
# Spark执行应用时上报"Connection to ip:port has been quiet for xxx ms while there are outstanding requests"并导致应用结束
#### 问题
Spark执行应用时上报如下类似错误并导致应用结束。
```
2016-04-20 10:42:00,557 | ERROR | [shuffle-server-2] | Connection to 10-91-8-208/10.18.0.115:57959 has been quiet for 180000 ms while there are outstanding requests. Assuming connection is dead; please adju
st spark.network.timeout if this is wrong. | org.apache.spark.network.server.TransportChannelHandler.userEventTriggered(TransportChannelHandler.java:128)
2016-04-20 10:42:00,558 | ERROR | [shuffle-server-2] | Still have 1 requests outstanding when connection from 10-91-8-208/10.18.0.115:57959 is closed | org.apache.spark.network.client.TransportResponseHandl
er.channelUnregistered(TransportResponseHandler.java:102)
2016-04-20 10:42:00,562 | WARN  | [yarn-scheduler-ask-am-thread-pool-160] | Error sending message [message = DoShuffleClean(application_1459995017785_0108,319)] in 1 attempts | org.apache.spark.Logging$clas
s.logWarning(Logging.scala:92)
java.io.IOException: Connection from 10-91-8-208/10.18.0.115:57959 closed
        at org.apache.spark.network.client.TransportResponseHandler.channelUnregistered(TransportResponseHandler.java:104)
        at org.apache.spark.network.server.TransportChannelHandler.channelUnregistered(TransportChannelHandler.java:94)
        at io.netty.channel.AbstractChannelHandlerContext.invokeChannelUnregistered(AbstractChannelHandlerContext.java:158)
        at io.netty.channel.AbstractChannelHandlerContext.fireChannelUnregistered(AbstractChannelHandlerContext.java:144)
        at io.netty.channel.ChannelInboundHandlerAdapter.channelUnregistered(ChannelInboundHandlerAdapter.java:53)
        at io.netty.channel.AbstractChannelHandlerContext.invokeChannelUnregistered(AbstractChannelHandlerContext.java:158)
        at io.netty.channel.AbstractChannelHandlerContext.fireChannelUnregistered(AbstractChannelHandlerContext.java:144)
        at io.netty.channel.ChannelInboundHandlerAdapter.channelUnregistered(ChannelInboundHandlerAdapter.java:53)
        at io.netty.channel.AbstractChannelHandlerContext.invokeChannelUnregistered(AbstractChannelHandlerContext.java:158)
        at io.netty.channel.AbstractChannelHandlerContext.fireChannelUnregistered(AbstractChannelHandlerContext.java:144)
        at io.netty.channel.ChannelInboundHandlerAdapter.channelUnregistered(ChannelInboundHandlerAdapter.java:53)
        at io.netty.channel.AbstractChannelHandlerContext.invokeChannelUnregistered(AbstractChannelHandlerContext.java:158)
        at io.netty.channel.AbstractChannelHandlerContext.fireChannelUnregistered(AbstractChannelHandlerContext.java:144)
        at io.netty.channel.DefaultChannelPipeline.fireChannelUnregistered(DefaultChannelPipeline.java:739)
        at io.netty.channel.AbstractChannel$AbstractUnsafe$8.run(AbstractChannel.java:659)
        at io.netty.util.concurrent.SingleThreadEventExecutor.runAllTasks(SingleThreadEventExecutor.java:357)
        at io.netty.channel.nio.NioEventLoop.run(NioEventLoop.java:357)
        at io.netty.util.concurrent.SingleThreadEventExecutor$2.run(SingleThreadEventExecutor.java:111)
        at java.lang.Thread.run(Thread.java:745)
2016-04-20 10:42:00,573 | INFO  | [dispatcher-event-loop-14] | Starting task 177.0 in stage 1492.0 (TID 1996351, linux-254, PROCESS_LOCAL, 2106 bytes) | org.apache.spark.Logging$class.logInfo(Logging.scala:
59)
2016-04-20 10:42:00,574 | INFO  | [task-result-getter-0] | Finished task 85.0 in stage 1492.0 (TID 1996259) in 191336 ms on linux-254 (106/3000) | org.apache.spark.Logging$class.logInfo(Logging.scala:59)
2016-04-20 10:42:00,811 | ERROR | [Yarn application state monitor] | Yarn application has already exited with state FINISHED! | org.apache.spark.Logging$class.logError(Logging.scala:75)
```
#### 回答
当配置channel过期时间（spark.rpc.io.connectionTimeout） \< RPC响应超时时间（spark.rpc.askTimeout），在特殊条件下（Full GC，网络延时等）消息响应时间较长，消息还没有反馈，channel又达到了过期时间，该channel就被终止了，AM端感知到channel被终止后认为driver失联，然后整个应用停止。
解决办法：在Spark客户端的"spark-defaults.conf"文件中或通过set命令行进行设置。参数配置时要保证channel过期时间（spark.rpc.io.connectionTimeout）大于或等于RPC响应超时时间（spark.rpc.askTimeout）。
表1参数说明 
| 参数                   | 描述                                          | 默认值  |
|:---|:---|:---|
| spark.rpc.askTimeout | RPC响应超时时间，不配置的话默认使用spark.network.timeout的值。 | 120s |
   
