Changeset 9129 for trunk/MagicSoft/Mars/datacenter/scripts/jobmanager

Timestamp:

08/21/08 18:49:35 (16 years ago)

Author:

Daniela Dorner

Message:

*** empty log message ***

File:

• trunk/MagicSoft/Mars/datacenter/scripts/jobmanager (modified) (5 diffs)

trunk/MagicSoft/Mars/datacenter/scripts/jobmanager

@@ -37 +37 @@
 echo "starting jobmanager ("`date`")" >> $jmscriptlog 2>&1
 
-if [ "$1" == "" ]
-then
-   echo "running jobmanager for data" >> $jmscriptlog 2>&1
-   scripts=( ${datascripts[@]} )
-   scriptscolname=( ${datascriptscolname[@]} )
-elif [ "$1" = "mc" ]
-then
-   echo "running jobmanager for mc" >> $jmscriptlog 2>&1
-   scripts=( ${mcscripts[@]} )
-   scriptscolname=( ${mcscriptscolname[@]} )
-else
-   echo "$1 is awrong commandline option for jobmanager -> exit" >> $jmscriptlog 2>&1
-   printprocesslog "WARN $1 is wrong commandline option for jobmanager"
-   finish >> $jmscriptlog 2>&1
-fi
+# decide which jobmanager you want to run
+# setup of the different jobmanagers (which scripts they start)
+#  the number of jobs are defined in the file setup
+case $1 in
+   data) echo "running jobmanager for data" >> $jmscriptlog 2>&1
+         scripts=( "runganymed" "runstar" "runcallisto" ) # not used: "dodatacheck" "cutslices"
+         scriptscolname=( "fGanymed" "fStar" "fCallisto" ) # not used: "fDataCheckDone" "fCompmux" 
+         pnosweek=( ${pnoganymed[@]} ${pnostar[@]} ${pnocallisto[@]} ) # not used: ${pnodatacheck[@]} ${pnocutslices[@]}
+         pnoswe=( ${pnoganymedwe[@]} ${pnostarwe[@]} ${pnocallistowe[@]} ) # not used: ${pnodatacheckwe[@]} ${pnocutsliceswe[@]}
+         break
+         ;;
+     mc) echo "running jobmanager for mc" >> $jmscriptlog 2>&1
+         scripts=( "runcorsika" "runreflector" "runcamera" )
+         scriptscolname=( "fCorsikaFileAvail" "fReflectorFileAvail" "fCameraFileAvail" )
+         pnosweek=( ${pnocorsika[@]} ${pnoreflector[@]} ${pnocamera[@]} )
+         pnoswe=( ${pnocorsikawe[@]} ${pnoreflectorwe[@]} ${pnocamerawe[@]} )
+         break
+         ;;
+      *) echo "$1 is a wrong commandline option for jobmanager -> exit" >> $jmscriptlog 2>&1
+         printprocesslog "WARN $1 is wrong commandline option for jobmanager"
+         finish >> $jmscriptlog 2>&1
+         break
+         ;;
+esac
+
+# choose commands according to queueing system (defined in setup)
+case $queuesys in
+      sge)  echo "setting commands for sun grid engine" >> $jmscriptlog 2>&1
+            alias 'queuesubmit'='/opt/gridengine/bin/lx26-amd64/qsub -b y -e `echo $runlogpath`/error-`echo $date`.log -o `echo $runlogpath`/log-`echo $date`.log `echo $scriptspath`/`echo ${scripts[$i]}` '
+#            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 '
+            alias 'checkqueue'="/opt/gridengine/bin/lx26-amd64/qstat | awk ' { print \"Owner\"\$4\" \" \$3\"Jobstatus\"\$5 } '"
+            break
+            ;;
+   condor)  echo "setting commands for condor" >> $jmscriptlog 2>&1
+            alias 'queuesubmit'='/usr/local/bin/condor_submit -a path=`echo $scriptspath` -a prog=`echo ${scripts[$i]}` -a date=`echo $date` -a dir=`echo $runlogpath` `echo $scriptspath`/run.condor'
+            alias 'checkqueue'='/usr/local/bin/condor_q -global -format "Owner%s " Owner -format "%s" CMD -format "Jobstatus%s\n" Jobstatus '
+            break
+            ;;
+        *)  finish >> $jmscriptlog 2>&1
+            ;;
+esac
 
