您可以使用字典解析在架构中导航。

{x.name: x.dataType[0].name for x in df.schema if x.dataType.typeName() == 'struct'}

试验1

df = spark.createDataFrame([], 'field_1 int, field_2 int, field_3 struct<date:date,timestamp:timestamp>, field_4 int')

df.printSchema()
# root
#  |-- field_1: integer (nullable = true)
#  |-- field_2: integer (nullable = true)
#  |-- field_3: struct (nullable = true)
#  |    |-- date: date (nullable = true)
#  |    |-- timestamp: timestamp (nullable = true)
#  |-- field_4: integer (nullable = true)

{x.name: x.dataType[0].name for x in df.schema if x.dataType.typeName() == 'struct'}
# {'field_3': 'date'}

测试2

df = spark.createDataFrame([], 'field_1 int, field_2 struct<col_int:int,col_long:long>, field_3 struct<date:date,timestamp:timestamp>')

df.printSchema()
# root
#  |-- field_1: integer (nullable = true)
#  |-- field_2: struct (nullable = true)
#  |    |-- col_int: integer (nullable = true)
#  |    |-- col_long: long (nullable = true)
#  |-- field_3: struct (nullable = true)
#  |    |-- date: date (nullable = true)
#  |    |-- timestamp: timestamp (nullable = true)

{x.name: x.dataType[0].name for x in df.schema if x.dataType.typeName() == 'struct'}
# {'field_2': 'col_int', 'field_3': 'date'}