kefu_cloud/server/internal/handler/ops_metrics.go

package handler

import (
	"context"
	"fmt"
	"runtime"
	"time"

	"github.com/gin-gonic/gin"
	"github.com/shirou/gopsutil/v4/cpu"
	"github.com/shirou/gopsutil/v4/disk"
	"github.com/shirou/gopsutil/v4/mem"
	"kefu-sys/server/internal/metrics"
	"kefu-sys/server/internal/middleware"
	"kefu-sys/server/internal/model"
	"kefu-sys/server/internal/storage"
	"kefu-sys/server/internal/ws"
)

// OpsMetrics 平台运维实时指标（当前 API 实例视角）。
// GET /api/admin/ops/metrics
func (h *AdminHandler) OpsMetrics(c *gin.Context) {
	now := time.Now()

	// —— 系统负载 ——
	load := collectSystemLoad()

	// —— 请求统计 ——
	reqSeries, errSeries, hourLabels, winReq, winErr := metrics.Default.Snapshot24h()
	totalReq, totalErr, total5xx, startedAt := metrics.Default.Totals()
	var errNow float64
	if winReq > 0 {
		errNow = float64(winErr) / float64(winReq) * 100
	}
	var qps float64
	uptimeSec := now.Sub(startedAt).Seconds()
	if uptimeSec > 0 {
		qps = float64(totalReq) / uptimeSec
	}

	// —— 服务探测 ——
	services := collectServiceHealth(h.store)

	// —— 整体状态 ——
	overall := "normal"
	for _, s := range services {
		if st, _ := s["status"].(string); st == "error" {
			overall = "error"
			break
		}
		if st, _ := s["status"].(string); st == "warning" && overall == "normal" {
			overall = "warning"
		}
	}
	if load.CPUPercent >= 90 || load.MemPercent >= 90 || load.DiskPercent >= 95 {
		if overall == "normal" {
			overall = "warning"
		}
	}
	if errNow >= 5 {
		overall = "error"
	} else if errNow >= 2 && overall == "normal" {
		overall = "warning"
	}

	wsTotal, wsAgents, wsVisitors := ws.DefaultHub.Stats()

	middleware.JSON(c, gin.H{
		"overall":    overall,
		"updated_at": now.Format(time.RFC3339),
		"instance":   "api", // 当前进程实例
		"uptime_sec": int64(uptimeSec),
		"services":   services,
		"load": gin.H{
			"cpu_percent":  load.CPUPercent,
			"mem_percent":  load.MemPercent,
			"disk_percent": load.DiskPercent,
			"cpu_cores":    load.CPUCores,
			"mem_total_mb": load.MemTotalMB,
			"mem_used_mb":  load.MemUsedMB,
			"disk_total_gb": load.DiskTotalGB,
			"disk_used_gb":  load.DiskUsedGB,
			"go_goroutines": runtime.NumGoroutine(),
			"go_heap_mb":    load.GoHeapMB,
		},
		"api": gin.H{
			"requests_last_24h":   reqSeries,
			"error_rate_last_24h": errSeries,
			"hours":               hourLabels,
			"qps":                 round2(qps),
			"error_rate_now":      round2(errNow),
			"window_requests":     winReq,
			"window_errors":       winErr,
			"total_requests":      totalReq,
			"total_errors":        totalErr,
			"total_5xx":           total5xx,
			"in_flight":           metrics.Default.InFlight(),
		},
		"websocket": gin.H{
			"connections": wsTotal,
			"agents":      wsAgents,
			"visitors":    wsVisitors,
		},
	})
}

type systemLoad struct {
	CPUPercent  float64
	MemPercent  float64
	DiskPercent float64
	CPUCores    int
	MemTotalMB  uint64
	MemUsedMB   uint64
	DiskTotalGB float64
	DiskUsedGB  float64
	GoHeapMB    float64
}

func collectSystemLoad() systemLoad {
	out := systemLoad{
		CPUCores: runtime.NumCPU(),
	}
	var ms runtime.MemStats
	runtime.ReadMemStats(&ms)
	out.GoHeapMB = round2(float64(ms.HeapAlloc) / 1024 / 1024)

	if percents, err := cpu.Percent(120*time.Millisecond, false); err == nil && len(percents) > 0 {
		out.CPUPercent = round2(percents[0])
	}
	if vm, err := mem.VirtualMemory(); err == nil {
		out.MemPercent = round2(vm.UsedPercent)
		out.MemTotalMB = vm.Total / 1024 / 1024
		out.MemUsedMB = vm.Used / 1024 / 1024
	}
	if du, err := disk.Usage("/"); err == nil {
		out.DiskPercent = round2(du.UsedPercent)
		out.DiskTotalGB = round2(float64(du.Total) / 1024 / 1024 / 1024)
		out.DiskUsedGB = round2(float64(du.Used) / 1024 / 1024 / 1024)
	}
	return out
}

func collectServiceHealth(store storage.ObjectStorage) []gin.H {
	services := make([]gin.H, 0, 5)

	// API：本进程存活
	services = append(services, gin.H{
		"name":    "API服务",
		"status":  "normal",
		"detail":  "进程运行中",
		"latency_ms": 0,
	})

	// WebSocket Hub
	wsTotal, wsAgents, wsVisitors := ws.DefaultHub.Stats()
	services = append(services, gin.H{
		"name":   "WebSocket",
		"status": "normal",
		"detail": fmt.Sprintf("连接 %d（坐席 %d / 访客 %d）", wsTotal, wsAgents, wsVisitors),
		"connections": wsTotal,
	})

	// 数据库
	dbStatus, dbDetail, dbLatency := probeDatabase()
	services = append(services, gin.H{
		"name":       "数据库",
		"status":     dbStatus,
		"detail":     dbDetail,
		"latency_ms": dbLatency,
	})

	// 对象存储
	stStatus, stDetail, stLatency := probeStorage(store)
	services = append(services, gin.H{
		"name":       "对象存储",
		"status":     stStatus,
		"detail":     stDetail,
		"latency_ms": stLatency,
	})

	return services
}

func probeDatabase() (status, detail string, latencyMs int64) {
	if model.DB == nil {
		return "error", "数据库未初始化", 0
	}
	sqlDB, err := model.DB.DB()
	if err != nil {
		return "error", "获取连接失败: " + err.Error(), 0
	}
	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
	defer cancel()
	start := time.Now()
	if err := sqlDB.PingContext(ctx); err != nil {
		return "error", "Ping 失败: " + err.Error(), time.Since(start).Milliseconds()
	}
	lat := time.Since(start).Milliseconds()
	status = "normal"
	detail = fmt.Sprintf("Ping 正常 · %dms", lat)
	if lat >= 200 {
		status = "warning"
		detail = fmt.Sprintf("Ping 偏慢 · %dms", lat)
	}
	return status, detail, lat
}

func probeStorage(store storage.ObjectStorage) (status, detail string, latencyMs int64) {
	if store == nil {
		return "warning", "未配置对象存储", 0
	}
	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
	defer cancel()
	start := time.Now()
	if err := store.HealthCheck(ctx); err != nil {
		return "error", "不可达: " + err.Error(), time.Since(start).Milliseconds()
	}
	lat := time.Since(start).Milliseconds()
	status = "normal"
	detail = fmt.Sprintf("Bucket 可达 · %dms", lat)
	if lat >= 500 {
		status = "warning"
		detail = fmt.Sprintf("探测偏慢 · %dms", lat)
	}
	return status, detail, lat
}

func round2(v float64) float64 {
	return float64(int(v*100+0.5)) / 100
}