| 1 | #!/bin/sh
|
|---|
| 2 | #
|
|---|
| 3 | # ========================================================================
|
|---|
| 4 | #
|
|---|
| 5 | # *
|
|---|
| 6 | # * This file is part of MARS, the MAGIC Analysis and Reconstruction
|
|---|
| 7 | # * Software. It is distributed to you in the hope that it can be a useful
|
|---|
| 8 | # * and timesaving tool in analysing Data of imaging Cerenkov telescopes.
|
|---|
| 9 | # * It is distributed WITHOUT ANY WARRANTY.
|
|---|
| 10 | # *
|
|---|
| 11 | # * Permission to use, copy, modify and distribute this software and its
|
|---|
| 12 | # * documentation for any purpose is hereby granted without fee,
|
|---|
| 13 | # * provided that the above copyright notice appear in all copies and
|
|---|
| 14 | # * that both that copyright notice and this permission notice appear
|
|---|
| 15 | # * in supporting documentation. It is provided "as is" without express
|
|---|
| 16 | # * or implied warranty.
|
|---|
| 17 | # *
|
|---|
| 18 | #
|
|---|
| 19 | #
|
|---|
| 20 | # Author(s): Daniela Dorner 05/2006 <mailto:dorner@astro.uni-wuerzburg.de>
|
|---|
| 21 | #
|
|---|
| 22 | # Copyright: MAGIC Software Development, 2000-2010
|
|---|
| 23 | #
|
|---|
| 24 | #
|
|---|
| 25 | # ========================================================================
|
|---|
| 26 | #
|
|---|
| 27 | # This a script, which launches other scripts (all scripts, that are run
|
|---|
| 28 | # on primary basis)
|
|---|
| 29 | #
|
|---|
| 30 |
|
|---|
| 31 | source `dirname $0`/sourcefile
|
|---|
| 32 | printprocesslog "INFO starting $0"
|
|---|
| 33 |
|
|---|
| 34 | set -C
|
|---|
| 35 |
|
|---|
| 36 | # function to continue in loop and go to next script
|
|---|
| 37 | function nextscript()
|
|---|
| 38 | {
|
|---|
| 39 | echo `date +%F\ %T`" sleeping \$$1 = $2 seconds... " >> $jmscriptlog 2>&1
|
|---|
| 40 | sleep $2
|
|---|
| 41 | echo "" >> $jmscriptlog 2>&1
|
|---|
| 42 | continue
|
|---|
| 43 | }
|
|---|
| 44 |
|
|---|
| 45 | echo "" >> $jmscriptlog 2>&1
|
|---|
| 46 | echo "" >> $jmscriptlog 2>&1
|
|---|
| 47 | echo -n `date +%F\ %T`" starting jobmanager for setup "$AUTOMATIONSETUP >> $jmscriptlog 2>&1
|
|---|
| 48 |
|
|---|
| 49 | user=`whoami`
|
|---|
| 50 |
|
|---|
| 51 | # choose commands according to queueing system (defined in setup)
|
|---|
| 52 | case $queuesys in
|
|---|
| 53 | sge) echo " on queuing system 'sun grid engine'" >> $jmscriptlog 2>&1
|
|---|
| 54 | # (-hard) -l hostname=compute-*
|
|---|
| 55 | # for qstat this returns the jobs running on that node + all jobs in the queue
|
|---|
| 56 | alias 'queuesubmit'='/opt/gridengine/bin/lx26-amd64/qsub -b y -v AUTOMATIONSETUP=$AUTOMATIONSETUP -e `echo $runlogpath`/error-`echo $date`.log -o `echo $runlogpath`/log-`echo $date`.log `echo $noderequirementsub` `echo $scriptspath`/`echo ${scripts[$i]}` '
|
|---|
| 57 | # alias 'queuesubmit'='/opt/gridengine/bin/lx26-amd64/qsub -b y -v AUTOMATIONSETUP=$AUTOMATIONSETUP -e `echo $runlogpath`/error-`echo $date`.log -o `echo $runlogpath`/log-`echo $date`.log `echo $scriptspath`/`echo ${scripts[$i]}` '
|
|---|
| 58 | # alias 'queuesubmit'='/opt/gridengine/bin/lx26-amd64/qsub -sc runlogpath=`echo $runlogpath` -sc date=`echo $date` -sc scriptspath=`echo $scriptspath` -sc script=`echo ${scripts[$i]}` `echo $scriptspath`/job.sge '
