背景
在io密集型的web 应用,如何能更好地提升后台性能,jetty continuation是一个选择
现在特别流行的说法就是事件驱动,看看node.js以及redis,
jetty continuation也不例外
一个例子
package org.kaka.web;
import java.io.IOException;
import java.io.PrintWriter;
import javax.servlet.ServletException;
import javax.servlet.http.HttpServlet;
import javax.servlet.http.HttpServletRequest;
import javax.servlet.http.HttpServletResponse;
import org.eclipse.jetty.continuation.Continuation;
import org.eclipse.jetty.continuation.ContinuationSupport;
public class SimpleSuspendResumeServlet extends HttpServlet {
/**
*
*/
private static final long serialVersionUID = 6112996063962978130L;
private MyAsyncHandler myAsyncHandler;
public void init() throws ServletException {
myAsyncHandler = new MyAsyncHandler() {
public void register(final MyHandler myHandler) {
new Thread(new Runnable() {
public void run() {
try {
Thread.sleep(10000);
myHandler.onMyEvent("complete!");
} catch (InterruptedException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
}).start();
}
};
}
public void doGet(HttpServletRequest request, HttpServletResponse response)
throws ServletException, IOException {
// if we need to get asynchronous results
//Object results = request.getAttribute("results");
final PrintWriter writer = response.getWriter();
final Continuation continuation = ContinuationSupport
.getContinuation(request);
//if (results == null) {
if (continuation.isInitial()) {
//request.setAttribute("results","null");
sendMyFirstResponse(response);
// suspend the request
continuation.suspend(); // always suspend before registration
// register with async service. The code here will depend on the
// the service used (see Jetty HttpClient for example)
myAsyncHandler.register(new MyHandler() {
public void onMyEvent(Object result) {
continuation.setAttribute("results", result);
continuation.resume();
}
});
return; // or continuation.undispatch();
}
if (continuation.isExpired()) {
sendMyTimeoutResponse(response);
return;
}
//Send the results
Object results = request.getAttribute("results");
if(results==null){
response.getWriter().write("why reach here??");
continuation.resume();
return;
}
sendMyResultResponse(response, results);
}
private interface MyAsyncHandler {
public void register(MyHandler myHandler);
}
private interface MyHandler {
public void onMyEvent(Object result);
}
private void sendMyFirstResponse(HttpServletResponse response) throws IOException {
//必须加上这一行,否者flush也没用,为什么?
response.setContentType("text/html");
response.getWriter().write("start");
response.getWriter().flush();
}
private void sendMyResultResponse(HttpServletResponse response,
Object results) throws IOException {
//response.setContentType("text/html");
response.getWriter().write("results:" + results);
response.getWriter().flush();
}
private void sendMyTimeoutResponse(HttpServletResponse response)
throws IOException {
response.getWriter().write("timeout");
}
}
这个例子到底干了什么?
出于演示的目的,这个例子格外简单。模拟了在web请求中如何处理一个耗时阻塞10s的io操作
- 常规的程序遇到这种情况(假设没有设置超时),通常会等待10s,线程会白白浪费在哪,一旦请求一多,线程池及等待队列会满掉,从而导致网站无法服务
- 稍微好点的程序恐怕会使用futuretask来解决问题,但无论怎样,futuretask带有轮询的性质,或多或少会带有阻塞
- jetty continuation更进了一步, 采用事件驱动的方式来通知请求完成,丝毫不浪费一点io时间,一旦遇到阻塞,当前worker线程会结束,这样就可以服务其他请求,等耗时操作处理完毕通知jetty,此时jetty会再动用一个新的worker线程再次处理请求
常见流程
- 具体流程1,耗时操作在AsyncContinuation默认超时(30s,可以设置)之内完成
- 当请求到来,被selector线程感知
- selector线程,会实例化org.eclipse.jetty.server.Request以及org.eclipse.jetty.server.AsyncContinuation
- AsyncContinuation实例此时状态为__IDLE
- selector线程,会实例化org.eclipse.jetty.server.Request以及org.eclipse.jetty.server.AsyncContinuation
