Spark SQL 从入门到精通 - Spark SQL 行转列、列转行案例

Spark SQL 行转列、列转行案例

需求分析：

⾏列之间的互相转换是ETL中的常见需求，在Spark SQL中，⾏转列有内建的PIVOT函数可⽤，没什么特别之处。
⽽列转⾏要稍微⿇烦点。

Execution Environment

Spark 3.2.0

以下将通过 Spark SQL DSL 与 Spark SQL 两种方案进行 案例分享。

1、行转列 -> PIOVT

data-file : city.csv
city,yearm,count
北京,202001,1000
北京,202004,1023
北京,202007,1980
北京,202010,1098
北京,202101,988
北京,202104,976
北京,202107,1098
北京,202110,1221
上海,202001,1222
上海,202004,800
上海,202007,908
上海,202010,1009
上海,202101,709
上海,202104,799
上海,202107,980
上海,202110,897

   // 创建执行环境
   val session: SparkSession = SparkSession.builder()
     .appName(this.getClass.getSimpleName)
     .master("local[2]").getOrCreate()
   // 加载数据文件，注册DF
   val piovtCsvDF = session.read.format("csv")
     .option("header", true)
     .load("/Users/zhoulei/Documents/workspaces/zholei-core/ToolsLibrary/src/main/resources/city.csv")

Spark SQL DSL

   // Spark DSL Method
   piovtCsvDF
     .groupBy("city")
     .pivot("yearm")
     .agg(sum("count"))
     .show()

Spark SQL

   // 注册临时视图
   piovtCsvDF.createOrReplaceTempView("city_main")
   // Spark SQL Method
   session.sql(
     """
       |select * from city_main
       |pivot(
       |    SUM(count) for yearm in (
       |        '202001' as Q202001,
       |        '202004' as Q202004,
       |        '202007' as Q202007,
       |        '202010' as Q202010,
       |        '202101' as Q202101,
       |        '202104' as Q202104,
       |        '202107' as Q202107,
       |        '202110' as Q202110
       |        )
       |    )
       |""".stripMargin).show(false)

2、列转行 -> UNPIOVT

studentscore.csv

un,chinese,math,English
张三,91,92,93
李四,80,81,32
王五,70,78,80

   val unpiovtCsvDF = session.read.format("csv")
     .option("header", true)
     .load(getClass.getResource("/studentscore.csv"))

Spark SQL DSL

stack(n, expr1, …, exprk) - 会将expr1, …, exprk 分割为n⾏

   unpiovtCsvDF
     .selectExpr("un", "stack(3,'chinese',chinese,'math',math,'english',english) as (subject,score)")
     .show(false)

Spark SQL

   // 注册临时视图
   unpiovtCsvDF.createOrReplaceTempView("student_score_main")

   // Spark SQL Method (stack function)
   session.sql(
     """
       |select
       |un,stack(3,'chinese',chinese,'math',math,'english',english) as (subject,score)
       |from student_score_main
       |""".stripMargin).show(false)

// Spark SQL Method (lateral view explode)
session.sql(
  """
    |select
    |un,split(temp1,':')[0] as subject,split(temp1,':')[1] as score
    |from (
    |select un,concat('chinese:',chinese,',','math:',math,',','english:',english) temp
    |from student_score_main
    |) lateral view explode(split(temp,',')) sub as temp1
    |""".stripMargin).show(false)

知识小课堂：

• MySQL无unpiovt(列转行)、pivot(行转列)这两个函数，unpiovt 用 union all 实现，pivot用聚合函数+ case when/if 等判断语句 实现。