|
|---|
| 59 | # FIXME: get complete scriptname (including command line option), needed for runstereo
|
|---|
| 60 | alias 'checkqueue'="/opt/gridengine/bin/lx26-amd64/qstat \`echo \$noderequirementstat\` | awk ' { print \"Owner\"\$4\" \" \$3\"Jobstatus\"\$5 } '"
|
|---|
| 61 | break
|
|---|
| 62 | ;;
|
|---|
| 63 | pbs) echo " on queuing system 'pbs'" >> $jmscriptlog 2>&1
|
|---|
| 64 | alias 'queuesubmit'='$pbspath/qsub -l walltime=$walltime -l pmem=$pmem -v AUTOMATIONSETUP=$AUTOMATIONSETUP,SOURCEFILEPATH=$SOURCEFILEPATH -e `echo $runlogpath`/error-`echo $date`.log -o `echo $runlogpath`/log-`echo $date`.log `echo $noderequirementsub` `echo $scriptspath`/`echo ${scripts[$i]}` '
|
|---|
| 65 | # check queue (restricted to current user only)
|
|---|
| 66 | alias 'checkqueue'="$pbspath/qstat -a -u $user | awk ' { print \"Owner\"\$2\" \" \$4\"Jobstatus\"\$10 } '"
|
|---|
| 67 | break
|
|---|
| 68 | ;;
|
|---|
| 69 | condor) echo " on queuing system 'condor'" >> $jmscriptlog 2>&1
|
|---|
| 70 | alias 'queuesubmit'='$condorpath/condor_submit -a path=`echo $scriptspath` -a prog=`echo ${scripts[$i]}` -a date=`echo $date` -a dir=`echo $runlogpath` -a automationsetup=$AUTOMATIONSETUP `echo $scriptspath`/run.condor'
|
|---|
| 71 | alias 'checkqueue'='$condorpath/condor_q -global -format "Owner%s " Owner -format "%s" CMD -format "Jobstatus%s\n" Jobstatus '
|
|---|
| 72 | break
|
|---|
| 73 | ;;
|
|---|
| 74 | *) echo "" >> $jmscriptlog 2>&1
|
|---|
| 75 | finish >> $jmscriptlog 2>&1
|
|---|
| 76 | ;;
|
|---|
| 77 | esac
|
|---|
| 78 |
|
|---|
| 79 | echo "" >> $jmscriptlog 2>&1
|
|---|
| 80 |
|
|---|
| 81 | prev=$max
|
|---|
| 82 | currentnode=$minnode
|
|---|
| 83 | numevaluated=0
|
|---|
| 84 | # endless loop
|
|---|
| 85 | notcount=0
|
|---|
| 86 | nothingtodocount=0
|
|---|
| 87 | nothingtodosleeptime=0
|
|---|
| 88 | errorsleeptime=$errorsleeptimedefault
|
|---|
| 89 | while (( $notcount < 100 ))
|
|---|
| 90 | do
|
|---|
| 91 | for (( i=0 ; i < ${#scripts[@]} ; i++ ))
|
|---|
| 92 | do
|
|---|
| 93 | source `dirname $0`/sourcefile
|
|---|
| 94 | echo `date +%F\ %T`" Evaluating processing status for script '${scripts[$i]}'" >> $jmscriptlog 2>&1
|
|---|
| 95 |
|
|---|
| 96 | # check if there's something to do
|
|---|
| 97 | step=${scriptscolname[$i]}
|
|---|
| 98 | getstepinfo
|
|---|
| 99 | # check if walltime has to be set
|
|---|
| 100 | if [ "$setwalltime" = "yes" ]
|
|---|
| 101 | then
|
|---|
| 102 | walltime=${walltimes[$i]}
|
|---|
| 103 | fi
|
|---|
| 104 | # check if memory has to be set
|
|---|
| 105 | if [ "$setpmem" = "yes" ]
|
|---|
| 106 | then
|
|---|
| 107 | pmem=${pmems[$i]}
|
|---|
| 108 | fi
|
|---|
| 109 | # check if the script is restricted to one node
|
|---|
| 110 | # (i.e. where output of previous step(s) is stored)
|
|---|
| 111 | # this information is taken from the steps.rc file
|
|---|
| 112 | if [ "$noderestricted" = "yes" ]
|
|---|
| 113 | then
|
|---|
| 114 | # get number of next node
|
|---|
| 115 | if [ $numevaluated -ge $numrestrictedscripts ]
|
|---|
| 116 | then
|
|---|
| 117 | currentnode=`echo $currentnode + 1 | bc -l`
|
|---|
| 118 | numevaluated=1
|
|---|
| 119 | else
|
|---|
