大数据Spark DataFrame/DataSet常用操作

Posted 赵广陆

tags:

篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了大数据Spark DataFrame/DataSet常用操作相关的知识，希望对你有一定的参考价值。

目录

1 一般操作：查找和过滤

1.1 读取数据源

1.1.1读取json

使用spark.read。注意：路径默认是从HDFS，如果要读取本机文件，需要加前缀file://，如下

scala> val people = spark.read.format("json").load("file:///opt/software/data/people.json")
people: org.apache.spark.sql.DataFrame = [age: bigint, name: string]
scala> people.show
+----+-------+
| age|   name|
+----+-------+
|null|Michael|
|  30|   Andy|
|  19| Justin|
+----+-------+

spark.read.format(“json”).load(“file:///opt/software/data/people.json”)
等价于spark.read.json(“file:///opt/software/data/people.json”)
如要要读取其它格式文件，只需修改format(“json”)即可，如format(“parquet”)

1.1.2 读取Hive表

使用spark.sql。其中hive数据库名default(默认数据库名可省略)，表为people

scala> val peopleDF=spark.sql("select * from default.people")
peopleDF: org.apache.spark.sql.DataFrame = [name: string, age: int ... 1 more field]
scala> peopleDF.show
+--------+---+--------+
|    name|age| address|
+--------+---+--------+
|zhangsan| 22| chengdu|
|  wangwu| 33| beijing|
|    lisi| 28|shanghai|
+--------+---+--------+
scala> peopleDF.printSchema
root
|-- name: string (nullable = true)
|-- age: integer (nullable = true)
|-- address: string (nullable = true)

1.2 取数据列

取列的三种方式如下

scala> peopleDF.select("name","age").show
+--------+---+
|    name|age|
+--------+---+
|zhangsan| 22|
|  wangwu| 33|
|    lisi| 28|
+--------+---+
scala> peopleDF.select($"name",$"age").show
+--------+---+
|    name|age|
+--------+---+
|zhangsan| 22|
|  wangwu| 33|
|    lisi| 28|
+--------+---+
scala> peopleDF.select(peopleDF.col("name"),peopleDF.col("age")).show
+--------+---+
|    name|age|
+--------+---+
|zhangsan| 22|
|  wangwu| 33|
|    lisi| 28|
+--------+---+

注意：如果在IDEA中编辑代码，使用 ， 则 必 须 增 加 语 句 ： i m p o r t s p a r k . i m p l i c i t s . ， 否 则 ，则必须增加语句：import spark.implicits._，否则 ，则必须增加语句：importspark.implicits.，​否则表达式会报错。spark-shell默认已经导入了的
$”列名”这个是语法糖，返回Column对象

1.3 过滤算子filter(filter等价于where算子)

DF.col("id")等价于$"id"，取列ColumnName
DF.filter("name=''") 过滤name等于空的行
DF.filter($"age" > 21).show() 过滤age大于21的行，必须增加语句：import spark.implicits._，否则$表达式会报错
DF.filter($"age" === 21) 取等于时必须用===，否则报错，对应的不等于是=!=。等价于DF.filter("age=21")
DF.filter("substring(name,0,1) = 'M'").show 显示name以M开头的行，其中substring是functions.scala,functions.scala包含很多函数方法，等价于DF.filter("substr(name,0,1) = 'M'").show

scala> peopleDF.printSchema
root
|-- name: string (nullable = true)
|-- age: integer (nullable = true)
|-- address: string (nullable = true)
scala> peopleDF.show
+--------+---+--------+
|    name|age| address|
+--------+---+--------+
|zhangsan| 22| chengdu|
|  wangwu| 33| beijing|
|    lisi| 28|shanghai|
+--------+---+--------+
scala> peopleDF.filter($"name" === "wangwu").show
+------+---+-------+
|  name|age|address|
+------+---+-------+
|wangwu| 33|beijing|
+------+---+-------+
scala> peopleDF.filter($"name" =!= "wangwu").show
+--------+---+--------+
|    name|age| address|
+--------+---+--------+
|zhangsan| 22| chengdu|
|    lisi| 28|shanghai|
+--------+---+--------+
scala> peopleDF.filter("age > 30").show
+------+---+-------+
|  name|age|address|
+------+---+-------+
|wangwu| 33|beijing|
+------+---+-------+
scala> peopleDF.filter($"age" > 30).show
+------+---+-------+
|  name|age|address|
+------+---+-------+
|wangwu| 33|beijing|
+------+---+-------+

2 聚合操作：groupBy和agg

2.1 排序算子sort(sort等价于orderBy)

DF.sort(DF.col(“id”).desc).show 以DF中字段id降序，指定升降序的方法。另外可指定多个字段排序
=DF.sort($“id”.desc).show
DF.sort 等价于DF.orderBy

scala> peopleDF.sort($"age").show
+--------+---+--------+
|    name|age| address|
+--------+---+--------+
|zhangsan| 22| chengdu|
|    lisi| 28|shanghai|
|  wangwu| 33| beijing|
+--------+---+--------+
scala> peopleDF.sort($"age".desc).show
+--------+---+--------+
|    name|age| address|
+--------+---+--------+
|  wangwu| 33| beijing|
|    lisi| 28|shanghai|
|zhangsan| 22| chengdu|
+--------+---+--------+
scala> peopleDF.sort($"age".asc).show
+--------+---+--------+
|    name|age| address|
+--------+---+--------+
|zhangsan| 22| chengdu|
|    lisi| 28|shanghai|
|  wangwu| 33| beijing|
+--------+---+--------+
scala> peopleDF.orderBy($"age".asc).show
+--------+---+--------+
|    name|age| address|
+--------+---+--------+
|zhangsan| 22| chengdu|
|    lisi| 28|shanghai|
|  wangwu| 33| beijing|
+--------+---+--------+
scala> peopleDF.orderBy($"age".desc).show
+--------+---+--------+
|    name|age| address|
+--------+---+--------+
|  wangwu| 33| beijing|
|    lisi| 28|shanghai|
|zhangsan| 22| chengdu|
+--------+---+--------+

