ASP.NET Core 6框架揭秘实例演示[28]：自定义一个服务器

作为ASP.NET Core请求处理管道的“龙头”的服务器负责监听和接收请求并最终完成对请求的响应。它将原始的请求上下文描述为相应的特性（Feature），并以此将HttpContext上下文创建出来，中间件针对HttpContext上下文的所有操作将借助于这些特性转移到原始的请求上下文上。学习ASP.NET Core框架最有效的方式就是按照它的原理“再造”一个框架，了解服务器的本质最好的手段就是试着自定义一个服务器。现在我们自定义一个真正的服务器。在此之前，我们再来回顾一下表示服务器的IServer接口。(本篇提供的实例已经汇总到《ASP.NET Core 6框架揭秘-实例演示版》）

一、IServer

二、请求和响应特性

三、StreamBodyFeature

四、HttpListenerServer

一、IServer

作为服务器的IServer对象利用如下所示的Features属性提供了与自身相关的特性。除了利用StartAsync<TContext>和StopAsync方法启动和关闭服务器之外，它还实现了IDisposable接口，资源的释放工作可以通过实现的Dispose方法来完成。StartAsync<TContext>方法将IHttpApplication<TContext>类型的参数作为处理请求的“应用”，该对象是对中间件管道的封装。从这个意义上讲，服务器就是传输层和这个IHttpApplication<TContext>对象之间的“中介”。

public interface IServer : IDisposable
{
    IFeatureCollection Features { get; }

    Task StartAsync<TContext>(IHttpApplication<TContext> application, CancellationToken cancellationToken) where TContext : notnull;
    Task StopAsync(CancellationToken cancellationToken);
}

虽然不同服务器类型的定义方式千差万别，但是背后的模式基本上与下面这个以伪代码定义的服务器类型一致。如下这个Server利用IListener对象来监听和接收请求，该对象是利用构造函数中注入的IListenerFactory工厂根据指定的监听地址创建出来的。StartAsync<TContext>方法从Features特性集合中提取出IServerAddressesFeature特性，并针对它提供的每个监听地址创建一个IListener对象。该方法为每个IListener对象开启一个“接收和处理请求”的循环，循环中的每次迭代都会调用IListener对象的AcceptAsync方法来接收请求，我们利用RequestContext对象来表示请求上下文。

public class Server : IServer
{
    private readonly IListenerFactory _listenerFactory;
    private readonly List<IListener> _listeners = new();

    public IFeatureCollection Features { get; } = new FeatureCollection();

    public Server(IListenerFactory listenerFactory) => _listenerFactory = listenerFactory;

    public async Task StartAsync<TContext>(IHttpApplication<TContext> application, CancellationToken cancellationToken) where TContext : notnull
    {
        var addressFeature = Features.Get<IServerAddressesFeature>()!;
        foreach (var address in addressFeature.Addresses)
        {
            var listener = await _listenerFactory.BindAsync(address);
            _listeners.Add(listener);
            _ = StartAcceptLoopAsync(listener);
        }

        async Task StartAcceptLoopAsync(IListener listener)
        {
            while (true)
            {
                var requestContext = await listener.AcceptAsync();
                _ = ProcessRequestAsync(requestContext);
            }
        }

        async Task ProcessRequestAsync(RequestContext requestContext)
        {
            var feature = new RequestContextFeature(requestContext);
            var contextFeatures = new FeatureCollection();
            contextFeatures.Set<IHttpRequestFeature>(feature);
            contextFeatures.Set<IHttpResponseFeature>(feature);
            contextFeatures.Set<IHttpResponseBodyFeature>(feature);

            var context = application.CreateContext(contextFeatures);
            Exception? exception = null;
            try
            {
                await application.ProcessRequestAsync(context);
            }
            catch (Exception ex)
            {
                exception = ex;
            }
            finally
            {
                application.DisposeContext(context, exception);
            }
        }
    }
    public Task StopAsync(CancellationToken cancellationToken) => Task.WhenAll(_listeners.Select(listener => listener.StopAsync()));

    public void Dispose() => _listeners.ForEach(listener => listener.Dispose());
}

public interface IListenerFactory
{
    Task<IListener> BindAsync(string listenAddress);
}

public interface IListener : IDisposable
{

    Task<RequestContext> AcceptAsync();
    Task StopAsync();
}

public class RequestContext
{
     ...
}

public class RequestContextFeature : IHttpRequestFeature, IHttpResponseFeature, IHttpResponseBodyFeature
{
    public RequestContextFeature(RequestContext requestContext);
    ...
}

