go - 包装 net.Conn 会导致挂起读取

标签 go networking dns udp

我是一名 Go 新手,试图在生产环境中调试一些网络问题,该应用程序遇到了一些我怀疑与 DNS 解析相关的问题。长话短说,在成功连接到 Redis 集群并与之交互后,连接失败,并且应用程序开始记录大量 dial: tcp :0: 连接被拒绝 错误。此消息中的 :0: 是我怀疑 dns 的原因,因为从我收集到的信息来看,这表明它正在尝试连接到零值远程地址,我真的不知道为什么除非主机名查找出现问题,否则就会发生这种情况。

无论如何,如果有人对这个特定问题有一些见解,我很想听听,但这里的主要问题与我在尝试调试此问题时偶然发现的问题有关:

为了了解主机名解析的实际情况,我将自定义解析器传递到拨号器中,该解析器从其自己的拨号器返回包装的连接。连接包装器本质上只是记录在 native 连接上读取和写入的字节,而不记录其他任何内容。简单的可重现示例如下:

package main

import (
    "context"
    "fmt"
    "net"
    "time"
)

type ConnWrapper struct {
    net.Conn
}

func (c *ConnWrapper) Write(bytes []byte) (int, error) {
    n, err := c.Conn.Write(bytes)
    if err != nil {
        fmt.Printf("Failed to write bytes %s: %v\n", string(bytes), err)
    } else {
        fmt.Printf("Successfully wrote %d of bytes %x\n", n, bytes[:n])
    }
    return n, err
}

func (c *ConnWrapper) Read(bytes []byte) (int, error) {
    n, err := c.Conn.Read(bytes)
    if err != nil {
        fmt.Printf("Failed to read bytes: %v\n", err)
    } else {
        fmt.Printf("Successfully read %d of bytes %x\n", n, bytes[:n])
    }
    return n, err
}

func (c *ConnWrapper) SetDeadline(t time.Time) error {
    fmt.Printf("Setting deadline %v %s %s\n", t, c.LocalAddr(), c.RemoteAddr())
    return c.Conn.SetDeadline(t)
}

func (c *ConnWrapper) SetReadDeadline(t time.Time) error {
    fmt.Printf("Setting read deadline %v %s %s\n", t, c.LocalAddr(), c.RemoteAddr())
    return c.Conn.SetReadDeadline(t)
}

func (c *ConnWrapper) SetWriteDeadline(t time.Time) error {
    fmt.Printf("Setting write deadline %v %s %s\n", t, c.LocalAddr(), c.RemoteAddr())
    return c.Conn.SetWriteDeadline(t)
}

func main() {
    var d *net.Dialer
    d = &net.Dialer{
        Timeout: time.Duration(30) * time.Minute,
        Resolver: &net.Resolver{
            PreferGo: true,
            Dial: func(ctx context.Context, network, address string) (net.Conn, error) {
                fmt.Printf("Redis dialing (from resolver) %s %s\n", network, address)
                conn, err := d.DialContext(ctx, network, address)

                if err != nil {
                    fmt.Printf("Redis resolver failed %v\n", err)
                }

                // When I return c2, the reads time out
                c2 := &ConnWrapper{conn}
                return c2, err

                // When I return conn, everything is fine
                //return conn, err
            },
        },
    }
    conn, err := d.DialContext(context.Background(), "tcp", "redis-node-0:6379")
    if err != nil {
        fmt.Printf("Redis dial failed %v\n", err)
    } else {
        fmt.Printf("Successfully dialed %s\n", conn.RemoteAddr())
    }
}

我不明白的是,当我返回 native 连接时,一切都会按预期进行。当我返回包装的连接时,连接在解析 DNS 名称时挂起/超时(特别是在 conn.Read 调用中)。我不知道为什么会发生这种情况,但我对 Go 的实践经验也很少,所以我想知道是否有一个我没有看到的简单解释。我尝试向 Google 和 ChatGPT 询问,但没有成功,因此希望社区中有人能提出一些我可以采用的想法。

执行上面的程序会在我的系统上产生以下输出(使用 golang 1.21)

