 | # export reportFile="${USER_OUTPUT_PROGRESS}/report.log" |
---|
120 | |
---|
121 | ########## Si l'option -f est activee, on detruit tous les fichiers status ########## |
---|
122 | if [ "x$restartExecFromZero" == "x1" ] |
---|
123 | then |
---|
124 | echo "Deleting all status files ..." |
---|
125 | for CONFIG in $( awk '{print $1}' ${IGCM_DEM}/config_card.liste ) |
---|
126 | do |
---|
127 | PATH_SIMU=$( dirname $CONFIG ) |
---|
128 | ListStatFileInSimu=`find $PATH_SIMU -type f -name "*status" ` |
---|
129 | if [ "x$ListStatFileInSimu" != "x" ] |
---|
130 | then |
---|
131 | for file in $ListStatFileInSimu |
---|
132 | do |
---|
133 | if [ -e $file ] |
---|
134 | then |
---|
135 | rm -f $file |
---|
136 | fi |
---|
137 | |
---|
138 | done |
---|
139 | fi |
---|
140 | done |
---|
141 | echo "Deletion done." |
---|
142 | fi |
---|
143 | |
---|
144 | |
---|
145 | # Par ailleurs, on vide les fichiers de cmds, d'output et de rapport si option -f activee |
---|
146 | if [ "x${restartExecFromZero}" == "x1" ] |
---|
147 | then |
---|
148 | rm -rf ${USER_OUTPUT_PROGRESS}/TRY__* |
---|
149 | fi |
---|
150 | |
---|
151 | # exit 0 # a virer |
---|
152 | |
---|
153 | ##################################################################################### |
---|
154 | |
---|
155 | export JOB_DIR=${LS_SUBCWD:-${PWD}} |
---|
156 | export EXE_DIR=${JOB_DIR} |
---|
157 | source ${EXE_DIR}/DEM_utilities.sh |
---|
158 | |
---|
159 | ############## gpdebug : Pour gestion des erreurs ######################################### |
---|
160 | # Execute la function errorReceive() si reception d'un signal TERM |
---|
161 | trap 'DEM_errorReceive' TERM |
---|
162 | # liste des processus à détruire en cas d'erreur bloquante |
---|
163 | export listPID=$$ |
---|
164 | # fichier pour la transmission des messages d'erreur d'un fils au prog principal |
---|
165 | export errorMsgFile="${PWD}/errorMsg.txt" |
---|
166 | ############## gpdebug : fin ############################################################## |
---|
167 | |
---|
168 | check_nco_version # verification de la version de nco ####################### |
---|
169 | |
---|
170 | # penser a arreter le present script si le module "libccc_user" n'a pas ete charge |
---|
171 | |
---|
172 | export badFailureFile=${USER_OUTPUT_PROGRESS}/badFailure.txt |
---|
173 | > $badFailureFile # nouvel essai, on vide le fichier d'erreur |
---|
174 | tryNumFile=${USER_OUTPUT_PROGRESS}/numero_current_try.txt |
---|
175 | numPreviousTry=$( getNumeroOfPreviousTry ) |
---|
176 | # echo "numPreviousTry=$numPreviousTry" |
---|
177 | numNewTry=$(( $numPreviousTry + 1 )) |
---|
178 | |
---|
179 | # Pas de 4eme essai permis |
---|
180 | # ---------------------------- |
---|
181 | if [ $numNewTry -ge 3 ] |
---|
182 | then |
---|
183 | echo "Pas de 4eme essai permis" |
---|
184 | exit 1 |
---|
185 | fi |
---|
186 | |
---|
187 | # echo "numNewTry=$numNewTry" |
---|
188 | mkdir -p ${USER_OUTPUT_PROGRESS}/TRY__${numNewTry} |
---|
189 | echo $numNewTry > $tryNumFile |
---|
190 | |
---|
191 | # exit 0 # a virer |
---|
192 | |
---|
193 | |
---|
194 | |
---|
195 | export timeHandlingFile="${USER_OUTPUT_PROGRESS}/timeHandlingFile.txt" |
---|
196 | > $timeHandlingFile |
---|
197 | startTime=$( getDateMilliSeconds ) |
---|
198 | echo "launch time:$startTime" >> $timeHandlingFile |
---|
199 | |
---|
200 | SCRIPT_NAME=$(basename ${0} ) |
---|
201 | |
---|
202 | DEM_log -0 "Demarrage de ${SCRIPT_NAME}" |
---|
203 | |
---|
204 | # ccc_msub -A tgcc0013 launch_and_measureTime.sh |
---|
205 | ccc_msub launch_and_measureTime.sh |
---|
206 | |
---|
207 | DEM_log -0 "Fin de ${SCRIPT_NAME}" |
---|
208 | |
---|
209 | |
---|
210 | elapsedTime=0 |
---|
211 | while [ $elapsedTime -lt 500 ] |
---|
212 | do |
---|
213 | sleep 5 |
---|
214 | elapsedTime=$(( $elapsedTime + 5 )) |
---|
215 | isFailureFileEmpty=`cat $badFailureFile | wc -l ` |
---|
216 | if [ "x${isFailureFileEmpty}" != "x0" ] |
---|
217 | then |
---|
218 | sleep 2 # on attend que le fichier soit correctement rempli |
---|
219 | cat $badFailureFile |
---|
220 | exit 0 |
---|
221 | fi |
---|
222 | done |
---|
223 | |
---|
224 | |
---|
225 | |
---|
226 | |
---|