SparkSQL JSON数据操作(1.3->1.4)

1.用户自定义schema

data

json串格式如下:

{
    "partner_code": "demo",
    "app_name": "web",
    "person_info": {
        "name": "张三",
        "age": 18
    },
    "items": [
        {
            "item_id": 1,
            "item_name": "王家村",
            "group": "group1"
        },
        {
            "item_id": 2,
            "item_name": "李家澡堂",
            "item_detail": {
                "platform_count": 2
            },
            "group": "group2"
        }
    ]
}

spark1.3

在spark1.3我们是这样处理的

//定义schema
val struct =StructType(
    StructField("partner_code", StringType, true) ::
    StructField("app_name", StringType, true)::
    StructField("person_info",MapType(StringType,StringType,true)) ::                    
    StructField("items",ArrayType(MapType(StringType,StringType,true))) ::
        Nil)

 val data = sc.textFile("path/jsonFile")
 val df = sqlContext.jsonRDD(data,struct)
 df.printSchema
 df.show

spark1.4

//定义schema
val struct =StructType(
    StructField("partner_code", StringType, true) ::
    StructField("app_name", StringType, true)::
    StructField("person_info",MapType(StringType,StringType,true)) ::                    
    StructField("items",ArrayType(MapType(StringType,StringType,true))) ::
        Nil)

val df = sqlContext.read.schema(struct).json("path/jsonFile")

输出结果

//df.printSchema
root
 |-- partner_code: string (nullable = true)
 |-- app_name: string (nullable = true)
 |-- person_info: map (nullable = true)
 |    |-- key: string
 |    |-- value: string (valueContainsNull = true)
 |-- items: array (nullable = true)
 |    |-- element: map (containsNull = true)
 |    |    |-- key: string
 |    |    |-- value: string (valueContainsNull = true)

//df.show
+------------+--------+--------------------+--------------------+
|partner_code|app_name|         person_info|               items|
+------------+--------+--------------------+--------------------+
|        demo|     web|Map(name -> 张三, a...|List(Map(item_id ...|
+------------+--------+--------------------+--------------------+

系统自动生成schema

直接使用自带的解析会更方便，不过那样会产生大量的struct结构，同时如果结构复杂多变将会产生大量的空值。

//不需要定义schema,系统自动判断生成
val df = sqlContext.read.json("path/jsonFile")
df.printSchema
df.show

输出结果

//df.printSchema
root
 |-- app_name: string (nullable = true)
 |-- items: array (nullable = true)
 |    |-- element: struct (containsNull = true)
 |    |    |-- group: string (nullable = true)
 |    |    |-- item_detail: struct (nullable = true)
 |    |    |    |-- platform_count: long (nullable = true)
 |    |    |-- item_id: long (nullable = true)
 |    |    |-- item_name: string (nullable = true)
 |-- partner_code: string (nullable = true)
 |-- person_info: struct (nullable = true)
 |    |-- age: long (nullable = true)
 |    |-- name: string (nullable = true)

//df.show
+--------+--------------------+------------+-----------+
|app_name|               items|partner_code|person_info|
+--------+--------------------+------------+-----------+
|     web|List([group1,null...|        demo|    [18,张三]|
+--------+--------------------+------------+-----------+