从打印精美的文本文件加载 spark 数据框

Question

我有几千个json个文件，每个文件里面的内容都类似于下面这样：

{
"1" : { "key":"key1", "val":"val1" },
"2" : { "key":"key2", "val":"val2" },
"3" : { "key":"key3", "val":"val3" }
.
.
.
}

将这些文件加载到 spark 数据帧中的正确方法是什么，因此我将拥有

+------+----------------------------------+
|id    | val.                             |
+------+----------------------------------+
|1     | { "key":"key1", "val":"val1" }   |
|2     | { "key":"key2", "val":"val2" }   |
|3     | { "key":"key2", "val":"val2" }   |
+------+----------------------------------+

我尝试将 json 加载为多行

val df= spark.read.option("multiline","true").json(small_file)

但结果是一行三列

+------------------------+------------------------+----------------+
|1                       |2                       |3               |
+------------------------+------------------------+----------------+
|{ "key":"key1", "val..} ||{"key":"key2", "val..} |{"key":"key3"...|
+------------------------+------------------------+----------------+

我所做的也是将文件加载到地图中

 val keys = df.columns
 val values = df.collect().last.toSeq
 val myMap = keys.zip(values).toMap
 
 println(myMap)
 // output
 // Map(1-> [key1, val1], 2-> [key2, val2], 3-> [key3, val3])

但我不知道如何从这个地图创建数据框

Answer 1

这是一个多行 JSON 文件，您可以像这样指定 multiline 选项来读取此类文件：

val spark = SparkSession
  .builder()
  .appName("JSONReader")
  .master("local")
  .getOrCreate()

val multiline_df = spark.read.option("multiline","true")
  .json("multiline-file.json")

multiline_df.show(false)

结果将是这样的：

[info] +------------+------------+------------+
[info] |1           |2           |3           |
[info] +------------+------------+------------+
[info] |[key1, val1]|[key2, val2]|[key3, val3]|
[info] +------------+------------+------------+
[info]

Answer 2

我能够通过以下步骤获得结果：

如问题中所述，加载后的结果 df 将如下所示

+------------------------+------------------------+----------------+
|1                       |2                       |3               |
+------------------------+------------------------+----------------+
|{ "key":"key1", "val..} ||{"key":"key2", "val..} |{"key":"key3"...|
+------------------------+------------------------+----------------+

1- 将列转换为字符串

val cols=df.columns.map(x => col(s"${x}").cast("string").alias(s"${x}"))

2- 创建列字符串

val str_cols=df.columns.mkString(",")

3- 使用步骤 1 中的转换值创建一个新的 df

val df1 = df.withColumn("temp",
                        explode(arrays_zip(array(cols:_*),
                        split(lit(str_cols),","))))
                        .select("temp.*")
                        .toDF("vals","index")

4- 生成的数据框将与所需的一样

df1.select($"index",$"vals").show()
+------+----------------------------------+
|index | vals                             |
+------+----------------------------------+
|1     | { "key":"key1", "val":"val1" }   |
|2     | { "key":"key2", "val":"val2" }   |
|3     | { "key":"key2", "val":"val2" }   |
+------+----------------------------------+

从打印精美的文本文件加载 spark 数据框

Load spark data frame from a pretty-printed text file

json

scala

pretty-print

dataframe

apache-spark