在Spark中创建分层JSON

Question

我有一个火花数据框，我需要写入MongoDB。我想知道如何在mongoDB中将数据框的一些列写成嵌套/分层JSON。假设数据框有6列，col1，col2，…… col5，col6我想要col1，col2，col3作为第一层次结构，其余列col4到col6作为第二层次结构。像这样的东西，

{
    "col1": 123,
    "col2": "abc",
    "col3": 45,
    "fields": {
        "col4": "ert",
        "col5": 45,
        "col6": 56
    }
}

我如何在pyspark中实现这一点？

Answer 1

对于这种情况，请在内置函数中使用< code>to_json struct。

示例：

df.show()                                                                                                         
#+----+----+----+----+----+----+
#|col1|col2|col3|col4|col5|col6|
#+----+----+----+----+----+----+
#| 123| abc|  45| ert|  45|  56|
#+----+----+----+----+----+----+

from pyspark.sql.functions import *
df.withColumn("jsn",to_json(struct("col1","col2","col3",struct("col4","col5","col6").alias("fields")))).show(10,False)
#+----+----+----+----+----+----+---------------------------------------------------------------------------------------+
#|col1|col2|col3|col4|col5|col6|jsn                                                                                    |
#+----+----+----+----+----+----+---------------------------------------------------------------------------------------+
#|123 |abc |45  |ert |45  |56  |{"col1":"123","col2":"abc","col3":"45","fields":{"col4":"ert","col5":"45","col6":"56"}}|
#+----+----+----+----+----+----+---------------------------------------------------------------------------------------+

cols=df.columns

df.withColumn("jsn",to_json(struct("col1","col2","col3",struct("col4","col5","col6").alias("fields")))).drop(*cols).show(10,False)
#+---------------------------------------------------------------------------------------+
#|jsn                                                                                    |
#+---------------------------------------------------------------------------------------+
#|{"col1":"123","col2":"abc","col3":"45","fields":{"col4":"ert","col5":"45","col6":"56"}}|
#+---------------------------------------------------------------------------------------+

#using toJSON
df.withColumn("jsn",struct("col1","col2","col3",struct("col4","col5","col6").alias("fields"))).drop(*cols).toJSON().collect()
#[u'{"jsn":{"col1":"123","col2":"abc","col3":"45","fields":{"col4":"ert","col5":"45","col6":"56"}}}']

#to write as json file
df.withColumn("jsn",struct("col1","col2","col3",struct("col4","col5","col6").alias("fields"))).\
drop(*cols).\
write.\
format("json").\
save("<path>")

更新：

jsn列表示为json结构

df.withColumn("jsn",struct("col1","col2","col3",struct("col4","col5","col6").alias("fields"))).drop(*cols).printSchema()
#root
# |-- jsn: struct (nullable = false)
# |    |-- col1: string (nullable = true)
# |    |-- col2: string (nullable = true)
# |    |-- col3: string (nullable = true)
# |    |-- fields: struct (nullable = false)
# |    |    |-- col4: string (nullable = true)
# |    |    |-- col5: string (nullable = true)
# |    |    |-- col6: string (nullable = true)