monitor_java_cpu.sh

#!/bin/bash

# Configuration
JAVA_PID=$1
CPU_THRESHOLD=${2:-50}      # CPU threshold percentage, default: 50%
CHECK_INTERVAL=${3:-5}      # Check interval in seconds, default: 5s
TOP_THREADS=${4:-10}        # Number of top CPU threads, default: 10
STACK_LINES=${5:-20}        # Stack trace lines per thread, default: 20
SNAPSHOT_DIR="./thread_snapshots"

if [ -z "$JAVA_PID" ]; then
    echo "Usage: $0 <Java_PID> [CPU_Threshold] [Check_Interval] [Top_Threads] [Stack_Lines]"
    echo "Example: $0 12345 50 5 10 20"
    echo ""
    echo "Parameters:"
    echo "  Java_PID        - Java process ID (required)"
    echo "  CPU_Threshold   - CPU threshold percentage (default: 50)"
    echo "  Check_Interval  - Check interval in seconds (default: 5)"
    echo "  Top_Threads     - Number of top CPU threads to capture (default: 10)"
    echo "  Stack_Lines     - Number of stack trace lines per thread (default: 20)"
    exit 1
fi

mkdir -p "$SNAPSHOT_DIR"

echo "Monitoring Java process: $JAVA_PID"
echo "CPU Threshold: ${CPU_THRESHOLD}%"
echo "Check Interval: ${CHECK_INTERVAL}s"
echo "Top Threads: ${TOP_THREADS}"
echo "Stack Lines: ${STACK_LINES}"
echo "Snapshot Directory: $SNAPSHOT_DIR"
echo "Press Ctrl+C to stop"
echo "================================"

while true; do
    # Get process total CPU usage
    TOTAL_CPU=$(top -b -n 1 -p $JAVA_PID | awk -v pid=$JAVA_PID '$1 == pid {print $9}')

    if [ -z "$TOTAL_CPU" ]; then
        echo "[$(date '+%Y-%m-%d %H:%M:%S')] Process $JAVA_PID not found"
        exit 1
    fi

    # Convert to integer for comparison
    CPU_INT=${TOTAL_CPU%.*}

    echo "[$(date '+%Y-%m-%d %H:%M:%S')] CPU: ${TOTAL_CPU}%"

    if [ "$CPU_INT" -ge "$CPU_THRESHOLD" ]; then
        TIMESTAMP=$(date '+%Y%m%d_%H%M%S')
        SNAPSHOT_FILE="$SNAPSHOT_DIR/jstack_${TIMESTAMP}_cpu${TOTAL_CPU}.txt"
        THREAD_FILE="$SNAPSHOT_DIR/threads_${TIMESTAMP}_cpu${TOTAL_CPU}.txt"
        STATS_FILE="$SNAPSHOT_DIR/stats_${TIMESTAMP}_cpu${TOTAL_CPU}.txt"

        echo "⚠️  HIGH CPU DETECTED: ${TOTAL_CPU}% - Capturing snapshot..."

        # Save full thread dump
        jstack $JAVA_PID > "$SNAPSHOT_FILE"

        # Save top N high CPU threads detail
        {
            echo "=== High CPU Threads at $(date) ==="
            echo "Total Process CPU: ${TOTAL_CPU}%"
            echo "Top ${TOP_THREADS} threads by CPU usage"
            echo ""

            top -H -p $JAVA_PID -b -n 1 | awk 'NR>7 {print $1,$9}' | head -${TOP_THREADS} | while read tid cpu; do
                hex_tid=$(printf "0x%x" $tid)
                echo "┌─────────────────────────────────────────────"
                echo "│ Thread ID: $tid ($hex_tid) - CPU: ${cpu}%"
                echo "└─────────────────────────────────────────────"
                grep -A ${STACK_LINES} "nid=$hex_tid" "$SNAPSHOT_FILE" | head -$((STACK_LINES + 1))
                echo ""
            done
        } > "$THREAD_FILE"

        # Generate thread statistics report
        {
            echo "=== Thread Statistics Report ==="
            echo "Generated at: $(date)"
            echo "Process PID: $JAVA_PID"
            echo "Total CPU: ${TOTAL_CPU}%"
            echo ""

