| 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, 2000-2007
|
|---|
| 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 | echo "" >> $jmscriptlog 2>&1
|
|---|
| 37 | echo "starting jobmanager ("`date`")" >> $jmscriptlog 2>&1
|
|---|
| 38 |
|
|---|
| 39 | if [ "$1" == "" ]
|
|---|
| 40 | then
|
|---|
| 41 | echo "running jobmanager for data" >> $jmscriptlog 2>&1
|
|---|
| 42 | scripts=( ${datascripts[@]} )
|
|---|
| 43 | scriptscolname=( ${datascriptscolname[@]} )
|
|---|
| 44 | elif [ "$1" = "mc" ]
|
|---|
| 45 | then
|
|---|
| 46 | echo "running jobmanager for mc" >> $jmscriptlog 2>&1
|
|---|
| 47 | scripts=( ${mcscripts[@]} )
|
|---|
| 48 | scriptscolname=( ${mcscriptscolname[@]} )
|
|---|
| 49 | else
|
|---|
| 50 | echo "$1 is awrong commandline option for jobmanager -> exit" >> $jmscriptlog 2>&1
|
|---|
| 51 | printprocesslog "WARN $1 is wrong commandline option for jobmanager"
|
|---|
| 52 | finish >> $jmscriptlog 2>&1
|
|---|
| 53 | fi
|
|---|
| 54 |
|
|---|
| 55 | prev=$max
|
|---|
| 56 | user=`whoami`
|
|---|
| 57 | # endless loop
|
|---|
| 58 | notcount=0
|
|---|
| 59 | nothingtodocount=0
|
|---|
| 60 | nothingtodosleeptime=0
|
|---|
| 61 | while (( $notcount < 100 ))
|
|---|
| 62 | do
|
|---|
| 63 | for (( i=0 ; i < ${#scripts[@]} ; i++ ))
|
|---|
| 64 | do
|
|---|
| 65 | date >> $jmscriptlog 2>&1
|
|---|
| 66 | source `dirname $0`/sourcefile
|
|---|
| 67 | echo "script: ${scripts[$i]}" >> $jmscriptlog 2>&1
|
|---|
| 68 |
|
|---|
| 69 | # check if there's something to do
|
|---|
| 70 | getstatus >> $jmscriptlog 2>&1
|
|---|
| 71 | echo " $numproc ${scripts[$i]} still do to" >> $jmscriptlog 2>&1
|
|---|
| 72 | if [ "$numproc" = "" ]
|
|---|
| 73 | then
|
|---|
| 74 | prev=0
|
|---|
| 75 | nothingtodocount=`expr $nothingtodocount + 1`
|
|---|
| 76 | if [ $nothingtodocount -lt ${#scripts[@]} ]
|
|---|
| 77 | then
|
|---|
| 78 | cont >> $jmscriptlog 2>&1
|
|---|
| 79 | else
|
|---|
| 80 | if [ $nothingtodosleeptime -lt $sleeptimelimit ]
|
|---|
| 81 | then
|
|---|
| 82 | nothingtodosleeptime=`echo " $nothingtodocount * $sleeptime " | bc`
|
|---|
| 83 | fi
|
|---|
| 84 | echo "sleeping $nothingtodosleeptime" >> $jmscriptlog 2>&1
|
|---|
| 85 | sleep $nothingtodosleeptime
|
|---|
| 86 | cont >> $jmscriptlog 2>&1
|
|---|
| 87 | fi
|
|---|
| 88 | else
|
|---|
| 89 | nothingtodocount=0
|
|---|
| 90 | nothingtodosleeptime=0
|
|---|
| 91 | fi
|
|---|
| 92 |
|
|---|
| 93 | echo "sleeping $sleeptime..." >> $jmscriptlog 2>&1
|
|---|
| 94 | sleep $sleeptime
|
|---|
| 95 |
|
|---|
| 96 | # get processes in queue
|
|---|
| 97 | q=(`/usr/local/bin/condor_q -global -format "Owner%s " Owner -format "%s" CMD -format "Jobstatus%s\n" Jobstatus 2>&1 `)
|
|---|
| 98 | if echo $q | egrep \(Error\|failed\)
|
|---|
| 99 | then
|
|---|
| 100 | echo `date`" WARN condor_q failed" >> $jmscriptlog 2>&1
|
|---|
| 101 | printprocesslog "WARN condor_q failed"
|
|---|
| 102 | echo `date`" WARN condor_q failed" >> $jmerrorlog
|
|---|
| 103 | cont >> $jmscriptlog 2>&1
|
|---|
| 104 | fi
|
|---|
| 105 | # get processes of user in queue
|
|---|
| 106 | q1=(`echo ${q[@]} | egrep -o Owner$user`)
|
|---|
| 107 | queued=${#q1[@]}
|
|---|
| 108 | # get scripts in queue
|
|---|
| 109 | q2=(`echo ${q[@]} | egrep -o ${scripts[$i]}`)
