bin/foamCheckJobs

   1 #!/bin/sh
   2 #---------------------------------*- sh -*-------------------------------------
   3 # =========                 |
   4 # \\      /  F ield         | OpenFOAM: The Open Source CFD Toolbox
   5 #  \\    /   O peration     |
   6 #   \\  /    A nd           | Copyright (C) 1991-2010 OpenCFD Ltd.
   7 #    \\/     M anipulation  |
   8 #------------------------------------------------------------------------------
   9 # License
  10 #     This file is part of OpenFOAM.
  11 #
  12 #     OpenFOAM is free software: you can redistribute it and/or modify it
  13 #     under the terms of the GNU General Public License as published by
  14 #     the Free Software Foundation, either version 3 of the License, or
  15 #     (at your option) any later version.
  16 #
  17 #     OpenFOAM is distributed in the hope that it will be useful, but WITHOUT
  18 #     ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  19 #     FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  20 #     for more details.
  21 #
  22 #     You should have received a copy of the GNU General Public License
  23 #     along with OpenFOAM.  If not, see <http://www.gnu.org/licenses/>.
  24 #
  25 # Script
  26 #     foamCheckJobs
  27 #
  28 # Description
  29 #     Uses runningJobs/, finishedJobs/ and foamProcessInfo to create stateFile.
  30 #     stateFile contains per pid information on state of process. Format:
  31 #       pid state command
  32 #
  33 #     where state is one of 'RUNN', 'SUSP', 'OTHR', 'FINI', 'ABRT' ('PEND')
  34 #     (first three are from foamProcessInfo, others from jobInfo files)
  35 #     (PEND is special state from when user has submitted but no jobInfo
  36 #     file yet. Not supported by this script yet)
  37 #
  38 #------------------------------------------------------------------------------
  39
  40 PROGNAME=${0##*/}
  41
  42 #-------------------------------------------------------------------------------
  43 #- User settings
  44
  45 #- Number of days for files to be considered old
  46 NDAYSLIMIT=7
  47 #-------------------------------------------------------------------------------
  48
  49 #- work file
  50 TMPFILE=/tmp/${PROGNAME}$$.tmp
  51 #- work dir. Needs to be accessible for all machines
  52 MACHDIR=$HOME/.OpenFOAM/${PROGNAME}
  53 DEFSTATEFILE=$HOME/.OpenFOAM/foamCheckJobs.out
  54
  55
  56 if [ `uname -s` = Linux ]
  57 then
  58     ECHO='echo -e'
  59 else
  60     ECHO='echo'
  61 fi
  62
  63
  64 #-------------------------------------------------------------------------------
  65 #
  66 # Functions
  67 #
  68 #-------------------------------------------------------------------------------
  69
  70 # getRawEntry dictionary entry
  71 # Prints value of dictionary entry
  72 getRawEntry() {
  73     grep -v '^//' $1 | grep "^[ \t]*$2 " | sed -e "s/^[ \t]*$2 [ ]*//"
  74 }
  75
  76 # getEntry dictionary entry
  77 # Like getRawEntry but strips " and ending ';'
  78 getEntry() {
  79     getRawEntry $1 $2 | sed -e 's/^"//'  -e 's/;$//' -e 's/"$//'
  80 }
  81
  82 # notEmpty directory
  83 # Returns 0 if directory contains files/directories
  84 notEmpty() {
  85     if [ "`ls $1`" ]; then
  86         return 0
  87     else
  88         return 1
  89     fi
  90 }
  91
  92 # dayDiff <date string 1>  <date string 2>
  93 # Prints number of days between the two
  94 # Eg. dayDiff "Jan 10 2002" "Dec 28 1999"
  95 #     ==> 13
  96 dayDiff() {
  97     date -d "$1" > /dev/null 2>&1
  98     if [ $? -ne 0 ]; then
  99         #- option '-d' on date not supported. Give up.
