| 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 $0"
|
|---|
| 33 |
|
|---|
| 34 | set -C
|
|---|
| 35 |
|
|---|
| 36 | # function to continue in loop and go to next script
|
|---|
| 37 | function sleepawhile()
|
|---|
| 38 | {
|
|---|
| 39 | usedsleeptime=$sleeptime
|
|---|
| 40 | case $1 in
|
|---|
| 41 | "error") if ! [ "$errorsleeptime" = "" ]
|
|---|
| 42 | then
|
|---|
| 43 | if [ $errorsleeptime -lt $sleeptimelimit ]
|
|---|
| 44 | then
|
|---|
| 45 | errorsleeptime=`echo " $errorsleeptime + $errorsleeptimedefault " | bc`
|
|---|
| 46 | fi
|
|---|
| 47 | usedsleeptime=$errorsleeptime
|
|---|
| 48 | fi
|
|---|
| 49 | ;;
|
|---|
| 50 | "ok") errorsleeptime=$errorsleeptimedefault
|
|---|
| 51 | ;;
|
|---|
| 52 | esac
|
|---|
| 53 | echo `date +%F\ %T`" sleeping "$usedsleeptime" seconds... (status: "$1")" >> $jmscriptlog 2>&1
|
|---|
| 54 | echo "" >> $jmscriptlog 2>&1
|
|---|
| 55 | sleep $usedsleeptime
|
|---|
| 56 | continue
|
|---|
| 57 | }
|
|---|
| 58 |
|
|---|
| 59 | echo "" >> $jmscriptlog 2>&1
|
|---|
| 60 | echo "" >> $jmscriptlog 2>&1
|
|---|
| 61 | echo -n `date +%F\ %T`" starting jobmanager for setup "$AUTOMATIONSETUP >> $jmscriptlog 2>&1
|
|---|
| 62 |
|
|---|
| 63 | user=`whoami`
|
|---|
| 64 |
|
|---|
| 65 | # choose commands according to queueing system (defined in setup)
|
|---|
| 66 | case $queuesys in
|
|---|
| 67 | sge) echo " on queuing system 'sun grid engine'" >> $jmscriptlog 2>&1
|
|---|
| 68 | # (-hard) -l hostname=compute-*
|
|---|
| 69 | # for qstat this returns the jobs running on that node + all jobs in the queue
|
|---|
| 70 | alias 'queuesubmit'='/opt/gridengine/bin/lx26-amd64/qsub -b y -t 1-`echo $tosubmit` -v AUTOMATIONSETUP=$AUTOMATIONSETUP -e `echo $runlogpath`/error-`echo $date`.log -o `echo $runlogpath`/log-`echo $date`.log `echo $noderequirementsub` `echo $scriptspath`/`echo ${scripts[$i]}` '
|
|---|
| 71 | # 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]}` '
|
|---|
| 72 | # 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 '
|
|---|
| 73 | # FIXME: get complete scriptname (including command line option), needed for runstereo
|
|---|
| 74 | alias 'checkqueue'="/opt/gridengine/bin/lx26-amd64/qstat \`echo \$noderequirementstat\` | awk ' { print \"Owner\"\$4\" \" \$3\"Jobstatus\"\$5 } '"
|
|---|
| 75 | break
|
|---|
| 76 | ;;
|
|---|
| 77 | pbs) echo " on queuing system 'pbs'" >> $jmscriptlog 2>&1
|
|---|
| 78 | alias 'queuesubmit'='$pbspath/qsub -t 1-`echo $tosubmit` -l walltime=`echo $walltime` -l pmem=`echo $pmem` -v AUTOMATIONSETUP=$AUTOMATIONSETUP,SOURCEFILEPATH=$SOURCEFILEPATH,SCRIPTNAME=`echo ${scripts[$i]}` -e `echo $runlogpath`/error-`echo $date`.log -o `echo $runlogpath`/log-`echo $date`.log `echo $noderequirementsub` `echo $scriptspath`/`echo ${scripts[$i]}` '
|
|---|
| 79 | # check queue (restricted to current user only)
|
|---|
| 80 | alias 'checkqueue'="$pbspath/qstat -a -u $user | awk ' { print \"Owner\"\$2\" \" \$4\"Jobstatus\"\$10 } '"
|
|---|
| 81 | break
|
|---|
| 82 | ;;
|
|---|
| 83 | condor) echo " on queuing system 'condor'" >> $jmscriptlog 2>&1
|
|---|
| 84 | alias 'queuesubmit'='$condorpath/condor_submit -a path=`echo $scriptspath` -a prog=`echo ${scripts[$i]}` -a date=`echo $date` -a dir=`echo $runlogpath` -a num=`echo $tosubmit` -a automationsetup=$AUTOMATIONSETUP `echo $scriptspath`/run.condor'
