1 |
jmc |
1.1 |
#! /usr/bin/env bash |
2 |
|
|
|
3 |
jmc |
1.15 |
# $Header: /u/gcmpack/MITgcm/tools/example_scripts/ACESgrid/aces_test_all,v 1.14 2010/04/01 15:46:03 jmc Exp $ |
4 |
jmc |
1.1 |
# $Name: $ |
5 |
|
|
|
6 |
jmc |
1.9 |
# action =2 : submit test jobs ; =1 : get jobs output ; =3 : do both |
7 |
|
|
case $1 in |
8 |
|
|
'-subOnly') action=2 ; shift ;; |
9 |
|
|
'-getOnly') action=1 ; shift ;; |
10 |
|
|
'-sub+get') action=3 ; shift ;; |
11 |
|
|
*) action=3 ;; |
12 |
|
|
esac |
13 |
|
|
#echo "action= $action , Arg= $# " |
14 |
|
|
|
15 |
jmc |
1.4 |
echo -n '-- Starting: ' |
16 |
|
|
date |
17 |
jmc |
1.8 |
start=`date +%Y%m%d` |
18 |
jmc |
1.9 |
|
19 |
jmc |
1.1 |
if test $# = 0 |
20 |
|
|
then |
21 |
jmc |
1.13 |
tst_list='gnu ifc pgi adm mp2 mth tuv' |
22 |
jmc |
1.1 |
else |
23 |
|
|
tst_list=$* |
24 |
|
|
fi |
25 |
|
|
|
26 |
|
|
. /etc/profile.d/modules.sh |
27 |
|
|
module list > ~/tmp/.module_list 2>&1 |
28 |
|
|
|
29 |
|
|
# QSUB="/usr/local/bin/qsub" |
30 |
|
|
# QSTAT="/usr/local/bin/qstat" |
31 |
|
|
QSUB=qsub |
32 |
|
|
QSTAT=qstat |
33 |
jmc |
1.14 |
TST_DIR="/home/jmc/test_ACES" |
34 |
jmc |
1.4 |
MPACK="MITgcm_tools/mpack-1.6/mpack" |
35 |
jmc |
1.14 |
SUB_DIR="MITgcm_tools/example_scripts/ACESgrid" |
36 |
|
|
|
37 |
jmc |
1.15 |
LOG_FIL="$TST_DIR/output/tst_all.log" |
38 |
|
|
rm -f $LOG_FIL |
39 |
|
|
|
40 |
jmc |
1.14 |
cd $TST_DIR |
41 |
jmc |
1.1 |
|
42 |
|
|
nbtst=0 |
43 |
|
|
for i in $tst_list |
44 |
|
|
do |
45 |
jmc |
1.13 |
sfx=`echo ${i}"_mpi" | sed 's/mth_mpi/ifc_mth/' | sed 's/tuv_mpi/op64_tuv/' | sed 's/mp2_mpi/mp2_mth/'` |
46 |
jmc |
1.14 |
if test -f $SUB_DIR/aces_test_$sfx ; then |
47 |
jmc |
1.4 |
JOB="tst_"$i |
48 |
|
|
job_exist=`$QSTAT -a | grep $USER | grep $JOB | wc -l` |
49 |
jmc |
1.15 |
if [ $action -ge 2 ] ; then |
50 |
jmc |
1.9 |
#-- to submit job |
51 |
|
|
if test "x_$job_exist" = x_0 ; then |
52 |
jmc |
1.14 |
$QSUB $SUB_DIR/aces_test_$sfx |
53 |
jmc |
1.9 |
eval M_$i='submitted' |
54 |
|
|
nbtst=`expr $nbtst + 1` |
55 |
jmc |
1.15 |
else |
56 |
jmc |
1.9 |
echo $JOB |
57 |
|
|
$QSTAT -a | grep $USER | grep $JOB |
58 |
|
|
echo 'job already exist => skip this test' |
59 |
|
|
eval M_$i='skipped' |
60 |
|
|
fi |
61 |
|
|
else |
62 |
|
|
#-- to get outp back: |
63 |
|
|
if test "x_$job_exist" = x_0 ; then |
64 |
|
|
echo -n 'did not find any job: ' |
65 |
|
|
echo $JOB |
66 |
|
|
eval M_$i='skipped' |
67 |
jmc |
1.15 |
else |
68 |
jmc |
1.9 |
echo -n 'found a job: ' |
69 |
|
|
echo $JOB |
70 |
|
|
$QSTAT -a | grep $USER | grep $JOB |
71 |
|
|
eval M_$i='submitted' |
72 |
|
|
nbtst=`expr $nbtst + 1` |