            # Count total threads
            TOTAL_THREADS=$(grep -c "^\"" "$SNAPSHOT_FILE")
            echo "Total Threads: $TOTAL_THREADS"
            echo ""

            # Count threads by state
            echo "Thread States:"
            echo "─────────────────────────────────────────────"

            RUNNABLE=$(grep "java.lang.Thread.State: RUNNABLE" "$SNAPSHOT_FILE" | wc -l)
            WAITING=$(grep "java.lang.Thread.State: WAITING" "$SNAPSHOT_FILE" | wc -l)
            TIMED_WAITING=$(grep "java.lang.Thread.State: TIMED_WAITING" "$SNAPSHOT_FILE" | wc -l)
            BLOCKED=$(grep "java.lang.Thread.State: BLOCKED" "$SNAPSHOT_FILE" | wc -l)
            NEW=$(grep "java.lang.Thread.State: NEW" "$SNAPSHOT_FILE" | wc -l)
            TERMINATED=$(grep "java.lang.Thread.State: TERMINATED" "$SNAPSHOT_FILE" | wc -l)

            printf "%-20s %6d  (%5.1f%%)\n" "RUNNABLE" $RUNNABLE $(echo "scale=1; $RUNNABLE * 100 / $TOTAL_THREADS" | bc)
            printf "%-20s %6d  (%5.1f%%)\n" "TIMED_WAITING" $TIMED_WAITING $(echo "scale=1; $TIMED_WAITING * 100 / $TOTAL_THREADS" | bc)
            printf "%-20s %6d  (%5.1f%%)\n" "WAITING" $WAITING $(echo "scale=1; $WAITING * 100 / $TOTAL_THREADS" | bc)
            printf "%-20s %6d  (%5.1f%%)\n" "BLOCKED" $BLOCKED $(echo "scale=1; $BLOCKED * 100 / $TOTAL_THREADS" | bc)
            printf "%-20s %6d  (%5.1f%%)\n" "NEW" $NEW $(echo "scale=1; $NEW * 100 / $TOTAL_THREADS" | bc)
            printf "%-20s %6d  (%5.1f%%)\n" "TERMINATED" $TERMINATED $(echo "scale=1; $TERMINATED * 100 / $TOTAL_THREADS" | bc)
            echo ""

            # Count thread types
            echo "Thread Types (Top 10):"
            echo "─────────────────────────────────────────────"
            grep "^\"" "$SNAPSHOT_FILE" | sed 's/"//g' | awk '{print $1}' | \
                sed 's/-[0-9]*$//' | sort | uniq -c | sort -rn | head -10 | \
                awk '{printf "%-40s %6d\n", $2, $1}'
            echo ""

            # RUNNABLE threads detail
            if [ $RUNNABLE -gt 0 ]; then
                echo "RUNNABLE Threads Detail:"
                echo "─────────────────────────────────────────────"
                grep -B 1 "java.lang.Thread.State: RUNNABLE" "$SNAPSHOT_FILE" | \
                    grep "^\"" | sed 's/"//g' | awk '{print $1}' | head -10
                echo ""
            fi

            # BLOCKED threads detail
            if [ $BLOCKED -gt 0 ]; then
                echo "BLOCKED Threads Detail:"
                echo "─────────────────────────────────────────────"
                grep -B 1 "java.lang.Thread.State: BLOCKED" "$SNAPSHOT_FILE" | \
                    grep "^\"" | sed 's/"//g' | awk '{print $1}' | head -10
                echo ""
            fi

            # Deadlock detection
            echo "Deadlock Detection:"
            echo "─────────────────────────────────────────────"
            if grep -q "Found one Java-level deadlock" "$SNAPSHOT_FILE"; then
                echo "⚠️  DEADLOCK DETECTED!"
                grep -A 20 "Found one Java-level deadlock" "$SNAPSHOT_FILE"
            else
                echo "✓ No deadlock detected"
            fi

        } > "$STATS_FILE"

        echo "✓ Snapshots saved:"
        echo "  - Full stack:    $SNAPSHOT_FILE"
        echo "  - Top threads:   $THREAD_FILE"
        echo "  - Statistics:    $STATS_FILE"
        echo ""

        # Output statistics summary to terminal
        echo "Thread Statistics Summary:"
        grep -A 10 "Thread States:" "$STATS_FILE"
    fi

    sleep $CHECK_INTERVAL
done