其他
2 小时入门 SparkSQL 编程
The following article is from 算法美食屋 Author 梁云1991
不同于RDD编程的命令式编程范式，SparkSQL编程是一种声明式编程范式，我们可以通过SQL语句或者调用DataFrame的相关API描述我们想要实现的操作。
然后Spark会将我们的描述进行语法解析，找到相应的执行计划并对其进行流程优化，然后调用相应基础命令进行执行。
我们使用pyspark进行RDD编程时，在Excutor上跑的很多时候就是Python代码，当然，少数时候也会跑java字节码。
但我们使用pyspark进行SparkSQL编程时，在Excutor上跑的全部是java字节码，pyspark在Driver端就将相应的Python代码转换成了java任务然后放到Excutor上执行。
因此，使用SparkSQL的编程范式进行编程，我们能够取得几乎和直接使用scala/java进行编程相当的效率(忽略语法解析时间差异)。
此外SparkSQL提供了非常方便的数据读写API，我们可以用它和Hive表，HDFS，mysql表，Cassandra，Hbase等各种存储媒介进行数据交换。
美中不足的是，SparkSQL的灵活性会稍差一些，其默认支持的数据类型通常只有 Int，Long，Float，Double，String，Boolean 等这些标准SQL数据类型, 类型扩展相对繁琐。
对于一些较为SQL中不直接支持的功能，通常可以借助于用户自定义函数(UDF)来实现，如果功能更加复杂，则可以转成RDD来进行实现。
本节我们将主要介绍以下主要内容：
RDD和DataFrame的对比
创建DataFrame
DataFrame保存成文件
DataFrame的API交互
DataFrame的SQL交互
import findspark
#指定spark_home为刚才的解压路径,指定python路径
spark_home = "/Users/liangyun/ProgramFiles/spark-3.0.1-bin-hadoop3.2"
python_path = "/Users/liangyun/anaconda3/bin/python"
findspark.init(spark_home,python_path)
import pyspark
from pyspark.sql import SparkSession
#SparkSQL的许多功能封装在SparkSession的方法接口中
spark = SparkSession.builder \
.appName("test") \
.config("master","local[4]") \
.enableHiveSupport() \
.getOrCreate()
sc = spark.sparkContext
一，RDD，DataFrame
和DataSet对比
DataFrame参照了Pandas的思想，在RDD基础上增加了schma，能够获取列名信息。
DataSet在DataFrame基础上进一步增加了数据类型信息，可以在编译时发现类型错误。
DataFrame可以看成DataSet[Row]，两者的API接口完全相同。
DataFrame和DataSet都支持SQL交互式查询，可以和 Hive无缝衔接。
DataSet只有Scala语言和Java语言接口中才支持，在Python和R语言接口只支持DataFrame。
DataFrame数据结构本质上是通过RDD来实现的，但是RDD是一种行存储的数据结构，而DataFrame是一种列存储的数据结构。
二，创建DataFrame
1，通过toDF方法转换成DataFrame
可以将RDD用toDF方法转换成DataFrame
#将RDD转换成DataFrame
rdd = sc.parallelize([("LiLei",15,88),("HanMeiMei",16,90),("DaChui",17,60)])
df = rdd.toDF(["name","age","score"])
df.show()
df.printSchema()
+---------+---+-----+
| name|age|score|
+---------+---+-----+
| LiLei| 15| 88|
|HanMeiMei| 16| 90|
| DaChui| 17| 60|
+---------+---+-----+
root
|-- name: string (nullable = true)
|-- age: long (nullable = true)
|-- score: long (nullable = true)2, 通过createDataFrame方法将Pandas.DataFrame转换成pyspark中的DataFrame
import pandas as pd
pdf = pd.DataFrame([("LiLei",18),("HanMeiMei",17)],columns = ["name","age"])
df = spark.createDataFrame(pdf)
df.show()
+---------+---+
| name|age|
+---------+---+
| LiLei| 18|
|HanMeiMei| 17|
+---------+---+
# 也可以对列表直接转换
values = [("LiLei",18),("HanMeiMei",17)]
df = spark.createDataFrame(values,["name","age"])
df.show()
+---------+---+
| name|age|
+---------+---+
| LiLei| 18|
|HanMeiMei| 17|
+---------+---+4, 通过createDataFrame方法指定schema动态创建DataFrame
可以通过createDataFrame的方法指定rdd和schema创建DataFrame。
这种方法比较繁琐，但是可以在预先不知道schema和数据类型的情况下在代码中动态创建DataFrame。
from pyspark.sql.types import *
from pyspark.sql import Row
from datetime import datetime
schema = StructType([StructField("name", StringType(), nullable = False),
