码迷,mamicode.com
首页 > Web开发 > 详细

asp.net core mvc剖析:KestrelServer

时间:2017-05-04 10:59:33      阅读:260      评论:0      收藏:0      [点我收藏+]

标签:ati   else   时间   业务逻辑   int()   net   local   str   push   

KestrelServer是基于Libuv开发的高性能web服务器,那我们现在就来看一下它是如何工作的。在上一篇文章中提到了Program的Main方法,在这个方法里Build了一个WebHost,我们再来看一下代码:

1
2
3
4
5
6
7
8
9
10
11
public static void Main(string[] args)
   {
       var host = new WebHostBuilder()
           .UseKestrel()
           .UseContentRoot(Directory.GetCurrentDirectory())
           .UseIISIntegration()
           .UseStartup<Startup>()
           .Build();
  
       host.Run();
   }

  里面有一个UseKestrel方法调用,这个方法的作用就是使用KestrelServer作为web server来提供web服务。在WebHost启动的时候,调用了IServer的Start方法启动服务,由于我们使用KestrelServer作为web server,自然这里调用的就是KestrelServer.Start方法,那我们来看下KestrelServer的Start方法里主要代码:

 首先,我们发现在Start方法里创建了一个KestrelEngine对象,具体代码如下:

1
2
3
4
5
6
7
8
9
10
11
12
var engine = new KestrelEngine(new ServiceContext
{
       FrameFactory = context =>
       {
           return new Frame<TContext>(application, context);
       },
       AppLifetime = _applicationLifetime,
       Log = trace,
       ThreadPool = new LoggingThreadPool(trace),
       DateHeaderValueManager = dateHeaderValueManager,
       ServerOptions = Options
 });

  KestrelEngine构造方法接受一个ServiceContext对象参数,ServiceContext里包含一个FrameFactory,从名称上很好理解,就是Frame得工厂,Frame是什么?Frame是http请求处理对象,每个请求过来后,都会交给一个Frame对象进行受理,我们这里先记住它的作用,后面还会看到它是怎么实例化的。除了这个外,还有一个是AppLiftTime,它是一个IApplicationLifetime对象,它是整个应用生命周期的管理对象,前面没有说到,这里补充上。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
public interface IApplicationLifetime
    {
        /// <summary>
        /// Triggered when the application host has fully started and is about to wait
        /// for a graceful shutdown.
        /// </summary>
        CancellationToken ApplicationStarted { get; }
        /// <summary>
        /// Triggered when the application host is performing a graceful shutdown.
        /// Requests may still be in flight. Shutdown will block until this event completes.
        /// </summary>
        CancellationToken ApplicationStopping { get; }
        /// <summary>
        /// Triggered when the application host is performing a graceful shutdown.
        /// All requests should be complete at this point. Shutdown will block
        /// until this event completes.
        /// </summary>
        CancellationToken ApplicationStopped { get; }
        /// <summary>
        /// Requests termination the current application.
        /// </summary>
        void StopApplication();
    }

  IApplicationLifetime中提供了三个时间点,

  1,ApplicationStarted:应用程序已启动
  2,ApplicationStopping:应用程序正在停止
  3,ApplicationStopped:应用程序已停止

  我们可以通过CancellationToken.Register方法注册回调方法,在上面说到的三个时间点,执行我们特定的业务逻辑。IApplicationLifetime是在WebHost的Start方法里创建的,如果想在我们自己的应用程序获取这个对象,我们可以直接通过依赖注入的方式获取即可。

 我们继续回到ServiceContext对象,这里面还包含了Log对象,用于跟踪日志,一般我们是用来看程序执行的过程,并可以通过它发现程序执行出现问题的地方。还包含一个ServerOptions,它是一个KestrelServerOptions,里面包含跟服务相关的配置参数:

1,ThreadCount:服务线程数,表示服务启动后,要开启多少个服务线程,因为每个请求都会使用一个线程来进行处理,多线程会提高吞吐量,但是并不一定线程数越多越好,在系统里默认值是跟CPU内核数相等。

2,ShutdownTimeout:The amount of time after the server begins shutting down before connections will be forcefully closed(在应用程序开始停止到强制关闭当前请求连接所等待的时间,在这个时间段内,应用程序会等待请求处理完,如果还没处理完,将强制关闭)

