Pyspark: Parse a column of json strings

For Spark 2.1+, you can use from_json which allows the preservation of the other non-json columns within the dataframe as follows:

from pyspark.sql.functions import from_json, col
json_schema = spark.read.json(df.rdd.map(lambda row: row.json)).schema
df.withColumn('json', from_json(col('json'), json_schema))

You let Spark derive the schema of the json string column. Then the df.json column is no longer a StringType, but the correctly decoded json structure, i.e., nested StrucType and all the other columns of df are preserved as-is.

You can access the json content as follows:

df.select(col('json.header').alias('header'))

Converting a dataframe with json strings to structured dataframe is'a actually quite simple in spark if you convert the dataframe to RDD of strings before (see: http://spark.apache.org/docs/latest/sql-programming-guide.html#json-datasets)

For example:

>>> new_df = sql_context.read.json(df.rdd.map(lambda r: r.json))
>>> new_df.printSchema()
root
 |-- body: struct (nullable = true)
 |    |-- id: long (nullable = true)
 |    |-- name: string (nullable = true)
 |    |-- sub_json: struct (nullable = true)
 |    |    |-- id: long (nullable = true)
 |    |    |-- sub_sub_json: struct (nullable = true)
 |    |    |    |-- col1: long (nullable = true)
 |    |    |    |-- col2: string (nullable = true)
 |-- header: struct (nullable = true)
 |    |-- foo: string (nullable = true)
 |    |-- id: long (nullable = true)