Annotation of /MITgcm_contrib/test_scripts/ollie/mitgcmtestreport_ifort

#!/bin/bash -e
# new script for running testreport on ollie.awi.de
# - compile and run on compute nodes
# - use ssh to call mpack command from the head node ollie0
#$Header: /u/gcmpack/MITgcm_contrib/test_scripts/ollie/mitgcmtestreport_ifort,v 1.7 2018/02/13 15:47:57 mlosch Exp $
#$Name:  $

# needed for cron-job
source /usr/Modules/3.2.10/init/bash
source /etc/profile.d/cray_pe.sh
source /etc/profile.d/AWI.sh
#
module purge
module load intel.compiler
module load intel.mpi
module load netcdf
#export LD_LIBRARY_PATH=${CRAY_LD_LIBRARY_PATH}:${LD_LIBRARY_PATH}:

export MPI_ROOT=${I_MPI_ROOT}
#(dirname $(dirname `which mpiifort`))
export MPI_INC_DIR=${MPI_ROOT}/include64

# there is no slurm module anymore and this is the current recommendation to
# have sbatch in your path (rather than running /etc/profile.d./slurm.sh)
export PATH=${PATH}:/global/opt/slurm/default/bin

dNam='ollie'
TST_DIR="/work/ollie/mlosch/test_$dNam"
echo "start from TST_DIR='$TST_DIR' at: "`date`

umask 0022

sfx='ifort'
RUNIT="runit_"$sfx

OPTFILE=../tools/build_options/linux_ia64_${sfx}_ollie
options="-MPI 6"
options="$options -odir ${dNam}-i"
options="$options -j 6"
#options="$options -t global_ocean.cs32x15"

#EXE='mpiexec.hydra -bootstrap slurm -n TR_NPROC ./mitgcmuv'
#EXE='srun --mpi=pmi2 -n TR_NPROC ./mitgcmuv'
EXE='srun -n TR_NPROC ./mitgcmuv'

if [ -e $TST_DIR ]; then
    echo "$TST_DIR exists"
else 
    mkdir $TST_DIR
fi
cd $TST_DIR
HERE=$TST_DIR/output
if [ -e $HERE ]; then
    echo "$HERE"
else 
    mkdir $HERE
fi
OUTFILE=$HERE/slurm_${sfx}.out
MYOUTPUT=$HERE/out_$sfx
if [ -e $MYOUTPUT ]; then
  rm -rf $MYOUTPUT
fi
if [ -e $OUTFILE ]; then
  rm -r $OUTFILE
fi
gcmDIR="MITgcm_${sfx}"
git_repo='MITgcm'
git_code='MITgcm'

# checkOut determines how much checking out is being done
# checkOut = 3: new clone from GitHub and make a new copy
# checkOut = 2: update (git pull) existing repo and make a new copy
# checkOut = 1: skip update
# checkOut = 0: use existing test code (if available otherwise switch to 1)

checkOut=2
tdir=${TST_DIR}
today=`date +%Y%m%d`
TODAY=`date +%d`
#tmpFil="/tmp/"`basename $0`".$$"
tmpFil=$TST_DIR/error.out

if [ $checkOut -le 1 ] ; then
  if test -e $TST_DIR/${gcmDIR}/doc ; then
    echo $TST_DIR/${gcmDIR}/doc 'exist'
  else
    echo -n $TST_DIR/${gcmDIR} 'missing ; '
    checkOut=2
    echo "will make a new copy ( checkOut=$checkOut )"
  fi
fi

if [ $checkOut -ge 2 ] ; then
  #---- cleaning:
  cd $TST_DIR

  #---- Make a new clone or update existing one:
  if test -e ${gcmDIR}/.git/config ; then
    echo "${gcmDIR}/.git/config exist"
  else
    echo -n "${gcmDIR}/.git/config 'missing "
    checkOut=3
    echo "will get new clone ( checkOut=$checkOut )"
  fi
  if [ $checkOut -eq 3 ] ; then
    echo -n "Removing old clone: $TST_DIR/${gcmDIR} ..."
    test -e $TST_DIR/${gcmDIR}  &&  rm -rf $TST_DIR/${gcmDIR}
    echo "  done"
    echo -n "Make a new clone of $git_code from repo: $git_repo ..."
    git clone https://github.com/$git_repo/${git_code}.git ${gcmDIR} 2> $tmpFil
    retVal=$?
    if test $retVal = 0 ; then
       echo ' --> done!'
       rm -f $tmpFil
    else
       echo " Error: 'git clone' returned: $retVal"
       cat $tmpFil
       rm -f $tmpFil
       exit 2
    fi
  else
    echo "Updating current clone ( $git_code ) ..."
    ( cd ${gcmDIR}; git checkout master ; git pull )
    echo ' --> done!'
  fi
else
  cd $TST_DIR
fi