|
|---|
| 110 | queuedscript=${#q2[@]}
|
|---|
| 111 | # get running scripts
|
|---|
| 112 | q3=(`echo ${q[@]} | egrep -o ${scripts[$i]}Jobstatus2`)
|
|---|
| 113 | runningscript=${#q3[@]}
|
|---|
| 114 | stillinqueue=`echo $queuedscript - $runningscript | bc `
|
|---|
| 115 |
|
|---|
| 116 | #get total number of allowed process for current time
|
|---|
| 117 | hour=`date +%k`
|
|---|
| 118 | totalpno=${pnototal[$hour]}
|
|---|
| 119 | #choose array according to the day of the week
|
|---|
| 120 | dayofweek=`date +%u`
|
|---|
| 121 | case $dayofweek in
|
|---|
| 122 | 0 | 6) pnos=( ${pnoswe[@]} ) ;;
|
|---|
| 123 | *) pnos=( ${pnosweek[@]} ) ;;
|
|---|
| 124 | esac
|
|---|
| 125 | # get number of allowed scripts for current time
|
|---|
| 126 | num=`echo "((( $i + 1 ) * 24 ) + ( $hour + 1 ) ) - 24 - 1 " | bc `
|
|---|
| 127 | pnoscript=${pnos[$num]}
|
|---|
| 128 | # if there was nothing to do for previous script, more scripts can be allowed
|
|---|
| 129 | if [ $prev -eq 0 ]
|
|---|
| 130 | then
|
|---|
| 131 | echo " prev=0 => resetting pnoscript [$pnoscript] to max [$max]" >> $jmscriptlog 2>&1
|
|---|
| 132 | pnoscript=$max
|
|---|
| 133 | fi
|
|---|
| 134 | echo " found $queued jobs in the queue (incl. running jobs) [allowed $totalpno]" >> $jmscriptlog 2>&1
|
|---|
| 135 | echo " found $queuedscript ${scripts[$i]} in the queue (incl. running jobs [$runningscript]) [allowed $pnoscript] - not running: $stillinqueue" >> $jmscriptlog 2>&1
|
|---|
| 136 |
|
|---|
| 137 | # continue if there are already enough processes or scripts in the queue
|
|---|
| 138 | if [ "$queued" -ge "$totalpno" ] || [ "$queuedscript" -ge "$pnoscript" ]
|
|---|
| 139 | then
|
|---|
| 140 | cont >> $jmscriptlog 2>&1
|
|---|
| 141 | fi
|
|---|
| 142 | # continue if the number of script is the queue is larger (or equal) than the number which still has to be done
|
|---|
| 143 | if [ $numproc -le $stillinqueue ]
|
|---|
| 144 | then
|
|---|
| 145 | echo " numproc($numproc) -le stillinqueue($stillinqueue)" >> $jmscriptlog 2>&1
|
|---|
| 146 | cont >> $jmscriptlog 2>&1
|
|---|
| 147 | fi
|
|---|
| 148 |
|
|---|
| 149 | # reset prev
|
|---|
| 150 | prev=$max
|
|---|
| 151 |
|
|---|
| 152 | # submit 1 script to condor
|
|---|
| 153 | date=`date +%Y-%m-%d`
|
|---|
| 154 | echo " committing 1 ${scripts[$i]} to condor" >> $jmscriptlog 2>&1
|
|---|
| 155 | if ! /usr/local/bin/condor_submit -a path=$scriptspath -a prog=${scripts[$i]} -a date=$date -a dir=$runlogpath $scriptspath/run.condor 2>> $jmerrorlog
|
|---|
| 156 | then
|
|---|
| 157 | echo `date`" WARN condor_submit failed" >> $jmerrorlog
|
|---|
| 158 | echo "condor is not working -> sleeping $errorsleeptime" >> $jmscriptlog 2>&1
|
|---|
| 159 | printprocesslog "WARN submitting ${scripts[$i]} to condor failed"
|
|---|
| 160 | if [ $errorsleeptime -lt $sleeptimelimit ]
|
|---|
| 161 | then
|
|---|
| 162 | errorsleeptime=`echo " $errorsleeptime + $errorsleeptimedefault " | bc`
|
|---|
| 163 | fi
|
|---|
| 164 | sleep $errorsleeptime
|
|---|
| 165 | else
|
|---|
| 166 | errorsleeptime=$errorsleeptimedefault
|
|---|
| 167 | fi
|
|---|
| 168 | date >> $jmscriptlog 2>&1
|
|---|
| 169 | echo "" >> $jmscriptlog 2>&1
|
|---|
| 170 | done
|
|---|
| 171 | done
|
|---|
| 172 |
|
|---|
