#!/bin/bash

# 增强版进程资源监控脚本（基于PID）
# 用法: ./pid_monitor.sh <进程PID> <监控时长(秒)>

# 检查参数是否正确
if [ $# -ne 2 ]; then
    echo "错误: 参数不正确!"
    echo "用法: $0 <进程PID> <监控时长(秒)>"
    exit 1
fi

TARGET_PID=$1
DURATION=$2
TIMESTAMP=$(date +%Y%m%d_%H%M%S)
LOG_PREFIX="pid_monitor_${TARGET_PID}"
DETAIL_LOG="${LOG_PREFIX}_detail_${TIMESTAMP}.log"
SUMMARY_LOG="${LOG_PREFIX}_summary_${TIMESTAMP}.log"

# 函数：检查进程是否存在
check_process_exists() {
    local pid=$1
    if [ ! -d /proc/$pid ]; then
        return 1
    fi
    return 0
}

# 函数：获取进程基本信息
get_process_info() {
    local pid=$1
    if [ -f /proc/$pid/status ]; then
        local process_name=$(cat /proc/$pid/comm 2>/dev/null || echo "N/A")
        local user=$(ps -o user= -p $pid 2>/dev/null || echo "N/A")
        local cmdline=$(cat /proc/$pid/cmdline | tr '\0' ' ' | head -c 100 2>/dev/null || echo "N/A")
        echo "$process_name" "$user" "$cmdline"
    else
        echo "N/A" "N/A" "N/A"
    fi
}

# 检查目标进程是否存在
if ! check_process_exists $TARGET_PID; then
    echo "错误: 进程 PID $TARGET_PID 不存在!"
    exit 1
fi

# 获取进程基本信息
read process_name process_user process_cmdline <<< $(get_process_info $TARGET_PID)

echo "开始监控进程..."
echo "进程PID: $TARGET_PID"
echo "进程名称: $process_name"
echo "运行用户: $process_user"
echo "命令行: $process_cmdline"
echo "监控时长: $DURATION 秒"
echo "开始时间: $(date)"
echo "详细日志: $DETAIL_LOG"
echo "汇总报告: $SUMMARY_LOG"

# 初始化日志文件
echo "进程监控报告 - PID: $TARGET_PID" > $DETAIL_LOG
echo "进程名称: $process_name" >> $DETAIL_LOG
echo "运行用户: $process_user" >> $DETAIL_LOG
echo "命令行: $process_cmdline" >> $DETAIL_LOG
echo "开始时间: $(date)" >> $DETAIL_LOG
echo "监控间隔: 1秒" >> $DETAIL_LOG
echo "===============================================================" >> $DETAIL_LOG
printf "%-20s %-8s %-8s %-12s %-12s %-12s\n" "时间戳" "CPU(%)" "内存(%)" "物理内存(MB)" "虚拟内存(MB)" "线程数" >> $DETAIL_LOG

# 初始化统计变量
COUNT=0
CPU_SUM=0
MEM_SUM=0
RSS_SUM=0
VSZ_SUM=0
THREADS_SUM=0

CPU_MAX=0
MEM_MAX=0
RSS_MAX=0
VSZ_MAX=0
THREADS_MAX=0

# 函数：获取精确的CPU使用率[8](@ref)
get_cpu_usage() {
    local pid=$1
    # 方法1: 使用top命令获取实时CPU使用率[1](@ref)
    local cpu_usage=$(top -bn1 -p $pid 2>/dev/null | awk -v pid=$pid '$1 == pid {print $9}')
    
    if [ -z "$cpu_usage" ]; then
        # 方法2: 通过/proc/stat计算[8](@ref)
        if [ -f /proc/$pid/stat ]; then
            # 获取进程CPU时间
            local utime=$(awk '{print $14}' /proc/$pid/stat)
            local stime=$(awk '{print $15}' /proc/$pid/stat)
            local total_time=$((utime + stime))
            
