我有这个测试代码。
一段时间后，所有gorutines都会锁定carbonwritebuf.flush（）而不会出现任何错误。停下来。
线程越多，间隔越短，停止越快。
如果我杀死接收到的应用程序（NC-L 127.0.0.1 2002-K-M 1000），Golang会重置连接并再次工作。但是，如果gorutine已经停止，那么就没有重启和任何消息。

package main

import (
    "bufio"
    "flag"
    "fmt"
    "net"
    "time"
)

var (
    SERVER   string
    INTERVAL int
    THREADS  int
)

func main() {
    flag.StringVar(&SERVER, "s", "localhost:2002", "server address:port (Default: localhost:2002)")
    flag.IntVar(&INTERVAL, "i", 500, "send interval (Default: 500")
    flag.IntVar(&THREADS, "t", 100, "num of threads (Default: 100")
    flag.Parse()

    for i := 0; i < THREADS; i++ {
        go sender(i)
    }

    for {
    }
}

func sender(id int) {
    var carbonErr error
    var carbonWriteBuf *bufio.Writer
    var carbonConn net.Conn
    defer fmt.Println("End 1")
    x := 0

    i := 10
    isUp := true

    updater := time.NewTicker(time.Duration(INTERVAL*3) * time.Millisecond)
    go func() {
        defer fmt.Println("End 2")
        for range updater.C {
            if isUp {
                i++
            } else {
                i--
            }

            if i == 20 {
                isUp = false
            }

            if i == 0 {
                isUp = true
            }
            fmt.Println(x)
        }
    }()

    dataSender := time.NewTicker(time.Duration(INTERVAL) * time.Millisecond)
    func() {
        defer fmt.Println("End 3")
        for range dataSender.C {
            x++
            if carbonConn != nil {
                //carbonConn.SetWriteDeadline(time.Now().Add(time.Second * 3))
                if err := carbonConn.SetDeadline(time.Now().Add(time.Second * 3)); err != nil {
                    fmt.Println("Cannot set deadline", err)
                }
                if _, writeErr := carbonWriteBuf.WriteString(fmt.Sprintf("test.%d.value %d %d\n", id, i, time.Now().Unix())); writeErr == nil {
                    fmt.Println("Buffered:", carbonWriteBuf.Buffered())

                    if flushErr := carbonWriteBuf.Flush(); flushErr == nil {
                        fmt.Println("Send", time.Now().Unix())
                    } else {
                        fmt.Printf("Cannot write data to carbon %s Error: %s\n", carbonConn.RemoteAddr(), flushErr)
                        carbonConn = nil
                    }
                } else {
                    fmt.Println("Cannot write data to buffer:", writeErr)
                    carbonConn = nil
                }
            } else {
                if carbonConn, carbonErr = net.Dial("tcp", SERVER); carbonErr != nil {
                    carbonConn = nil
                    fmt.Println("Cannot connect to carbon:", carbonErr)
                } else {
                    defer carbonConn.Close()
                    carbonWriteBuf = bufio.NewWriterSize(carbonConn, 4096)
                    //carbonWriteBuf = bufio.NewWriter(carbonConn)
                    fmt.Println("Successfully connected to carbon:", carbonConn.RemoteAddr())
                }
            }
        }
    }()
}

最佳答案

主线程中的忙循环出现问题。谢谢吉姆

关于go - Golang在TCP上死锁。一段时间后写入，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/50876267/