source: palm/trunk/SCRIPTS/hlrn_watchdog @ 1448

#!/bin/ksh
#--------------------------------------------------------------------------------#
# This file is part of PALM.
#
# PALM is free software: you can redistribute it and/or modify it under the terms
# of the GNU General Public License as published by the Free Software Foundation,
# either version 3 of the License, or (at your option) any later version.
#
# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
# A PARTICULAR PURPOSE.  See the GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License along with
# PALM. If not, see <http://www.gnu.org/licenses/>.
#
# Copyright 1997-2014  Leibniz Universitaet Hannover
#--------------------------------------------------------------------------------#
#
# Current revisions:
# -----------------
# Bugfix: removed debug mode
# 
# Former revisions:
# -----------------
# $Id: hlrn_watchdog 1448 2014-08-07 10:47:05Z maronga $
#
# 1446 2014-08-07 10:08:56Z maronga
# Adapted for HLRN III. Added windows if no runs are queued.
# 
# 1046 2012-11-09 14:38:45Z maronga
# code put under GPL (PALM 3.9)
#
# 1029 2012-10-17 15:33:56Z maronga
# Initial revision
#
# Description:
# ------------
# The hlrn_watchdog works for jobs on HLRN in Hannover/Berlin and with KDE.
# It can be used for monitoring currently submitted jobs and will display
# Running, Idle and Blocked jobs in a separate window, which is updated every
# 10 minutes. 
# Starting: "hlrn_watchdog start <your hlrn username>"
# Stopping: "hlrn_watchdog stop"
#------------------------------------------------------------------------------!

    check_hannover=true
    check_berlin=true
    debug=false

    cd $PALM_BIN

    gate_h="hlogin.hlrn.de"
    gate_b="blogin.hlrn.de"
    update_frequency=600

#   trap strg+c
    trap 'kill -9 $infoPID > /dev/null; kill -9 $dialogPID > /dev/null; exit' 2


#   start/stop routine
    if [[ $1 == "stop" ]]
    then
       result=`ps aux|grep -c "./hlrn_watchdog"`
       if (( $result > 2 ))
       then
          result=`ps aux|grep -m1 "./hlrn_watchdog"`
          killid=`echo $result | tr -s " " | cut -d" " -s -f2`
          kill -9 $killid > /dev/null
          killall kdialog

          if [[ -f .watchdog_report.x ]] then
             rm .watchdog_report.x
          fi
          if [[ -f .watchdog_status.x ]] then
             rm .watchdog_status.x
          fi
          echo "*** hlrn_watchdog stopped."
       else
          echo "+++ hlrn_watchdog is not running."
       fi
       exit
    elif [[ $1 == "start" ]]
    then
        result=`ps aux|grep -c "hlrn_watchdog"`
        if (( $result > 2 ))
        then
           echo "+++ hlrn_watchdog is already running."
        else
           if [[ $debug = true ]] then
              ./hlrn_watchdog $2 &
           else
              nohup ./hlrn_watchdog $2 1> /dev/null 2> /dev/null &
           fi
           echo "\n*** hlrn_watchdog starting..."
        fi
        exit
    else
#      login via ssh and collect information in .watchdog_report.x
       while true
       do
          touch .watchdog_report.x
          if [[ $check_hannover == true ]] then
             ssh  $gate_h -l $1 "showq | egrep \"($1)\"" > .watchdog_report.x
          fi
          if [[ $check_berlin == true ]] then
          ssh  $gate_b -l $1 "showq | egrep \"($1)\"" >> .watchdog_report.x
          fi

