| 1 |
mlosch |
1.1 |
#!/bin/bash |
| 2 |
|
|
# new script for running testreport on stan1.awi.de |
| 3 |
|
|
# - split the testreport into 3 steps: |
| 4 |
|
|
# 1/ compiling on head node (tx7.awi.de), with -norun option |
| 5 |
|
|
# 2/ running on compute node (using PBS qsub), with -runonly option |
| 6 |
|
|
# 3/ evaluating result on head node with -runonly option |
| 7 |
mlosch |
1.5 |
# |
| 8 |
|
|
# Notes: |
| 9 |
mlosch |
1.1 |
# - step 2 leads to many error messages, because the OS on the compute |
| 10 |
|
|
# nodes does not have the appropriate shell tools, modifying the |
| 11 |
|
|
# runonly option to skip the evalution step would be nice but not |
| 12 |
|
|
# necessary; you'll just have to live with the error messages |
| 13 |
|
|
# - step 3 assumes that all experiments have been run successfully, i.e. |
| 14 |
|
|
# that the output files are up-to-date. |
| 15 |
|
|
# if not, testreport will try to run the sx ace-code on the tx7 frontend |
| 16 |
|
|
# which will fail inevitably and produce more errors, maybe we can |
| 17 |
|
|
# have a flag that skips everything but the evaluation step to avoid this |
| 18 |
mlosch |
1.5 |
# $Header: /u/gcmpack/MITgcm_contrib/test_scripts/sxace/mitgcmtestreport,v 1.4 2018/02/08 15:50:19 mlosch Exp $ |
| 19 |
mlosch |
1.1 |
# $Name: $ |
| 20 |
|
|
|
| 21 |
|
|
# for some reason the module command is not available in a bash script on |
| 22 |
|
|
# this computer so we have to create it here |
| 23 |
|
|
#module () { eval `/usr/bin/modulecmd bash $*` ; } |
| 24 |
|
|
# alternatively we can source this script that contains all relevant |
| 25 |
|
|
# definitions |
| 26 |
|
|
source /usr/share/Modules/init/bash |
| 27 |
|
|
#module use --append /sx8/user2/awisoft/modulefiles |
| 28 |
|
|
# load latest compilers: |
| 29 |
|
|
#module load sxf90/460 |
| 30 |
|
|
#module load sxc++/094 |
| 31 |
|
|
module load sxf90 |
| 32 |
|
|
module load sxc++ |
| 33 |
|
|
module load sxmpi |
| 34 |
|
|
module load sxnetcdf |
| 35 |
|
|
# |
| 36 |
|
|
# make sure that we have qsub and qstat |
| 37 |
|
|
#export PATH=${PATH}:/usr/bin/nqsII |
| 38 |
|
|
source /etc/profile.d/nec.sh |
| 39 |
|
|
# |
| 40 |
|
|
VENDOR=sxf90 |
| 41 |
|
|
RUNIT="runit_"$VENDOR |
| 42 |
|
|
HERE=`pwd` |
| 43 |
|
|
EXE='mpirun -np TR_NPROC ./mitgcmuv' |
| 44 |
|
|
NPROCS=2 |
| 45 |
|
|
MPI="-MPI $NPROCS" |
| 46 |
|
|
OUTFILE=$HOME/out_${VENDOR} |
| 47 |
|
|
MYOUTPUT=$HOME/testreport_${VENDOR} |
| 48 |
|
|
OUTFILE=out_${VENDOR} |
| 49 |
|
|
JOBNAME=test_ace |
| 50 |
|
|
JOBSCRIPT=job_${VENDOR} |
| 51 |
|
|
selectexperiment='-t exp2' |
| 52 |
|
|
selectexperiment='' |
| 53 |
|
|
# download code into this directory |
| 54 |
mlosch |
1.2 |
TDIR=/ace/user/mlosch/tmp_$VENDOR |
| 55 |
mlosch |
1.4 |
gcmDIR=MITgcm |
| 56 |
|
|
git_repo='MITgcm' |
| 57 |
|
|
git_code='MITgcm' |
| 58 |
mlosch |
1.1 |
|
| 59 |
|
|
OPTFILE=../tools/build_options/SUPER-UX_SX-ACE_sxf90_awi |
| 60 |
mlosch |
1.2 |
#OPTFILE=/home/ace/mlosch/MITgcm/tools/build_options/SUPER-UX_SX-ACE_sxf90_awi |
| 61 |
mlosch |
1.1 |
|
| 62 |
|
|
RUNTESTREPORT="./testreport $MPI -of=${OPTFILE} $selectexperiment -small_f" |
| 63 |
|
|
# |
| 64 |
|
|
# create batch script |
| 65 |
|
|
# |
| 66 |
|
|
cat << EOF > $HERE/$JOBSCRIPT |
| 67 |
mlosch |
1.2 |
#PBS -q ace-r # job queue |
