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 |
# |
8 |
# Notes: |
9 |
# - 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 |
# $Header: /u/gcmpack/MITgcm_contrib/test_scripts/sxace/mitgcmtestreport,v 1.3 2018/01/09 10:11:43 mlosch Exp $ |
19 |
# $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_${VENDOR} |
51 |
selectexperiment='-t exp2' |
52 |
selectexperiment='' |
53 |
# download code into this directory |
54 |
TDIR=/ace/user/mlosch/tmp_$VENDOR |
55 |
gcmDIR=MITgcm |
56 |
git_repo='MITgcm' |
57 |
git_code='MITgcm' |
58 |
|
59 |
OPTFILE=../tools/build_options/SUPER-UX_SX-ACE_sxf90_awi |
60 |
#OPTFILE=/home/ace/mlosch/MITgcm/tools/build_options/SUPER-UX_SX-ACE_sxf90_awi |
61 |
|
62 |
RUNTESTREPORT="./testreport $MPI -of=${OPTFILE} $selectexperiment -small_f" |
63 |
# |
64 |
# create batch script |
65 |
# |
66 |
cat << EOF > $HERE/$JOBSCRIPT |
67 |
#PBS -q ace-r # job queue |
68 |
#PBS -N $JOBNAME # give the job a name |
69 |
#PBS -l cpunum_job=$NPROCS # cpus per node |
70 |
#PBS -l elapstim_req=2:00:00 |
71 |
#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 $OUTFILE |
89 |
fi |
90 |
|
91 |
# checkOut determines how much checking out is being done |
92 |
# checkOut = 3: new clone from GitHub and make a new copy |
93 |
# checkOut = 2: update (git pull) existing repo and make a new copy |
94 |
# checkOut = 1: skip update |
95 |
# checkOut = 0: use existing test code (if available otherwise switch to 1) |
96 |
|
97 |
checkOut=2 |
98 |
|
99 |
tmpFil=$TDIR/error.out |
100 |
if [ $checkOut -le 1 ] ; then |
101 |
if test -e $TDIR/${gcmDIR}/doc ; then |
102 |
echo $TDIR/${gcmDIR}/doc 'exist' |
103 |
else |
104 |
echo -n $TDIR/${gcmDIR} 'missing ; ' |
105 |
checkOut=2 |
106 |
echo "will make a new copy ( checkOut=$checkOut )" |
107 |
fi |
108 |
fi |
109 |
|
110 |
if [ $checkOut -ge 2 ] ; then |
111 |
#---- cleaning: |
112 |
cd $TDIR |
113 |
|
114 |
#---- Make a new clone or update existing one: |
115 |
if test -e ${gcmDIR}/.git/config ; then |
116 |
echo "${gcmDIR}/.git/config exist" |
117 |
else |
118 |
echo -n "${gcmDIR}/.git/config 'missing, " |
119 |
checkOut=3 |
120 |
echo "will get new clone ( checkOut=$checkOut )" |
121 |
fi |
122 |
if [ $checkOut -eq 3 ] ; then |
123 |
echo -n "Removing old clone: $TDIR/${gcmDIR} ..." |
124 |
test -e $TDIR/${gcmDIR} && rm -rf $TDIR/${gcmDIR} |
125 |
echo " done" |
126 |
echo -n "Make a new clone of $git_code from repo: $git_repo ..." |
127 |
git clone https://github.com/$git_repo/${git_code}.git ${gcmDIR} 2> $tmpFil |
128 |
retVal=$? |
129 |
if test $retVal = 0 ; then |
130 |
echo ' --> done!' |
131 |
rm -f $tmpFil |
132 |
else |
133 |
echo " Error: 'git clone' returned: $retVal" |
134 |
cat $tmpFil |
135 |
rm -f $tmpFil |
136 |
exit 2 |
137 |
fi |
138 |
else |
139 |
echo "Updating current clone ( $git_code ) ..." |
140 |
( cd ${gcmDIR}; git checkout master ; git pull ) |
141 |
echo ' --> done!' |
142 |
fi |
143 |
else |
144 |
cd $TDIR |
145 |
fi |
146 |
|
147 |
cd $TDIR/MITgcm/verification |
148 |
|
149 |
# make sure that we do not use the cross compiler for testreport |
150 |
unset CC |
151 |
# make sure that do use the cross compiler for testreport |
152 |
#export CC=sxcc |
153 |
|
154 |
$RUNTESTREPORT -j 8 -norun > $MYOUTPUT 2>&1 |
155 |
|
156 |
if [ "$?" != "0" ] |
157 |
then |
158 |
echo "something wrong with testreport" |
159 |
echo "keeping the working directory" |
160 |
#else |
161 |
# echo "check restarts" |
162 |
# echo ../tools/do_tst_2+2 -mpi -exe \"$HERE/$RUNIT\" -a NONE |
163 |
# ../tools/do_tst_2+2 -mpi -exe $HERE/$RUNIT -a NONE |
164 |
# everything OK: delete working directory |
165 |
# rm -rf $TDIR |
166 |
fi |
167 |
|
168 |
if [ ! -e $MYOUTPUT ] |
169 |
then |
170 |
touch $MYOUTPUT |
171 |
fi |
172 |
|
173 |
echo " " >> $MYOUTPUT |
174 |
echo "***********************************************************" >> $MYOUTPUT |
175 |
echo "Submitting this job script:" >> $MYOUTPUT |
176 |
echo "***********************************************************" >> $MYOUTPUT |
177 |
cat $HERE/$JOBSCRIPT >> $MYOUTPUT |
178 |
echo "***********************************************************" >> $MYOUTPUT |
179 |
echo "end of job script" >> $MYOUTPUT |
180 |
echo "***********************************************************" >> $MYOUTPUT |
181 |
echo " " >> $MYOUTPUT |
182 |
|
183 |
# now submit the job that actually runs all the experiments in one go |
184 |
echo "qsub $HERE/$JOBSCRIPT" |
185 |
qsub $HERE/$JOBSCRIPT |
186 |
# keep looking for the job in the job queues and wait until has disappeared |
187 |
jobruns=`qstat -n -u mlosch | grep "$JOBNAME"` |
188 |
while [ "${jobruns}"x != x ] |
189 |
do |
190 |
sleep 200 |
191 |
jobruns=`qstat -n -u mlosch | grep "$JOBNAME"` |
192 |
echo "waiting for job ${jobruns%% *} ($JOBNAME) to complete" |
193 |
currentexp=`grep Experiment $MYOUTPUT | tail -1` |
194 |
echo "currently running $currentexp" |
195 |
done |
196 |
|
197 |
# after running the experiments on the compute node run testreport |
198 |
# for a third time to evaluate results on the head node again |
199 |
echo " " >> $MYOUTPUT |
200 |
echo "now run testreport for a final time to evaluate results:" >> $MYOUTPUT |
201 |
echo "$RUNTESTREPORT -match 10 -runonly" >> $MYOUTPUT |
202 |
#$RUNTESTREPORT -match 10 -runonly >> $MYOUTPUT 2>&1 |
203 |
$RUNTESTREPORT -match 10 -runonly \ |
204 |
-a "jm_c@mitgcm.org" >> $MYOUTPUT 2>&1 |
205 |
# -a "jm_c@mitgcm.org, Martin.Losch@awi.de" >> $MYOUTPUT 2>&1 |
206 |
|
207 |
echo "end of mitgcmtestreport" |