Annotation of /MITgcm_contrib/test_scripts/ollie/mitgcmtestreport_cray

#!/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_cray,v 1.7 2017/03/13 09:31:51 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 craype-broadwell
module load PrgEnv-cray/1.0.2_test
# module load PrgEnv-cray
# to me this looks like a terrible hack, but that is what we are supposed 
# to use now
# module unload craype-network-infiniband cudatoolkit mvapich2_cce
# module load craype-network-opa
# module load mvapich2_cce
module load netcdf
#export LD_LIBRARY_PATH=${CRAY_LD_LIBRARY_PATH}:${LD_LIBRARY_PATH}:
# not sure why I have to set these paths here again
export MPI_ROOT=$(dirname $(dirname `which mpicc`))
# $(dirname `echo $LD_LIBRARY_PATH  | awk -F: '{print $1}'`)
export MPI_INC_DIR=${MPI_ROOT}/include

# 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='cray'
RUNIT="runit_"$sfx

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

#EXE='srun --mpi=pmi2 -n TR_NPROC ./mitgcmuv'
#EXE='srun -n TR_NPROC ./mitgcmuv'
EXE='srun -n TR_NPROC --cpu_bind=cores ./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=12:00:00
#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	1.4	# - compile and run on compute nodes
4	mlosch	1.3	# - use ssh to call mpack command from the head node ollie0
5	mlosch	1.8	#$Header: /u/gcmpack/MITgcm_contrib/test_scripts/ollie/mitgcmtestreport_cray,v 1.7 2017/03/13 09:31:51 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 craype-broadwell
15	mlosch	1.8	module load PrgEnv-cray/1.0.2_test
16			# module load PrgEnv-cray
17	mlosch	1.7	# to me this looks like a terrible hack, but that is what we are supposed
18			# to use now
19	mlosch	1.8	# module unload craype-network-infiniband cudatoolkit mvapich2_cce
20			# module load craype-network-opa
21			# module load mvapich2_cce
22	mlosch	1.7	module load netcdf
23	mlosch	1.6	#export LD_LIBRARY_PATH=${CRAY_LD_LIBRARY_PATH}:${LD_LIBRARY_PATH}:
24	mlosch	1.8	# not sure why I have to set these paths here again
25	mlosch	1.7	export MPI_ROOT=$(dirname $(dirname `which mpicc`))
26			# $(dirname `echo $LD_LIBRARY_PATH \| awk -F: '{print $1}'`)
27	mlosch	1.6	export MPI_INC_DIR=${MPI_ROOT}/include
28
29	mlosch	1.7	# there is no slurm module anymore and this is the current recommendation to
30			# have sbatch in your path (rather than running /etc/profile.d./slurm.sh)
31			export PATH=${PATH}:/global/opt/slurm/default/bin
32	mlosch	1.1
33			dNam='ollie'
34			TST_DIR="/work/ollie/mlosch/test_$dNam"
35			echo "start from TST_DIR='$TST_DIR' at: "`date`
36
37			umask 0022
38
39	mlosch	1.2	sfx='cray'
40	mlosch	1.1	RUNIT="runit_"$sfx
41
42	mlosch	1.2	OPTFILE=../tools/build_options/linux_ia64_${sfx}_ollie
