spark SQL之 DataFrame使用

pom.xml中

<!-- https://mvnrepository.com/artifact/com.google.guava/guava -->
    <dependency>
      <groupId>com.google.guava</groupId>
      <artifactId>guava</artifactId>
      <version>14.0.1</version>
    </dependency>

一、toDF

1.直接写全部列名

val spark = SparkSession.builder().master("local[2]").appName("app").getOrCreate()
val df = spark.read.format("csv").load("file:///D:/idea/ideaProjects/spark_projects/myspark8/src/main/scala/com/njbdqn/DSDF/orders.csv")
      .toDF("id", "orddate", "itemid", "status")
// 去表头的写法
spark.read.format("csv").option("header","true").load("hdfs://192.168.56.111:9000/party/data/users.csv").cache() dfUsers.show(false)

2.选择需要的列

val orders = spark.read.format("csv").load("file:///D:/idea/ideaProjects/spark_projects/myspark8/src/main/scala/com/njbdqn/DSDF/orders.csv")
      .select("_c0","_c2").toDF("orderid","userid")

二、使用DF-SQL语句方式

1.全局TempView需要 global_temp.orders 

//df.createOrReplaceGlobalTempView("orders")  
 df.createOrReplaceTempView("orders")

2.使用spark.sql()

spark.sql("select if(dayofweek(orddate)==1,'Sun',dayofweek(orddate)-1) as s ,count(id) from orders group by dayofweek(orddate)").show(false)

 spark.sql()中写多行：使用stripMargin，按照“ | ”识别开头，避免格式不一致的问题

    val sql="""select
              |userid,eventid,
              |max(case when statu='invited' then 1 else 0 end) invited,
              |max(case when statu='yes' then 1 else 0 end) yes,
              |max(case when statu='maybe' then 1 else 0 end) maybe,
              |max(case when statu='no' then 1 else 0 end) no
              |from
              |dwd_events.dwd_eventAttendees group by userid,eventid""".stripMargin
    spark.sql(sql)

三、使用DF-SQL API

【如何实现CASE WHEN1】

1.自定义UDF函数:

1. 如何调用？参数是Column对象 wdchange($"weekdate") 
2. 入参/出参 是什么类型？

   udf{(sc:Int)=>{返回String}} 
   udf{(sc:String)=>{返回Int}}
   // 都没有问题

import org.apache.spark.sql.functions._
    val wdchange = udf{(num:Int)=>
      num match {
        case 1 => "七"
        case 2 => "一"
        case 3 => "二"
        case 4 => "三"
        case 5 => "四"
        case 6 => "五"
        case _ => "六"
      }
    }

// 先groupby再agg
// 操作几个列，如果最后不select，都会显示、
// agg里面两种写法
// withColumn表示再添一列，$表示列对象，需要implicts._
df // 几周？date_format(,"u")
    .groupBy(date_format($"orddate","u").alias("weekdate"))
     // .agg("id"->"sum","id"->"count")  
    .agg(count("id").alias("countId"),sum("id").alias("sumId"))
    .withColumn("wd",wdchange($"weekdate"))
    .show(false)

【如何实现CASE WHEN2】

2.使用when.otherwise

    import spark.implicits._
    df
      .groupBy(when(dayofweek($"orddate")===1,"七")
        .when(dayofweek($"orddate")===2,"一")
        .when(dayofweek($"orddate")===3,"二")
        .when(dayofweek($"orddate")===4,"三")
        .when(dayofweek($"orddate")===5,"四")
        .when(dayofweek($"orddate")===6,"五")
        .otherwise("六").alias("day"))
      .agg(count("id").alias("countId"),sum("id").alias("sumId"))
      .show(false)

结果：

+---+-------+------------+
|day|countId|sumId |
+---+-------+------------+
|六 |9984 |3.35875381E8|
|三 |9758 |3.46913986E8|
|四 |9862 |3.32166696E8|
|二 |9964 |3.45180682E8|
|一 |9292 |3.23174788E8|
|五 |10288 |3.53314523E8|
|七 |9735 |3.3584223E8 |
+---+-------+------------+

3.另一个例子

分析：根据city分组sum(countprice)

orderid  - countprice ： 1：n   =>   orderid  - sum(countprice) : 1： 1

              - userid      ：  n：1   =>    sum(countprice) - userid   : n：1 

 　　　　 userid - city: 1 ：1   =>   sum(countprice)  - city : n：1      => sum(countprice)  - groupby(city): 1 : 1

// groupBy(列)必须是当表有的列

// groupBy可以加列对象也可以加列名，返回的是 RelationalGroupedDataset，只有DataFrame格式的可以show()

def groupBy(col1 : scala.Predef.String, cols : scala.Predef.String*)

def groupBy(cols : org.apache.spark.sql.Column*)

 orderitem
      //.select($"orderid",$"countprice".cast(DataTypes.DoubleType))
      .groupBy("orderid").agg(sum("countprice").alias("sumPrice"))
      .join(orders,Seq("orderid"),"inner")
      .join(customers,Seq("userid"),"inner").groupBy($"city").agg(sum("sumPrice").alias("cityPrice"))
      .orderBy(desc("cityPrice"))
      .show()

4. 使用窗口函数

 row_number().over(Window.orderBy("good_id")).alias("gid")) 

import org.apache.spark.sql.expressions.Window
val wnd = Window.partitionBy(month($"date"))
.orderBy("date") 
df.select($"name",$"date",$"count",sum("count") over(wnd)).show(false)

 为什么使用窗口函数？因为使用了sum/min聚合函数，必须有group by；如果不想group by，就需要over（）成为单独的一列

.withColumn("min_login_time",min($"last_login_time")over())　　　　　　　// 聚合函数，不是groupBy，就是over()
.withColumn("lasttime",datediff($"last_login_time",$"min_login_time")) // 普通函数，不需要窗口

四、使用JAVA的Calendar + RDD 形式

// date转星期 JAVA
  def daychange(dateStr:String): String ={
    val arr = Array("日","一","二","三","四","五","六")
    val cal = Calendar.getInstance()
    val sdf = new SimpleDateFormat("yyyy-MM-dd hh:mm:ss")
    cal.setTime(sdf.parse(dateStr))
    val week = arr(cal.get(Calendar.DAY_OF_WEEK)-1)
   // println(week)
    week
  }

// Java + RDD
// dataFrame.map => dataSet 转变成DataSet

    val str = df.select("id", "orddate")
      .map(x => {
        (daychange(x(1).toString), x(0).toString)
      })
     .rdd.groupByKey().foreach(x=>println(x._1,x._2.size))