kvm-manager/backend/app/routers/monitor.py

"""资源监控路由"""
from fastapi import APIRouter, HTTPException
from app.libvirt_conn import libvirt_conn
import libvirt
import time
import threading

router = APIRouter()

# 简易内存缓存
_stats_cache = {}
_cache_lock = threading.Lock()


@router.get("/overview")
async def monitor_overview():
    """宿主机总览监控"""
    conn = libvirt_conn.conn

    # 宿主机信息
    host_info = conn.getInfo()
    hostname = conn.getHostname()

    # CPU 使用率（通过 node info）
    cpu_stats = conn.getCPUStats(-1, 0)  # 全局 CPU 统计
    cpu_total = cpu_stats.get("user", 0) + cpu_stats.get("system", 0) + cpu_stats.get("idle", 0)
    cpu_used = cpu_stats.get("user", 0) + cpu_stats.get("system", 0)
    cpu_percent = round(cpu_used / cpu_total * 100, 1) if cpu_total > 0 else 0

    # 内存
    memory_total_kb = host_info[1]
    # 获取可用内存
    try:
        with open("/proc/meminfo", "r") as f:
            meminfo = {}
            for line in f:
                parts = line.split()
                if len(parts) >= 2:
                    meminfo[parts[0].rstrip(":")] = int(parts[1])

        mem_total_mb = meminfo.get("MemTotal", 0) // 1024
        mem_available_mb = meminfo.get("MemAvailable", 0) // 1024
        mem_used_mb = mem_total_mb - mem_available_mb
        mem_percent = round(mem_used_mb / mem_total_mb * 100, 1) if mem_total_mb > 0 else 0
    except Exception:
        mem_total_mb = memory_total_kb // 1024
        mem_used_mb = 0
        mem_available_mb = mem_total_mb
        mem_percent = 0

    # 虚拟机统计
    domains = conn.listAllDomains(0)
    running = sum(1 for d in domains if d.isActive())
    stopped = len(domains) - running

    return {
        "hostname": hostname,
        "cpu": {
            "cores": host_info[2],
            "speed_mhz": host_info[3],
            "usage_percent": cpu_percent,
        },
        "memory": {
            "total_mb": mem_total_mb,
            "used_mb": mem_used_mb,
            "available_mb": mem_available_mb,
            "usage_percent": mem_percent,
        },
        "vms": {
            "total": len(domains),
            "running": running,
            "stopped": stopped,
        },
    }


@router.get("/vm/{name}")
async def monitor_vm(name: str):
    """获取虚拟机实时监控数据"""
    conn = libvirt_conn.conn
    try:
        dom = conn.lookupByName(name)
    except libvirt.libvirtError:
        raise HTTPException(status_code=404, detail=f"虚拟机 '{name}' 不存在")

    if not dom.isActive():
        return {"name": name, "state": "stopped", "cpu_percent": 0, "memory": {}}

    # CPU 百分比
    cpu_percent = _get_vm_cpu_percent(dom)

    # 内存
    mem_stats = {}
    try:
        raw = dom.memoryStats()
        mem_stats = {
            "rss_mb": raw.get("rss", 0) // 1024 if "rss" in raw else 0,
            "actual_mb": raw.get("actual", 0) // 1024 if "actual" in raw else 0,
            "available_mb": raw.get("available", 0) // 1024 if "available" in raw else 0,
            "usage_percent": round(
                raw.get("rss", 0) / raw.get("actual", 1) * 100, 1
            ) if "rss" in raw and "actual" in raw else 0,
        }
    except Exception:
        pass

    # 磁盘IO
    disk_stats = _get_vm_disk_stats(dom)

    # 网络IO
    net_stats = _get_vm_net_stats(dom)

    return {
        "name": name,
        "state": "running",
        "cpu_percent": cpu_percent,
        "memory": mem_stats,
        "disk": disk_stats,
        "network": net_stats,
    }


def _get_vm_cpu_percent(dom) -> float:
    """计算虚拟机 CPU 使用率"""
    cache_key = f"cpu_{dom.name()}"

    try:
        # 第一次采样
        info1 = dom.info()
        cpu_time1 = info1[2]
        t1 = time.time()

        # 从缓存获取上一次数据
        with _cache_lock:
            prev = _stats_cache.get(cache_key)

        if prev:
            cpu_time0, t0 = prev
            elapsed = t1 - t0
            cpu_diff = cpu_time1 - cpu_time0
            # CPU时间单位是纳秒
            cpu_percent = round((cpu_diff / 1e9) / elapsed * 100, 1)
            cpu_percent = min(cpu_percent, 100.0)
        else:
            cpu_percent = 0.0

        # 更新缓存
        with _cache_lock:
            _stats_cache[cache_key] = (cpu_time1, t1)

        return cpu_percent
    except Exception:
        return 0.0


def _get_vm_disk_stats(dom) -> list:
    """获取虚拟机磁盘IO"""
    from lxml import etree
    stats = []
    try:
        xml = etree.fromstring(dom.XMLDesc(0).encode())
        for disk in xml.findall(".//disk[@device='disk']"):
            target = disk.find("target")
            if target is not None:
                dev = target.get("dev", "")
                try:
                    s = dom.blockStats(dev)
                    stats.append({
                        "dev": dev,
                        "read_bytes": s[1],
                        "write_bytes": s[3],
                        "read_requests": s[0],
                        "write_requests": s[2],
                    })
                except Exception:
                    pass
    except Exception:
        pass
    return stats


def _get_vm_net_stats(dom) -> list:
    """获取虚拟机网络IO"""
    from lxml import etree
    stats = []
    try:
        xml = etree.fromstring(dom.XMLDesc(0).encode())
        for iface in xml.findall(".//interface"):
            target = iface.find("target")
            if target is not None:
                dev = target.get("dev", "")
                try:
                    s = dom.interfaceStats(dev)
                    stats.append({
                        "dev": dev,
                        "rx_bytes": s[0],
                        "tx_bytes": s[4],
                        "rx_packets": s[1],
                        "tx_packets": s[5],
                        "rx_errors": s[2],
                        "tx_errors": s[6],
                    })
                except Exception:
                    pass
    except Exception:
        pass
    return stats