2.2 分组函数groupBy

2.2.1 分组计数

select address,count(1) from people group by address; 等价的算子如下

scala> peopleDF.show()
+--------+---+--------+
|    name|age| address|
+--------+---+--------+
|zhangsan| 22| chengdu|
|  wangwu| 33| beijing|
|    lisi| 28|shanghai|
|xiaoming| 28| beijing|
|      mm| 21| chengdu|
|xiaoming| 18| beijing|
|      mm| 11| chengdu|
+--------+---+--------+
scala> peopleDF.groupBy("address").count().show
+--------+-----+
| address|count|
+--------+-----+
| beijing|    3|
| chengdu|    3|
|shanghai|    1|
+--------+-----+

2.2.2 分组后求最值、平均值、求和的方法

//等价于select address,max(age) from people group by address;
scala> peopleDF.groupBy("address").max("age").show
+--------+--------+
| address|max(age)|
+--------+--------+
| beijing|      33|
| chengdu|      22|
|shanghai|      28|
+--------+--------+
//等价于select address,avg(age) from people group by address;
scala> peopleDF.groupBy("address").avg("age").show
+--------+------------------+
| address|          avg(age)|
+--------+------------------+
| beijing|26.333333333333332|
| chengdu|              18.0|
|shanghai|              28.0|
+--------+------------------+
//等价于select address,min(age) from people group by address;
scala> peopleDF.groupBy("address").min("age").show
+--------+--------+
| address|min(age)|
+--------+--------+
| beijing|      18|
| chengdu|      11|
|shanghai|      28|
+--------+--------+
//等价于select address,sum(age) from people group by address;
scala> peopleDF.groupBy("address").sum("age").show
+--------+--------+
| address|sum(age)|
+--------+--------+
| beijing|      79|
| chengdu|      54|
|shanghai|      28|
+--------+--------+

2.2.3 分组后，求多个聚合值（最值、平均值等）。使用算子groupBy+agg

//等价于select address,count(age),max(age),min(age),avg(age),sum(age) from people group by address;
scala> peopleDF.groupBy("address").agg(count("age"),max("age"),min("age"),avg("age"),sum("age")).show
+--------+----------+--------+--------+------------------+--------+
| address|count(age)|max(age)|min(age)|          avg(age)|sum(age)|
+--------+----------+--------+--------+------------------+--------+
| beijing|         3|      33|      18|26.333333333333332|      79|
| chengdu|         3|      22|      11|              18.0|      54|
|shanghai|         1|      28|      28|              28.0|      28|
+--------+----------+--------+--------+------------------+--------+

2.2.4 分组聚合后取别名

scala> peopleDF.groupBy("address").agg(count("age").as("cnt"),avg("age").as("avg")).show
+--------+---+------------------+
| address|cnt|               avg|
+--------+---+------------------+
| beijing|  3|26.333333333333332|
| chengdu|  3|              18.0|
|shanghai|  1|              28.0|
+--------+---+------------------+

2.2.5 分组后行转列，使用pivot

//求同名用户在同一个地址的平均年龄
//把name的不同值作为列名
scala> peopleDF.groupBy("address").pivot("name").avg("age").show
+--------+----+----+------+--------+--------+
| address|lisi|  mm|wangwu|xiaoming|zhangsan|
+--------+----+----+------+--------+--------+
| beijing|null|null|  33.0|    23.0|    null|
| chengdu|null|16.0|  null|    null|    22.0|
|shanghai|28.0|null|  null|    null|    null|
+--------+----+----+------+--------+--------+

2.3 案例

3 多表操作Join






3.1 数据准备

先构建两个DataFrame

scala>  val df1 = spark.createDataset(Seq(("a", 1,2), ("b",2,3) )).toDF("k1","k2","k3")
df1: org.apache.spark.sql.DataFrame = [k1: string, k2: int ... 1 more field]
scala>  val df2 = spark.createDataset(Seq(("a", 2,2), ("b",3,3), ("b", 2,1), ("c", 1,1)) ).toDF("k1","k2","k4")
df2: org.apache.spark.sql.DataFrame = [k1: string, k2: int ... 1
以上是关于大数据Spark DataFrame/DataSet常用操作的主要内容，如果未能解决你的问题，请参考以下文章
 

 大数据入门核心技术-Spark执行Spark任务的两种方式:spark-submit和spark-shell


 大数据（spark sql 和 spark dataframes 连接）


 大数据之Spark:Spark 基础


 大数据中的Spark指的是啥？


 数分-理论-大数据7-Spark


 Spark一出，Hadoop必死？Spark才是大数据的未来？