StructField("score", IntegerType(), nullable = True),
StructField("birthday", DateType(), nullable = True)])
rdd = sc.parallelize([Row("LiLei",87,datetime(2010,1,5)),
Row("HanMeiMei",90,datetime(2009,3,1)),
Row("DaChui",None,datetime(2008,7,2))])
dfstudent = spark.createDataFrame(rdd, schema)
dfstudent.show()
+---------+-----+----------+
| name|score| birthday|
+---------+-----+----------+
| LiLei| 87|2010-01-05|
|HanMeiMei| 90|2009-03-01|
| DaChui| null|2008-07-02|
+---------+-----+----------+
4，通过读取文件创建
可以读取json文件，csv文件，hive数据表或者mysql数据表得到DataFrame。
#读取json文件生成DataFrame
df = spark.read.json("data/people.json")
df.show()
+----+-------+
| age| name|
+----+-------+
|null|Michael|
| 30| Andy|
| 19| Justin|
+----+-------+
#读取csv文件
df = spark.read.option("header","true") \
.option("inferSchema","true") \
.option("delimiter", ",") \
.csv("data/iris.csv")
df.show(5)
df.printSchema()
+-----------+----------+-----------+----------+-----+
|sepallength|sepalwidth|petallength|petalwidth|label|
+-----------+----------+-----------+----------+-----+
| 5.1| 3.5| 1.4| 0.2| 0|
| 4.9| 3.0| 1.4| 0.2| 0|
| 4.7| 3.2| 1.3| 0.2| 0|
| 4.6| 3.1| 1.5| 0.2| 0|
| 5.0| 3.6| 1.4| 0.2| 0|
+-----------+----------+-----------+----------+-----+
only showing top 5 rows
root
|-- sepallength: double (nullable = true)
|-- sepalwidth: double (nullable = true)
|-- petallength: double (nullable = true)
|-- petalwidth: double (nullable = true)
|-- label: integer (nullable = true)
#读取csv文件
df = spark.read.format("com.databricks.spark.csv") \
.option("header","true") \
.option("inferSchema","true") \
.option("delimiter", ",") \
.load("data/iris.csv")
df.show(5)
df.printSchema()
+-----------+----------+-----------+----------+-----+
|sepallength|sepalwidth|petallength|petalwidth|label|
+-----------+----------+-----------+----------+-----+
| 5.1| 3.5| 1.4| 0.2| 0|
| 4.9| 3.0| 1.4| 0.2| 0|
| 4.7| 3.2| 1.3| 0.2| 0|
| 4.6| 3.1| 1.5| 0.2| 0|
| 5.0| 3.6| 1.4| 0.2| 0|
+-----------+----------+-----------+----------+-----+
only showing top 5 rows
root
|-- sepallength: double (nullable = true)
|-- sepalwidth: double (nullable = true)
|-- petallength: double (nullable = true)
|-- petalwidth: double (nullable = true)
|-- label: integer (nullable = true)#读取parquet文件
df = spark.read.parquet("data/users.parquet")
df.show()
+------+--------------+----------------+
| name|favorite_color|favorite_numbers|
+------+--------------+----------------+
|Alyssa| null| [3, 9, 15, 20]|
| Ben| red| []|
+------+--------------+----------------+
#读取hive数据表生成DataFrame
spark.sql("CREATE TABLE IF NOT EXISTS src (key INT, value STRING) USING hive")
spark.sql("LOAD DATA LOCAL INPATH 'data/kv1.txt' INTO TABLE src")
df = spark.sql("SELECT key, value FROM src WHERE key < 10 ORDER BY key")
df.show(5)
+---+-----+
|key|value|
+---+-----+
| 0|val_0|
| 0|val_0|
| 0|val_0|
| 0|val_0|
| 0|val_0|
+---+-----+
