Changeset 1446 for palm/trunk/SCRIPTS/hlrn_watchdog

Timestamp:

Aug 7, 2014 10:08:56 AM (10 years ago)

Author:

maronga

Message:

removed blanks from <> strings in .mrun.config files / adapted hlrn_watchdog for HLRN-III

File:

• palm/trunk/SCRIPTS/hlrn_watchdog (modified) (8 diffs)

palm/trunk/SCRIPTS/hlrn_watchdog

@@ -1 +1 @@
 #!/bin/ksh
-
 #--------------------------------------------------------------------------------#
 # This file is part of PALM.
@@ -20 +19 @@
 # Current revisions:
 # -----------------
-#
+# Adapted for HLRN III. Added windows if no runs are queued.
+# 
 # Former revisions:
 # -----------------
@@ -41 +41 @@
 #------------------------------------------------------------------------------!
 
+    check_hannover=false
+    check_berlin=true
+    debug=true
+    PALM_BIN="/home/maronga/palm/tmp_version/trunk/SCRIPTS"
+
     cd $PALM_BIN
 
-    gate_h="hicegate.hlrn.de"
-    gate_b="bicegate.hlrn.de"
+    gate_h="hlogin.hlrn.de"
+    gate_b="blogin.hlrn.de"
     update_frequency=600
 
 #   trap strg+c
-    trap 'killall kdialog; exit' 2
+    trap 'kill -9 $infoPID > /dev/null; kill -9 $dialogPID > /dev/null; exit' 2
 
 
@@ -61 +66 @@
           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
@@ -73 +85 @@
            echo "+++ hlrn_watchdog is already running."
         else
-           nohup ./hlrn_watchdog $2 1> /dev/null 2> /dev/null &
+           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
-          ssh  $gate_h -l $1 "showq | egrep \"($1)\"" > .watchdog_report.x
+          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
@@ -104 +124 @@
           done
           rm .watchdog_report.x
+          touch .watchdog_status.x
 
 #         check for terminated jobs and status changes
@@ -138 +159 @@
           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 ))
@@ -162 +190 @@
 
 #         kill all windows
-          killall kdialog
+          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)" &
+          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" &
+             kdialog --msgbox "$info" --title "HLRN Job Information" & infoPID=$!
           fi