|
|---|
| 85 | alias 'checkqueue'='$condorpath/condor_q -global -format "Owner%s " Owner -format "%s" CMD -format "Jobstatus%s\n" Jobstatus '
|
|---|
| 86 | break
|
|---|
| 87 | ;;
|
|---|
| 88 | *) echo "" >> $jmscriptlog 2>&1
|
|---|
| 89 | finish >> $jmscriptlog 2>&1
|
|---|
| 90 | ;;
|
|---|
| 91 | esac
|
|---|
| 92 |
|
|---|
| 93 | echo "" >> $jmscriptlog 2>&1
|
|---|
| 94 |
|
|---|
| 95 | # for processing with local storage on different nodes
|
|---|
| 96 | currentnode=$minnode
|
|---|
| 97 | numevaluated=0
|
|---|
| 98 |
|
|---|
| 99 | # endless loop
|
|---|
| 100 | notcount=0
|
|---|
| 101 | errorsleeptime=$errorsleeptimedefault
|
|---|
| 102 | while (( $notcount < 100 ))
|
|---|
| 103 | do
|
|---|
| 104 | # get and set some information for the processing
|
|---|
| 105 | source `dirname $0`/sourcefile
|
|---|
| 106 | # reset some values
|
|---|
| 107 | tosubmit=0
|
|---|
| 108 | idleratio=0
|
|---|
| 109 | addtoscript=
|
|---|
| 110 |
|
|---|
| 111 | # get processes in queue
|
|---|
| 112 | q=(`checkqueue 2>&1 `)
|
|---|
| 113 | if echo $q | egrep \(Error\|failed\)
|
|---|
| 114 | then
|
|---|
| 115 | echo `date +%F\ %T`" WARN checking query ($queuesys) failed" >> $jmscriptlog 2>&1
|
|---|
| 116 | printprocesslog "WARN checking query ($queuesys) failed"
|
|---|
| 117 | echo `date +%F\ %T`" WARN checking query ($queuesys) failed" >> $jmerrorlog
|
|---|
| 118 | sleepawhile "error"
|
|---|
| 119 | fi
|
|---|
| 120 |
|
|---|
| 121 | # general check whether one should submit something depending on chosen algorithm
|
|---|
| 122 | # algorithm 1:
|
|---|
| 123 | # submit new jobs in case there are less than $limitidle idle jobs in the queue
|
|---|
| 124 | # algorithm 2:
|
|---|
| 125 | # submit new jobs in case the total number of jobs in the queue has fallen below $totalpno
|
|---|
| 126 | case $algorithm in
|
|---|
| 127 | 1) # algorithm 1
|
|---|
| 128 | # get number of idle jobs in the queue
|
|---|
| 129 | q5=( `echo ${q[@]} | egrep -o \(Jobstatus1\|Jobstatusq\|JobstatusQ\)` )
|
|---|
| 130 | idle=${#q5[@]}
|
|---|
| 131 | if [ $idle -gt $limitidle ]
|
|---|
| 132 | then
|
|---|
| 133 | echo `date +%F\ %T`" more than "$limitidle" jobs waiting ("$idle")" >> $jmscriptlog 2>&1
|
|---|
| 134 | sleepawhile "ok"
|
|---|
| 135 | fi
|
|---|
| 136 | ;;
|
|---|
| 137 | 2) # algorithm 2
|
|---|
| 138 | # get processes of user in queue
|
|---|
| 139 | q1=( `echo ${q[@]} | egrep -o "Owner$user"`)
|
|---|
| 140 | queued=${#q1[@]}
|
|---|
| 141 | hour=`date +%k`
|
|---|
| 142 | # choose array of total number of jobs to be done
|
|---|
| 143 | # according to the day of the week
|
|---|
| 144 | dayofweek=`date +%u`
|
|---|
| 145 | case $dayofweek in
|
|---|
| 146 | 0 | 6) totalpno=${pnototal[$hour]} ;;
|
|---|
| 147 | *) totalpno=${pnototalwe[$hour]} ;;
|
|---|
| 148 | esac
|
|---|
| 149 | # get total number of jobs to be submitted
|
|---|
| 150 | if [ $queued -gt $totalpno ]
|
|---|
| 151 | then
|
|---|
| 152 | echo `date +%F\ %T`" more than "$totalpno" jobs waiting ("$queued")" >> $jmscriptlog 2>&1
|
|---|
| 153 | sleepawhile "ok"
|
|---|
| 154 | else
|
|---|
| 155 | tosubmittotal=`echo "$totalpno - $queued" | bc -l`
|
|---|
| 156 | fi
|
|---|
| 157 | ;;
|
|---|
| 158 | *) echo "Please give an algorithm to calculate the number of allowed jobs."
