change to centralize controller design

Author: Qi Li (leeli4)

collect.sh

@@ -7,6 +7,8 @@ OUTPUT_FILE=$2
 DURATION=$3
 VERBOSE=${4:-0}
 SIGNAL_FILE=$5
+SIGNALBACK_FILE="/tmp/signalback/nso_collect_start_signalback_$$"
+
 
 if [ -z "$PID" ] || [ -z "$OUTPUT_FILE" ] || [ -z "$DURATION" ]; then
   echo "Usage: $0 <pid> <output_file> <duration> [verbose_flag] [signal_file]"
@@ -30,21 +32,30 @@ OUTPUT_DIR=$(dirname "$OUTPUT_FILE")
 mkdir -p "$OUTPUT_DIR"
 
 
-# Wait for all collection processes to be ready before starting
-if [ ! -z "$SIGNAL_FILE" ]; then
-  log_verbose "Waiting for start signal..."
-
-  while [ ! -f "$SIGNAL_FILE" ]; do
-    sleep 0.1
-  done
-
-  log_verbose "Start signal received. Beginning data collection for PID $PID..."
-fi
-
-# for (( i=0;i<=$DURATION;i++ ))
-# do
-#   START_TIME=$(date +%s%N)
 
+ 
+for (( i=0;i<=$DURATION;i++ ))
+do
+  
+  # Wait for centralized controller signal
+  if [ ! -z "$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
+      else
+        break
+      fi
+    done
+    
+
+    log_verbose "Start signal received. Beginning data collection for PID $PID..."
+  fi
+  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}')
 
@@ -58,15 +69,9 @@ fi
     echo $TIME" "$PHY" "$ALO_PID" "$ALO_TOTAL" "$Limit  >> "$OUTPUT_FILE"
     log_verbose "$i second is collected to $OUTPUT_FILE"
   fi
+  touch $SIGNALBACK_FILE
 
-  # END_TIME=$(date +%s%N)
-  # ELAPSED=$(($END_TIME - $START_TIME))
-  # SLEEP_TIME=$(($NS - $ELAPSED))
 
-#   if (( SLEEP_TIME > 0 )); then
-#     SLEEP_SECONDS=$(awk "BEGIN {printf \"%.9f\", $SLEEP_TIME/$NS}")
-#     sleep $SLEEP_SECONDS
-#   fi
-# done
+ done
 
-#echo "Collection for PID $PID done"
+log_verbose "Collection for PID $PID done"

compare_mem_alloc.plt

@@ -56,8 +56,8 @@ set format x '%H:%M:%S'
 
 show style line
 
-plot "data/python3/mem_ncs.smp.log" using 1:5 with lines axes x1y1 lc 'red' lw 2 title "CommitLimit", \
-     "data/python3/mem_ncs.smp.log" using 1:4 with lines axes x1y1 title "Commited_AS", \
+plot "data/ncs.smp/mem_ncs.smp.log" using 1:5 with lines axes x1y1 lc 'red' lw 2 title "CommitLimit", \
+     "data/ncs.smp/mem_ncs.smp.log" using 1:4 with lines axes x1y1 title "Commited_AS", \
      "data/ncs.smp/mem_ncs.smp.log" using 1:3 with lines axes x1y1 title "ncs.smp", \
      "data/python3/mem_total.log" using 1:3 with lines axes x1y1 title "PythonVM(Total)", \
      "data/NcsJVMLauncher/mem_NcsJVMLauncher.log" using 1:3 with lines axes x1y1 title "JavaVM"

plot.sh

@@ -76,33 +76,39 @@ echo "====================================== Collection for for all process ====
 SIGNAL_FILE="/tmp/nso_collect_start_signal_$$"
 rm -f "$SIGNAL_FILE"
 
+rm -rf "/tmp/signalback"
+mkdir "/tmp/signalback"
+
 # Find and collect for each process type
 echo "Starting collection processes..."
 
 NS=1000000000
+counter=0
 
 mkdir -p data/python3
 for (( i=0;i<=$DURATION;i++ ))
 do
   START_TIME=$(date +%s%N)
   PYTHON_PIDS=$(pgrep -f "python.* .*startup\.py")
-  JVM_PID=$(pgrep -f NcsJVMLauncher)
+  JVM_PID=$(pgrep -f com.tailf.ncs.NcsJVMLauncher)
   NCS_PID=$(pgrep -f "\.smp.*-ncs true")
 
   # Collect ncs.smp or beam.smp NSO process
   if [ ! -z "$NCS_PID" ]; then
-    #echo "Starting collection for ncs.smp PID $NCS_PID"
     COLLECT_PIDS=$(pgrep -f ".*collect.sh.* $NCS_PID")
     if [ -z "$COLLECT_PIDS" ]; then
+      echo "New ncs.smp process PID $NCS_PID: ncs.smp. Start Collection"
+      counter=$((counter+1))
       bash collect.sh $NCS_PID "data/ncs.smp/mem_ncs.smp.log" $DURATION $VERBOSE "$SIGNAL_FILE" &
     fi
   fi
 
   # Collect NcsJVMLauncher process
   if [ ! -z "$JVM_PID" ]; then
-    #echo "Starting collection for NcsJVMLauncher PID $JVM_PID"
     COLLECT_PIDS=$(pgrep -f ".*collect.sh.* $JVM_PID")
     if [ -z "$COLLECT_PIDS" ]; then
+      echo "New JVM process PID $JVM_PID: NcsJVMLauncher. Start Collection"
+      counter=$((counter+1))
       bash collect.sh $JVM_PID "data/NcsJVMLauncher/mem_NcsJVMLauncher.log" $DURATION $VERBOSE "$SIGNAL_FILE" &
     fi
   fi
@@ -112,19 +118,16 @@ do
     for pid in $PYTHON_PIDS; do
       COLLECT_PIDS=$(pgrep -f ".*collect.sh.* $pid")
       if [ -z "$COLLECT_PIDS" ]; then
-        #echo "Not Found Collection Process for Python process PID $pid. Spwaning new Collection Process."
         PYTHON_SCRIPT=$(ps -p $pid -o command | tail -n 1 | awk -F' ' '{print $9}')
         SCRIPT_NAME=$(basename "$PYTHON_SCRIPT" .py 2>/dev/null || echo "python_$pid")
         if [ ! -z "$PYTHON_SCRIPT" ]; then
-          #echo "Starting collection for Python process PID $pid: $SCRIPT_NAME"
+          echo "New Python process PID $pid: $SCRIPT_NAME. Start Collection"
+          counter=$((counter+1))
           bash collect.sh $pid "data/python3/mem_$SCRIPT_NAME.log" $DURATION-$i $VERBOSE "$SIGNAL_FILE" &
         fi
-      # else
-      #   echo "Collection Process already running for Python process PID $pid"
+
       fi
     done
-  # else
-  #   echo "No Python processes found to collect. for second $i"
   fi
 
 
@@ -139,17 +142,26 @@ do
   fi
 
   # Signal all processes to start collecting
-  touch "$SIGNAL_FILE"
-  wait
-  echo -ne "Data Collection - $i second out of $DURATION second"\\r
+  echo "$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"
-  pkill -f collect.sh
-#  # Give a moment for all processes to register
-#  sleep 1
+
+  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"
 echo ""
 echo "Data Collection - OK!"