- worker线程A会获取到selector线程派发的这个请求
- 流程进入org.eclipse.jetty.server.AsyncHttpConnection(AbstractHttpConnection).handleRequest()
- 调用_request._async.handling(),AsyncContinuation实例的状态__IDLE-->__DISPATCHED
- 调用真正的业务逻辑server.handle(this);业务逻辑中调用AsyncContinuation.suspend(ServletContext, ServletRequest, ServletResponse) ,会将AsyncContinuation实例的状态__DISPATCHED-->__ASYNCSTARTED
- 最后调用_request._async.unhandle(),AsyncContinuation实例的状态__ASYNCSTARTED-->__ASYNCWAIT
- 会将suspend的那个请求任务放入selector线程的超时队列(和redis 事件框架的做法非常类似)
- 为了简单期间,假设在上诉过程中,耗时操作还没完成,此时worker线程A被回收将会服务其他请求,虽然当前请求并未完成
- 流程进入org.eclipse.jetty.server.AsyncHttpConnection(AbstractHttpConnection).handleRequest()
- 匿名应用线程(在jetty woker线程另外启动的线程)完成耗时操作
- 调用AsyncContinuation.resume() ,此时AsyncContinuation实例的状态__ASYNCWAIT-->__REDISPATCH
- 然后会往worker线程队列或者selector线程队列中添加一个请求task,以此来触发jetty再次完成未完成的请求
- 调用AsyncContinuation.resume() ,此时AsyncContinuation实例的状态__ASYNCWAIT-->__REDISPATCH
- worker线程B再次处理那个请求
- 流程进入org.eclipse.jetty.server.AsyncHttpConnection(AbstractHttpConnection).handleRequest()
- 调用_request._async.handling(),AsyncContinuation实例的状态__REDISPATCH-->__REDISPATCHED
- 调用server.handleAsync(this);再次进入业务逻辑,此时耗时操作已完成,可以输出最后的结果
- 调用_request._async.unhandle() AsyncContinuation实例的状态__REDISPATCHED-->__UNCOMPLETED
- 调用_request._async.doComplete(),AsyncContinuation实例的状态__UNCOMPLETED-->__COMPLETED
- 请求结束
- 具体流程2,耗时操作超过了AsyncContinuation默认超时时间
- 同上
- 同上
- selector线程轮询中感知到耗时任务超时
- 此时在轮询中能感知并获取这个task,见org.eclipse.jetty.io.nio.SelectorManager$SelectSet.doSelect()
- 将此task丢入worker线程队列
- 此时在轮询中能感知并获取这个task,见org.eclipse.jetty.io.nio.SelectorManager$SelectSet.doSelect()
- worker线程C处理超时任务
- 调用AsyncContinuation.expired() ,此时AsyncContinuation实例的状态__ASYNCWAIT-->__REDISPATCH
- 然后会往worker线程队列或者selector线程队列中添加一个请求task,以此来触发jetty再次完成未完成的请求
- 调用AsyncContinuation.expired() ,此时AsyncContinuation实例的状态__ASYNCWAIT-->__REDISPATCH
- 同上面的4
- 具体流程3,耗时操作非常快
- 同上
- worker线程A会获取到selector线程派发的这个请求
- 流程进入org.eclipse.jetty.server.AsyncHttpConnection(AbstractHttpConnection).handleRequest()
- 调用_request._async.handling(),AsyncContinuation实例的状态__IDLE-->__DISPATCHED
- 调用真正的业务逻辑server.handle(this);业务逻辑中调用AsyncContinuation.suspend(ServletContext, ServletRequest, ServletResponse) ,会将AsyncContinuation实例的状态__DISPATCHED-->__ASYNCSTARTED
- 流程进入org.eclipse.jetty.server.AsyncHttpConnection(AbstractHttpConnection).handleRequest()
- 匿名应用线程(在jetty woker线程另外启动的线程)完成耗时操作,在2)中调用_request._async.unhandle()之前完成
- 调用AsyncContinuation.resume() ,此时AsyncContinuation实例的状态__ASYNCSTARTED-->__REDISPATCHING
- 还是那个work线程A,注意此时3已经完成
- 调用_request._async.unhandle(),,AsyncContinuation实例的状态__REDISPATCHING-->__REDISPATCHED
- 由于此时_request._async.unhandle()返回false,再次进入循环调用server.handleAsync(this);再次进入业务逻辑,此时耗时操作已完成,可以输出最后的结果
- 调用_request._async.unhandle() AsyncContinuation实例的状态__REDISPATCHED-->__UNCOMPLETED
- 调用_request._async.doComplete(),AsyncContinuation实例的状态__UNCOMPLETED-->__COMPLETED
- 请求结束
- 调用_request._async.unhandle(),,AsyncContinuation实例的状态__REDISPATCHING-->__REDISPATCHED
小结
总体来讲
- jetty continuation需要在应用中使用应用级的线程池来完成一些io任务,这个在普通的web编程并不常见
- 在应用的第一次请求中需要
- 调用AsyncContinuation.suspend(完成一个状态的转换,以及产生一个超时任务)
- 将耗时任务派发给应用线程池
- 完毕后,jetty回收该请求线程
- io任务在应用线程中完成后,然后通过AsyncContinuation.resume或者complete等方法通知jetty任务完成
- jetty然后会再次分配一个worker线程处理该请求(如果逻辑复杂,如并行的多次Io会分配多个worker线程)
- 引入jetty continuation带来的负面作用是编程模型会变得复杂,需要仔细的切割各类io任务
此例子在jetty7.2和jetty7.6中测试通过,但在jetty8中运行失败(因为调用flush导致request中_header为空从而引发NPE[jetty8的bug?],但如果不掉用有达不大展示效果)