            # 获取系统启动时间
            local uptime=$(awk '{print $1}' /proc/uptime)
            local hz=$(getconf CLK_TCK)
            
            # 计算进程启动时间
            local starttime=$(awk '{print $22}' /proc/$pid/stat)
            local process_uptime=$(echo "$uptime - $starttime / $hz" | bc -l)
            
            if [ $COUNT -gt 0 ]; then
                local prev_total=$PREV_TOTAL
                local prev_seconds=$PREV_SECONDS
                local current_total=$total_time
                local current_seconds=$(echo "$uptime - $process_uptime" | bc -l)
                
                if [ $(echo "$current_seconds > $prev_seconds" | bc) -eq 1 ]; then
                    cpu_usage=$(echo "scale=2; ($current_total - $prev_total) / ($current_seconds - $prev_seconds) / $hz * 100" | bc -l)
                fi
                
                PREV_TOTAL=$current_total
                PREV_SECONDS=$current_seconds
            else
                PREV_TOTAL=$total_time
                PREV_SECONDS=$(echo "$uptime - $process_uptime" | bc -l)
            fi
        fi
    fi
    
    echo ${cpu_usage:-0}
}

# 函数：获取内存信息[3,8](@ref)
get_memory_info() {
    local pid=$1
    local mem_usage=0
    local rss_kb=0
    local vsz_kb=0
    local threads=0

    if [ -f /proc/$pid/status ]; then
        # 获取内存使用百分比[1](@ref)
        mem_usage=$(ps -o %mem= -p $pid 2>/dev/null || echo 0)
        
        # 获取RSS和VSZ[3](@ref)
        rss_kb=$(grep VmRSS /proc/$pid/status 2>/dev/null | awk '{print $2}' || echo 0)
        vsz_kb=$(grep VmSize /proc/$pid/status 2>/dev/null | awk '{print $2}' || echo 0)
        
        # 获取线程数[7](@ref)
        threads=$(ls /proc/$pid/task 2>/dev/null | wc -l || echo 1)
    fi
    
    # 转换为MB
    local rss_mb=0
    local vsz_mb=0
    if [ $rss_kb -gt 0 ]; then
        rss_mb=$((rss_kb / 1024))
    fi
    if [ $vsz_kb -gt 0 ]; then
        vsz_mb=$((vsz_kb / 1024))
    fi
    
    echo "$mem_usage" "$rss_mb" "$vsz_mb" "$threads"
}

# 主监控循环
START_TIME=$(date +%s)
END_TIME=$((START_TIME + DURATION))

echo "开始监控进程 PID: $TARGET_PID ..."

while [ $(date +%s) -lt $END_TIME ]; do
    CURRENT_TIME=$(date +%s)
    ELAPSED=$((CURRENT_TIME - START_TIME))
    REMAINING=$((DURATION - ELAPSED))
    
    if [ $REMAINING -le 0 ]; then
        break
    fi
    
    # 检查进程是否仍然存在
    if ! check_process_exists $TARGET_PID; then
        echo "警告: 进程 $TARGET_PID 在监控期间退出!"
        break
    fi
    
    TIMESTAMP=$(date +"%Y-%m-%d %H:%M:%S")
    
    # 获取CPU使用率
    CPU_USAGE=$(get_cpu_usage $TARGET_PID)
    
    # 获取内存信息
    read MEM_USAGE RSS_MB VSZ_MB THREADS <<< $(get_memory_info $TARGET_PID)
    
    # 记录到详细日志
    printf "%-20s %-8.1f %-8.1f %-12s %-12s %-12s\n" "$TIMESTAMP" $CPU_USAGE $MEM_USAGE "${RSS_MB}MB" "${VSZ_MB}MB" "$THREADS" >> $DETAIL_LOG
    
    # 更新统计信息
    CPU_SUM=$(echo "$CPU_SUM + $CPU_USAGE" | bc)
    MEM_SUM=$(echo "$MEM_SUM + $MEM_USAGE" | bc)
    RSS_SUM=$((RSS_SUM + RSS_MB))
    VSZ_SUM=$((VSZ_SUM + VSZ_MB))
    THREADS_SUM=$((THREADS_SUM + THREADS))
    
