Thrift 提供了如图五种模式:TSimpleServer、TNonblockingServer、THsHaServer、TThreadPoolServer、TThreadSelectorServer
TSimpleServer、TThreadPoolServer 属于阻塞模型
TNonblockingServer、THsHaServer、TThreadedSelectorServer 属于非阻塞模型
TServer
TServer 为抽象类
public static class Args extends AbstractServerArgs<Args> {
public Args(TServerTransport transport) {
super(transport);
}
} public static abstract class AbstractServerArgs<T extends AbstractServerArgs<T>> {
final TServerTransport serverTransport;
// 处理层工厂
TProcessorFactory processorFactory;
// 传输层工厂
TTransportFactory inputTransportFactory = new TTransportFactory();
TTransportFactory outputTransportFactory = new TTransportFactory();
// 协议层工厂
TProtocolFactory inputProtocolFactory = new TBinaryProtocol.Factory();
TProtocolFactory outputProtocolFactory = new TBinaryProtocol.Factory();
}
TServer 定义的对外方法
/**
* The run method fires up the server and gets things going.
*/
public abstract void serve();
/**
* Stop the server. This is optional on a per-implementation basis. Not
* all servers are required to be cleanly stoppable.
*/
public void stop() {}
stop 并不是每个服务都需要优雅的退出,所以没有定义为抽象方法
抽象方法 serve() 由具体的 TServer 实例实现
TSimpleServer
TSimpleServer 实现比较简单,是单线程阻塞模型,只适合测试开发使用
serve 方法源码分析
public void serve() {
// 启动监听 socket
serverTransport.listen();
// 设置服务状态
setServing(true);
// 不断的等待与处理 socket 请求
while(!stopped) {
// accept 一个业务 socket 请求
client = serverTransport_.accept();
if (client != null) {
// 通过工厂获取 server 定义的处理层、传输层和协议层
processor = processorFactory_.getProcessor(client);
inputTransport = inputTransportFactory_.getTransport(client);
outputTransport = outputTransportFactory_.getTransport(client);
inputProtocol = inputProtocolFactory_.getProtocol(inputTransport);
outputProtocol = outputProtocolFactory_.getProtocol(outputTransport);
if (eventHandler_ != null) {
connectionContext = eventHandler_.createContext(inputProtocol, outputProtocol);
}
// 阻塞式处理
while (true) {
// 处理请求事件
if (eventHandler_ != null) {
eventHandler_.processContext(connectionContext, inputTransport, outputTransport);
}
// 如果处理层为异步,则退出
if(!processor.process(inputProtocol, outputProtocol)) {
break;
}
}
}
// 关闭
eventHandler_.deleteContext(connectionContext, inputProtocol, outputProtocol);
inputTransport.close();
outputTransport.close();
setServing(false);
}
}
TSimpleServer 工作图
TThreadPoolServer
ThreadPoolServer 解决了 TSimple 不支持并发和多连接的问题,引入了线程池
与 TSimple 相同,主线程负责阻塞式监听 socket,而剩下的业务处理则全部交由线程池去处理
public void serve() {
// 主线程启动监听 socket
serverTransport_.listen();
// 设置服务状态
stopped_ = false;
setServing(true);
// 等待并处理 socket 请求
while (!stopped_) {
TTransport client = serverTransport_.accept();
// Runnable run 逻辑与 TSimpleServer 类似
WorkerProcess wp = new WorkerProcess(client);
int retryCount = 0;
long remainTimeInMillis = requestTimeoutUnit.toMillis(requestTimeout);
while(true) {
// 交由线程池来处理
executorService_.execute(wp);
break;
}
}
executorService_.shutdown();
setServing(false);
}
TThreadPoolServer 的缺点:
处理能力的好坏受限于线程池的设置
TNoblockingServer
TNoblockingServer 是单线程工作,但该模式采用了 NIO,所有的 socket 被注册到 selector 中,通过一个线程循环 selector 来监控所有 socket,当有就绪的 socket 时,根据不同的请求做不同的动作(读取、写入数据或 accept 新连接)
TNoblockingServer 的 serve 方法在其父类 AbstractNonblockingServer 中定义
/**
* Begin accepting connections and processing invocations.