 prev=$max
@@ -59 +85 @@
 nothingtodocount=0
 nothingtodosleeptime=0
+errorsleeptime=$errorsleeptimedefault
 while (( $notcount < 100 ))
 do
@@ -96 +123 @@
 
       # get processes in queue
-      q=(`/usr/local/bin/condor_q -global -format "Owner%s " Owner -format "%s" CMD -format "Jobstatus%s\n" Jobstatus 2>&1 `)
+#      q=(`/usr/local/bin/condor_q -global -format "Owner%s " Owner -format "%s" CMD -format "Jobstatus%s\n" Jobstatus 2>&1 `)
+#      q=(`/opt/gridengine/bin/lx26-amd64/qstat | awk ' { print "Owner"$4" " $3"Jobstatus"$5 } ' 2>&1 `)
+      q=(`checkqueue 2>&1 `)
       if echo $q | egrep \(Error\|failed\)
       then 
-         echo `date`" WARN condor_q failed" >> $jmscriptlog 2>&1
-         printprocesslog "WARN condor_q failed"
-         echo `date`" WARN condor_q failed" >> $jmerrorlog
+         echo `date`" WARN checking query ($queuesys) failed" >> $jmscriptlog 2>&1
+         printprocesslog "WARN checking query ($queuesys) failed"
+         echo `date`" WARN checking query ($queuesys) failed" >> $jmerrorlog
          cont >> $jmscriptlog 2>&1
       fi
@@ -111 +140 @@
       queuedscript=${#q2[@]}
       # get running scripts
-      q3=(`echo ${q[@]} | egrep -o ${scripts[$i]}Jobstatus2`)
+      q3=( `echo ${q[@]} | egrep -o \(${scripts[$i]}Jobstatus2\|${scripts[$i]}Jobstatusr\)` )
       runningscript=${#q3[@]}
       stillinqueue=`echo $queuedscript - $runningscript | bc `
@@ -151 +180 @@
       prev=$max
 
-      # submit 1 script to condor
+      # submit 1 script to queuing system
       date=`date +%Y-%m-%d`
-      echo " committing 1 ${scripts[$i]} to condor" >> $jmscriptlog 2>&1
-      if ! /usr/local/bin/condor_submit -a path=$scriptspath -a prog=${scripts[$i]} -a date=$date -a dir=$runlogpath $scriptspath/run.condor 2>> $jmerrorlog
-      then 
-         echo `date`" WARN condor_submit failed" >> $jmerrorlog
-         echo "condor is not working -> sleeping $errorsleeptime" >> $jmscriptlog 2>&1
-         printprocesslog "WARN submitting ${scripts[$i]} to condor failed"
+      echo " committing 1 ${scripts[$i]} to $queuesys" >> $jmscriptlog 2>&1
+#      if ! /usr/local/bin/condor_submit -a path=$scriptspath -a prog=${scripts[$i]} -a date=$date -a dir=$runlogpath $scriptspath/run.condor 2>> $jmerrorlog
+#      if ! /opt/gridengine/bin/lx26-amd64/qsub -e $runlogpath/error-$date.log -o $runlogpath/log-$(date).log 2>> $jmerrorlog
+      if ! queuesubmit 2>> $jmerrorlog
+      then 
+         echo `date`" WARN submitting job ($queuesys) failed" >> $jmerrorlog
+         echo "$queuesys is not working -> sleeping $errorsleeptime" >> $jmscriptlog 2>&1
+         printprocesslog "WARN submitting ${scripts[$i]} ($queuesys) failed"
          if [ $errorsleeptime -lt $sleeptimelimit ]
          then