gnet: 一个轻量级且高性能的 Go 网络框架 使用笔记

gnet: 一个轻量级且高性能的 Go 网络框架 使用笔记

一个偶然的机会接触到了golang，被它的高并发传说所吸引，就开始学这门语言，越学感觉越有意思^_^

注册了博客园这么多年，第一次写东西，年纪大了，脑子不好使了，就得写下来，记下来，为了自己以后查阅，同时也分享给golang的爱好者。

gnet 在CSDN上的介绍如下：https://blog.csdn.net/qq_31967569/article/details/103107707

github v1版本的开源地址为：https://github.com/panjf2000/gnet

详细介绍看这里，英文版的有各个模块的介绍，较详细：https://pkg.go.dev/github.com/panjf2000/gnet?GOOS=windows#Option

gnet作者潘少是国内一枚年轻大帅哥^_^，在知乎上对gnet这个框架有不同的观点，说什么的也有，有不同的声音，是好事，喜欢就用，不喜欢就不用，请尊重开源作者。开源作者对go社区的贡献有目共睹。下面记录一下我在使用中的一些笔记。

先记录一下Options，为便于查询，保留原始英文备注

type Options     struct {

    // Multicore indicates whether the server will be effectively created with multi-cores, if so,
    // then you must take care with synchronizing memory between all event callbacks, otherwise,
    // it will run the server with single thread. The number of threads in the server will be automatically
    // assigned to the value of logical CPUs usable by the current process.
    // Multicore 表示服务器是否将有效地使用多核创建，如果是，则必须注意在所有事件回调之间同步内存，否则它将以单线程运行服务器。 
    // 服务器中的线程数将自动分配给当前进程可用的逻辑 CPU 的值。 
    Multicore bool

    // NumEventLoop is set up to start the given number of event-loop goroutine.　　 
    // NumEventLoop 设置为启动给定数量的event-loop goroutine,也就是 sub reactor的数量----这个参数的说明我特意私下请教过潘少
    // Note: Setting up NumEventLoop will override Multicore.　　　　 
    // 这个要注意，如果设置了NumEventLoop的值，那么Multicored 的参数将会被覆盖，潘少给的答复是一般用NumEventLoop这个参数，Multicore和NumEventLoop用哪个参数都行，需要自行在生产压力测试的时候验证　　　　 
    // NumEventLoop的值设置多少合适，也是经验值，需根据实际情况调整，使用这可以放在配置文件中，便于生产环境中随时调整而不需要修改代码
    NumEventLoop int

    // LB represents the load-balancing algorithm used when assigning new connections.
    LB LoadBalancing

    // ReuseAddr indicates whether to set up the SO_REUSEADDR socket option.
    ReuseAddr bool

    // ReusePort indicates whether to set up the SO_REUSEPORT socket option.
    ReusePort bool

    // ReadBufferCap is the maximum number of bytes that can be read from the peer when the readable event comes.
    // The default value is 64KB, it can be reduced to avoid starving the subsequent connections.
    // Note that ReadBufferCap will always be converted to the least power of two integer value greater than
    // or equal to its real amount.
    ReadBufferCap int

    // LockOSThread is used to determine whether each I/O event-loop is associated to an OS thread, it is useful when you
    // need some kind of mechanisms like thread local storage, or invoke certain C libraries (such as graphics lib: GLib)
    // that require thread-level manipulation via cgo, or want all I/O event-loops to actually run in parallel for a
    // potential higher performance.
    LockOSThread bool

    // Ticker indicates whether the ticker has been set up.
    Ticker bool

    // TCPKeepAlive sets up a duration for (SO_KEEPALIVE) socket option.
    TCPKeepAlive time.Duration

    // TCPNoDelay controls whether the operating system should delay
    // packet transmission in hopes of sending fewer packets (Nagle's algorithm).
    // The default is true (no delay), meaning that data is sent
    // as soon as possible after a write operation.
    TCPNoDelay TCPSocketOpt

    // SocketRecvBuffer sets the maximum socket receive buffer in bytes.
    SocketRecvBuffer int

    // SocketSendBuffer sets the maximum socket send buffer in bytes.
    SocketSendBuffer int