| 68 |
mlosch |
1.1 |
#PBS -N $JOBNAME # give the job a name |
| 69 |
|
|
#PBS -l cpunum_job=$NPROCS # cpus per node |
| 70 |
mlosch |
1.2 |
#PBS -l elapstim_req=2:00:00 |
| 71 |
mlosch |
1.1 |
#PBS -l cputim_job=2:00:00 # time limit |
| 72 |
|
|
#PBS -l memsz_job=32gb # max accumulated memory, we need this much because of many netcdf files |
| 73 |
|
|
#PBS -j o # join i/o |
| 74 |
|
|
#PBS -S /bin/sh |
| 75 |
|
|
#PBS -o $OUTFILE # o Where to write output |
| 76 |
|
|
# |
| 77 |
|
|
|
| 78 |
|
|
cd \${PBS_O_WORKDIR} |
| 79 |
|
|
$RUNTESTREPORT -runonly -command "$EXE" >> $MYOUTPUT 2>&1 |
| 80 |
|
|
|
| 81 |
|
|
EOF |
| 82 |
|
|
|
| 83 |
|
|
# clean up old testreport output |
| 84 |
|
|
if [ -e $MYOUTPUT ]; then |
| 85 |
|
|
rm -rf $MYOUTPUT |
| 86 |
|
|
fi |
| 87 |
|
|
if [ -e $OUTFILE ]; then |
| 88 |
|
|
rm -r $OUTFILE |
| 89 |
|
|
fi |
| 90 |
mlosch |
1.4 |
|
| 91 |
|
|
# checkOut determines how much checking out is being done |
| 92 |
|
|
# checkOut = 3: new clone from GitHub and make a new copy |
| 93 |
|
|
# checkOut = 2: update (git pull) existing repo and make a new copy |
| 94 |
|
|
# checkOut = 1: skip update |
| 95 |
|
|
# checkOut = 0: use existing test code (if available otherwise switch to 1) |
| 96 |
|
|
|
| 97 |
|
|
checkOut=2 |
| 98 |
|
|
|
| 99 |
mlosch |
1.5 |
gitcmd=$HOME/git/git |
| 100 |
mlosch |
1.4 |
tmpFil=$TDIR/error.out |
| 101 |
|
|
if [ $checkOut -le 1 ] ; then |
| 102 |
|
|
if test -e $TDIR/${gcmDIR}/doc ; then |
| 103 |
|
|
echo $TDIR/${gcmDIR}/doc 'exist' |
| 104 |
|
|
else |
| 105 |
|
|
echo -n $TDIR/${gcmDIR} 'missing ; ' |
| 106 |
|
|
checkOut=2 |
| 107 |
|
|
echo "will make a new copy ( checkOut=$checkOut )" |
| 108 |
|
|
fi |
| 109 |
|
|
fi |
| 110 |
|
|
|
| 111 |
|
|
if [ $checkOut -ge 2 ] ; then |
| 112 |
|
|
#---- cleaning: |
| 113 |
|
|
cd $TDIR |
| 114 |
|
|
|
| 115 |
|
|
#---- Make a new clone or update existing one: |
| 116 |
|
|
if test -e ${gcmDIR}/.git/config ; then |
| 117 |
|
|
echo "${gcmDIR}/.git/config exist" |
| 118 |
|
|
else |
| 119 |
|
|
echo -n "${gcmDIR}/.git/config 'missing, " |
| 120 |
|
|
checkOut=3 |
| 121 |
|
|
echo "will get new clone ( checkOut=$checkOut )" |
| 122 |
|
|
fi |
| 123 |
|
|
if [ $checkOut -eq 3 ] ; then |
| 124 |
|
|
echo -n "Removing old clone: $TDIR/${gcmDIR} ..." |
| 125 |
|
|
test -e $TDIR/${gcmDIR} && rm -rf $TDIR/${gcmDIR} |
| 126 |
|
|
echo " done" |
| 127 |
|
|
echo -n "Make a new clone of $git_code from repo: $git_repo ..." |
| 128 |
mlosch |
1.5 |
${gitcmd} clone https://github.com/$git_repo/${git_code}.git ${gcmDIR} 2> $tmpFil |
| 129 |
mlosch |
1.4 |
retVal=$? |
| 130 |
|
|
if test $retVal = 0 ; then |
| 131 |
|
|
echo ' --> done!' |
| 132 |
|
|
rm -f $tmpFil |
| 133 |
|
|
else |
| 134 |
|
|
echo " Error: 'git clone' returned: $retVal" |
| 135 |
|
|
cat $tmpFil |
| 136 |
|
|
rm -f $tmpFil |
| 137 |
|
|
exit 2 |
| 138 |
mlosch |
1.1 |
fi |
| 139 |
mlosch |
1.4 |
else |
| 140 |
|
|
echo "Updating current clone ( $git_code ) ..." |
| 141 |
mlosch |
1.5 |
( cd ${gcmDIR}; ${gitcmd} checkout master ; ${gitcmd} pull ) |
| 142 |
|
|
retVal=$? |
| 143 |
|
|
if test $retVal = 0 ; then |
| 144 |
|
|
echo ' --> done!' |
| 145 |
|
|
else |
| 146 |
|
|
echo " Error: 'git pull' returned: $retVal" |
| 147 |
|
|
