fix sync bug

Author: Qi Li (leeli4)

collect.sh

@@ -7,7 +7,6 @@ OUTPUT_FILE=$2
 DURATION=$3
 VERBOSE=${4:-0}
 SIGNAL_FILE=$5
-SIGNALBACK_FILE="/tmp/signalback/nso_collect_start_signalback_$$"
 STARTI=$6
 
 if [ -z "$PID" ] || [ -z "$OUTPUT_FILE" ] || [ -z "$DURATION" ]; then
@@ -36,25 +35,32 @@ mkdir -p "$OUTPUT_DIR"
 
 for (( i=$STARTI;i<=$DURATION;i++ ))
 do
-  
+  SIGNALBACK_FILE="/tmp/signalback/nso_collect_start_signalback_$$_$i"
+
   # Wait for centralized controller signal
   if [ ! -z "$SIGNAL_FILE" ]; then
     log_verbose "Waiting for start signal..."
     #echo "Waiting for start signal..."
+
     while true; do
-      if [ ! -f "$SIGNAL_FILE" ]; then
-        sleep 0.1
-      elif [[ $(cat $SIGNAL_FILE) -ne $i ]];then
-        sleep 0.1
+      if [ -f "$SIGNAL_FILE" ]; then
+        Data=$(cat $SIGNAL_FILE 2>/dev/null || echo 12345678999999  )
+        if  [[ $Data -eq $i ]]; then
+              break
+              log_verbose "Start signal received. Beginning data collection for PID $PID..."
+        else
+           sleep 0.1 
+        fi
       else
-        break
+         sleep 0.1
       fi
     done
-    
 
-    log_verbose "Start signal received. Beginning data collection for PID $PID..."
+  else 
+      echo "SIGNAL_FILE Parameter not provided"
   fi
-  rm -f $SIGNALBACK_FILE
+  
+  #rm -f $SIGNALBACK_FILE
   # Tick!
   ALO_TOTAL=$(cat /proc/meminfo | grep 'Committed_AS' | awk -F' ' '{print $2}')
   Limit=$(cat /proc/meminfo | grep 'CommitLimit' | awk -F' ' '{print $2}')
@@ -63,11 +69,14 @@ do
 
   log_verbose "Monitoring PID: $PID"
   ALO_PID=$(pmap -d $PID | grep "writeable/private" | awk -F' ' '{print $4}' | egrep -o '[0-9.]+'  )
-  PHY=$(cat /proc/$PID/status | grep VmRSS | awk -F' ' '{print $2}')
+  PHY=$(cat /proc/$PID/status 2>/dev/null || echo "" | grep VmRSS | awk -F' ' '{print $2}')
 
   if [ ! -z "$ALO_PID" ] && [ ! -z "$PHY" ]; then
     echo $TIME" "$PHY" "$ALO_PID" "$ALO_TOTAL" "$Limit  >> "$OUTPUT_FILE"
     log_verbose "$i second is collected to $OUTPUT_FILE"
+  else
+    echo $TIME" "0" "0" "$ALO_TOTAL" "$Limit  >> "$OUTPUT_FILE"
+    log_verbose "$i second is collected to $OUTPUT_FILE"
   fi
   touch $SIGNALBACK_FILE
 

plot.sh

@@ -76,7 +76,7 @@ echo "====================================== Collection for for all process ====
 SIGNAL_FILE="/tmp/nso_collect_start_signal_$$"
 rm -f "$SIGNAL_FILE"
 
-rm -rf "/tmp/signalback"
+rm -rf /tmp/signalback
 mkdir "/tmp/signalback"
 
 # Find and collect for each process type
@@ -132,6 +132,19 @@ do
 
 
 
+  sleep 0.1
+  # Signal all processes to start collecting
+  echo $i > $SIGNAL_FILE
+
+  while [[ $(ps -aux | grep "collect.sh" | wc -l) -gt $(($(ls "/tmp/signalback/" | grep "nso_collect_start_signalback_.*_$i" | wc -l)+1)) ]]; do
+      sleep 0.1
+    done
+
+  # Clean up signal file
+  rm -f $SIGNAL_FILE
+  rm -rf /tmp/signalback/*
+  
+
   END_TIME=$(date +%s%N)
   ELAPSED=$(($END_TIME - $START_TIME))
   SLEEP_TIME=$(($NS - $ELAPSED))
@@ -141,27 +154,16 @@ do
     sleep $SLEEP_SECONDS
   fi
 
-  # Signal all processes to start collecting
-  echo "$i" > $SIGNAL_FILE
-
-  while [[ $(ps -aux | grep "collect.sh" | wc -l) -gt $(($(ls "/tmp/signalback" | wc -l)+1)) ]]; do
-
-      sleep 0.1
-    done
-  # Clean up signal file
-  rm -f "$SIGNAL_FILE"
-
-  rm -rf "/tmp/signalback"
-  mkdir "/tmp/signalback"
-
   echo -ne "Data Collection - $i second out of $DURATION second"\\r
 
-
 done
 
+
 wait
+
+
 pkill -f collect.sh
-rm -rf "/tmp/signalback"
+rm -rf /tmp/signalback
 echo ""
 echo "Data Collection - OK!"