Spark Streaming xml文件

我需要处理流到S3文件夹中的xml文件.目前，我已将其实现如下.

I have a requirement to process xml files streamed into a S3 folder. Currently, I have implemented it as follows.

首先，使用Spark的fileStream读取文件

First, Read files using Spark's fileStream

val data = ssc.fileStream[LongWritable, Text, TextInputFormat]("s3://myfolder/",(t: org.apache.hadoop.fs.Path) => true, newFilesOnly = true, hadoopConf).map(_._2.toString())

对于每个RDD，请检查是否已读取任何文件

For each RDD, check if any file has been read

if (data.count() !=0)

将字符串写入新的HDFS目录

Write the string to a new HDFS directory

data.coalesce(1).saveAsTextFile(sdir);

从上面的HDFS目录创建一个读取的数据框

Create a Dataframe reading from the above HDFS directory

val loaddata = sqlContext.read.format("com.databricks.spark.xml").option("rowTag", "Trans").load(sdir)

对Dataframe进行一些处理并另存为JSON

Do some processing on Dataframe and save as JSON

loaddata.write.mode("append").json("s3://mybucket/somefolder")

不知何故，我认为上述方法效率低下，坦率地说很像个男孩子. 有更好的解决方案吗?任何帮助将不胜感激.

Somehow, I feel that the above approach is very inefficient and frankly quite school boyish. Is there a better solution? Any help would be greatly appreciated.

一个后续问题: 如何操作数据框中的字段(不是列)? 我有一个vey复杂的嵌套xml，当我使用上述方法时，我将获得一个包含9列和50个奇数内部Struct数组的Dataframe.很好，除了需要修剪某些字段名称.有没有办法在不分解数据帧的情况下实现这一目标，因为我需要再次构造相同的结构?

A follow up question: How to manipulate fields (not Columns) in a dataframe? I have a vey complex nested xml and when I use the above described method, I am getting a Dataframe with 9 columns and 50 odd inner Struct arrays. That is fine except for the need to trim certain field names. Is there a way to achieve that without exploding the dataframe, as I need to construct the same structure again?