| 120 | numevaluated=`echo $numevaluated + 1 | bc -l`
|
|---|
| 121 | fi
|
|---|
| 122 | if [ $currentnode -gt $maxnode ]
|
|---|
| 123 | then
|
|---|
| 124 | currentnode=$minnode
|
|---|
| 125 | fi
|
|---|
| 126 | # check if node is excluded
|
|---|
| 127 | for excludednode in ${excludednodes[@]}
|
|---|
| 128 | do
|
|---|
| 129 | if [ $currentnode -eq $excludednode ]
|
|---|
| 130 | then
|
|---|
| 131 | echo `date +%F\ %T`" Node compute-0-$currentnode is currently excluded." >> $jmscriptlog 2>&1
|
|---|
| 132 | continue 2
|
|---|
| 133 | fi
|
|---|
| 134 | done
|
|---|
| 135 | # define requirement for submission
|
|---|
| 136 | # FIXME: currently only for sge at isdc
|
|---|
| 137 | echo `date +%F\ %T`" Checking for node $currentnode. " >> $jmscriptlog 2>&1
|
|---|
| 138 | noderequirementsub=" -hard -l hostname=compute-0-${currentnode}"
|
|---|
| 139 | noderequirementstat=" -l hostname=compute-0-${currentnode}"
|
|---|
| 140 | getstatus $currentnode >> $jmscriptlog 2>&1
|
|---|
| 141 | else
|
|---|
| 142 | noderequirementsub=""
|
|---|
| 143 | noderequirementstat=""
|
|---|
| 144 | getstatus >> $jmscriptlog 2>&1
|
|---|
| 145 | fi
|
|---|
| 146 |
|
|---|
| 147 | # check number of processes to be done
|
|---|
| 148 | echo `date +%F\ %T`" Database: $numproc ${scripts[$i]} still to be done (incl. idle jobs) [DB/step $db/$step]" >> $jmscriptlog 2>&1
|
|---|
| 149 | if [ "$numproc" = "0" ]
|
|---|
| 150 | then
|
|---|
| 151 | prev=0
|
|---|
| 152 | nothingtodocount=`expr $nothingtodocount + 1`
|
|---|
| 153 | if [ $nothingtodocount -lt ${#scripts[@]} ]
|
|---|
| 154 | then
|
|---|
| 155 | nextscript 0 0
|
|---|
| 156 | else
|
|---|
| 157 | if [ $nothingtodosleeptime -lt $sleeptimelimit ]
|
|---|
| 158 | then
|
|---|
| 159 | nothingtodosleeptime=`echo " $nothingtodocount * $sleeptime " | bc`
|
|---|
| 160 | fi
|
|---|
| 161 | nextscript nothingtodosleeptime $nothingtodosleeptime
|
|---|
| 162 | fi
|
|---|
| 163 | else
|
|---|
| 164 | nothingtodocount=0
|
|---|
| 165 | nothingtodosleeptime=0
|
|---|
| 166 | fi
|
|---|
| 167 |
|
|---|
| 168 | # get processes in queue
|
|---|
| 169 | q=(`checkqueue 2>&1 `)
|
|---|
| 170 | if echo $q | egrep \(Error\|failed\)
|
|---|
| 171 | then
|
|---|
| 172 | echo `date +%F\ %T`" WARN checking query ($queuesys) failed" >> $jmscriptlog 2>&1
|
|---|
| 173 | printprocesslog "WARN checking query ($queuesys) failed"
|
|---|
| 174 | echo `date +%F\ %T`" WARN checking query ($queuesys) failed" >> $jmerrorlog
|
|---|
| 175 | nextscript sleeptime $sleeptime
|
|---|
| 176 | fi
|
|---|
| 177 | # FIXME: sge cuts scriptname to 8 digits in qstat
|
|---|
| 178 | # get processes of user in queue
|
|---|
| 179 | q1=( `echo ${q[@]} | egrep -o "Owner$user"`)
|
|---|
| 180 | queued=${#q1[@]}
|
|---|
| 181 | # get number of scripts in queue
|
|---|
| 182 | q2=( `echo ${q[@]} | egrep -o "${scripts[$i]}"`)
|
|---|
| 183 | queuedscript=${#q2[@]}
|
|---|
| 184 | # get running scripts
|
|---|
| 185 | # condor: 2
|
|---|
| 186 | # sge: r
|
|---|
| 187 | # pbs: R
|
|---|
| 188 | q3=( `echo ${q[@]} | egrep -o \("${scripts[$i]}"Jobstatus2\|"${scripts[$i]}"Jobstatusr\|"${scripts[$i]}"JobstatusR\)` )
|
|---|
| 189 | runningscript=${#q3[@]}
|
|---|
