feat: Agent采集uptime, 后端持久化kernel/uptime/diskIO, 前端磁盘BPS+IOPS展示

- Agent: system.go 添加 readUptime(), SystemStat 携带 Uptime - Backend: Server 模型新增 Kernel 字段, ReceiveReport 持久化 kernel+uptime - Backend: /agent/latest 返回 kernel/uptime/port_info/disk_io_info/disk_total - Frontend: 磁盘监控改为 BPS+IOPS (4张卡片), 系统信息修复内核/运行时长/监听端口/磁盘容量 - 图表单位从 MB/s 改为 KB/s, 添加 formatBPS 工具函数
2026-06-19 01:10:00 +00:00
1 parent 31c2e338a3
commit 5949ec5d0c
@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"math"
 	"net/http"
+	"sort"
 	"strconv"
 	"strings"
 	"time"
@@ -77,12 +78,14 @@ func (h *AgentHandler) ReceiveReport(c *gin.Context) {
 	// 同步更新 servers 表的实时指标（三级匹配：hostname → client IP → hostname LIKE）
 	diskAvg := avgDiskUsage(req.Disks)

-	// 计算硬件规格
+	// 计算硬件规格（四舍五入到整数，避免 31.34→31GB 的显示偏差）
 	totalDiskGB := 0.0
 	for _, d := range req.Disks {
 		totalDiskGB += d.TotalGB
 	}
-	memMB := int64(req.Memory.TotalGB * 1024) // GB → MB
+	memGB := math.Ceil(req.Memory.TotalGB)        // 31.34 → 32, 62.79 → 63（物理内存向上取整）
+	memMB := int64(memGB * 1024)                    // GB → MB
+	diskGB := int64(totalDiskGB + 0.5)              // 39.99 → 40

 	clientIP := c.ClientIP()

@@ -92,8 +95,8 @@ func (h *AgentHandler) ReceiveReport(c *gin.Context) {
 		matchIP = clientIP
 	}

-	// 直接 SQL 更新，避免 GORM soft delete 自动添加 deleted_at 条件干扰
-	updateSQL := `UPDATE servers SET 
+	// 直接 SQL 更新 asset_servers 监控字段
+	updateSQL := `UPDATE asset_servers SET 
 		cpu_usage = ?, memory_usage = ?, disk_usage = ?,
 		cpu = ?, memory = ?, disk = ?,
 		status = ?, agent_status = ?, agent_version = ?,
@@ -101,14 +104,24 @@ func (h *AgentHandler) ReceiveReport(c *gin.Context) {
 		updated_at = NOW()`
 	args := []interface{}{
 		req.CPU.UsagePercent, req.Memory.UsagePercent, diskAvg,
-		req.CPU.CoreCount, memMB, int64(totalDiskGB),
+		req.CPU.CoreCount, memMB, diskGB,
 		"running", "online", req.AgentVersion,
 		// CASE WHEN hostname != ? AND ip = ? THEN ? ELSE hostname END
 		req.Hostname, matchIP, req.Hostname,
 	}
-	if req.System != nil && req.System.OSVersion != "" {
-		updateSQL += ", os = ?"
-		args = append(args, req.System.OSVersion)
+	if req.System != nil {
+		if req.System.OSVersion != "" {
+			updateSQL += ", os = ?"
+			args = append(args, req.System.OSVersion)
+		}
+		if req.System.Kernel != "" {
+			updateSQL += ", kernel = ?"
+			args = append(args, req.System.Kernel)
+		}
+		if req.System.Uptime > 0 {
+			updateSQL += ", uptime = ?"
+			args = append(args, req.System.Uptime)
+		}
 	}
 	updateSQL += " WHERE hostname = ? OR ip = ?"
 	args = append(args, req.Hostname, matchIP)
@@ -122,9 +135,44 @@ func (h *AgentHandler) ReceiveReport(c *gin.Context) {
 		result = h.db.Exec(updateSQL2, args2...)
 	}