43	mlosch	1.1	options="-MPI 6"
44	mlosch	1.2	options="$options -odir ${dNam}-c"
45	mlosch	1.8	options="$options -j 6"
46	mlosch	1.3	#options="$options -t global_ocean.cs32x15"
47
48	mlosch	1.8	#EXE='srun --mpi=pmi2 -n TR_NPROC ./mitgcmuv'
49			#EXE='srun -n TR_NPROC ./mitgcmuv'
50			EXE='srun -n TR_NPROC --cpu_bind=cores ./mitgcmuv'
51	mlosch	1.1
52			if [ -e $TST_DIR ]; then
53			echo "$TST_DIR exists"
54			else
55			mkdir $TST_DIR
56			fi
57			cd $TST_DIR
58			HERE=$TST_DIR/output
59			if [ -e $HERE ]; then
60			echo "$HERE"
61			else
62			mkdir $HERE
63			fi
64	mlosch	1.2	OUTFILE=$HERE/slurm_${sfx}.out
65	mlosch	1.1	MYOUTPUT=$HERE/out_$sfx
66			if [ -e $MYOUTPUT ]; then
67			rm -rf $MYOUTPUT
68			fi
69			if [ -e $OUTFILE ]; then
70			rm -r $OUTFILE
71			fi
72	mlosch	1.8	gcmDIR="MITgcm_${sfx}"
73			git_repo='MITgcm'
74			git_code='MITgcm'
75
76			# checkOut determines how much checking out is being done
77			# checkOut = 3: new clone from GitHub and make a new copy
78			# checkOut = 2: update (git pull) existing repo and make a new copy
79			# checkOut = 1: skip update
80			# checkOut = 0: use existing test code (if available otherwise switch to 1)
81
82			checkOut=2
83			tdir=${TST_DIR}
84			today=`date +%Y%m%d`
85			TODAY=`date +%d`
86			#tmpFil="/tmp/"`basename $0`".$$"
87			tmpFil=$TST_DIR/error.out
88
89			if [ $checkOut -le 1 ] ; then
90			if test -e $TST_DIR/${gcmDIR}/doc ; then
91			echo $TST_DIR/${gcmDIR}/doc 'exist'
92			else
93			echo -n $TST_DIR/${gcmDIR} 'missing ; '
94			checkOut=2
95			echo "will make a new copy ( checkOut=$checkOut )"
96	mlosch	1.1	fi
97	mlosch	1.8	fi
98
99			if [ $checkOut -ge 2 ] ; then
100			#---- cleaning:
101			cd $TST_DIR
102
103			#---- Make a new clone or update existing one:
104			if test -e ${gcmDIR}/.git/config ; then
105			echo "${gcmDIR}/.git/config exist"
106			else
107			echo -n "${gcmDIR}/.git/config 'missing "
108			checkOut=3
109			echo "will get new clone ( checkOut=$checkOut )"
110	mlosch	1.1	fi
111	mlosch	1.8	if [ $checkOut -eq 3 ] ; then
112			echo -n "Removing old clone: $TST_DIR/${gcmDIR} ..."
113			test -e $TST_DIR/${gcmDIR} && rm -rf $TST_DIR/${gcmDIR}
114			echo " done"
115			echo -n "Make a new clone of $git_code from repo: $git_repo ..."
116			git clone https://github.com/$git_repo/${git_code}.git ${gcmDIR} 2> $tmpFil
117			retVal=$?
118			if test $retVal = 0 ; then
119			echo ' --> done!'
120			rm -f $tmpFil
121			else
122			echo " Error: 'git clone' returned: $retVal"
123			cat $tmpFil
124			rm -f $tmpFil
125			exit 2
126			fi
127			else
128			echo "Updating current clone ( $git_code ) ..."
129			( cd ${gcmDIR}; git checkout master ; git pull )
130			echo ' --> done!'