StartAsync<TContext>方法接下来利用此RequestContext上下文将RequestContextFeature特性创建出来。RequestContextFeature特性类型同时实现了IHttpRequestFeature, IHttpResponseFeature和 IHttpResponseBodyFeature这三个核心接口，我们特性针对这三个接口将特性对象添加到创建的FeatureCollection集合中。特性集合随后作为参数调用IHttpApplication<TContext>的CreateContext方法将TContext上下文创建出来，后者将进一步作为参数调用另一个ProcessRequestAsync方法将请求分发给中间件管道进行处理。待处理结束，IHttpApplication<TContext>对象的DisposeContext方法被调用，创建的TContext上下文承载的资源得以释放。

二、请求和响应特性

接下来我们将采用类似的模式来定义一个基于HttpListener的服务器。提供的HttpListenerServer的思路就是利用自定义特性来封装表示原始请求上下文的HttpListenerContext对象，我们使用HttpRequestFeature和HttpResponseFeature这个两个现成特性。

public class HttpRequestFeature : IHttpRequestFeature
{
    public string 		Protocol { get; set; }
    public string 		Scheme { get; set; }
    public string 		Method { get; set; }
    public string 		PathBase { get; set; }
    public string 		Path { get; set; }
    public string 		QueryString { get; set; }

    public string 		RawTarget { get; set; }
    public IHeaderDictionary 	Headers { get; set; }
    public Stream 		Body { get; set; }
}

public class HttpResponseFeature : IHttpResponseFeature
{
    public int 		StatusCode { get; set; }
    public string? 		ReasonPhrase { get; set; }
    public IHeaderDictionary 	Headers { get; set; }
    public Stream 		Body { get; set; }
    public virtual bool 	HasStarted => false;

    public HttpResponseFeature()
    {
        StatusCode = 200;
        Headers = new HeaderDictionary();
        Body = Stream.Null;
    }

    public virtual void OnStarting(Func<object, Task> callback, object state){}
    public virtual void OnCompleted(Func<object, Task> callback, object state){}
}

如果我们使用HttpRequestFeature来描述请求，意味着HttpListener在接受到请求之后需要将请求信息从HttpListenerContext上下文转移到该特性上。如果使用HttpResponseFeature来描述响应，待中间件管道在完成针对请求的处理后，我们还需要将该特性承载的响应数据应用到HttpListenerContext上下文上。

三、StreamBodyFeature

现在我们有了描述请求和响应的两个特性，还需要一个描述响应主体的特性，为此我们定义了如下这个StreamBodyFeature特性类型。StreamBodyFeature直接使用构造函数提供的Stream对象作为响应主体的输出流，并根据该对象创建出Writer属性返回的PipeWriter对象。本着“一切从简”的原则，我们并没有实现用来发送文件的SendFileAsync方法，其他成员也采用最简单的方式进行了实现。

public class StreamBodyFeature : IHttpResponseBodyFeature
{
    public Stream 	Stream { get; }
    public PipeWriter 	Writer { get; }

    public StreamBodyFeature(Stream stream)
    {
        Stream = stream;
        Writer = PipeWriter.Create(Stream);
    }

    public Task CompleteAsync() => Task.CompletedTask;
    public void DisableBuffering() { }
    public Task SendFileAsync(string path, long offset, long? count, CancellationToken cancellationToken = default)=> throw new NotImplementedException();
    public Task StartAsync(CancellationToken cancellationToken = default) => Task.CompletedTask;
}

四、HttpListenerServer

在如下这个自定义的HttpListenerServer服务器类型中，与传输层交互的HttpListener体现在_listener字段上。服务器在初始化过程中，它的Features属性返回的IFeatureCollection对象中添加了一个ServerAddressesFeature特性，因为我们需要用它来存放注册的监听地址。实现StartAsync<TContext>方法将监听地址从这个特性中取出来应用到HttpListener对象上。

public class HttpListenerServer : IServer
{
    private readonly HttpListener _listener = new();
    public IFeatureCollection Features { get; } = new FeatureCollection();

    public HttpListenerServer() => Features.Set<IServerAddressesFeature>(new ServerAddressesFeature());
    public Task StartAsync<TContext>(IHttpApplication<TContext> application,CancellationToken cancellationToken) where TContext : notnull
    {
        var pathbases = new HashSet<string>(StringComparer.OrdinalIgnoreCase);
        var addressesFeature = Features.Get<IServerAddressesFeature>()!;
        foreach (string address in addressesFeature.Addresses)
        {
            _listener.Prefixes.Add(address.TrimEnd('/') + "/");
            pathbases.Add(new Uri(address).AbsolutePath.TrimEnd('/'));
        }
        _listener.Start();

        while (true)
        {
            var listenerContext = _listener.GetContext();
            _ = ProcessRequestAsync(listenerContext);
        }

