red5源码分析---2

red5源码分析—服务器连接

和red5客户端类似，red5服务器端也使用了mina框架进行底层Socket的封装。另外，red5服务器端使用Spring管理相关的bean，在分析red5服务器端如何处理客户端的连接之前，下面先简单看一下red5 standalone的启动流程。

一. red5服务器启动

red5服务器端代码的main函数在org.red5.server.Bootstrap中，main函数加载org.red5.server.Launcher，Launcher通过Spring解析red5.xml文件，red5.xml文件会继续根据red5-common.xml和red5-core.xml创建相应的bean，并根据red5.globals文件创建一个GlobalScope，后面会提到。
本文从red5-core.xml文件中构造的RTMPMinaTransport说起，该类对应的Spring配置文件如下，

    <bean id="rtmpTransport" class="org.red5.server.net.rtmp.RTMPMinaTransport" init-method="start" destroy-method="stop">        <property name="ioHandler" ref="rtmpMinaIoHandler" />        <property name="addresses">            <list>                <value>${rtmp.host}:${rtmp.port}</value>            </list>        </property>        <property name="ioThreads" value="${rtmp.io_threads}" />        <property name="sendBufferSize" value="${rtmp.send_buffer_size}" />        <property name="receiveBufferSize" value="${rtmp.receive_buffer_size}" />        <property name="trafficClass" value="${rtmp.traffic_class}" />        <property name="backlog" value="${rtmp.backlog}" />        <property name="tcpNoDelay" value="${rtmp.tcp_nodelay}" />        <property name="keepAlive" value="${rtmp.tcp_keepalive}" />        <property name="thoughputCalcInterval" value="${rtmp.thoughput_calc_interval}" />        <property name="enableDefaultAcceptor" value="${rtmp.default_acceptor}" />        <property name="initialPoolSize" value="${rtmp.initial_pool_size}" />        <property name="maxPoolSize" value="${rtmp.max_pool_size}" />        <property name="maxProcessorPoolSize" value="${rtmp.max_processor_pool_size}" />        <property name="executorKeepAliveTime" value="${rtmp.executor_keepalive_time}" />        <property name="minaPollInterval" value="${jmx.mina.poll.interval}" />        <property name="enableMinaMonitor" value="${jmx.mina.monitor.enable}" />        <property name="enableMinaLogFilter" value="${mina.logfilter.enable}" />    </bean>

根据RTMPMinaTransport的配置，Spring框架会调用RTMPMinaTransport的start函数进行初始化，代码如下，

    public void start() throws Exception {        initIOHandler();        IoBuffer.setUseDirectBuffer(!useHeapBuffers);        if (useHeapBuffers) {            IoBuffer.setAllocator(new SimpleBufferAllocator());        }        if (enableDefaultAcceptor) {            acceptor = new NioSocketAcceptor(ioThreads);        } else {            SimpleIoProcessorPool<NioSession> pool = new SimpleIoProcessorPool<NioSession>(NioProcessor.class, maxProcessorPoolSize);            executor = new ThreadPoolExecutor(initialPoolSize, maxPoolSize, executorKeepAliveTime, TimeUnit.MILLISECONDS, new LinkedBlockingQueue<Runnable>(Short.MAX_VALUE));            acceptor = new NioSocketAcceptor(executor, pool);        }        if (enableMinaLogFilter) {            DefaultIoFilterChainBuilder chain = acceptor.getFilterChain();            LoggingFilter logFilter = new LoggingFilter(RTMPMinaTransport.class);            chain.addLast("logger", logFilter);        }        acceptor.setCloseOnDeactivation(true);        acceptor.setHandler(ioHandler);        acceptor.setBacklog(backlog);        SocketSessionConfig sessionConf = acceptor.getSessionConfig();        sessionConf.setReuseAddress(true);        sessionConf.setTcpNoDelay(tcpNoDelay);        sessionConf.setSendBufferSize(sendBufferSize);        sessionConf.setReceiveBufferSize(receiveBufferSize);        sessionConf.setMaxReadBufferSize(receiveBufferSize);        sessionConf.setThroughputCalculationInterval(thoughputCalcInterval);        sessionConf.setReaderIdleTime(readerIdleTime);        sessionConf.setKeepAlive(keepAlive);        if (trafficClass == -1) {        } else {            sessionConf.setTrafficClass(trafficClass);        }        acceptor.setReuseAddress(true);        try {            Set<InetSocketAddress> socketAddresses = new HashSet<InetSocketAddress>();            for (String addr : addresses) {                if (addr.indexOf(':') != -1) {                    String[] parts = addr.split(":");                    socketAddresses.add(new InetSocketAddress(parts[0], Integer.valueOf(parts[1])));                } else {                    socketAddresses.add(new InetSocketAddress(addr, 1935));                }            }            acceptor.bind(socketAddresses);            String cName = this.getClass().getName();            if (cName.indexOf('.') != -1) {                cName = cName.substring(cName.lastIndexOf('.')).replaceFirst("[\\.]", "");            }            if (enableMinaMonitor) {                stats = new IoServiceStatistics((AbstractIoService) acceptor);                stats.setThroughputCalculationInterval(minaPollInterval);                MBeanServer mbs = ManagementFactory.getPlatformMBeanServer();                try {                    serviceManagerObjectName = new ObjectName("org.red5.server:type=RTMPMinaTransport");                    mbs.registerMBean(new StandardMBean(this, RTMPMinaTransportMXBean.class, true), serviceManagerObjectName);                } catch (Exception e) {                }            }        } catch (Exception e) {        }    }

start函数主要是mina框架的初始化，开头的initIOHandler函数用于初始化mina框架中的IoHandler，根据red5-core.xml，其已经被初始化为RTMPMinaIoHandler，并设置其成员变量handler为RTMPHandler，和上一章分析类似，RTMPMinaIoHandler属于mina框架调用，RTMPHandler属于RTMP业务调用，两个handler执行的逻辑不同。

