通用ipc超时控制

除了DataNode外的超时，比如Yarn自己通信超时那么也会报如下错误。


failed on socket timeout exception: java.net.SocketTimeoutException: 60000 millis timeout while waiting for channel to be ready for read. ch : java.nio.channels.SocketChannel

增大core-site.xml中的ipc.ping.interval。
如果关闭ipc.client.ping。则要增大ipc.client.rpc-timeout.ms。
如下是Hadoop3.1.1的源码。由源码可知。如果开启ipc.client.ping心跳，则没有RPC超时。以心跳超时为准。

  1.   /**
  2.    * The time after which a RPC will timeout.
  3.    * If ping is not enabled (via ipc.client.ping), then the timeout value is the 
  4.    * same as the pingInterval.
  5.    * If ping is enabled, then there is no timeout value.
  6.    * 
  7.    * @param conf Configuration
  8.    * @return the timeout period in milliseconds. -1 if no timeout value is set
  9.    * @deprecated use {@link #getRpcTimeout(Configuration)} instead
  10.    */
  11.   @Deprecated
  12.   final public static int getTimeout(Configuration conf) {
  13.     int timeout = getRpcTimeout(conf);
  14.     if (timeout > 0)  {
  15.       return timeout;
  16.     }
  17.     if (!conf.getBoolean(CommonConfigurationKeys.IPC_CLIENT_PING_KEY,
  18.         CommonConfigurationKeys.IPC_CLIENT_PING_DEFAULT)) {
  19.       return getPingInterval(conf);
  20.     }
  21.     return -1;
  22.   }
  24.   /**
  25.    * The time after which a RPC will timeout.
  26.    *
  27.    * @param conf Configuration
  28.    * @return the timeout period in milliseconds.
  29.    */
  30.   public static final int getRpcTimeout(Configuration conf) {
  31.     int timeout =
  32.         conf.getInt(CommonConfigurationKeys.IPC_CLIENT_RPC_TIMEOUT_KEY,
  33.             CommonConfigurationKeys.IPC_CLIENT_RPC_TIMEOUT_DEFAULT);
  34.     return (timeout < 0) ? 0 : timeout;
  35.   }

参考资料

https://www.cnblogs.com/yjt1993/p/11164492.html
https://yq.aliyun.com/articles/476766