|
|---|
| 159 | exit
|
|---|
| 160 | ;;
|
|---|
| 161 | esac
|
|---|
| 162 | echo `date +%F\ %T`" Total number of jobs to be submitted: "$tosubmittotal >> $jmscriptlog 2>&1
|
|---|
| 163 |
|
|---|
| 164 |
|
|---|
| 165 | # first loop to determine
|
|---|
| 166 | # a) how many jobs of this script have to be done
|
|---|
| 167 | # b) how many jobs of this script are running or queued
|
|---|
| 168 | for (( i=0 ; i < ${#scripts[@]} ; i++ ))
|
|---|
| 169 | do
|
|---|
| 170 | # set the step to be evaluated
|
|---|
| 171 | step=${scriptscolname[$i]}
|
|---|
| 172 | getstepinfo
|
|---|
| 173 |
|
|---|
| 174 | # check if the script is restricted to one node
|
|---|
| 175 | # (i.e. where output of previous step(s) is stored)
|
|---|
| 176 | # this information is taken from the steps.rc file
|
|---|
| 177 | # currently this is implemented for sge only
|
|---|
| 178 | # then get number of jobs to be done
|
|---|
| 179 | if [ "$noderestricted" = "yes" ]
|
|---|
| 180 | then
|
|---|
| 181 | # get number of next node
|
|---|
| 182 | if [ $numevaluated -ge $numrestrictedscripts ]
|
|---|
| 183 | then
|
|---|
| 184 | currentnode=`echo " $currentnode + 1 " | bc -l`
|
|---|
| 185 | numevaluated=1
|
|---|
| 186 | else
|
|---|
| 187 | numevaluated=`echo " $numevaluated + 1 " | bc -l`
|
|---|
| 188 | fi
|
|---|
| 189 | if [ $currentnode -gt $maxnode ]
|
|---|
| 190 | then
|
|---|
| 191 | currentnode=$minnode
|
|---|
| 192 | fi
|
|---|
| 193 | # check if node is excluded
|
|---|
| 194 | for excludednode in ${excludednodes[@]}
|
|---|
| 195 | do
|
|---|
| 196 | if [ $currentnode -eq $excludednode ]
|
|---|
| 197 | then
|
|---|
| 198 | echo `date +%F\ %T`" Node compute-0-$currentnode is currently excluded." >> $jmscriptlog 2>&1
|
|---|
| 199 | continue 2
|
|---|
| 200 | fi
|
|---|
| 201 | done
|
|---|
| 202 | # define requirement for submission
|
|---|
| 203 | # FIXME: currently only for sge at isdc
|
|---|
| 204 | echo `date +%F\ %T`" Checking for node $currentnode. " >> $jmscriptlog 2>&1
|
|---|
| 205 | noderequirementsub=" -hard -l hostname=compute-0-${currentnode}"
|
|---|
| 206 | noderequirementstat=" -l hostname=compute-0-${currentnode}"
|
|---|
| 207 | # get number of jobs to be done from the DB
|
|---|
| 208 | getstatus $currentnode >> $jmscriptlog 2>&1
|
|---|
| 209 | else
|
|---|
| 210 | noderequirementsub=""
|
|---|
| 211 | noderequirementstat=""
|
|---|
| 212 | # get number of jobs to be done from the DB
|
|---|
| 213 | getstatus >> $jmscriptlog 2>&1
|
|---|
| 214 | fi
|
|---|
| 215 | # store the number of processes to be done for this script
|
|---|
| 216 | todo[$i]=$numproc
|
|---|
| 217 | tododb[$i]=$numproc
|
|---|
| 218 |
|
|---|
