⚡️ improve: 服务监控相关优化

2021-01-16 18:04:47 +08:00 · 2021-01-16 18:04:47 +08:00 · df0eca5a74
commit df0eca5a74
parent 161102a37b
6 changed files with 109 additions and 107 deletions
--- a/README.md
+++ b/README.md
@ -182,27 +182,7 @@ URL 里面也可放置占位符，请求时会进行简单的字符串替换。

 ## 变更日志

-`dashboard 0.2.2` `agent 0.2.2`
-
- `dashboard 0.2.2`
-
-    - 修复一直报警SSL证书问题
-
- `agent 0.2.2`
-
-    - 修复双栈IP只能获取到v6的问题 #61
-
- `dashboard 0.2.1` `agent 0.2.1`
-   
-   - dashboard
-     - 修复了默认开启IP变更通知
-     - hotaru 主题的服务状态页面
-     - **新增可以指定服务器忽略监控规则**
-     - 修复info透明 @ilay1678
-
-   - agent
-     - 优化了 IPv6/IPv4 双栈问题
-     - 增加 SSL 证书过期、即将过期提醒
+最新：`dashboard 0.2.3` `agent 0.2.3`，只记录最后一次更新导致必须更新面板的说明。

 - `dashboard 0.2.0` `agent 0.2.0` **重大更新**

--- a/cmd/agent/main.go
+++ b/cmd/agent/main.go
@ -147,81 +147,85 @@ func run(cmd *cobra.Command, args []string) {
 			continue
 		}
 		err = receiveTasks(tasks)
-		log.Printf("receiveCommand exit to main: %v", err)
+		log.Printf("receiveTasks exit to main: %v", err)
 		retry()
 	}
 }

 func receiveTasks(tasks pb.NezhaService_RequestTaskClient) error {
 	var err error
-	var task *pb.Task
-
-	defer log.Printf("receiveTasks exit %v %v => %v", time.Now(), task, err)
+	defer log.Printf("receiveTasks exit %v => %v", time.Now(), err)
 	for {
+		var task *pb.Task
 		task, err = tasks.Recv()
 		if err != nil {
 			return err
 		}
-		var result pb.TaskResult
-		result.Id = task.GetId()
-		result.Type = task.GetType()
-		switch task.GetType() {
-		case model.MonitorTypeHTTPGET:
-			start := time.Now()
-			resp, err := httpClient.Get(task.GetData())
-			if err == nil {
-				result.Delay = float32(time.Now().Sub(start).Microseconds()) / 1000.0
-				if resp.StatusCode > 299 || resp.StatusCode < 200 {
-					err = errors.New("\n应用错误：" + resp.Status)
-				}
-			}
-			if err == nil {
-				if strings.HasPrefix(task.GetData(), "https://") {
-					c := cert.NewCert(task.GetData()[8:])
-					if c.Error != "" {
-						if strings.Contains(c.Error, "expired") {
-							result.Data = "SSL证书错误：证书已过期"
-						} else {
-							result.Data = "SSL证书错误：" + c.Error
-						}
-					} else {
-						result.Data = c.Issuer + "|" + c.NotAfter
-						result.Successful = true
-					}
-				}
-			} else {
-				result.Data = err.Error()
-			}
-		case model.MonitorTypeICMPPing:
-			pinger, err := ping.NewPinger(task.GetData())
-			if err == nil {
-				pinger.Count = 10
-				err = pinger.Run() // Blocks until finished.
-			}
-			if err == nil {
-				stat := pinger.Statistics()
-				result.Delay = float32(stat.AvgRtt.Microseconds()) / 1000.0
-				result.Successful = true
-			} else {
-				result.Data = err.Error()
-			}
-		case model.MonitorTypeTCPPing:
-			start := time.Now()
-			conn, err := net.DialTimeout("tcp", task.GetData(), time.Second*10)
-			if err == nil {
-				conn.Close()
-				result.Delay = float32(time.Now().Sub(start).Microseconds()) / 1000.0
-				result.Successful = true
-			} else {
-				result.Data = err.Error()
-			}
-		default:
-			log.Printf("Unknown action: %v", task)
-		}
-		client.ReportTask(ctx, &result)
+		go doTask(task)
 	}
 }