    <bean id="rtmpHandler" class="org.red5.server.net.rtmp.RTMPHandler">        <property name="server" ref="red5.server" />        <property name="statusObjectService" ref="statusObjectService" />    </bean>    <bean id="rtmpMinaIoHandler" class="org.red5.server.net.rtmp.RTMPMinaIoHandler">        <property name="handler" ref="rtmpHandler" />    </bean>

回到start函数中，再往下设置了IoBuffer，构造NioSocketAcceptor用于监听客户端连接，并对其进行相应的设置，最后调用NioSocketAcceptor的bind开始监听了。

二. sessionCreated和sessionOpened

当有客户端连接到来时，mina框架会回调RTMPMinaIoHandler的sessionCreated和sessionOpened函数，下面来看，

    public void sessionCreated(IoSession session) throws Exception {        session.getFilterChain().addFirst("rtmpeFilter", new RTMPEIoFilter());        RTMPMinaConnection conn = createRTMPMinaConnection();        conn.setIoSession(session);        conn.setHandler(handler);        session.setAttribute(RTMPConnection.RTMP_SESSION_ID, conn.getSessionId());        session.setAttribute(RTMPConnection.RTMP_HANDSHAKE, new InboundHandshake());    }

sessionCreated函数首先添加了RTMPEIoFilter过滤器用于握手，接着通过createRTMPMinaConnection创建一个RTMPMinaConnection，然后对RTMPMinaConnection进行相应的设置，最后向session中添加InboundHandshake用于服务器端的握手。

下面来看sessionOpened函数，

    public void sessionOpened(IoSession session) throws Exception {        String sessionId = (String) session.getAttribute(RTMPConnection.RTMP_SESSION_ID);        RTMPConnManager connManager = (RTMPConnManager) RTMPConnManager.getInstance();        session.setAttribute(RTMPConnection.RTMP_CONN_MANAGER, new WeakReference<IConnectionManager<RTMPConnection>>(connManager));        RTMPMinaConnection conn = (RTMPMinaConnection) connManager.getConnectionBySessionId(sessionId);        handler.connectionOpened(conn);    }

sessionOpened函数首先获取RTMPConnManager，然后通过RTMPConnManager获得前面在sessionCreated函数中创建的RTMPMinaConnection，并调用RTMPHandler的connectionOpened函数，

    public void connectionOpened(RTMPConnection conn) {        conn.open();        conn.startWaitForHandshake();    }

connectionOpened函数依次调用了RTMPMinaConnection的open和startWaitForHandshake函数，open函数为空，startWaitForHandshake定义在RTMPMinaConnection的父类RTMPConnection中，代码如下，

    public void startWaitForHandshake() {        try {            waitForHandshakeTask = scheduler.schedule(new WaitForHandshakeTask(), new Date(System.currentTimeMillis() + maxHandshakeTimeout));        } catch (TaskRejectedException e) {        }    }

startWaitForHandshake函数启动了一个WaitForHandshakeTask线程，并延迟maxHandshakeTimeout时间启动，maxHandshakeTimeout的默认值为10秒。从客户端与服务器建立TCP连接完成开始，如果10秒内没有完成RTMP的连接，red5服务器端就会关闭该连接。看一下WaitForHandshakeTask执行的任务就一目了然了，

    public void run() {        if (state.getState() != RTMP.STATE_CONNECTED) {            onInactive();        }    }    protected void onInactive() {        close();    }

刚刚在sessionOpened中提到过，该函数会获取RTMPConnManager，RTMPConnManager是一个单例，用来管理服务器端的连接，其中有一段静态代码块值得一看，

    {        executor.scheduleAtFixedRate(new Runnable() {            public void run() {                int closedConnections = 0;                Collection<RTMPConnection> allConns = getAllConnections();                for (RTMPConnection conn : allConns) {                    String sessionId = conn.getSessionId();                    RTMP rtmp = conn.getState();                    switch (rtmp.getState()) {                        case RTMP.STATE_DISCONNECTED:                        case RTMP.STATE_DISCONNECTING:                            removeConnection(sessionId);                            break;                        default:                            long ioTime = 0L;                            IoSession session = conn.getIoSession();                            if (conn instanceof RTMPMinaConnection) {                                ioTime = System.currentTimeMillis() - session.getLastIoTime();                            } else if (conn instanceof RTMPTConnection) {                                ioTime = System.currentTimeMillis() - ((RTMPTConnection) conn).getLastDataReceived();                            }                            if (ioTime >= conn.maxInactivity) {                                if (session != null) {                                    if (session.isConnected()) {                                        session.getWriteRequestQueue().clear(session);                                    }                                }                                conn.onInactive();                                if (!conn.isClosed()) {                                } else {                                    closedConnections++;                                }                            }                    }                }                if (closedConnections > 0) {                    System.gc();                }            }        }, 7000, 30000, TimeUnit.MILLISECONDS);    }

可以看出，RTMPConnManager在初始化时会创建一个线程，该线程每30秒运行一次，用来判断服务器端管理的连接是否过期。判断的方式是检测每个连接有多长时间没有被访问了，如果超过了系统设定的最大时间，就关闭该连接。

下一章开始分析red5客户端的握手过程。