    // ICodec encodes and decodes TCP stream.
    Codec ICodec

    // LogPath the local path where logs will be written, this is the easiest way to set up logging,
    // gnet instantiates a default uber-go/zap logger with this given log path, you are also allowed to employ
    // you own logger during the lifetime by implementing the following log.Logger interface.
    // Note that this option can be overridden by the option Logger.
    LogPath string

    // LogLevel indicates the logging level, it should be used along with LogPath.
    LogLevel logging.Level

    // Logger is the customized logger for logging info, if it is not set,
    // then gnet will use the default logger powered by go.uber.org/zap.
    Logger logging.Logger
}

　再看一下EventHandler

type EventHandler     interface {
    // OnInitComplete fires when the server is ready for accepting connections.
    // The parameter server has information and various utilities.
    OnInitComplete(server Server) (action Action)

    // OnShutdown fires when the server is being shut down, it is called right after
    // all event-loops and connections are closed.
    OnShutdown(server Server)

    // OnOpened fires when a new connection has been opened.
    // The Conn c has information about the connection such as it's local and remote address.
    // The parameter out is the return value which is going to be sent back to the peer.
    // It is usually not recommended to send large amounts of data back to the peer in OnOpened.
    // Note that the bytes returned by OnOpened will be sent back to the peer without being encoded.
    OnOpened(c Conn) (out []byte, action Action)

    // OnClosed fires when a connection has been closed.
    // The parameter err is the last known connection error.
    OnClosed(c Conn, err error) (action Action)

    // PreWrite fires just before a packet is written to the peer socket, this event function is usually where
    // you put some code of logging/counting/reporting or any fore operations before writing data to the peer.
    PreWrite(c Conn)

    // AfterWrite fires right after a packet is written to the peer socket, this event function is usually where
    // you put the []byte returned from React() back to your memory pool.
    AfterWrite(c Conn, b []byte)

    // React fires when a socket receives data from the peer.
    // Call c.Read() or c.ReadN(n) of Conn c to read incoming data from the peer.
    // The parameter out is the return value which is going to be sent back to the peer.
    // Note that the parameter packet returned from React() is not allowed to be passed to a new goroutine,
    // as this []byte will be reused within event-loop after React() returns.
    // If you have to use packet in a new goroutine, then you need to make a copy of buf and pass this copy
    // to that new goroutine.　　　　 
    // 我们的业务处理应该就是对packet中数据的处理，需要注意的是，如果业务处理放在新的goroutine中执行，那么需要将复制packet的副本传递进去        
    // 具体为什么，作者没说，我猜测可能是因为切片属于引用类型，新的gorountine的生命周期会影响整个gnet的工作
    React(packet []byte, c Conn) (out []byte, action Action)

    // Tick fires immediately after the server starts and will fire again
    // following the duration specified by the delay return value.
    Tick() (delay time.Duration, action Action)
}

下面是官方给的带阻塞的使用方法，使用了pool

package main

import (
    "log"
    "time"

    "github.com/panjf2000/gnet"
    "github.com/panjf2000/gnet/pool"
)

type echoServer    struct {
    *gnet.EventServer
    pool *pool.WorkerPool
}

func (es *echoServer) React(c gnet.Conn) (out []byte, action gnet.Action) {
    data := append([]byte{}, c.Read()...)
    c.ResetBuffer()

    // Use ants pool to unblock the event-loop.
    _ = es.pool.Submit(   func () {
        time.Sleep(1 * time.Second)
        c.AsyncWrite(data)
    })

    return
}

func main() {
    p := pool.NewWorkerPool()
    defer p.Release()
 
    echo := &echoServer{pool: p}
    log.Fatal(gnet.Serve(echo,    "tcp://:9000" , gnet.WithMulticore(true)))
}

在Ubuntu上压测过，单台连接过6万结点，很轻松，当然，这个连接是没有任何业务处理的。仅仅是一个echo的测试。注意，在Ubuntu上测试的时候 需要修改 ulimite的值，默认1024太小。修改ulimite值的方法自行搜索。

目前遇到的问题是，options参数为 gnet.WithMulticore(true) 时，在windows上压测的时候，从任务管理器上看，gnet创建了很多线程，即使所有客户端都断开，这些线程也不减少，这个不知道什么原因。我私下请教过作者，作者说，windows平台下仅用于开发调试，不能用于生产。