Redis dialing (from resolver) udp 1.1.1.1:53
Redis dialing (from resolver) udp 1.1.1.1:53
Setting deadline 2023-10-15 10:43:44.031475 -0400 AST m=+5.001741085 192.168.68.61:62862 1.1.1.1:53
Setting deadline 2023-10-15 10:43:44.031494 -0400 AST m=+5.001759751 192.168.68.61:57358 1.1.1.1:53
Successfully wrote 49 of bytes 002f53c8010000010000000000010c72656469732d6e6f64652d30056c6f63616c00001c000100002904d0000000000000
Successfully wrote 49 of bytes 002f29c8010000010000000000010c72656469732d6e6f64652d30056c6f63616c000001000100002904d0000000000000
Successfully read 2 of bytes 002f
Successfully read 2 of bytes 002f
Failed to read bytes: read udp 192.168.68.61:57358->1.1.1.1:53: i/o timeout
Redis dialing (from resolver) udp 1.0.0.1:53
Failed to read bytes: read udp 192.168.68.61:62862->1.1.1.1:53: i/o timeout
Redis dialing (from resolver) udp 1.0.0.1:53
Setting deadline 2023-10-15 10:43:49.033465 -0400 AST m=+10.003676876 192.168.68.61:64641 1.0.0.1:53
Setting deadline 2023-10-15 10:43:49.033817 -0400 AST m=+10.004027543 192.168.68.61:59264 1.0.0.1:53
Successfully wrote 49 of bytes 002f6612010000010000000000010c72656469732d6e6f64652d30056c6f63616c00001c000100002904d0000000000000
Successfully wrote 49 of bytes 002fd26c010000010000000000010c72656469732d6e6f64652d30056c6f63616c000001000100002904d0000000000000
Successfully read 2 of bytes 002f
Failed to read bytes: read udp 192.168.68.61:59264->1.0.0.1:53: i/o timeout
Redis dialing (from resolver) udp 1.1.1.1:53
Setting deadline 2023-10-15 10:43:54.036073 -0400 AST m=+15.006228210 192.168.68.61:56757 1.1.1.1:53
Failed to read bytes: read udp 192.168.68.61:64641->1.0.0.1:53: i/o timeout
Redis dialing (from resolver) udp 1.1.1.1:53
Successfully wrote 49 of bytes 002f4f6d010000010000000000010c72656469732d6e6f64652d30056c6f63616c00001c000100002904d0000000000000
Setting deadline 2023-10-15 10:43:54.036487 -0400 AST m=+15.006642126 192.168.68.61:58296 1.1.1.1:53
Successfully wrote 49 of bytes 002f404c010000010000000000010c72656469732d6e6f64652d30056c6f63616c000001000100002904d0000000000000
Successfully read 2 of bytes 002f
Successfully read 2 of bytes 002f
Failed to read bytes: read udp 192.168.68.61:58296->1.1.1.1:53: i/o timeout
Failed to read bytes: read udp 192.168.68.61:56757->1.1.1.1:53: i/o timeout

当我更新程序以简单地返回内置 Conn 实例时,我得到以下信息(请注意,我在 redis 主机不可用的环境中运行此程序,因此实际上期望“没有这样的主机”这里的行为。重点是 dns 查找实际上完成了。)

Redis dialing (from resolver) udp 1.1.1.1:53
Redis dialing (from resolver) udp 1.1.1.1:53
Redis dialing (from resolver) udp 1.1.1.1:53
Redis dialing (from resolver) udp 1.1.1.1:53
Redis dial failed dial tcp: lookup redis-node-0 on 1.1.1.1:53: no such host

最佳答案

解析器区分 net.PacketConnnet.Conn (documentation)。如果 dial 返回数据包连接,则返回数据包连接包装器。

这是包装器:

type PacketConnWrapper struct {
    ConnWrapper
    pc net.PacketConn
}

func (c PacketConnWrapper) ReadFrom(bytes []byte) (int, net.Addr, error) {
    n, addr, err := c.pc.ReadFrom(bytes)
    if err != nil {
        fmt.Printf("Failed to read from bytes: %v\n", err)
    } else {
        fmt.Printf("Successfully read %d bytes %x from %s\n", n, bytes[:n], addr)
    }
    return n, addr, err
}

func (c PacketConnWrapper) WriteTo(bytes []byte, addr net.Addr) (int, error) {
    n, err := c.pc.WriteTo(bytes, addr)
    if err != nil {
        fmt.Printf("Failed to write bytes %x to %v: %s\n", bytes, addr, err)
    } else {
        fmt.Printf("Successfully write bytes %x to %s\n", bytes, addr)
    }
    return n, err
}

在拨号函数中使用数据包包装器,如下所示:

d = &net.Dialer{
    Timeout: time.Duration(30) * time.Minute,
    Resolver: &net.Resolver{
        PreferGo: true,
        Dial: func(ctx context.Context, network, address string) (net.Conn, error) {
            fmt.Printf("Redis dialing (from resolver) %s %s\n", network, address)
            conn, err := d.DialContext(ctx, network, address)

            if err != nil {
                fmt.Printf("Redis resolver failed %v\n", err)
                return conn, err
            }

            if pc, ok := conn.(net.PacketConn); ok {
                return PacketConnWrapper{ConnWrapper{conn}, pc}, nil
            }

            return ConnWrapper{conn}, nil
        },
    },
 }

https://go.dev/play/p/1KOr95FbDKD

关于go - 包装 net.Conn 会导致挂起读取,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/77295050/

相关文章:

c# - Dns.GetHostEntry 能否返回一个地址列表为空的 IPHostEntry?

go - 如何修改golang请求对象?

c# - 点对点去中心化网络——向所有点发送消息

dns - 具有单独区域的 Google Cloud DNS 子域委派

java - 无法通过 ssh 隧道通过 Jedis 连接到 redis

c++ - 从 sockaddr * 转换为 sockaddr_in * 增加了所需的对齐方式

DNS响应答案&权威版 block

go - 如何从Google Cloud VM部署Hugo网站?

javascript - golang api 请求在浏览器中有效,但在从本地运行的 JavaScript 请求时返回 401(未经授权)

google-app-engine - 为什么 GAE 标准环境中没有 Cloud Endpoints for Golang?