    # 更新最大值
    if [ $(echo "$CPU_USAGE > $CPU_MAX" | bc) -eq 1 ]; then
        CPU_MAX=$CPU_USAGE
    fi
    
    if [ $(echo "$MEM_USAGE > $MEM_MAX" | bc) -eq 1 ]; then
        MEM_MAX=$MEM_USAGE
    fi
    
    if [ $RSS_MB -gt $RSS_MAX ]; then
        RSS_MAX=$RSS_MB
    fi
    
    if [ $VSZ_MB -gt $VSZ_MAX ]; then
        VSZ_MAX=$VSZ_MB
    fi
    
    if [ $THREADS -gt $THREADS_MAX ]; then
        THREADS_MAX=$THREADS
    fi
    
    COUNT=$((COUNT + 1))
    
    # 进度显示
    if [ $((ELAPSED % 10)) -eq 0 ]; then
        echo "进度: ${ELAPSED}/${DURATION}秒, CPU: ${CPU_USAGE}%, 内存: ${MEM_USAGE}%, 物理内存: ${RSS_MB}MB"
    fi
    
    sleep 1
done

# 计算平均值
if [ $COUNT -gt 0 ]; then
    CPU_AVG=$(echo "scale=2; $CPU_SUM / $COUNT" | bc)
    MEM_AVG=$(echo "scale=2; $MEM_SUM / $COUNT" | bc)
    RSS_AVG=$((RSS_SUM / COUNT))
    VSZ_AVG=$((VSZ_SUM / COUNT))
    THREADS_AVG=$((THREADS_SUM / COUNT))
else
    CPU_AVG=0
    MEM_AVG=0
    RSS_AVG=0
    VSZ_AVG=0
    THREADS_AVG=0
fi

# 生成汇总报告
echo "===============================================================" > $SUMMARY_LOG
echo "进程监控统计报告" >> $SUMMARY_LOG
echo "===============================================================" >> $SUMMARY_LOG
echo "进程PID: $TARGET_PID" >> $SUMMARY_LOG
echo "进程名称: $process_name" >> $SUMMARY_LOG
echo "运行用户: $process_user" >> $SUMMARY_LOG
echo "监控时长: $DURATION 秒" >> $SUMMARY_LOG
echo "数据采样数: $COUNT 次" >> $SUMMARY_LOG
echo "开始时间: $(date -d @$START_TIME)" >> $SUMMARY_LOG
echo "结束时间: $(date)" >> $SUMMARY_LOG
echo "" >> $SUMMARY_LOG

echo "CPU使用率统计:" >> $SUMMARY_LOG
echo "  平均值: $CPU_AVG%" >> $SUMMARY_LOG
echo "  最大值: $CPU_MAX%" >> $SUMMARY_LOG
echo "" >> $SUMMARY_LOG

echo "内存使用率统计:" >> $SUMMARY_LOG
echo "  平均值: $MEM_AVG%" >> $SUMMARY_LOG
echo "  最大值: $MEM_MAX%" >> $SUMMARY_LOG
echo "" >> $SUMMARY_LOG

echo "物理内存(RSS)统计:" >> $SUMMARY_LOG
echo "  平均值: ${RSS_AVG}MB" >> $SUMMARY_LOG
echo "  最大值: ${RSS_MAX}MB" >> $SUMMARY_LOG
echo "" >> $SUMMARY_LOG

echo "虚拟内存(VSZ)统计:" >> $SUMMARY_LOG
echo "  平均值: ${VSZ_AVG}MB" >> $SUMMARY_LOG
echo "  最大值: ${VSZ_MAX}MB" >> $SUMMARY_LOG
echo "" >> $SUMMARY_LOG

echo "线程数统计:" >> $SUMMARY_LOG
echo "  平均值: $THREADS_AVG" >> $SUMMARY_LOG
echo "  最大值: $THREADS_MAX" >> $SUMMARY_LOG
echo "" >> $SUMMARY_LOG

# 屏幕输出总结
echo ""
echo "===== 监控完成 ====="
echo "详细监控数据: $DETAIL_LOG"
echo "统计报告: $SUMMARY_LOG"
echo ""
echo "监控统计摘要:"
echo "  - 数据采样数: $COUNT 次"
echo "  - CPU平均使用率: $CPU_AVG%"
echo "  - CPU最大使用率: $CPU_MAX%"
echo "  - 内存平均使用率: $MEM_AVG%"
echo "  - 内存最大使用率: $MEM_MAX%"
echo "  - 物理内存平均使用: ${RSS_AVG}MB"
echo "  - 物理内存峰值使用: ${RSS_MAX}MB"