k8s容器内资源监控统计脚本

可直接复制以下命令到容器内shell中执行：

cat << "EOF" > /tmp/mem-monitor.sh
#!/bin/bash
DURATION=${1:-60}
INTERVAL=${2:-5}
# 自动检测容器 CPU 核心数
if [ -f /sys/fs/cgroup/cpu/cpu.cfs_quota_us ] && [ -f /sys/fs/cgroup/cpu/cpu.cfs_period_us ]; then
    QUOTA=$(cat /sys/fs/cgroup/cpu/cpu.cfs_quota_us)
    PERIOD=$(cat /sys/fs/cgroup/cpu/cpu.cfs_period_us)
    if [ "$QUOTA" -gt 0 ]; then
        CPU_CORES=$(awk "BEGIN {printf \"%.0f\", $QUOTA/$PERIOD}")
    else
        CPU_CORES=$(nproc 2>/dev/null || echo 1)
    fi
else
    CPU_CORES=$(nproc 2>/dev/null || echo 1)
fi
# 获取容器 CPU 使用时间（纳秒）
get_cpu_usage() {
    if [ -f /sys/fs/cgroup/cpuacct/cpuacct.usage ]; then
        cat /sys/fs/cgroup/cpuacct/cpuacct.usage
    elif [ -f /sys/fs/cgroup/cpu/cpuacct.usage ]; then
        cat /sys/fs/cgroup/cpu/cpuacct.usage
    else
        echo "0"
    fi
}
echo "=========================================="
echo "      进程内存 & CPU 监控脚本"
echo "=========================================="
echo "监控时长: ${DURATION}秒 | 采集间隔: ${INTERVAL}秒"
echo "容器 CPU: ${CPU_CORES} 核 (100% = 用满配额)"
echo "开始时间: $(date +"%Y-%m-%d %H:%M:%S")"
echo "=========================================="
echo ""
echo ">>> 请在监控期间触发你的操作 <<<"
echo ""
TMP_FILE="/tmp/mem_monitor_$$.csv"
echo "timestamp,pid,rss_kb,cpu_raw,cmd" > "$TMP_FILE"
if [ -f /sys/fs/cgroup/memory/memory.usage_in_bytes ]; then
    container_usage=$(cat /sys/fs/cgroup/memory/memory.usage_in_bytes)
    echo "容器内存基线: $((container_usage/1024/1024))MB"
fi
# 初始化 CPU 基线
PREV_CPU_USAGE=$(get_cpu_usage)
PREV_TIME=$(date +%s%N)
to_mb() {
    awk "BEGIN {printf \"%.1f\", $1/1024}"
}
# 计算容器总 CPU 占用率
calc_container_cpu() {
    local current_usage=$1
    local current_time=$2
    
    local usage_diff=$((current_usage - PREV_CPU_USAGE))
    local time_diff=$((current_time - PREV_TIME))
    
    if [ $time_diff -gt 0 ]; then
        # CPU% = (usage_diff / time_diff) / CPU_CORES * 100
        awk "BEGIN {printf \"%.1f\", ($usage_diff / $time_diff) / $CPU_CORES * 100}"
    else
        echo "0.0"
    fi
}
show_snapshot() {
    local timestamp=$1
    local label=$2
    
    # 计算容器总 CPU
    local current_cpu_usage=$(get_cpu_usage)
    local current_time=$(date +%s%N)
    local container_cpu=$(calc_container_cpu $current_cpu_usage $current_time)
    
    # 更新基线
    PREV_CPU_USAGE=$current_cpu_usage
    PREV_TIME=$current_time
    
    echo "[$timestamp] $label"
    echo "--------------------------------------------------------------------"
    
    if [ -f /sys/fs/cgroup/memory/memory.usage_in_bytes ]; then
        local usage=$(cat /sys/fs/cgroup/memory/memory.usage_in_bytes)
        echo "  容器内存: $((usage/1024/1024))MB | 容器总CPU: ${container_cpu}%"
    else
        echo "  容器总CPU: ${container_cpu}%"
    fi
    
    printf "  %-8s  %10s  %12s  %s\n" "PID" "RSS(MB)" "CPU(实际%)" "命令"
    printf "  %-8s  %10s  %12s  %s\n" "---" "-------" "----------" "----"
    
    ps -eo pid,rss,%cpu,args --sort=-%cpu | grep -iE "java|node" | grep -v "grep" | head -20 | while IFS= read -r line; do
        pid=$(echo "$line" | awk "{print \$1}")
        rss=$(echo "$line" | awk "{print \$2}")
        cpu_raw=$(echo "$line" | awk "{print \$3}")
        cmd=$(echo "$line" | awk "{for(i=4;i<=NF;i++) printf \"%s \", \$i}" | cut -c1-55)
        
