这几天在解决一个用mina开发的高并发通信过程中产生的一个bug。
模拟场景为:
通过定时触发启动线程模拟高并发短连接测试,测试的服务端有2个,一个是服务有起,一个没起,客户端和服务端均在同一服务器上。执行一段时间后linux主机上通过lsof命令查看,发现有递增的文件句柄,pipe和eventpoll。
抛出的异常如下:
2012 - 10 - 13 10 : 09 : 48 -org.apache.mina.core.service.SimpleIoProcessorPool.<init>(SimpleIoProcessorPool.java: 197 )
2. Failed to create a new instance of org.apache.mina.transport.socket.nio.NioProcessor: null
3.java.lang.reflect.InvocationTargetException
4. at sun.reflect.GeneratedConstructorAccessor110.newInstance(Unknown Source)
5. at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:27 )
6. at java.lang.reflect.Constructor.newInstance(Constructor.java:513 )
7. at org.apache.mina.core.service.SimpleIoProcessorPool.<init>(SimpleIoProcessorPool.java:180 )
8. at org.apache.mina.core.service.SimpleIoProcessorPool.<init>(SimpleIoProcessorPool.java:112 )
9. at org.apache.mina.core.polling.AbstractPollingIoConnector.<init>(AbstractPollingIoConnector.java:93 )
10. at org.apache.mina.transport.socket.nio.NioSocketConnector.<init>(NioSocketConnector.java:56 )
11. at com.develop.webplatform.funnel.client.JobClient.sendMessage(JobClient.java:39 )
12. at com.develop.webplatform.funnel.client.JobClient.sendJob(JobClient.java:126 )
13. at com.develop.webplatform.funnel.extend.JobExecRemotelyBySocket.execJobByTask(JobExecRemotelyBySocket.java:66 )
14. at com.develop.webplatform.funnel.JobManager.execJobByTask(JobManager.java:27 )
15. at com.develop.webplatform.quartz.job.TaskJob.executeInternal(TaskJob.java:38 )
16. at org.springframework.scheduling.quartz.QuartzJobBean.execute(QuartzJobBean.java:86 )
17. at org.quartz.core.JobRunShell.run(JobRunShell.java:223 )
18. at org.quartz.simpl.SimpleThreadPool$WorkerThread.run(SimpleThreadPool.java:549 )
19.Caused by: org.apache.mina.core.RuntimeIoException: Failed to open a selector.
20. at org.apache.mina.transport.socket.nio.NioProcessor.<init>(NioProcessor.java:61 )
21. ... 15 more
22.Caused by: java.io.IOException: Too many open files
23. at sun.nio.ch.IOUtil.initPipe(Native Method)
24. at sun.nio.ch.EPollSelectorImpl.<init>(EPollSelectorImpl.java:49 )
25. at sun.nio.ch.EPollSelectorProvider.openSelector(EPollSelectorProvider.java:18 )
26. at java.nio.channels.Selector.open(Selector.java:209 )
27. at org.apache.mina.transport.socket.nio.NioProcessor.<init>(NioProcessor.java:59 )
28. ... 15 more
源代码中,关于客户端连接的代码如下:
final NioSocketConnector connector = new NioSocketConnector();
2. final String[] result = new String[ 1 ];
3. connector.getFilterChain().addLast("codec" ,
4. new ProtocolCodecFilter( new ObjectSerializationCodecFactory()));
5. connector.setHandler(handler);
6.
7. //设置超时
8. connector.setConnectTimeoutMillis(defaultConnectTimeOut);
9. ConnectFuture connectFuture = connector.connect(address);
10. connectFuture.awaitUninterruptibly(); //同步,等待,直到连接完成
11. if (connectFuture.isDone()) {
12. if (!connectFuture.isConnected()) { //若在指定时间内没连接成功,则抛出异常
13. logger.info("fail to connect " + logInfo);
14.
15. throw new Exception();
16. }
17. }
经过分析,导致主机文件句柄泄露的原因为,客户端发起服务端连接时,会请求系统分配相关的文件句柄,在原代码中,仅仅判断是否连接成功,而未对连接失败进 行资源释放,从而造成文件句柄泄露。当总的文件句柄数超过系统设置值(ulimit -n 查看同一个进程允许的最大文件句柄数),则抛出异常“java.io.IOException: Too many open files",导致无法创建新的连接,服务器挂掉。
更改后的代码如下:
final NioSocketConnector connector = new NioSocketConnector();
2. final String[] result = new String[ 1 ];
3. connector.getFilterChain().addLast("codec" ,
4. new ProtocolCodecFilter( new ObjectSerializationCodecFactory()));
5. connector.setHandler(handler);
6.
7. //设置超时
8. connector.setConnectTimeoutMillis(defaultConnectTimeOut);
9. ConnectFuture connectFuture = connector.connect(address);
10. connectFuture.awaitUninterruptibly(); //同步,等待,直到连接完成
11. if (connectFuture.isDone()) {
12. if (!connectFuture.isConnected()) { //若在指定时间内没连接成功,则抛出异常
13. logger.info("fail to connect " + logInfo);
14. connector.dispose(); //不关闭的话会运行一段时间后抛出,too many open files异常,导致无法连接
15.
16. throw new Exception();
17. }
18. }