131			fi
132			else
133			cd $TST_DIR
134	mlosch	1.1	fi
135
136	mlosch	1.8	cd ${TST_DIR}/${gcmDIR}/verification
137	mlosch	1.1
138	mlosch	1.6	cwd=\`pwd\`
139	mlosch	1.3	SENDCMD="ssh ollie0 ${TST_DIR}/${gcmDIR}/tools/mpack-1.6/mpack"
140			runtestreport="./testreport $options -of $OPTFILE -command \"${EXE}\" -send \"${SENDCMD}\" -sd ${cwd}"
141	mlosch	1.8	emailaddress="-a jm_c@mitgcm.org"
142	mlosch	1.1
143			if [ ! -e $MYOUTPUT ]
144			then
145			touch $MYOUTPUT
146			fi
147
148	mlosch	1.2	# echo "running testreport like this:"
149			# echo ${runtestreport} -norun
150			# eval "${runtestreport} -norun >> $MYOUTPUT 2>&1"
151	mlosch	1.1
152			# create batch script
153			#
154	mlosch	1.2	JOBNAME=tst$sfx
155	mlosch	1.1	echo "creating batch script $HERE/$RUNIT"
156			cat << EOF >\| $HERE/$RUNIT
157	mlosch	1.3	#!/bin/bash
158	mlosch	1.1	#SBATCH --job-name=${JOBNAME}
159			#SBATCH -o ${OUTFILE}
160	mlosch	1.3	#SBATCH --time=12:00:00
161	mlosch	1.1	#SBATCH --ntasks=6
162
163	mlosch	1.3	# still need this to be able to run a few experiments:
164			# cfc_example, fizhi, tutorial_deep_convection
165			ulimit -s unlimited
166	mlosch	1.1
167	mlosch	1.8	# binds OpenMP task to given cores
168			export OMP_PROC_BIND=TRUE
169
170			# for debugging
171	mlosch	1.7	# export FLEXLM_DIAGNOSTICS=2
172			# export FNP_IP_ENV=1
173			# export LM_A_CONN_TIMEOUT=99
174
175	mlosch	1.1	cd \${SLURM_SUBMIT_DIR}
176
177	mlosch	1.3	cwd=`pwd`
178			echo "running testreport like this in \${cwd}:"
179			echo "${runtestreport} -devel -match 10"
180	mlosch	1.8	${runtestreport} -devel -match 10 ${emailaddress} >> $MYOUTPUT 2>&1
181	mlosch	1.3
182			echo "running testreport like this in \${cwd}:"
183			echo "./testreport -clean"
184			./testreport -clean
185
186			echo "running testreport like this in \${cwd}:"
187			echo "${runtestreport} -fast -match 10"
188	mlosch	1.8	${runtestreport} -fast -match 10 ${emailaddress} >> $MYOUTPUT 2>&1
189	mlosch	1.1
190			EOF
191
192			chmod a+x $HERE/$RUNIT
193
194			echo " " >> $MYOUTPUT
195			echo "***********************************************************" >> $MYOUTPUT
196			echo "Submitting this job script:" >> $MYOUTPUT
197			echo "***********************************************************" >> $MYOUTPUT
198			cat $HERE/$RUNIT >> $MYOUTPUT
199			echo "***********************************************************" >> $MYOUTPUT
200			echo "end of job script" >> $MYOUTPUT
201			echo "***********************************************************" >> $MYOUTPUT
202			echo " " >> $MYOUTPUT
203
204	mlosch	1.8	echo "sbatch $HERE/$RUNIT"
205	mlosch	1.1	sbatch $HERE/$RUNIT
206
207	mlosch	1.3	# # keep looking for the job in the job queues and wait until it has disappeared
208			# jobruns=somedummy
209			# while [ "${jobruns}"x != x ]
210			# do
211			# sleep 200
212			# jobruns=`squeue --noheader -u mlosch \| grep "$JOBNAME" \| awk '{print $1}'`
213			# echo "waiting for job ${jobruns%% *} ($JOBNAME) to complete"
214			# currentexp=`grep Experiment $MYOUTPUT \| tail -1`
215			# echo "currently running $currentexp"
216			# done
217
218			# # workaround for mailing the stuff
219			# echo "mail the stuff"
220
221			# MPACKCMD=../tools/mpack-1.6/mpack
222			# fn=`ls -dtr tr_$dNam* \| grep -v tar.gz \| tail -1`
223			# echo "fname ${fn}"
224			# tar cf - $fn \| gzip > "${fn}.tar.gz"
225	mlosch	1.8	# $MPACKCMD -s MITgcm-test -m 3555000 ${fn}.tar.gz jm_c@mitgcm.org
226	mlosch	1.3	# sleep 2
227			# rm -rf "${fn}.tar.gz"
228	mlosch	1.1
229			echo "end of mitgcmtestreport"
230
231
232
233
234