3,Limits:KestrelServerLimits对象,里面包含了服务限制参数,比如MaxRequestBufferSize,MaxResponseBufferSize

其他参数就不再一个一个说明了。

KestrelEngine对象创建好后,通过调用 engine.Start(threadCount),根据配置的threadcount进行服务线程KestrelThread实例化,代码如下:
技术分享
     public void Start(int count)
        {
            for (var index = 0; index < count; index++)
            {
                Threads.Add(new KestrelThread(this));
            }

            foreach (var thread in Threads)
            {
                thread.StartAsync().Wait();
            }
        }
技术分享

 上面的代码会创建指定数量的Thread对象,然后开始等待任务处理。KestrelThread是对libuv线程处理的封装。

这些工作都准备好后,就开始启动监听服务了

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
foreach (var endPoint in listenOptions)
               {
                   try
                   {
                       _disposables.Push(engine.CreateServer(endPoint));
                   }
                   catch (AggregateException ex)
                   {
                       if ((ex.InnerException as UvException)?.StatusCode == Constants.EADDRINUSE)
                       {
                           throw new IOException($"Failed to bind to address {endPoint}: address already in use.", ex);
                       }
                       throw;
                   }
                   // If requested port was "0", replace with assigned dynamic port.
                   _serverAddresses.Addresses.Add(endPoint.ToString());
               }

  上面红色字体代码,就是创建监听服务的方法,我们再详细看下里面的详细情况:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
      public IDisposable CreateServer(ListenOptions listenOptions)
        {
            var listeners = new List<iasyncdisposable>();
            try
            {                //如果前面创建的线程数量为1,直接创建listener对象,启动监听
                if (Threads.Count == 1)
                {
                    var listener = new Listener(ServiceContext);
                    listeners.Add(listener);
                    listener.StartAsync(listenOptions, Threads[0]).Wait();
                }
                else
                {            //如果线程数不为1的时候
                    var pipeName = (Libuv.IsWindows ? @"\\.\pipe\kestrel_" "/tmp/kestrel_") + Guid.NewGuid().ToString("n");
                    var pipeMessage = Guid.NewGuid().ToByteArray();
             //先创建一个主监听对象,这个Listenerprimary就是一个Listener,监听socket就是在这里面创建的
                    var listenerPrimary = new ListenerPrimary(ServiceContext);
                    listeners.Add(listenerPrimary);           //启动监听
                    listenerPrimary.StartAsync(pipeName, pipeMessage, listenOptions, Threads[0]).Wait();
                    //为剩余的每个服务线程关联一个ListenerSecondary对象,这个对象使用命名Pipe与主监听对象通信,在主监听对象接收到请求后,通过pipe把接受的socket对象发送给特定的线程处理
                    foreach (var thread in Threads.Skip(1))
                    {
                        var listenerSecondary = new ListenerSecondary(ServiceContext);
                        listeners.Add(listenerSecondary);
                        listenerSecondary.StartAsync(pipeName, pipeMessage, listenOptions, thread).Wait();
                    }
                }
                return new Disposable(() =>
                {
                    DisposeListeners(listeners);
                });
            }
            catch
            {
                DisposeListeners(listeners);
                throw;
            }
        }
</iasyncdisposable>

  这个时候服务就开始接受http请求了,我们前面说到了,监听socket在listener类中创建(ListenerPrimary也是一个Listener),下面是listener的start方法

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
      public Task StartAsync(
            ListenOptions listenOptions,
            KestrelThread thread)
        {
            ListenOptions = listenOptions;
            Thread = thread;
            var tcs = new TaskCompletionSource<int>(this);
            Thread.Post(state =>
            {
                var tcs2 = (TaskCompletionSource<int>) state;
                try
                {
                    var listener = ((Listener) tcs2.Task.AsyncState);                    //创建监听socket
                    listener.ListenSocket = listener.CreateListenSocket();                    //开始监听,当有连接请求过来后,触发ConnectionCallback方法
                    ListenSocket.Listen(Constants.ListenBacklog, ConnectionCallback, this);
                    tcs2.SetResult(0);
                }
                catch (Exception ex)
                {
                    tcs2.SetException(ex);
                }
            }, tcs);
            return tcs.Task;
        }
</int></int>

  ConnectionCallback:当连接请求过来后被触发,在回调方法里,进行连接处理分发,连接分发代码如下:

