线程模型

SocketAsyncEventArgs编程模式不支持设置同时工作线程个数，使用的NET的IO线程，由NET底层提供，这点和直接使用完成端口API编程不同。NET底层IO线程也是每个异步事件都是由不同的线程返回到Completed事件，因此在Completed事件需要对用户对象进行加锁，避免同一个用户对象同时触发两个Completed事件。

        void IO_Completed(object sender, SocketAsyncEventArgs asyncEventArgs)
        {
            AsyncSocketUserToken userToken = asyncEventArgs.UserToken as AsyncSocketUserToken;
            userToken.ActiveDateTime = DateTime.Now;
            try
            {                
                lock (userToken) //避免同一个userToken同时有多个线程操作
                {
                    if (asyncEventArgs.LastOperation == SocketAsyncOperation.Receive)
                        ProcessReceive(asyncEventArgs);
                    else if (asyncEventArgs.LastOperation == SocketAsyncOperation.Send)
                        ProcessSend(asyncEventArgs);
                    else
                        throw new ArgumentException("The last operation completed on the socket was not a receive or send");
                }   
            }
            catch (Exception E)
            {
                Program.Logger.ErrorFormat("IO_Completed {0} error, message: {1}", userToken.ConnectSocket, E.Message);
                Program.Logger.Error(E.StackTrace);
            }                     
        }

使用ProceXP可以看到服务端在比较忙的时候，服务的线程会越多。在Completed事件加锁有好处是后续逻辑处理都是串行的，可以不用考虑线程同步。还有一个地方需要注意的是断开超时连接，由于超时连接会调用Shutdown函数来强行中断SOCKET，因此在守护线程调用时，也需要锁住userToken对象。

        public void DaemonThreadStart()
        {
            while (m_thread.IsAlive)
            {
                AsyncSocketUserToken[] userTokenArray = null;
                m_asyncSocketServer.AsyncSocketUserTokenList.CopyList(ref userTokenArray);
                for (int i = 0; i < userTokenArray.Length; i++)
                {
                    if (!m_thread.IsAlive)
                        break;
                    try
                    {
                        if ((DateTime.Now - userTokenArray[i].ActiveDateTime).Milliseconds > m_asyncSocketServer.SocketTimeOutMS) //超时Socket断开
                        {
                            lock (userTokenArray[i])
                            {
                                m_asyncSocketServer.CloseClientSocket(userTokenArray[i]);
                            }
                        }
                    }                    
                    catch (Exception E)
                    {
                        Program.Logger.ErrorFormat("Daemon thread check timeout socket error, message: {0}", E.Message);
                        Program.Logger.Error(E.StackTrace);
                    }
                }

                for (int i = 0; i < 60 * 1000 / 10; i++) //每分钟检测一次
                {
                    if (!m_thread.IsAlive)
                        break;
                    Thread.Sleep(10);
                }
            }
        }

在CloseClientSocket方法中，对m_asyncSocketUserTokenPool和m_asyncSocketUserTokenList进行处理的时候都有加锁，代码如下：

        public void CloseClientSocket(AsyncSocketUserToken userToken)
        {
            if (userToken.ConnectSocket == null)
                return;
            string socketInfo = string.Format("Local Address: {0} Remote Address: {1}", userToken.ConnectSocket.LocalEndPoint,
                userToken.Conn