        async Task ProcessRequestAsync( HttpListenerContext listenerContext)
        {
            FeatureCollection features = new();
            var requestFeature = CreateRequestFeature(pathbases, listenerContext);
            var responseFeature = new HttpResponseFeature();
            var body = new MemoryStream();
            var bodyFeature = new StreamBodyFeature(body);
            features.Set<IHttpRequestFeature>(requestFeature);
            features.Set<IHttpResponseFeature>(responseFeature);
            features.Set<IHttpResponseBodyFeature>(bodyFeature);

            var context = application.CreateContext(features);
            Exception? exception = null;
            try
            {
                await application.ProcessRequestAsync(context);

                var response = listenerContext.Response;
                response.StatusCode = responseFeature.StatusCode;
                if (responseFeature.ReasonPhrase is not null)
                {
                    response.StatusDescription = responseFeature.ReasonPhrase;
                }
                foreach (var kv in responseFeature.Headers)
                {
                    response.AddHeader(kv.Key, kv.Value);
                }
                body.Position = 0;
                await body.CopyToAsync(listenerContext.Response.OutputStream);
            }
            catch (Exception ex)
            {
                exception = ex;
            }
            finally
            {
                body.Dispose();
                application.DisposeContext(context, exception);
                listenerContext.Response.Close();
            }
        }
    }
    public void Dispose() => _listener.Stop();

    private static HttpRequestFeature CreateRequestFeature(HashSet<string> pathbases,HttpListenerContext listenerContext)
    {
        var request 		= listenerContext.Request;
        var url 		= request.Url!;
        var absolutePath 	= url.AbsolutePath;
        var protocolVersion 	= request.ProtocolVersion;
        var requestHeaders 	= new HeaderDictionary();
        foreach (string key in request.Headers)
        {
            requestHeaders.Add(key, request.Headers.GetValues(key));
        }

        var requestFeature = new HttpRequestFeature
        {
            Body 		= request.InputStream,
            Headers 		= requestHeaders,
            Method 		= request.HttpMethod,
            QueryString 	                = url.Query,
            Scheme 		= url.Scheme,
            Protocol 		= $"{url.Scheme.ToUpper()}/{protocolVersion.Major}.{protocolVersion.Minor}"
        };
        var pathBase = pathbases.First(it => absolutePath.StartsWith(it, StringComparison.OrdinalIgnoreCase));
        requestFeature.Path = absolutePath[pathBase.Length..];
        requestFeature.PathBase = pathBase;
        return requestFeature;
    }

    public Task StopAsync(CancellationToken cancellationToken)
    {
        _listener.Stop();
        return Task.CompletedTask;
    }
}

在调用Start方法将HttpListener启动后，StartAsync<TContext>方法开始“请求接收处理”循环。接收到的请求上下文被封装成HttpListenerContext上下文，其承载的请求信息利用CreateRequestFeature方法转移到创建的HttpRequestFeature特性上。StartAsync<TContext>方法创建的“空”HttpResponseFeature对象来描述响应，另一个描述响应主体的StreamBodyFeature特性则根据创建的MemoryStream对象构建而成，意味着中间件管道写入的响应主体的内容将暂存到这个内存流中。我们将这三个特性注册到创建的FeatureCollection集合上，并将后者作为参数调用了IHttpApplication<TContext>对象的CreateContext方法将TContext上下文创建出来。此上下文进一步作为参数调用了IHttpApplication<TContext>对象的ProcessRequestAsync方法，中间件管道得以接管请求。

待中间件管道的处理工作完成后，响应的内容还暂存在两个特性中，我们还需要将它们应用到代表原始HttpListenerContext上下文上。StartAsync<TContext>方法从HttpResponseFeature特性提取出响应状态码和响应报头转移到HttpListenerContext上下文上，然后上述这个MemoryStream对象“拷贝”到HttpListenerContext上下文承载的响应主体输出流中。

using App;
using Microsoft.AspNetCore.Hosting.Server;
using Microsoft.Extensions.DependencyInjection.Extensions;

var builder = WebApplication.CreateBuilder(args);
builder.Services.Replace(ServiceDescriptor.Singleton<IServer, HttpListenerServer>());
var app = builder.Build();
app.Run(context => context.Response.WriteAsync("Hello World!"));
app.Run("http://localhost:5000/foobar/");

我们采用上面的演示程序来检测HttpListenerServer能否正常工作。我们为HttpListenerServer类型创建了一个ServiceDescriptor对象将现有的服务器的服务注册替换掉。在调用WebApplication对象的Run方法时显式指定了具有PathBase（“/foobar”）的监听地址“http://localhost:5000/foobar/”，如图1所示的浏览器以此地址访问应用，会得到我们希望的结果。

图1 HttpListenerServer返回的结果