-	// 如果还是没匹配到，记录日志（Agent 上报了，但服务器列表里没有对应条目）
+	// 如果还是没匹配到 —— 自动注册新服务器（UPSERT 防重复）
 	if result.RowsAffected == 0 {
-		fmt.Printf("[agent] 上报主机 %q (IP: %s) 未匹配任何 servers 条目，指标仅写入 agent_reports 表\n", req.Hostname, matchIP)
+		osName := ""
+		if req.System != nil {
+			osName = req.System.OSVersion
+		}
+		newServer := models.Server{
+			Hostname:     req.Hostname,
+			IP:           matchIP,
+			Port:         22,
+			Username:     "root",
+			AuthType:     "password",
+			OS:           osName,
+			CPU:          req.CPU.CoreCount,
+			Memory:       memMB,
+			Disk:         diskGB,
+			Status:       "running",
+			AgentStatus:  "online",
+			AgentVersion: req.AgentVersion,
+			CPUUsage:     req.CPU.UsagePercent,
+			MemoryUsage:  req.Memory.UsagePercent,
+			DiskUsage:    diskAvg,
+		}
+		// ON DUPLICATE KEY UPDATE — 若 hostname 已存在则更新字段（防重复注册）
+		if err := h.db.Clauses(clause.OnConflict{
+			Columns: []clause.Column{{Name: "hostname"}},
+			DoUpdates: clause.AssignmentColumns([]string{
+				"ip", "os", "cpu", "memory", "disk",
+				"status", "agent_status", "agent_version",
+				"cpu_usage", "memory_usage", "disk_usage",
+				"updated_at",
+			}),
+		}).Create(&newServer).Error; err != nil {
+			fmt.Printf("[agent] 自动注册主机 %q (IP: %s) 失败: %v\n", req.Hostname, matchIP, err)
+		} else {
+			fmt.Printf("[agent] 自动注册新主机: %q (IP: %s, CPU=%d核, 内存=%dMB, 磁盘=%dGB, OS=%q)\n",
+				req.Hostname, matchIP, req.CPU.CoreCount, memMB, diskGB, osName)
+		}
 	}

 	// WebSocket 广播：向所有前端客户端推送该主机的最新指标 // Logan.Li
@@ -183,38 +231,51 @@ func (h *AgentHandler) writeHotMetric(c *gin.Context, key string, value float64,
 	_ = h.cache.ZRemRangeByScore(ctx, key, "-inf", fmt.Sprintf("%.0f", cutoff))
 }

-// GetLatestReports 获取各主机最新上报（仅 3 分钟内活跃的 hostname）
+// GetLatestReports 获取所有服务器的 Agent 最新状态（含未安装 Agent 的主机）
 // GET /api/v1/agent/latest
 func (h *AgentHandler) GetLatestReports(c *gin.Context) {
 	var results []models.AgentLatestResponse
 	threeMinAgo := time.Now().Add(-3 * time.Minute)
+	// 以 servers 表为主，LEFT JOIN 最近 3 分钟内的 Agent 上报
+	// 未装 Agent 的主机也会显示（agent_status='not_installed'），打破"不显示→无法安装"的循环
 	h.db.Raw(`
 		SELECT
-			a.hostname,
+			s.id as server_id,
+			s.hostname,
 			COALESCE(s.ip, '') as ip,
-			a.cpu_usage,
-			a.cpu_cores,
-			a.mem_usage,
-			a.mem_total,
-			a.mem_used,
+			COALESCE(a.cpu_usage, 0) as cpu_usage,
+			COALESCE(a.cpu_cores, 0) as cpu_cores,
+			COALESCE(a.mem_usage, 0) as mem_usage,
+			COALESCE(a.mem_total, 0) as mem_total,
+			COALESCE(a.mem_used, 0) as mem_used,
 			COALESCE(s.disk_usage, 0) as disk_usage,
 			a.disk_info,
+			a.port_info,
 			a.net_info,
+			a.disk_io_info,
 			a.report_time,
 			COALESCE(s.agent_version, '') as agent_version,
-			COALESCE(s.agent_status, 'offline') as agent_status,
+			CASE
+				WHEN a.hostname IS NOT NULL THEN COALESCE(s.agent_status, 'online')
+				ELSE 'not_installed'
+			END as agent_status,
 			COALESCE(s.os, '') as os,
-			COALESCE(r.name, '') as region
-		FROM agent_reports a
-		LEFT JOIN servers s ON s.hostname = a.hostname
-		LEFT JOIN regions r ON r.id = s.region_id
-		WHERE a.id IN (
-			SELECT MAX(id) FROM agent_reports
-			WHERE report_time >= ?
-			GROUP BY hostname
-		)
-		AND s.id IS NOT NULL  -- 只显示匹配 servers 表的主机（首次上线后自动同步 hostname）
-		ORDER BY a.report_time DESC
+			COALESCE(s.kernel, '') as kernel,
+			COALESCE(s.uptime, 0) as uptime,
+			COALESCE(s.disk, 0) as disk_total,
+			COALESCE(r.name, '') as region,
+			CASE WHEN s.password != '' OR s.ssh_key != '' THEN true ELSE false END as has_auth
+		FROM asset_servers s
+		LEFT JOIN (
+			SELECT * FROM monitor_agent_reports
+			WHERE id IN (
+				SELECT MAX(id) FROM monitor_agent_reports
+				WHERE report_time >= ?
+				GROUP BY hostname
+			)
+		) a ON a.hostname = s.hostname
+		LEFT JOIN asset_regions r ON r.id = s.region_id
+		ORDER BY a.report_time DESC, s.id ASC
 	`, threeMinAgo).Scan(&results)