 100         echo "0"
 101     else
 102         year1=`echo "$1" | awk '{print $3}'`
 103         year2=`echo "$2" | awk '{print $3}'`
 104         day1=`date -d "$1" "+%j"`
 105         day2=`date -d "$2" "+%j"`
 106
 107         nYears=`expr $year1 - $year2`
 108         tmp1=`expr $nYears \* 365`
 109         tmp2=`expr $day1 - $day2`
 110         expr $tmp1 + $tmp2
 111     fi
 112 }
 113 #dayDiff  "`date '+%b %d %Y'`" "Dec 28 2001"
 114
 115
 116 # getAllJobs jobInfoDirectory
 117 # Prints list of all jobs in directory (e.g. runningJobs/)
 118 # Also handles 'slaves' entries in jobInfo:
 119 #   slaves           1 ( penfold.23766 );
 120 getAllJobs() {
 121     if notEmpty $1; then
 122         jobs=$1/*
 123         for f in $jobs
 124         do
 125             line=`grep '^[ ]*slaves' $f 2>/dev/null`
 126             if [ $? -eq 0 ]; then
 127                 slaveJobs=`echo "$line" | sed -e 's/.*(\(.*\)).*/\1/'`
 128                 jobs="$jobs $slaveJobs"
 129             fi
 130         done
 131     else
 132         jobs=''
 133     fi
 134     echo "$jobs"
 135 }
 136
 137 # releaseLock jobId lockFile
 138 # Releases lock on jobId
 139 releaseLock () {
 140     if [ -f $2 ]; then
 141         #- move lock to finishedJobs
 142         mv $2 $FOAM_JOB_DIR/finishedJobs/
 143     fi
 144     $ECHO "Lock on job $1 released."
 145 }
 146
 147
 148 printUsage() {
 149 cat << LABEL
 150 Usage: $PROGNAME [stateFile]
 151
 152 This program checks all the locks in the license directory to see if
 153 their processes are still running. Processes will not release their
 154 lock if they exit abnormally. This program will try to obtain process
 155 information on the machine the process ran on and release the lock
 156 if the program is no longer running.
 157
 158 Requirements: the environment variable FOAM_JOB_DIR needs to point to the
 159 license directory and all machines have to be reachable using ssh.
 160
 161 The output from checking all running jobs is collected in an optional
 162 file.
 163
 164 FILES:
 165     \$FOAM_JOB_DIR/runningJobs    locks for running processes
 166                   /finishedJobs     ,,      finished processes
 167 LABEL
 168 }
 169
 170
 171 #-------------------------------------------------------------------------------
 172 #
 173 # Main
 174 #
 175 #-------------------------------------------------------------------------------
 176
 177 #- Check a few things
 178
 179 if [ ! "$FOAM_JOB_DIR" ]; then
 180     $ECHO "$PROGNAME : FOAM_JOB_DIR environment variable not set."
 181     $ECHO "This should point to your central license directory."
 182     exit 1
 183 fi
 184
 185 if [ ! -d "$FOAM_JOB_DIR" ]; then
 186     $ECHO "$PROGNAME : The license directory accoring to FOAM_JOB_DIR is not valid."
 187     $ECHO "FOAM_JOB_DIR: $FOAM_JOB_DIR"
 188     exit 1
 189 fi
 190 if [ ! -d "$FOAM_JOB_DIR/runningJobs" -o ! -d "$FOAM_JOB_DIR/finishedJobs" ]; then
 191     $ECHO "$PROGNAME : The license directory according to FOAM_JOB_DIR is not valid."
 192     $ECHO "FOAM_JOB_DIR: $FOAM_JOB_DIR"
 193     exit 1
 194 fi
 195
 196
 197 if [ $# -eq 1 ]; then
 198     STATEFILE=$1
 199 elif [ $# -eq 0 ]; then
 200     STATEFILE=${STATEFILE:-$DEFSTATEFILE}
 201 else
 202     printUsage
 203     exit 1
 204 fi
 205
 206 #- obtain rsh method
 207 RSH='ssh'
 208 echo "Using remote shell type : $RSH"
 209
 210 echo ""
 211 echo "Collecting information on jobs in"
 212 echo "  $FOAM_JOB_DIR"
 213 echo ""
 214
 215
 216 #- Collect machine names into $TMPFILE
 217 # Also handles 'slaves' entry in jobInfo:
 218
 219 rm -f $TMPFILE; touch $TMPFILE
 220 RUNJOBS=`getAllJobs $FOAM_JOB_DIR/runningJobs`
 221 for f in $RUNJOBS
 222 do
 223     machinePid=`basename $f`
 224     machine=`echo $machinePid | sed -e 's/\.[0-9][0-9]*$//'`
 225     pid=`echo $machinePid | sed -e 's/.*\.\([0-9][0-9]*\)$/\1/'`
 226
 227     fgrep "$machine" $TMPFILE >/dev/null 2>&1
 228     if [ $? -ne 0 ]; then
 229         $ECHO "$machine" >> $TMPFILE
 230     fi
 231 done
 232 $ECHO "Found machines:"
 233 cat $TMPFILE
 234 $ECHO ""
 235
 236
 237
 238 #- Collect process info on all machines, one file per machine
 239
 240 mkdir -p $MACHDIR
 241 cnt=1
 242 while true
 243 do
 244     machine=`sed -n -e "${cnt}p" $TMPFILE`
 245     if [ ! "$machine" ]; then
 246         break
 247     fi
 248
 249     machFile=$MACHDIR/$machine
 250     rm -f $machFile
 251     $ECHO "Contacting $machine to collect process information:"
 252     if [ $machine = `hostname` ]; then
 253         $ECHO "  foamProcessInfo $machFile"
 254         foamProcessInfo $machFile >/dev/null 2>&1
 255     else
 256         $ECHO "  $RSH $machine foamProcessInfo $machFile"
 257         $RSH $machine foamProcessInfo $machFile >/dev/null 2>&1
 258     fi
 259     if [ $? -ne 0 -o ! -s $machFile ]; then
 260         $ECHO "** Failed collecting process information on $machine."