          i=0
          j=0
          cat .watchdog_report.x|while read variable
          do
#            analyze output
             comid[$i]=`echo $variable | tr -s " " | cut -d" " -s -f1`
             jobid[$i]=`echo ${comid[$i]} | tr -s " " | cut -d"." -s -f2`
             complex[$i]=`echo ${comid[$i]} | tr -s " " | cut -d"." -s -f1`
             username[$i]=`echo $variable | tr -s " " | cut -d" " -s -f2`
             status[$i]=`echo $variable | tr -s " " | cut -d" " -s -f3`
             nodes[$i]=`echo $variable | tr -s " " | cut -d" " -s -f4`
             walltime[$i]=`echo $variable | tr -s " " | cut -d" " -s -f5`  
             day[$i]=`echo $variable | tr -s " " | cut -d" " -s -f6`   
             month[$i]=`echo $variable | tr -s " " | cut -d" " -s -f7`   
             date[$i]=`echo $variable | tr -s " " | cut -d" " -s -f8`
             subtime[$i]=`echo $variable | tr -s " " | cut -d" " -s -f9`   
             ((i = i + 1))
          done
          rm .watchdog_report.x
          touch .watchdog_status.x

#         check for terminated jobs and status changes
          k=0
          cat .watchdog_status.x|while read variable
          do
#            analyze output
             old_comid[$k]=`echo $variable | tr -s " " | cut -d" " -s -f1`
             old_status[$k]=`echo $variable | tr -s " " | cut -d" " -s -f2` 
             ((k = k + 1))
          done
          rm .watchdog_status.x

          info=""
          for ((m=0;m<$k;m++))
          do
             found=0
             for ((n=0;n<$i;n++))
             do        
                if [[ ${old_comid[$m]} == ${comid[$n]} ]]
                then
                   if [[ ${old_status[$m]} != ${status[$n]} ]]
                   then
                       info="$info ${old_comid[$m]} has changed status from ${old_status[$m]} to ${status[$n]}.\n"
                   fi
                   found=1
                   break
                fi
             done
             if (( $found == 0 ))
             then
                info="$info ${old_comid[$m]} has been terminated (Status was ${old_status[$m]}).\n"
             fi
          done

#         check whether any jobs are queued
          touch .watchdog_report.x
          file_size=`ls -l .watchdog_report.x | tr -s " " | cut -d " " -f 5`
          if [[ $file_size == 0 ]] then
            printf "No jobs queued.\n" >> .watchdog_report.x
          fi

#         get estimated starting time for all idle jobs and write watchdog output in .watchdog_report.x
          while (( $j < $i ))
          do
             if [[ ${status[$j]} == "Idle" ]]
             then
                if [[ ${complex[$j]} == "hannover" ]]
                then
                   eststart[$j]=`ssh $gate_h -l $1 "showstart ${comid[$j]}|grep \"based start in\""`
                else
                   eststart[$j]=`ssh $gate_b -l $1 "showstart ${comid[$j]}|grep \"based start in\""`
                fi
                eststart[$j]="Start in: "`echo ${eststart[$j]} | tr -s " " | cut -d" " -s -f6`   
             else
                eststart[$j]=""
             fi

#            write final output line
             ((k = j + 1))
             printf "%-9s%8s%3s%04i%13s%11s%2s%-s%3s%-20s\n" "Job: $k:" "${status[$j]}" " @ " "${nodes[$j]}" " nodes, time:" "${walltime[$j]}" " (" "${comid[$j]}" "). " "${eststart[$j]}" >> .watchdog_report.x
             printf "${comid[$j]} ${status[$j]}\n" >> .watchdog_status.x
             ((j = j + 1))
          done

#         kill all windows
          if  [[ "$infoPID" -ne "" ]] then
             kill -9 $infoPID > /dev/null
             unset $infoPID
          fi
          if  [[ "$dialogPID" -ne "" ]] then
             kill -9 $dialogPID > /dev/null
             unset $dialogPID
          fi

          timestamp=`date`

#         create window and show information
          kdialog --textbox .watchdog_report.x 550 150 --title "HLRN watchdog  (last update: $timestamp)" & dialogPID=$!

#         in case of status changes and terminated jobs, inform the user
          if [[ $info != "" ]]
          then
             kdialog --msgbox "$info" --title "HLRN Job Information" & infoPID=$!
          fi

          sleep $update_frequency
       done

    fi