        [ "$rss" = "0" ] && continue
        
        cpu_real=$(awk "BEGIN {printf \"%.1f\", $cpu_raw/$CPU_CORES}")
        
        cpu_mark=""
        awk "BEGIN {exit ($cpu_real > 80) ? 0 : 1}" && cpu_mark="🔴"
        awk "BEGIN {exit ($cpu_real > 50 && $cpu_real <= 80) ? 0 : 1}" && cpu_mark="🟡"
        
        printf "  %-8s  %10s  %10s%%  %s %s\n" "$pid" "$(to_mb $rss)" "$cpu_real" "$cpu_mark" "$cmd"
        csv_cmd=$(echo "$cmd" | tr "," ";")
        echo "$timestamp,$pid,$rss,$cpu_raw,$csv_cmd" >> "$TMP_FILE"
    done
    echo ""
}
START_TIME=$(date +%s)
ITERATION=0
show_snapshot "$(date +"%H:%M:%S")" "开始"
while true; do
    CURRENT_TIME=$(date +%s)
    ELAPSED=$((CURRENT_TIME - START_TIME))
    [ $ELAPSED -ge $DURATION ] && break
    
    sleep $INTERVAL
    ITERATION=$((ITERATION + 1))
    show_snapshot "$(date +"%H:%M:%S")" "第 ${ITERATION} 次采集 (${ELAPSED}s)"
done
echo "=========================================="
echo "              变化分析"
echo "=========================================="
echo ""
declare -A first_rss last_rss max_rss cpu_sum cpu_count
while IFS=, read -r timestamp pid rss cpu_raw cmd; do
    [ "$pid" = "pid" ] && continue
    [ -z "${first_rss[$pid]}" ] && first_rss[$pid]=$rss
    last_rss[$pid]=$rss
    [ -z "${max_rss[$pid]}" ] || [ $rss -gt ${max_rss[$pid]} ] && max_rss[$pid]=$rss
    
    cpu_sum[$pid]=$(awk "BEGIN {print ${cpu_sum[$pid]:-0} + $cpu_raw}")
    cpu_count[$pid]=$(( ${cpu_count[$pid]:-0} + 1 ))
done < "$TMP_FILE"
printf "%-8s  %10s  %10s  %10s  %12s  %s\n" "PID" "开始(MB)" "结束(MB)" "最大(MB)" "平均CPU(实际%)" "内存变化"
printf "%-8s  %10s  %10s  %10s  %12s  %s\n" "---" "--------" "--------" "--------" "------------" "--------"
for key in "${!first_rss[@]}"; do
    start=${first_rss[$key]}
    end=${last_rss[$key]}
    max=${max_rss[$key]}
    change=$((end - start))
    
    avg_cpu_raw=$(awk "BEGIN {printf \"%.1f\", ${cpu_sum[$key]} / ${cpu_count[$key]}}")
    avg_cpu_real=$(awk "BEGIN {printf \"%.1f\", $avg_cpu_raw / $CPU_CORES}")
    
    if [ $change -gt 10240 ]; then marker="🔴 +$(to_mb $change)"
    elif [ $change -gt 5120 ]; then marker="🟡 +$(to_mb $change)"
    elif [ $change -lt -5120 ]; then marker="🟢 $(to_mb $change)"
    else marker="⚪ $(to_mb $change)"
    fi
    
    printf "%-8s  %10s  %10s  %10s  %12s  %sMB\n" "$key" "$(to_mb $start)" "$(to_mb $end)" "$(to_mb $max)" "${avg_cpu_real}%" "$marker"
done
echo ""
if [ -f /sys/fs/cgroup/memory/memory.usage_in_bytes ]; then
    container_end=$(cat /sys/fs/cgroup/memory/memory.usage_in_bytes)
    echo "容器内存变化: $(( (container_end - container_usage) / 1024 / 1024 ))MB"
fi
echo ""
echo "========== 结论 =========="
max_increase=0 max_mem_pid=""
max_cpu=0 max_cpu_pid=""
for key in "${!first_rss[@]}"; do
    increase=$((last_rss[$key] - first_rss[$key]))
    [ $increase -gt $max_increase ] && { max_increase=$increase; max_mem_pid=$key; }
    
    avg_raw=$(awk "BEGIN {print ${cpu_sum[$key]} / ${cpu_count[$key]}}")
    avg_real=$(awk "BEGIN {print $avg_raw / $CPU_CORES}")
    awk "BEGIN {exit ($avg_real > $max_cpu) ? 0 : 1}" && { max_cpu=$avg_real; max_cpu_pid=$key; }
done
echo "📊 内存增长最多: PID $max_mem_pid (+$(to_mb $max_increase)MB)"
echo "📊 CPU 占用最高: PID $max_cpu_pid (${max_cpu}% 容器配额)"
if [ $max_increase -gt 10240 ]; then
    echo ""
    echo "🔴 PID $max_mem_pid 内存增长较大，可能存在泄漏"
fi
awk "BEGIN {exit ($max_cpu > 80) ? 0 : 1}" && {
    echo "🔴 PID $max_cpu_pid CPU 占用超过 80%，检查是否有死循环或密集计算"
}
awk "BEGIN {exit ($max_cpu > 100) ? 0 : 1}" && {
    echo "🔴🔴 PID $max_cpu_pid CPU 已超配额，会被 K8s 限流！"
}
echo "=========================================="
echo "监控完成: $(date +"%Y-%m-%d %H:%M:%S")"
echo "=========================================="
EOF
chmod +x /tmp/mem-monitor.sh && /tmp/mem-monitor.sh 60 5