* 开始接受并处理调用
*/
public void serve() {
// start any IO threads
// 注册一些监听 socket 的线程到 selector 中
if (!startThreads()) {
return;
}
// start listening, or exit
// 开始监听
if (!startListening()) {
return;
}
// 设置服务状态
setServing(true);
// this will block while we serve
// TNonblocking 中实现为 selectAcceptThread_.join();
// 主线程等待 selectAcceptThread 执行完毕
// SelectAcceptThread 的 run 方法为 select();
// 取出一个就绪的 socket
waitForShutdown(); setServing(false); // do a little cleanup
stopListening();
} // SelectAcceptThread run 方法
public void run() {
while (!stopped_) {
select();
processInterestChanges();
}
for (SelectionKey selectionKey : selector.keys()) {
cleanupSelectionKey(selectionKey);
}
} // SelectAcceptThread Select 过程
private void select() {
try {
// wait for io events.
// NIO 取出一个
selector.select();
// process the io events we received
Iterator<SelectionKey> selectedKeys = selector.selectedKeys().iterator();
// 遍历就绪的 socket
while (!stopped_ && selectedKeys.hasNext()) {
SelectionKey key = selectedKeys.next();
selectedKeys.remove();
// if the key is marked Accept, then it has to be the server
// transport.
// accept 新 socket 并将其注册到 selector 中
if (key.isAcceptable()) {
handleAccept();
} else if (key.isReadable()) {
// deal with reads
// 处理读数据的 socket 请求
handleRead(key);
} else if (key.isWritable()) {
// deal with writes
// 处理写数据的 socket 请求
handleWrite(key);
} else {
LOGGER.warn("Unexpected state in select! " + key.interestOps());
}
}
} catch (IOException e) {
LOGGER.warn("Got an IOException while selecting!", e);
}
} // 接收新的连接
private void handleAccept() throws IOException {
SelectionKey clientKey = null;
TNonblockingTransport client = null;
// accept the connection
client = (TNonblockingTransport)serverTransport.accept();
// 注册到 selector 中
clientKey = client.registerSelector(selector, SelectionKey.OP_READ);
// add this key to the map
FrameBuffer frameBuffer = createFrameBuffer(client, clientKey, SelectAcceptThread.this);
clientKey.attach(frameBuffer);
}
TNonblockingServer 模式的缺点:
其还是采用单线程顺序来完成,当业务处理比较复杂耗时,该模式的效率将会下降
TNonblockingServer 工作图:
THsHaServer
THsHaServer 是 TNoblockingServer 的子类,处理逻辑基本相同,不同的是,在处理读取请求时,THsHaServer 将处理过程交由线程池来完成,主线程直接返回进行下一次循环,提高了效率
THsHaServer 模式的缺点:
其主线程需要完成对所有 socket 的监听一级数据的写操作,当大请求量时,效率较低
TThreadedSelectorServer
TThreadedSelectorServer 是 Thrift 目前提供的*模式,生产环境的首选,其对 TNonblockingServer 进行了扩展
TThreadedSelectorServer 源码中一些关键的属性
public static class Args extends AbstractNonblockingServerArgs<Args> {
// 在已接收的连接中选择线程的个数
public int selectorThreads = 2;
// 执行线程池 ExecutorService 的线程个数
private int workerThreads = 5;
// 执行请求具体任务的线程池
private ExecutorService executorService = null;
}
// The thread handling all accepts
private AcceptThread acceptThread;
// Threads handling events on client transports
private final Set<SelectorThread> selectorThreads = new HashSet<SelectorThread>();
// This wraps all the functionality of queueing and thread pool management
// for the passing of Invocations from the selector thread(s) to the workers
// (if any).