73 |
|
|
fi |
74 |
jmc |
1.4 |
fi |
75 |
jmc |
1.15 |
else |
76 |
jmc |
1.1 |
echo 'no file:' aces_test_$sfx 'to submit' |
77 |
jmc |
1.4 |
eval M_$i='skipped' |
78 |
jmc |
1.1 |
fi |
79 |
|
|
done |
80 |
|
|
|
81 |
jmc |
1.15 |
if test $action = 2 ; then |
82 |
jmc |
1.9 |
count=0 |
83 |
|
|
else |
84 |
jmc |
1.4 |
#date_str=`date +%Y%m%d`"_0" |
85 |
jmc |
1.1 |
|
86 |
jmc |
1.9 |
MPACKDIR=`dirname $MPACK` |
87 |
jmc |
1.7 |
#- build the mpack utility (from build_mpack in testreport): |
88 |
jmc |
1.9 |
( cd $MPACKDIR && ./configure && make ) > TTT.build_mpack.$$ 2>&1 |
89 |
|
|
RETVAL=$? |
90 |
|
|
if test "x$RETVAL" != x0 ; then |
91 |
jmc |
1.7 |
echo |
92 |
|
|
echo "Error building the mpack tools at: $MPACK_DIR" |
93 |
|
|
if test -x $MPACK ; then |
94 |
|
|
HAVE_MPACK=t |
95 |
|
|
echo " use (old ?) executable:" ; ls -l $MPACK |
96 |
|
|
else |
97 |
|
|
HAVE_MPACK=f |
98 |
|
|
fi |
99 |
jmc |
1.9 |
else |
100 |
jmc |
1.7 |
if test -x $MPACK ; then |
101 |
|
|
rm -f TTT.build_mpack.$$ |
102 |
|
|
HAVE_MPACK=t |
103 |
|
|
echo " building mpack: OK" |
104 |
|
|
else |
105 |
|
|
echo " $MPACK not executable" |
106 |
|
|
HAVE_MPACK=f |
107 |
|
|
fi |
108 |
jmc |
1.9 |
fi |
109 |
|
|
echo |
110 |
|
|
|
111 |
|
|
count=$nbtst |
112 |
jmc |
1.7 |
fi |
113 |
|
|
|
114 |
jmc |
1.9 |
#- when it's done, retrieve output and send e-mail |
115 |
jmc |
1.15 |
minutes=0 ; freq=10 |
116 |
|
|
fsec=`expr $freq \* 60` |
117 |
|
|
echo "Check every $freq mn for $count test(s) to finish" | tee -a $LOG_FIL |
118 |
|
|
echo "- start at :" `date` | tee -a $LOG_FIL |
119 |
jmc |
1.1 |
while test $count != 0 ; do |
120 |
|
|
|
121 |
jmc |
1.15 |
sleep $fsec |
122 |
|
|
minutes=$(( $minutes + $freq )) |
123 |
jmc |
1.1 |
|
124 |
|
|
for i in $tst_list ; do |
125 |
|
|
|
126 |
|
|
eval comm=\$M_$i |
127 |
|
|
if test $comm = 'submitted' ; then |
128 |
|
|
JOB="tst_"$i |
129 |
|
|
ready_to_send=`$QSTAT -a | grep $USER | grep $JOB | wc -l` |
130 |
|
|
|
131 |
|
|
if test "x_$ready_to_send" = x_0 ; then |
132 |
jmc |
1.10 |
run_dir=${TST_DIR}"/MITgcm_"$i"/verification" |
133 |
|
|
#- results output: |
134 |
|
|
tdir=`ls -1 -t $run_dir | grep -v tr_out | grep '^tr_aces' | head -1` |
135 |
|
|
if test "x$tdir" != x ; then |
136 |
jmc |
1.8 |
#- check this is the right output |
137 |
|
|
chk=`echo $tdir | grep -c $start` |
138 |
|
|
if test $chk = '0' ; then |
139 |
|
|
curday=`date +%Y%m%d` |
140 |
|
|
chk=`echo $tdir | grep -c $curday` |
141 |
|
|
fi |
142 |
jmc |
1.10 |
if test $chk = '0' ; then |
143 |
jmc |
1.15 |
echo "tdir='$tdir'" | tee -a $LOG_FIL |
144 |
|
|