1
2
3
4
5
protected virtual void DispatchConnection(UvStreamHandle socket)
   {
       var connection = new Connection(this, socket);
       connection.Start();
   }

  这个是listener类中的实现,我们前面看到,只有在线程数为1的情况下,才创建Listener对象进行监听,否则创建ListenerPrimary监听,ListenerPrimay里重写了方法,它的实现如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
protected override void DispatchConnection(UvStreamHandle socket)
   {            //这里采用轮询的方式,把连接请求依次分发给不同的线程进行处理
       var index = _dispatchIndex++ % (_dispatchPipes.Count + 1);
       if (index == _dispatchPipes.Count)
       {              //
           base.DispatchConnection(socket);
       }
       else
       {
           DetachFromIOCP(socket);
           var dispatchPipe = _dispatchPipes[index];                //这里就是通过命名pipe,传递socket给特定的线程
           var write = new UvWriteReq(Log);
           write.Init(Thread.Loop);
           write.Write2(
               dispatchPipe,
               _dummyMessage,
               socket,
               (write2, status, error, state) =>
               {
                   write2.Dispose();
                   ((UvStreamHandle)state).Dispose();
               },
               socket);
       }
   }

  好了,连接请求找到处理线程后,后面就可以开始处理工作了。ListenerSecondary里的代码比较复杂,其实最终都会调用下面的代码完成Connection对象的创建

1
2
var connection = new Connection(this, socket);
connection.Start();

  Connection表示的就是当前连接,下面是它的构造方法

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
public Connection(ListenerContext context, UvStreamHandle socket) : base(context)
        {
            _socket = socket;
            _connectionAdapters = context.ListenOptions.ConnectionAdapters;
            socket.Connection = this;
            ConnectionControl = this;
            ConnectionId = GenerateConnectionId(Interlocked.Increment(ref _lastConnectionId));
            if (ServerOptions.Limits.MaxRequestBufferSize.HasValue)
            {
                _bufferSizeControl = new BufferSizeControl(ServerOptions.Limits.MaxRequestBufferSize.Value,this);
            }
        //创建输入输出socket流
            Input = new SocketInput(Thread.Memory, ThreadPool, _bufferSizeControl);
            Output = new SocketOutput(Thread, _socket, this, ConnectionId, Log, ThreadPool);
            var tcpHandle = _socket as UvTcpHandle;
            if (tcpHandle != null)
            {
                RemoteEndPoint = tcpHandle.GetPeerIPEndPoint();
                LocalEndPoint = tcpHandle.GetSockIPEndPoint();
            }
        //创建处理frame,这里的framefactory就是前面创建KestrelEngine时创建的工厂
            _frame = FrameFactory(this);
            _lastTimestamp = Thread.Loop.Now();
        }

  然后调用Connection的Start方法开始进行处理,这里面直接把处理任务交给Frame处理,Start方法实现:

1
2
3
4
5
6
7
8
9
10
11
12
public void Start()
        {
            Reset();       //启动了异步处理任务开始进行处理
            _requestProcessingTask =
                Task.Factory.StartNew(
                    (o) => ((Frame)o).RequestProcessingAsync(),//具体的处理方法
                    this,
                    default(CancellationToken),
                    TaskCreationOptions.DenyChildAttach,
                    TaskScheduler.Default).Unwrap();
            _frameStartedTcs.SetResult(null);
        }

  

1 RequestProcessingAsync方法里不再详细介绍了,把主要的代码拿出来看一下:
1
2
3
4
5
。。。。。//_application就是上一篇文章提到的HostApplication,首先调用CreateContext创建HttpContext对象
var context = _application.CreateContext(this);
。。。。。。//进入处理管道
await _application.ProcessRequestAsync(context).ConfigureAwait(false);
。。。。。。

  

1 ProcessRequestAsync完成处理后,把结果输出给客户端,好到此介绍完毕。如果有问题,欢迎大家指点。

asp.net core mvc剖析:KestrelServer

标签:ati   else   时间   业务逻辑   int()   net   local   str   push   

原文地址:http://www.cnblogs.com/xieweikai/p/6805673.html

(0)
(0)
   
举报
评论 一句话评论(0
登录后才能评论!
© 2014 mamicode.com 版权所有  联系我们:gaon5@hotmail.com
迷上了代码!