| 219 | # FIXME: sge cuts scriptname to 8 digits in qstat
|
|---|
| 220 | # number of idle jobs, i.e. jobs waiting in the queue to run
|
|---|
| 221 | # condor: 1
|
|---|
| 222 | # sge: q
|
|---|
| 223 | # pbs: Q
|
|---|
| 224 | q4=( `echo ${q[@]} | egrep -o \("${scripts[$i]}"Jobstatus1\|"${scripts[$i]}"Jobstatusq\|"${scripts[$i]}"JobstatusQ\)` )
|
|---|
| 225 | idlescript[$i]=${#q4[@]}
|
|---|
| 226 |
|
|---|
| 227 | q2=( `echo ${q[@]} | egrep -o "${scripts[$i]}"`)
|
|---|
| 228 | queuedscript[$i]=${#q2[@]}
|
|---|
| 229 |
|
|---|
| 230 | stillfree[$i]=`echo "${maxjobs[$i]} - ${queuedscript[$i]} " | bc -l`
|
|---|
| 231 |
|
|---|
| 232 | if [ $numproc -eq 0 ] || [ ${todo[$i]} -le ${idlescript[$i]} ] || [ ${maxjobs[$i]} -le ${queuedscript[$i]} ]
|
|---|
| 233 | then
|
|---|
| 234 | # store the fraction of cpus to add it to another process
|
|---|
| 235 | idleratio=`echo " ${ratio[$i]} + $idleratio " | bc -l`
|
|---|
| 236 | ratio[$i]=0
|
|---|
| 237 | todo[$i]=0
|
|---|
| 238 | idlenum=$i
|
|---|
| 239 | continue
|
|---|
| 240 | fi
|
|---|
| 241 | done
|
|---|
| 242 | echo `date +%F\ %T`" Evaluated scripts: "${scripts[@]} >> $jmscriptlog 2>&1
|
|---|
| 243 | echo `date +%F\ %T`" Running scripts: "${queuedscript[@]}" (max: "${maxjobs[@]}")" >> $jmscriptlog 2>&1
|
|---|
| 244 | echo `date +%F\ %T`" Number of jobs to be done (from DB): "${tododb[@]} >> $jmscriptlog 2>&1
|
|---|
| 245 | echo `date +%F\ %T`" Number of jobs to be done (updated): "${todo[@]} >> $jmscriptlog 2>&1
|
|---|
| 246 | echo `date +%F\ %T`" Ratio: "${ratio[@]}" (idle: "$idleratio")" >> $jmscriptlog 2>&1
|
|---|
| 247 |
|
|---|
| 248 | # loop to update the ratio taking into account the ratio of
|
|---|
| 249 | # a) steps where nothing has to done
|
|---|
| 250 | # b) steps where already enough jobs are in the queue
|
|---|
| 251 | # sum up this idle ratio
|
|---|
| 252 | # determine for which step still most jobs have to be done
|
|---|
| 253 | if ! [ "$idleratio" = "0" ]
|
|---|
| 254 | then
|
|---|
| 255 | addtoscript=
|
|---|
| 256 | for (( i=0 ; i < ${#scripts[@]} ; i++ ))
|
|---|
| 257 | do
|
|---|
| 258 | if [ ${todo[$i]} -gt ${todo[$idlenum]} ] && [ ${todo[$i]} -gt 0 ]
|
|---|
| 259 | then
|
|---|
| 260 | if ! [ "$addtoscript" = "" ]
|
|---|
| 261 | then
|
|---|
| 262 | if [ ${todo[$i]} -lt ${todo[$addtoscript]} ]
|
|---|
| 263 | then
|
|---|
| 264 | continue
|
|---|
| 265 | fi
|
|---|
| 266 | fi
|
|---|
| 267 | addtoscript=$i
|
|---|
| 268 | fi
|
|---|
| 269 | done
|
|---|
| 270 |
|
|---|
| 271 | # continue in case nothing has to be done for all steps
|
|---|
| 272 | # else: update the ratio for the step where most jobs have to be done
|
|---|
| 273 | # by adding the idle ratio
|
|---|