cd ${TST_DIR}/${gcmDIR}/verification

cwd=\`pwd\`
SENDCMD="ssh ollie0 ${TST_DIR}/${gcmDIR}/tools/mpack-1.6/mpack"
runtestreport="./testreport $options -of $OPTFILE -command \"${EXE}\" -send \"${SENDCMD}\" -sd ${cwd}" 
emailaddress="-a jm_c@mitgcm.org"

if [ ! -e $MYOUTPUT ]
    then
    touch $MYOUTPUT
fi

# echo "running testreport like this:"
# echo ${runtestreport} -norun
# eval "${runtestreport} -norun >> $MYOUTPUT 2>&1"

# create batch script
#
JOBNAME=tst$sfx
echo "creating batch script $HERE/$RUNIT"
cat << EOF >| $HERE/$RUNIT
#!/bin/bash
#SBATCH --job-name=${JOBNAME}
#SBATCH  -o ${OUTFILE}
#SBATCH --time=6:00:00
#SBATCH --ntasks=6

# still need this to be able to run a few experiments: 
# cfc_example, fizhi, tutorial_deep_convection
ulimit -s unlimited

# binds OpenMP task to given cores
export OMP_PROC_BIND=TRUE

# for debugging
export FLEXLM_DIAGNOSTICS=2
export FNP_IP_ENV=1
export LM_A_CONN_TIMEOUT=99

cd \${SLURM_SUBMIT_DIR}

cwd=`pwd`
echo "running testreport like this in \${cwd}:"
echo "${runtestreport} -devel -match 10"
${runtestreport} -devel -match 10 ${emailaddress} >> $MYOUTPUT 2>&1

echo "running testreport like this in \${cwd}:"
echo "./testreport -clean"
./testreport -clean

echo "running testreport like this in \${cwd}:"
echo "${runtestreport} -fast -match 10"
${runtestreport} -fast -match 10 ${emailaddress} >> $MYOUTPUT 2>&1

EOF

chmod a+x $HERE/$RUNIT

echo " " >> $MYOUTPUT
echo "***********************************************************" >> $MYOUTPUT
echo "Submitting this job script:" >> $MYOUTPUT
echo "***********************************************************" >> $MYOUTPUT
cat $HERE/$RUNIT >> $MYOUTPUT
echo "***********************************************************" >> $MYOUTPUT
echo "end of job script" >> $MYOUTPUT
echo "***********************************************************" >> $MYOUTPUT
echo " " >> $MYOUTPUT

echo "sbatch $HERE/$RUNIT"
sbatch $HERE/$RUNIT

# # keep looking for the job in the job queues and wait until it has disappeared
# jobruns=somedummy
# while [ "${jobruns}"x != x ]
# do
#   sleep 200
#   jobruns=`squeue --noheader -u mlosch | grep "$JOBNAME" | awk '{print $1}'`
#   echo "waiting for job ${jobruns%% *} ($JOBNAME) to complete"
#   currentexp=`grep Experiment $MYOUTPUT | tail -1`
#   echo "currently running $currentexp"
# done

# # workaround for mailing the stuff
# echo "mail the stuff"

# MPACKCMD=../tools/mpack-1.6/mpack
# fn=`ls -dtr tr_$dNam* | grep -v tar.gz | tail -1`
# echo "fname ${fn}"
# tar cf - $fn | gzip > "${fn}.tar.gz"
# $MPACKCMD -s MITgcm-test -m 3555000 ${fn}.tar.gz jm_c@mitgcm.org
# sleep 2
# rm -rf "${fn}.tar.gz"

echo "end of mitgcmtestreport"


