Changeset 9492 for trunk/MagicSoft/Mars/datacenter/scripts/jobmanager

Timestamp:

08/14/09 17:14:50 (15 years ago)

Author:

Daniela Dorner

Message:

*** empty log message ***

File:

• trunk/MagicSoft/Mars/datacenter/scripts/jobmanager (modified) (6 diffs)

trunk/MagicSoft/Mars/datacenter/scripts/jobmanager

@@ -50 +50 @@
 case $queuesys in
       sge)  echo " on queuing system '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]}` '
+            # (-hard) -l hostname=compute-*
+            #   for qstat this returns the jobs running on that node + all jobs in the queue
+            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]}` '
+#            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]}` '
 #            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 $scriptspath`/job.sge '
-            alias 'checkqueue'="/opt/gridengine/bin/lx26-amd64/qstat | awk ' { print \"Owner\"\$4\" \" \$3\"Jobstatus\"\$5 } '"
+            # FIXME: get complete scriptname (including command line option), needed for runstereo
+            alias 'checkqueue'="/opt/gridengine/bin/lx26-amd64/qstat \`echo \$noderequirementstat\`  | awk ' { print \"Owner\"\$4\" \" \$3\"Jobstatus\"\$5 } '"
             break
             ;;
@@ -69 +73 @@
 prev=$max
 user=`whoami`
+currentnode=$minnode
+numevaluated=0
 # endless loop
 notcount=0
@@ -83 +89 @@
       # check if there's something to do
       column=${scriptscolname[$i]}
-      getstatus >> $jmscriptlog 2>&1
+      getstepinfo
+      if [ "$noderestricted" = "yes" ]
+      then
+         # get number of next node 
+         if [ $numevaluated -ge $numrestrictedscripts ]
+         then 
+            currentnode=`echo $currentnode + 1 | bc -l`
+            numevaluated=1
+         else
+            numevaluated=`echo $numevaluated + 1 | bc -l`
+         fi
+         if [ $currentnode -gt $maxnode ]
+         then 
+            currentnode=$minnode
+         fi
+         # check if node is excluded
+         for excludednode in ${excludednodes[@]}
+         do
+            if [ $currentnode -eq $excludednode ]
+            then
+               echo `date +%F\ %T`" Node compute-0-$currentnode is currently excluded." >> $jmscriptlog 2>&1
+               continue 2
+            fi
+         done
+         # define requirement for submission 
+         # FIXME: currently only for sge at isdc
+         echo `date +%F\ %T`" Checking for node $currentnode. " >> $jmscriptlog 2>&1
+         noderequirementsub=" -hard -l hostname=compute-0-${currentnode}"
+         noderequirementstat=" -l hostname=compute-0-${currentnode}"
+         getstatus $currentnode >> $jmscriptlog 2>&1
+      else
+         noderequirementsub=""
+         noderequirementstat=""
+         getstatus >> $jmscriptlog 2>&1
+      fi
+      
+      # check number of processes to be done
       echo `date +%F\ %T`" Database: $numproc ${scripts[$i]} still to be done (incl. idle jobs) [DB/table/column $db/$table/$column]" >> $jmscriptlog 2>&1
       if [ "$numproc" = "" ]
@@ -105 +147 @@
 
       # 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=(`/opt/gridengine/bin/lx26-amd64/qstat | awk ' { print "Owner"$4" " $3"Jobstatus"$5 } ' 2>&1 `)
       q=(`checkqueue 2>&1 `)
       if echo $q | egrep \(Error\|failed\)
@@ -115 +155 @@
          nextscript sleeptime $sleeptime
       fi
+      # FIXME: sge cuts scriptname to 8 digits in qstat
       # get processes of user in queue
-      q1=(`echo ${q[@]} | egrep -o Owner$user`)
+      q1=( `echo ${q[@]} | egrep -o "Owner$user"`)
       queued=${#q1[@]}
       # get scripts in queue
-      q2=(`echo ${q[@]} | egrep -o ${scripts[$i]}`)
+      q2=( `echo ${q[@]} | egrep -o "${scripts[$i]}"`)
       queuedscript=${#q2[@]}
       # get running scripts
-      q3=( `echo ${q[@]} | egrep -o \(${scripts[$i]}Jobstatus2\|${scripts[$i]}Jobstatusr\)` )
+      q3=( `echo ${q[@]} | egrep -o \("${scripts[$i]}"Jobstatus2\|"${scripts[$i]}"Jobstatusr\)` )
       runningscript=${#q3[@]}
       stillinqueue=`echo $queuedscript - $runningscript | bc `
@@ -171 +212 @@
       date=`date +%Y-%m-%d`
       echo `date +%F\ %T`" 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