+func doTask(task *pb.Task) {
+	var result pb.TaskResult
+	result.Id = task.GetId()
+	result.Type = task.GetType()
+	switch task.GetType() {
+	case model.MonitorTypeHTTPGET:
+		start := time.Now()
+		resp, err := httpClient.Get(task.GetData())
+		if err == nil {
+			result.Delay = float32(time.Now().Sub(start).Microseconds()) / 1000.0
+			if resp.StatusCode > 299 || resp.StatusCode < 200 {
+				err = errors.New("\n应用错误：" + resp.Status)
+			}
+		}
+		if err == nil {
+			if strings.HasPrefix(task.GetData(), "https://") {
+				c := cert.NewCert(task.GetData()[8:])
+				if c.Error != "" {
+					if strings.Contains(c.Error, "expired") {
+						result.Data = "SSL证书错误：证书已过期"
+					} else {
+						result.Data = "SSL证书错误：" + c.Error
+					}
+				} else {
+					result.Data = c.Issuer + "|" + c.NotAfter
+					result.Successful = true
+				}
+			}
+		} else {
+			result.Data = err.Error()
+		}
+	case model.MonitorTypeICMPPing:
+		pinger, err := ping.NewPinger(task.GetData())
+		if err == nil {
+			pinger.Count = 10
+			pinger.Timeout = time.Second * 20
+			err = pinger.Run() // Blocks until finished.
+		}
+		if err == nil {
+			result.Delay = float32(pinger.Statistics().AvgRtt.Microseconds()) / 1000.0
+			result.Successful = true
+		} else {
+			result.Data = err.Error()
+		}
+	case model.MonitorTypeTCPPing:
+		start := time.Now()
+		conn, err := net.DialTimeout("tcp", task.GetData(), time.Second*10)
+		if err == nil {
+			conn.Write([]byte("ping\n"))
+			conn.Close()
+			result.Delay = float32(time.Now().Sub(start).Microseconds()) / 1000.0
+			result.Successful = true
+		} else {
+			result.Data = err.Error()
+		}
+	default:
+		log.Printf("Unknown action: %v", task)
+	}
+	client.ReportTask(ctx, &result)
+}
+
 func reportState() {
 	var lastReportHostInfo time.Time
 	var err error
--- a/cmd/dashboard/main.go
+++ b/cmd/dashboard/main.go
@ -52,6 +52,6 @@ func initDB() {
 func main() {
 	go controller.ServeWeb(dao.Conf.HTTPPort)
 	go rpc.ServeRPC(5555)
-	go rpc.DispatchTask(time.Minute * 10)
+	go rpc.DispatchTask(time.Minute * 3)
 	alertmanager.Start()
 }
--- a/cmd/dashboard/rpc/rpc.go
+++ b/cmd/dashboard/rpc/rpc.go
@ -32,6 +32,7 @@ func DispatchTask(duration time.Duration) {
 		var hasAliveAgent bool
 		dao.DB.Find(&tasks)
 		dao.ServerLock.RLock()
+		startedAt := time.Now()
 		for i := 0; i < len(tasks); i++ {
 			if index >= uint64(len(dao.SortedServerList)) {
 				index = 0
@ -50,6 +51,6 @@ func DispatchTask(duration time.Duration) {
 			index++
 		}
 		dao.ServerLock.RUnlock()
-		time.Sleep(duration)
+		time.Sleep(time.Until(startedAt.Add(duration)))
 	}
 }
--- a/cmd/playground/main.go
+++ b/cmd/playground/main.go
@ -15,6 +15,35 @@ import (
 )

 func main() {
+	icmp()
+	// tcpping()
+	// httpWithSSLInfo()
+	// diskinfo()
+}
+
+func tcpping() {
+	start := time.Now()
+	conn, err := net.DialTimeout("tcp", "example.com:80", time.Second*10)
+	if err != nil {
+		panic(err)
+	}
+	conn.Write([]byte("ping\n"))
+	conn.Close()
+	fmt.Println(time.Now().Sub(start).Microseconds(), float32(time.Now().Sub(start).Microseconds())/1000.0)
+}
+
+func diskinfo() {
+	// 硬盘信息
+	dparts, _ := disk.Partitions(false)
+	for _, part := range dparts {
+		u, _ := disk.Usage(part.Mountpoint)
+		if u != nil {
+			log.Printf("%s %d %d", part.Device, u.Total, u.Used)
+		}
+	}
+}
+
+func httpWithSSLInfo() {
 	// 跳过 SSL 检查
 	transCfg := &http.Transport{
 		TLSClientConfig: &tls.Config{InsecureSkipVerify: true},
@ -25,31 +54,19 @@ func main() {
 	// SSL 证书信息获取
 	c := cert.NewCert("expired-ecc-dv.ssl.com")
 	fmt.Println(c.Error)
-	// TCP
-	conn, err := net.DialTimeout("tcp", "example.com:80", time.Second*10)
+}
+
+func icmp() {
+	pinger, err := ping.NewPinger("10.10.10.2")
 	if err != nil {
+		panic(err) // Blocks until finished.
+	}
+	pinger.Count = 3000
+	pinger.Timeout = 10 * time.Second
+	if err = pinger.Run(); err != nil {
 		panic(err)
 	}
-	println(conn)
-	// ICMP Ping
-	pinger, err := ping.NewPinger("example.com")
-	if err != nil {
-		panic(err)
-	}
-	pinger.Count = 3
-	err = pinger.Run() // Blocks until finished.
-	if err != nil {
-		panic(err)
-	}
-	fmt.Printf("%+v", pinger.Statistics())
-	// 硬盘信息
-	dparts, _ := disk.Partitions(false)
-	for _, part := range dparts {
-		u, _ := disk.Usage(part.Mountpoint)
-		if u != nil {
-			log.Printf("%s %d %d", part.Device, u.Total, u.Used)
-		}
-	}
+	fmt.Println(pinger.PacketsRecv, float32(pinger.Statistics().AvgRtt.Microseconds())/1000.0)
 }

 func cmdExec() {
--- a/service/dao/dao.go
+++ b/service/dao/dao.go
@ -26,7 +26,7 @@ var SortedServerList []*model.Server

 var ServerLock sync.RWMutex

-var Version = "v0.2.2"
+var Version = "v0.2.3"

 func ReSortServer() {
 	SortedServerList = []*model.Server{}