1	mlosch	1.1	#!/bin/bash -e
2			# new script for running testreport on ollie.awi.de
3	mlosch	1.3	# - compile and run on compute nodes
4			# - use ssh to call mpack command from the head node ollie0
5	mlosch	1.8	#$Header: /u/gcmpack/MITgcm_contrib/test_scripts/ollie/mitgcmtestreport_ifort,v 1.7 2018/02/13 15:47:57 mlosch Exp $
6	mlosch	1.2	#$Name: $
7	mlosch	1.1
8			# needed for cron-job
9			source /usr/Modules/3.2.10/init/bash
10			source /etc/profile.d/cray_pe.sh
11			source /etc/profile.d/AWI.sh
12			#
13			module purge
14			module load intel.compiler
15			module load intel.mpi
16	mlosch	1.5	module load netcdf
17			#export LD_LIBRARY_PATH=${CRAY_LD_LIBRARY_PATH}:${LD_LIBRARY_PATH}:
18	mlosch	1.1
19	mlosch	1.5	export MPI_ROOT=${I_MPI_ROOT}
20			#(dirname $(dirname `which mpiifort`))
21	mlosch	1.1	export MPI_INC_DIR=${MPI_ROOT}/include64
22
23	mlosch	1.5	# there is no slurm module anymore and this is the current recommendation to
24			# have sbatch in your path (rather than running /etc/profile.d./slurm.sh)
25			export PATH=${PATH}:/global/opt/slurm/default/bin
26
27	mlosch	1.1	dNam='ollie'
28			TST_DIR="/work/ollie/mlosch/test_$dNam"
29			echo "start from TST_DIR='$TST_DIR' at: "`date`
30
31			umask 0022
32
33			sfx='ifort'
34			RUNIT="runit_"$sfx
35
36	mlosch	1.2	OPTFILE=../tools/build_options/linux_ia64_${sfx}_ollie
37	mlosch	1.1	options="-MPI 6"
38	mlosch	1.2	options="$options -odir ${dNam}-i"
39	mlosch	1.1	options="$options -j 6"
40	mlosch	1.8	#options="$options -t global_ocean.cs32x15"
41	mlosch	1.3
42	mlosch	1.6	#EXE='mpiexec.hydra -bootstrap slurm -n TR_NPROC ./mitgcmuv'
43	mlosch	1.1	#EXE='srun --mpi=pmi2 -n TR_NPROC ./mitgcmuv'
44	mlosch	1.6	EXE='srun -n TR_NPROC ./mitgcmuv'
45	mlosch	1.1
46			if [ -e $TST_DIR ]; then
47			echo "$TST_DIR exists"
48			else
49			mkdir $TST_DIR
50			fi
51			cd $TST_DIR
52			HERE=$TST_DIR/output
53			if [ -e $HERE ]; then
54			echo "$HERE"
55			else
56			mkdir $HERE
57			fi
58	mlosch	1.2	OUTFILE=$HERE/slurm_${sfx}.out
59	mlosch	1.1	MYOUTPUT=$HERE/out_$sfx
60			if [ -e $MYOUTPUT ]; then
61			rm -rf $MYOUTPUT
62			fi
63			if [ -e $OUTFILE ]; then
64			rm -r $OUTFILE
65			fi
66	mlosch	1.7	gcmDIR="MITgcm_${sfx}"
67	mlosch	1.6	git_repo='MITgcm'
68			git_code='MITgcm'
69
70			# checkOut determines how much checking out is being done
71			# checkOut = 3: new clone from GitHub and make a new copy
72			# checkOut = 2: update (git pull) existing repo and make a new copy
73			# checkOut = 1: skip update
74			# checkOut = 0: use existing test code (if available otherwise switch to 1)
75
76			checkOut=2
77			tdir=${TST_DIR}
78			today=`date +%Y%m%d`
79			TODAY=`date +%d`
80			#tmpFil="/tmp/"`basename $0`".$$"
81			tmpFil=$TST_DIR/error.out
82
83			if [ $checkOut -le 1 ] ; then
84			if test -e $TST_DIR/${gcmDIR}/doc ; then
85			echo $TST_DIR/${gcmDIR}/doc 'exist'
86			else
87			echo -n $TST_DIR/${gcmDIR} 'missing ; '
88			checkOut=2
89			echo "will make a new copy ( checkOut=$checkOut )"
90	mlosch	1.1	fi
91	mlosch	1.6	fi
92
93			if [ $checkOut -ge 2 ] ; then
94			#---- cleaning:
95			cd $TST_DIR
96
97			#---- Make a new clone or update existing one:
98			if test -e ${gcmDIR}/.git/config ; then
99			echo "${gcmDIR}/.git/config exist"
100			else
101	mlosch	1.8	echo -n "${gcmDIR}/.git/config 'missing "
102	mlosch	1.6	checkOut=3
103			echo "will get new clone ( checkOut=$checkOut )"
104			fi
105			if [ $checkOut -eq 3 ] ; then
106			echo -n "Removing old clone: $TST_DIR/${gcmDIR} ..."
107			test -e $TST_DIR/${gcmDIR} && rm -rf $TST_DIR/${gcmDIR}
108			echo " done"
109			echo -n "Make a new clone of $git_code from repo: $git_repo ..."
110			git clone https://github.com/$git_repo/${git_code}.git ${gcmDIR} 2> $tmpFil
111			retVal=$?
112			if test $retVal = 0 ; then
113			echo ' --> done!'
114			rm -f $tmpFil