echo " Error: 'git pull' returned: $retVal" \ |
| 148 |
|
|
| mail -s "Git-error on Stan" Martin.Losch@awi.de |
| 149 |
|
|
exit 2 |
| 150 |
|
|
fi |
| 151 |
mlosch |
1.4 |
fi |
| 152 |
mlosch |
1.1 |
else |
| 153 |
mlosch |
1.4 |
cd $TDIR |
| 154 |
mlosch |
1.1 |
fi |
| 155 |
|
|
|
| 156 |
|
|
cd $TDIR/MITgcm/verification |
| 157 |
|
|
|
| 158 |
|
|
# make sure that we do not use the cross compiler for testreport |
| 159 |
|
|
unset CC |
| 160 |
|
|
# make sure that do use the cross compiler for testreport |
| 161 |
|
|
#export CC=sxcc |
| 162 |
|
|
|
| 163 |
|
|
$RUNTESTREPORT -j 8 -norun > $MYOUTPUT 2>&1 |
| 164 |
|
|
|
| 165 |
|
|
if [ "$?" != "0" ] |
| 166 |
|
|
then |
| 167 |
|
|
echo "something wrong with testreport" |
| 168 |
|
|
echo "keeping the working directory" |
| 169 |
|
|
#else |
| 170 |
|
|
# echo "check restarts" |
| 171 |
|
|
# echo ../tools/do_tst_2+2 -mpi -exe \"$HERE/$RUNIT\" -a NONE |
| 172 |
|
|
# ../tools/do_tst_2+2 -mpi -exe $HERE/$RUNIT -a NONE |
| 173 |
|
|
# everything OK: delete working directory |
| 174 |
|
|
# rm -rf $TDIR |
| 175 |
|
|
fi |
| 176 |
|
|
|
| 177 |
|
|
if [ ! -e $MYOUTPUT ] |
| 178 |
|
|
then |
| 179 |
|
|
touch $MYOUTPUT |
| 180 |
|
|
fi |
| 181 |
|
|
|
| 182 |
|
|
echo " " >> $MYOUTPUT |
| 183 |
|
|
echo "***********************************************************" >> $MYOUTPUT |
| 184 |
|
|
echo "Submitting this job script:" >> $MYOUTPUT |
| 185 |
|
|
echo "***********************************************************" >> $MYOUTPUT |
| 186 |
|
|
cat $HERE/$JOBSCRIPT >> $MYOUTPUT |
| 187 |
|
|
echo "***********************************************************" >> $MYOUTPUT |
| 188 |
|
|
echo "end of job script" >> $MYOUTPUT |
| 189 |
|
|
echo "***********************************************************" >> $MYOUTPUT |
| 190 |
|
|
echo " " >> $MYOUTPUT |
| 191 |
|
|
|
| 192 |
|
|
# now submit the job that actually runs all the experiments in one go |
| 193 |
mlosch |
1.4 |
echo "qsub $HERE/$JOBSCRIPT" |
| 194 |
mlosch |
1.1 |
qsub $HERE/$JOBSCRIPT |
| 195 |
|
|
# keep looking for the job in the job queues and wait until has disappeared |
| 196 |
|
|
jobruns=`qstat -n -u mlosch | grep "$JOBNAME"` |
| 197 |
|
|
while [ "${jobruns}"x != x ] |
| 198 |
mlosch |
1.5 |
do |
| 199 |
mlosch |
1.1 |
sleep 200 |
| 200 |
|
|
jobruns=`qstat -n -u mlosch | grep "$JOBNAME"` |
| 201 |
|
|
echo "waiting for job ${jobruns%% *} ($JOBNAME) to complete" |
| 202 |
|
|
currentexp=`grep Experiment $MYOUTPUT | tail -1` |
| 203 |
|
|
echo "currently running $currentexp" |
| 204 |
|
|
done |
| 205 |
|
|
|
| 206 |
|
|
# after running the experiments on the compute node run testreport |
| 207 |
|
|
# for a third time to evaluate results on the head node again |
| 208 |
|
|
echo " " >> $MYOUTPUT |
| 209 |
|
|
echo "now run testreport for a final time to evaluate results:" >> $MYOUTPUT |
| 210 |
|
|
echo "$RUNTESTREPORT -match 10 -runonly" >> $MYOUTPUT |
| 211 |
|
|
#$RUNTESTREPORT -match 10 -runonly >> $MYOUTPUT 2>&1 |
| 212 |
|
|
$RUNTESTREPORT -match 10 -runonly \ |
| 213 |
mlosch |
1.3 |
-a "jm_c@mitgcm.org" >> $MYOUTPUT 2>&1 |
| 214 |
|
|
# -a "jm_c@mitgcm.org, Martin.Losch@awi.de" >> $MYOUTPUT 2>&1 |
| 215 |
mlosch |
1.1 |
|
| 216 |
|
|
echo "end of mitgcmtestreport" |
Parent Directory
| 
Revision Log
| 
Revision Graph