| 274 | if [ "$addtoscript" = "" ]
|
|---|
| 275 | then
|
|---|
| 276 | echo `date +%F\ %T`" No jobs to be done for any step." >> $jmscriptlog 2>&1
|
|---|
| 277 | sleepawhile "ok"
|
|---|
| 278 | else
|
|---|
| 279 | ratio[$addtoscript]=`echo " ${ratio[$addtoscript]} + $idleratio " | bc -l`
|
|---|
| 280 | fi
|
|---|
| 281 | fi
|
|---|
| 282 | echo `date +%F\ %T`" Updated ratio: "${ratio[@]} >> $jmscriptlog 2>&1
|
|---|
| 283 |
|
|---|
| 284 |
|
|---|
| 285 | # loop to submit jobs to queueing system
|
|---|
| 286 | for (( i=0 ; i < ${#scripts[@]} ; i++ ))
|
|---|
| 287 | do
|
|---|
| 288 | # calculate number of jobs to be submitted
|
|---|
| 289 | tosubmit=`echo "scale=0; $tosubmittotal * ${ratio[$i]} / 1 " | bc -l`
|
|---|
| 290 | if [ ${todo[$i]} -lt $tosubmit ]
|
|---|
| 291 | then
|
|---|
| 292 | echo `date +%F\ %T`" Updating tosubmit for "${scripts[$i]}" from "$tosubmit" to "${todo[$i]} >> $jmscriptlog 2>&1
|
|---|
| 293 | tosubmit=${todo[$i]}
|
|---|
| 294 | fi
|
|---|
| 295 | if [ $tosubmit -eq 0 ]
|
|---|
| 296 | then
|
|---|
| 297 | echo `date +%F\ %T`" No jobs to be submitted for script '"${scripts[$i]}"'" >> $jmscriptlog 2>&1
|
|---|
| 298 | continue
|
|---|
| 299 | fi
|
|---|
| 300 | if [ $tosubmit -gt ${stillfree[$i]} ]
|
|---|
| 301 | then
|
|---|
| 302 | echo `date +%F\ %T`" Updating tosubmit for "${scripts[$i]}" from "$tosubmit" to "${stillfree[$i]} >> $jmscriptlog 2>&1
|
|---|
| 303 | tosubmit=${stillfree[$i]}
|
|---|
| 304 | fi
|
|---|
| 305 |
|
|---|
| 306 | # set the step to be evaluated
|
|---|
| 307 | step=${scriptscolname[$i]}
|
|---|
| 308 | # check if walltime has to be set
|
|---|
| 309 | if [ "$setwalltime" = "yes" ]
|
|---|
| 310 | then
|
|---|
| 311 | walltime=${walltimes[$i]}
|
|---|
| 312 | fi
|
|---|
| 313 | # check if memory has to be set
|
|---|
| 314 | if [ "$setpmem" = "yes" ]
|
|---|
| 315 | then
|
|---|
| 316 | pmem=${pmems[$i]}
|
|---|
| 317 | fi
|
|---|
| 318 |
|
|---|
| 319 | # submit $tosubmit scripts to queuing system
|
|---|
| 320 | echo `date +%F\ %T`" Submitting "$tosubmit" jobs for script '"${scripts[$i]}"' to "$queuesys >> $jmscriptlog 2>&1
|
|---|
| 321 | date=`date +%Y-%m-%d`
|
|---|
| 322 | if ! queuesubmit 2>> $jmerrorlog
|
|---|
| 323 | then
|
|---|
| 324 | echo `date`" WARN submitting job ($queuesys) failed" >> $jmerrorlog
|
|---|
| 325 | echo `date +%F\ %T`" WARN $queuesys is not working -> sleeping $errorsleeptime [\$errorsleeptime]" >> $jmscriptlog 2>&1
|
|---|
| 326 | printprocesslog "WARN submitting ${scripts[$i]} ($queuesys) failed"
|
|---|
| 327 | sleepawhile "error"
|
|---|
| 328 | fi
|
|---|
| 329 | done
|
|---|
| 330 | sleepawhile "ok"
|
|---|
| 331 | done
|
|---|
| 332 |
|
|---|