echo "Output do not match, no email sent for $i" | tee -a $LOG_FIL |
145 |
jmc |
1.8 |
else |
146 |
|
|
rm -f "/tmp/tr_aces-"$i".tar.gz" |
147 |
|
|
( cd $run_dir ; tar -czf "/tmp/tr_aces-"$i".tar.gz" ./$tdir ) |
148 |
jmc |
1.7 |
if test "x$HAVE_MPACK" = xt ; then |
149 |
jmc |
1.3 |
$MPACK -s MITgcm-test -m 3555000 "/tmp/tr_aces-"$i".tar.gz" jmc@mitgcm.org |
150 |
jmc |
1.15 |
echo "Email sent for $i at: " `date` | tee -a $LOG_FIL |
151 |
jmc |
1.7 |
else |
152 |
jmc |
1.15 |
echo " no email sent for $i (no mpack)" | tee -a $LOG_FIL |
153 |
jmc |
1.7 |
fi |
154 |
jmc |
1.1 |
fi |
155 |
jmc |
1.10 |
else |
156 |
jmc |
1.15 |
echo " no output found for $i" | tee -a $LOG_FIL |
157 |
jmc |
1.10 |
fi |
158 |
|
|
#- restart output: |
159 |
|
|
tdir=`ls -1 -t $run_dir | grep -v tr_out | grep '^rs_aces' | head -1` |
160 |
|
|
if test "x$tdir" != x ; then |
161 |
|
|
#- check this is the right output |
162 |
|
|
chk=`echo $tdir | grep -c $start` |
163 |
|
|
if test $chk = '0' ; then |
164 |
|
|
curday=`date +%Y%m%d` |
165 |
|
|
chk=`echo $tdir | grep -c $curday` |
166 |
|
|
fi |
167 |
|
|
if test $chk = '0' ; then |
168 |
jmc |
1.15 |
echo "tdir='$tdir'" | tee -a $LOG_FIL |
169 |
|
|
echo "Restart output do not match, no email sent for $i" | tee -a $LOG_FIL |
170 |
jmc |
1.10 |
else |
171 |
|
|
rm -f "/tmp/rs_aces-"$i".tar.gz" |
172 |
|
|
( cd $run_dir ; tar -czf "/tmp/rs_aces-"$i".tar.gz" ./$tdir ) |
173 |
|
|
if test "x$HAVE_MPACK" = xt ; then |
174 |
|
|
$MPACK -s MITgcm-test -m 3555000 "/tmp/rs_aces-"$i".tar.gz" jmc@mitgcm.org |
175 |
jmc |
1.15 |
echo "Email sent for $i restart:" `date` | tee -a $LOG_FIL |
176 |
jmc |
1.10 |
else |
177 |
jmc |
1.15 |
echo " no email sent for $i restart (no mpack)" | tee -a $LOG_FIL |
178 |
jmc |
1.10 |
fi |
179 |
|
|
fi |
180 |
|
|
else |
181 |
jmc |
1.15 |
echo " no restart output for $i" | tee -a $LOG_FIL |
182 |
jmc |
1.10 |
fi |
183 |
|
|
#- record successful sending |
184 |
|
|
eval M_$i=done |
185 |
|
|
count=`expr $count - 1` |
186 |
|
|
chmod 644 output/tst_$i.std* |
187 |
jmc |
1.1 |
fi |
188 |
|
|
fi |
189 |
|
|
done |
190 |
jmc |
1.15 |
|
191 |
jmc |
1.1 |
# "long" queue is 24hrs = 24*60min = 1440min |
192 |
jmc |
1.4 |
if test $minutes -gt 2160 ; then |
193 |
|
|
hrs=$(( $minutes / 60 )); |
194 |
jmc |
1.1 |
echo "Time expired after $minutes minutes ($hrs hours)" |
195 |
|
|
echo ' ' $count '/' $nbtst 'tests not yet finished' |
196 |
|
|
exit 1 |
197 |
|
|
fi |
198 |
|
|
|
199 |
|
|
done |
200 |
|
|
|
201 |
jmc |
1.15 |
echo "Retrieving $nbtst tests finish :" `date` | tee -a $LOG_FIL |
202 |
|
|
exit 0 |
203 |
jmc |
1.1 |
|