| 190 | stillinqueue=`echo $queuedscript - $runningscript | bc `
|
|---|
| 191 |
|
|---|
| 192 | #get total number of allowed process for current time
|
|---|
| 193 | hour=`date +%k`
|
|---|
| 194 | if [ ${pnototal[$hour]} -lt $totalmax ]
|
|---|
| 195 | then
|
|---|
| 196 | totalpno=${pnototal[$hour]}
|
|---|
| 197 | else
|
|---|
| 198 | totalpno=$totalmax
|
|---|
| 199 | fi
|
|---|
| 200 |
|
|---|
| 201 | #choose array according to the day of the week
|
|---|
| 202 | dayofweek=`date +%u`
|
|---|
| 203 | case $dayofweek in
|
|---|
| 204 | 0 | 6) pnos=( ${pnoswe[@]} ) ;;
|
|---|
| 205 | *) pnos=( ${pnosweek[@]} ) ;;
|
|---|
| 206 | esac
|
|---|
| 207 | # get number of allowed scripts for current time
|
|---|
| 208 | num=`echo "((( $i + 1 ) * 24 ) + ( $hour + 1 ) ) - 24 - 1 " | bc `
|
|---|
| 209 | pnoscript=${pnos[$num]}
|
|---|
| 210 | # if there was nothing to do for previous script, more scripts can be allowed
|
|---|
| 211 | if [ $prev -eq 0 ]
|
|---|
| 212 | then
|
|---|
| 213 | echo `date +%F\ %T`" \$prev=0 => resetting \$pnoscript from $pnoscript to $max [\$max]" >> $jmscriptlog 2>&1
|
|---|
| 214 | pnoscript=$max
|
|---|
| 215 | fi
|
|---|
| 216 | echo `date +%F\ %T`" queue for user '$user': total: $queued queued jobs [allowed \$totalpno = $totalpno]" >> $jmscriptlog 2>&1
|
|---|
| 217 | echo `date +%F\ %T`" queue for user '$user': ${scripts[$i]}: $queuedscript queued, $runningscript running, $stillinqueue idle [allowed \$pnoscript = $pnoscript]" >> $jmscriptlog 2>&1
|
|---|
| 218 |
|
|---|
| 219 | # continue if there are already enough processes or scripts in the queue
|
|---|
| 220 | if [ "$queued" -ge "$totalpno" ] || [ "$queuedscript" -ge "$pnoscript" ]
|
|---|
| 221 | then
|
|---|
| 222 | nextscript sleeptime $sleeptime
|
|---|
| 223 | fi
|
|---|
| 224 | # continue if the number of script is the queue is larger (or equal) than the number which still has to be done
|
|---|
| 225 | if [ $numproc -le $stillinqueue ]
|
|---|
| 226 | then
|
|---|
| 227 | echo `date +%F\ %T`" \$numproc ($numproc) < \$stillinqueue ($stillinqueue) " >> $jmscriptlog 2>&1
|
|---|
| 228 | nextscript sleeptime $sleeptime
|
|---|
| 229 | fi
|
|---|
| 230 |
|
|---|
| 231 | # reset prev
|
|---|
| 232 | prev=$max
|
|---|
| 233 |
|
|---|
| 234 | # submit 1 script to queuing system
|
|---|
| 235 | date=`date +%Y-%m-%d`
|
|---|
| 236 | echo `date +%F\ %T`" committing 1 ${scripts[$i]} to $queuesys" >> $jmscriptlog 2>&1
|
|---|
| 237 | if ! queuesubmit 2>> $jmerrorlog
|
|---|
| 238 | then
|
|---|
| 239 | echo `date`" WARN submitting job ($queuesys) failed" >> $jmerrorlog
|
|---|
| 240 | echo `date +%F\ %T`" WARN $queuesys is not working -> sleeping $errorsleeptime [\$errorsleeptime]" >> $jmscriptlog 2>&1
|
|---|
| 241 | printprocesslog "WARN submitting ${scripts[$i]} ($queuesys) failed"
|
|---|
| 242 | if [ $errorsleeptime -lt $sleeptimelimit ]
|
|---|
| 243 | then
|
|---|
| 244 | errorsleeptime=`echo " $errorsleeptime + $errorsleeptimedefault " | bc`
|
|---|
| 245 | fi
|
|---|
| 246 | nextscript errorsleeptime $errorsleeptime
|
|---|
| 247 | else
|
|---|
| 248 | errorsleeptime=$errorsleeptimedefault
|
|---|
| 249 | fi
|
|---|
| 250 | nextscript sleeptime $sleeptime
|
|---|
| 251 | done
|
|---|
| 252 | done
|
|---|
| 253 |
|
|---|
