我看过源代码，知道相关原理

/**
 * Given datanode address or host name, returns the DatanodeDescriptor for the
 * same, or if it doesn't find the datanode, it looks for a machine local and
 * then rack local datanode, if a rack local datanode is not possible either,
 * it returns the DatanodeDescriptor of any random node in the cluster.
 *
 * @param address hostaddress:transfer address
 * @return the best match for the given datanode
 */

// If we can't even choose rack local, just choose any node in the
      // cluster.
      if (node == null) {
        node = (DatanodeDescriptor)getNetworkTopology()
                                   .chooseRandom(NodeBase.ROOT);
      }

你不能这样做。这对HDFS客户端是透明的。块的放置正是你所要求的。文件被分成块。只有块被放置在数据节点中，而不是文件。创建“文件”的信息只作为元数据存储在名称节点中（数据节点上块列表的文件路径）
https://hadoop.apache.org/docs/current/hadoop-project-dist/hadoop-hdfs/HdfsDesign.html