private final ExecutorService invoker;
/**
* 循环模式的负载均衡器,用于为新的连接选择 SelectorThread
*/
protected static class SelectorThreadLoadBalancer {}
AcceptThread 线程对象,用于监听 socket 的新连接
多个 SelectorThread 线程对象,用于处理 socket 的读写操作
一个负载均衡对象 SelectorThreadLoadBalancer,用于决定将 AcceptThread 接收到的 socket 请求分配给哪个 SelectorThread 线程
SelectorThread 线程执行过读写操作后,通过 ExecutorService 线程池来完成此次调用的具体执行
SelectorThread 对象源码解析
/**
* 多个 SelectorThread 负责处理 socket 的 I/O 操作
*/
protected class SelectorThread extends AbstractSelectThread {
/**
* The work loop. Handles selecting (read/write IO), dispatching, and
* managing the selection preferences of all existing connections.
* 选择(处理 socket 的网络读写 IO),分配和管理现有连接
*/
public void run() {
while (!stopped_) {
select();
}
}
private void select() {
// process the io events we received
Iterator<SelectionKey> selectedKeys = selector.selectedKeys().iterator();
while (!stopped_ && selectedKeys.hasNext()) {
SelectionKey key = selectedKeys.next();
selectedKeys.remove();
// skip if not valid
if (!key.isValid()) {
cleanupSelectionKey(key);
continue;
}
if (key.isReadable()) {
// deal with reads
handleRead(key);
} else if (key.isWritable()) {
// deal with writes
handleWrite(key);
} else {
LOGGER.warn("Unexpected state in select! " + key.interestOps());
}
}
}
}
AcceptThread 对象源码解析
/**
* 在服务器传输中选择线程(监听 socket 请求)并向 IO 选择器(SelectorThread)提供新连接
*/
protected class AcceptThread extends Thread {
// The listen socket to accept on
private final TNonblockingServerTransport serverTransport;
private final Selector acceptSelector;
// 负载均衡器,决定将连接分配给哪个 SelectorThread
private final SelectorThreadLoadBalancer threadChooser;
public void run() {
while (!stopped_) {
select();
}
}
private void select() {
// process the io events we received
Iterator<SelectionKey> selectedKeys = acceptSelector.selectedKeys().iterator();
while (!stopped_ && selectedKeys.hasNext()) {
SelectionKey key = selectedKeys.next();
selectedKeys.remove();
// 处理接收的新情求
if (key.isAcceptable()) {
handleAccept();
} else {
LOGGER.warn("Unexpected state in select! " + key.interestOps());
}
}
}
/**
* Accept a new connection.
*/
private void handleAccept() {
final TNonblockingTransport client = doAccept();
if (client != null) {
// 从负载均衡器中,获取 SelectorThread 线程
final SelectorThread targetThread = threadChooser.nextThread();
if (args.acceptPolicy == Args.AcceptPolicy.FAST_ACCEPT || invoker == null) {
doAddAccept(targetThread, client);
} else {
// FAIR_ACCEPT
invoker.submit(new Runnable() {
public void run() {
// 将选择到的线程和连接放入 线程池 处理
// 用 targetThread 线程取处理一个给接受的链接 client,如果新连接的队列处于满的状态,则将处于阻塞状态
doAddAccept(targetThread, client);
}
});
}
}
}
private TNonblockingTransport doAccept() {
return (TNonblockingTransport) serverTransport.accept();
}
// 用 targetThread 线程取处理一个给接受的链接 client,如果新连接的队列处于满的状态,则将处于阻塞状态
private void doAddAccept(SelectorThread thread, TNonblockingTransport client) {
if (!thread.addAcceptedConnection(client)) {
client.close();
}
}
}
TThreadedSelectorServer 工作图
参考资料
- Thrift server端的几种工作模式分析:http://blog.csdn.net/houjixin/article/details/42779915
- Thrift 网络服务模型:http://www.cnblogs.com/mumuxinfei/p/3875165.html