 261         $ECHO "Check $machFile and run foamProcessInfo by hand"
 262         rm -f $machFile
 263     else
 264         $ECHO "Succesfully collected information in $machFile ..."
 265     fi
 266
 267     cnt=`expr $cnt + 1`
 268 done
 269 $ECHO ""
 270
 271
 272 #- Construct state for runningJobs; move non runnning jobs to finishedJobs
 273
 274 releaseAll=''
 275 rm -f $STATEFILE
 276 for f in $RUNJOBS
 277 do
 278     machinePid=`basename $f`
 279     machine=`echo $machinePid | sed -e 's/\.[0-9][0-9]*$//'`
 280     pid=`echo $machinePid | sed -e 's/.*\.\([0-9][0-9]*\)$/\1/'`
 281
 282     machFile=$MACHDIR/$machine
 283     if [ -r $machFile ]; then
 284         entry=`grep "^$pid " $machFile 2>/dev/null`
 285         if [ $? -ne 0 -o ! "$entry" ]; then
 286             if [ "$releaseAll" ]; then
 287                 releaseLock $machinePid $f
 288             else
 289                 $ECHO "Job $machinePid seems to be no longer running. Release lock? (y/a)\c"
 290                 read answ
 291                 if [ "${answ:-y}" = 'y' ]; then
 292                     releaseLock $machinePid $f
 293                 elif [ "${answ:-y}" = 'a' ]; then
 294                     releaseAll='yes'
 295                     releaseLock $machinePid $f
 296                 else
 297                     state='OTHR'
 298                     $ECHO "$machinePid $state" >> $STATEFILE
 299                 fi
 300             fi
 301         else
 302             state=`echo "$entry" | awk '{print $2}'`
 303             $ECHO "$machinePid $state" >> $STATEFILE
 304         fi
 305     fi
 306 done
 307
 308
 309
 310 #- Collect old jobs in finishedJobs
 311
 312 OLDFILES=`find $FOAM_JOB_DIR/finishedJobs -mtime +$NDAYSLIMIT -print`
 313
 314 #- Construct state for finishedJobs and check on date of files.
 315
 316 if notEmpty $FOAM_JOB_DIR/finishedJobs; then
 317     dateNow=`date '+%b %d %Y'`
 318     for f in $FOAM_JOB_DIR/finishedJobs/*
 319     do
 320         sz=`ls -s $f | awk '{print $1}'`
 321         if [ "$sz" -gt 0 ]; then
 322             machinePid=`basename $f`
 323             machine=`echo $machinePid | sed -e 's/\.[0-9][0-9]*$//'`
 324             pid=`echo $machinePid | sed -e 's/.*\.\([0-9][0-9]*\)$/\1/'`
 325
 326             end=`getEntry $f endDate`
 327             if [ ! "$end" ]; then
 328                 state='ABRT'
 329             else
 330                 nDaysOld=`dayDiff  "$dateNow" "$end"`
 331                 if [ "$nDaysOld" -gt $NDAYSLIMIT ]; then
 332                     OLDFILES="$OLDFILES $f"
 333                 fi
 334
 335                 state='FINI'
 336             fi
 337
 338             $ECHO "$machinePid $state" >> $STATEFILE
 339         fi
 340     done
 341 fi
 342
 343
 344 #- Remove old locks
 345
 346 nOldFiles=`echo "$OLDFILES" | wc -w`
 347 if [ "$nOldFiles" -gt 0 ]; then
 348     $ECHO "You seem to have $nOldFiles locks older than $NDAYSLIMIT days in finishedJobs/"
 349     $ECHO "Do you want to remove these? (y)\c"
 350     read answ
 351     if [ "${answ:-y}" = 'y' ]; then
 352         rm -f $OLDFILES
 353     fi
 354 fi
 355
 356
 357 rm -f $TMPFILE
 358 rm -r $MACHDIR
 359
 360 $ECHO ""
 361 $ECHO "Updated stateFile:"
 362 $ECHO "  $STATEFILE"
 363 $ECHO ""
 364
 365 #------------------------------------------------------------------------------