115			else
116			echo " Error: 'git clone' returned: $retVal"
117			cat $tmpFil
118			rm -f $tmpFil
119			exit 2
120			fi
121			else
122			echo "Updating current clone ( $git_code ) ..."
123			( cd ${gcmDIR}; git checkout master ; git pull )
124			echo ' --> done!'
125	mlosch	1.1	fi
126	mlosch	1.6	else
127			cd $TST_DIR
128	mlosch	1.1	fi
129
130	mlosch	1.7	cd ${TST_DIR}/${gcmDIR}/verification
131	mlosch	1.1
132	mlosch	1.4	cwd=\`pwd\`
133	mlosch	1.3	SENDCMD="ssh ollie0 ${TST_DIR}/${gcmDIR}/tools/mpack-1.6/mpack"
134			runtestreport="./testreport $options -of $OPTFILE -command \"${EXE}\" -send \"${SENDCMD}\" -sd ${cwd}"
135	mlosch	1.8	emailaddress="-a jm_c@mitgcm.org"
136	mlosch	1.1
137			if [ ! -e $MYOUTPUT ]
138			then
139			touch $MYOUTPUT
140			fi
141
142	mlosch	1.2	# echo "running testreport like this:"
143			# echo ${runtestreport} -norun
144			# eval "${runtestreport} -norun >> $MYOUTPUT 2>&1"
145	mlosch	1.1
146			# create batch script
147			#
148	mlosch	1.2	JOBNAME=tst$sfx
149	mlosch	1.1	echo "creating batch script $HERE/$RUNIT"
150			cat << EOF >\| $HERE/$RUNIT
151	mlosch	1.3	#!/bin/bash
152	mlosch	1.1	#SBATCH --job-name=${JOBNAME}
153			#SBATCH -o ${OUTFILE}
154	mlosch	1.3	#SBATCH --time=6:00:00
155	mlosch	1.1	#SBATCH --ntasks=6
156
157	mlosch	1.3	# still need this to be able to run a few experiments:
158			# cfc_example, fizhi, tutorial_deep_convection
159			ulimit -s unlimited
160	mlosch	1.1
161	mlosch	1.6	# binds OpenMP task to given cores
162			export OMP_PROC_BIND=TRUE
163
164			# for debugging
165			export FLEXLM_DIAGNOSTICS=2
166			export FNP_IP_ENV=1
167			export LM_A_CONN_TIMEOUT=99
168
169	mlosch	1.1	cd \${SLURM_SUBMIT_DIR}
170
171	mlosch	1.3	cwd=`pwd`
172			echo "running testreport like this in \${cwd}:"
173			echo "${runtestreport} -devel -match 10"
174	mlosch	1.8	${runtestreport} -devel -match 10 ${emailaddress} >> $MYOUTPUT 2>&1
175	mlosch	1.3
176			echo "running testreport like this in \${cwd}:"
177			echo "./testreport -clean"
178			./testreport -clean
179
180			echo "running testreport like this in \${cwd}:"
181			echo "${runtestreport} -fast -match 10"
182	mlosch	1.8	${runtestreport} -fast -match 10 ${emailaddress} >> $MYOUTPUT 2>&1
183	mlosch	1.1
184			EOF
185
186			chmod a+x $HERE/$RUNIT
187
188			echo " " >> $MYOUTPUT
189			echo "***********************************************************" >> $MYOUTPUT
190			echo "Submitting this job script:" >> $MYOUTPUT
191			echo "***********************************************************" >> $MYOUTPUT
192			cat $HERE/$RUNIT >> $MYOUTPUT
193			echo "***********************************************************" >> $MYOUTPUT
194			echo "end of job script" >> $MYOUTPUT
195			echo "***********************************************************" >> $MYOUTPUT
196			echo " " >> $MYOUTPUT
197
198	mlosch	1.6	echo "sbatch $HERE/$RUNIT"
199	mlosch	1.1	sbatch $HERE/$RUNIT
200
201	mlosch	1.3	# # keep looking for the job in the job queues and wait until it has disappeared
202			# jobruns=somedummy
203			# while [ "${jobruns}"x != x ]
204			# do
205			# sleep 200
206			# jobruns=`squeue --noheader -u mlosch \| grep "$JOBNAME" \| awk '{print $1}'`
207			# echo "waiting for job ${jobruns%% *} ($JOBNAME) to complete"
208			# currentexp=`grep Experiment $MYOUTPUT \| tail -1`
209			# echo "currently running $currentexp"
210			# done
211
212			# # workaround for mailing the stuff
213			# echo "mail the stuff"
214
215			# MPACKCMD=../tools/mpack-1.6/mpack
216			# fn=`ls -dtr tr_$dNam* \| grep -v tar.gz \| tail -1`
217			# echo "fname ${fn}"
218			# tar cf - $fn \| gzip > "${fn}.tar.gz"
219	mlosch	1.6	# $MPACKCMD -s MITgcm-test -m 3555000 ${fn}.tar.gz jm_c@mitgcm.org
220	mlosch	1.3	# sleep 2
221			# rm -rf "${fn}.tar.gz"
222	mlosch	1.1
223			echo "end of mitgcmtestreport"
224
225
226
227
228