 	if results == nil {
@@ -253,7 +314,7 @@ func (h *AgentHandler) GetHostHistory(c *gin.Context) {
 // AggregateMetrics 聚合 Redis 热数据到 MySQL klines
 // POST /api/v1/agent/aggregate
 func (h *AgentHandler) AggregateMetrics(c *gin.Context) {
-	var req models.AggregateKlineReq
+	var req models.AggregateReq
 	if err := c.ShouldBindJSON(&req); err != nil {
 		c.JSON(http.StatusBadRequest, models.Response{Code: 400, Message: "请求格式错误: " + err.Error()})
 		return
@@ -303,37 +364,47 @@ func (h *AgentHandler) AggregateMetrics(c *gin.Context) {
 			buckets[bucketStart] = append(buckets[bucketStart], val)
 		}

-		// 为每个桶计算 OHLC + Avg
+		// 计算 max/min/avg/p95
 		for bucketTs, values := range buckets {
 			if len(values) == 0 {
 				continue
 			}
-			open := values[0]
-			closeV := values[len(values)-1]
-			high := values[0]
-			low := values[0]
+			maxV := values[0]
+			minV := values[0]
 			sum := 0.0

 			for _, v := range values {
 				sum += v
-				if v > high {
-					high = v
+				if v > maxV {
+					maxV = v
 				}
-				if v < low {
-					low = v
+				if v < minV {
+					minV = v
 				}
 			}
 			avg := sum / float64(len(values))

-			kline := models.MetricsKline{
+			// 计算 P95（排序后取 95% 位置的值）
+			p95 := maxV
+			if len(values) >= 20 {
+				sorted := make([]float64, len(values))
+				copy(sorted, values)
+				sort.Float64s(sorted)
+				idx := int(float64(len(sorted)) * 0.95)
+				if idx >= len(sorted) {
+					idx = len(sorted) - 1
+				}
+				p95 = sorted[idx]
+			}
+
+			agg := models.MetricsAggregate{
 				Hostname:   req.Hostname,
 				Metric:     req.Metric,
 				Interval:   iv.name,
-				Open:       math.Round(open*100) / 100,
-				High:       math.Round(high*100) / 100,
-				Low:        math.Round(low*100) / 100,
-				Close:      math.Round(closeV*100) / 100,
 				Avg:        math.Round(avg*100) / 100,
+				MaxV:       math.Round(maxV*100) / 100,
+				MinV:       math.Round(minV*100) / 100,
+				P95:        math.Round(p95*100) / 100,
 				Count:      len(values),
 				BucketTime: time.Unix(bucketTs, 0),
 			}
@@ -347,9 +418,9 @@ func (h *AgentHandler) AggregateMetrics(c *gin.Context) {
 					{Name: "bucket_time"},
 				},
 				DoUpdates: clause.AssignmentColumns([]string{
-					"open", "high", "low", "close", "avg", "count",
+					"avg", "max_v", "min_v", "p95", "count",
 				}),
-			}).Create(&kline)
+			}).Create(&agg)

 			if result.Error == nil {
 				totalUpserted++
@@ -364,10 +435,10 @@ func (h *AgentHandler) AggregateMetrics(c *gin.Context) {
 	})
 }

-// GetKline 查询 K 线数据
-// GET /api/v1/agent/kline?hostname=&metric=&interval=&from=&to=
-func (h *AgentHandler) GetKline(c *gin.Context) {
-	var req models.KlineQueryReq
+// GetAggregate 查询聚合数据（原 GetKline）
+// GET /api/v1/agent/aggregate?hostname=&metric=&interval=&from=&to=
+func (h *AgentHandler) GetAggregate(c *gin.Context) {
+	var req models.AggregateQueryReq
 	if err := c.ShouldBindQuery(&req); err != nil {
 		c.JSON(http.StatusBadRequest, models.Response{Code: 400, Message: "参数错误: " + err.Error()})
 		return
@@ -384,25 +455,25 @@ func (h *AgentHandler) GetKline(c *gin.Context) {
 		return
 	}

-	var klines []models.MetricsKline
+	var items []models.MetricsAggregate
 	result := h.db.Where("hostname = ? AND metric = ? AND `interval` = ? AND bucket_time >= ? AND bucket_time <= ?",
 		req.Hostname, req.Metric, req.Interval, from, to).
 		Order("bucket_time ASC").
-		Find(&klines)
+		Find(&items)

 	if result.Error != nil {
 		c.JSON(http.StatusInternalServerError, models.Response{Code: 500, Message: "查询失败: " + result.Error.Error()})
 		return
 	}

-	if klines == nil {
-		klines = []models.MetricsKline{}
+	if items == nil {
+		items = []models.MetricsAggregate{}
 	}

 	c.JSON(http.StatusOK, models.Response{
 		Code:    200,
 		Message: "success",
-		Data:    klines,
+		Data:    items,
 	})
 }