only showing top 5 rows
#读取mysql数据表生成DataFrame
"""
url = "jdbc:mysql://localhost:3306/test"
df = spark.read.format("jdbc") \
.option("url", url) \
.option("dbtable", "runoob_tbl") \
.option("user", "root") \
.option("password", "0845") \
.load()\
df.show()
"""
三，DataFrame保存成文件
可以保存成csv文件，json文件，parquet文件或者保存成hive数据表
#保存成csv文件
df = spark.read.format("json").load("data/people.json")
df.write.format("csv").option("header","true").save("data/people_write.csv")
#先转换成rdd再保存成txt文件
df.rdd.saveAsTextFile("data/people_rdd.txt")
#保存成json文件
df.write.json("data/people_write.json")
#保存成parquet文件, 压缩格式, 占用存储小, 且是spark内存中存储格式，加载最快
df.write.partitionBy("age").format("parquet").save("data/namesAndAges.parquet")
df.write.parquet("data/people_write.parquet")
#保存成hive数据表
df.write.bucketBy(42, "name").sortBy("age").saveAsTable("people_bucketed")
四，DataFrame的API交互
from pyspark.sql import Row
from pyspark.sql.functions import *
df = spark.createDataFrame(
[("LiLei",15,"male"),
("HanMeiMei",16,"female"),
("DaChui",17,"male")]).toDF("name","age","gender")
df.show()
df.printSchema()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
+---------+---+------+
root
|-- name: string (nullable = true)
|-- age: long (nullable = true)
|-- gender: string (nullable = true)
1，Action操作
DataFrame的Action操作包括show，count，collect，describe，take，head，first等操作。
#show
df.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
+---------+---+------+
#show(numRows: Int, truncate: Boolean)
#第二个参数设置是否当输出字段长度超过20时进行截取
df.show(2,False)
+---------+---+------+
|name |age|gender|
+---------+---+------+
|LiLei |15 |male |
|HanMeiMei|16 |female|
+---------+---+------+
only showing top 2 rows
#count
df.count()
3
#collect
df.collect()
[Row(name='LiLei', age=15, gender='male'),
Row(name='HanMeiMei', age=16, gender='female'),
Row(name='DaChui', age=17, gender='male')]
#first
df.first()
Row(name='LiLei', age=15, gender='male')
#take
df.take(2)
[Row(name='LiLei', age=15, gender='male'),
Row(name='HanMeiMei', age=16, gender='female')]
#head
df.head(2)
[Row(name='LiLei', age=15, gender='male'),
Row(name='HanMeiMei', age=16, gender='female')]
2，类RDD操作
DataFrame支持RDD中一些诸如distinct，cache，sample，foreach，intersect，except等操作。
可以把DataFrame当做数据类型为Row的RDD来进行操作，必要时可以将其转换成RDD来操作。
df = spark.createDataFrame([("Hello World",),("Hello China",),("Hello Spark",)]).toDF("value")
df.show()
+-----------+
| value|
+-----------+
|Hello World|
|Hello China|
|Hello Spark|
+-----------+
#map操作，需要先转换成rdd
rdd = df.rdd.map(lambda x:Row(x[0].upper()))
dfmap = rdd.toDF(["value"]).show()
+-----------+
| value|
+-----------+
|HELLO WORLD|
|HELLO CHINA|
|HELLO SPARK|
+-----------+
#flatMap，需要先转换成rdd
df_flat = df.rdd.flatMap(lambda x:x[0].split(" ")).map(lambda x:Row(x)).toDF(["value"])
df_flat.show()
+-----+
|value|
+-----+
|Hello|
|World|
|Hello|
|China|
|Hello|
|Spark|
+-----+
#filter过滤
df_filter = df.rdd.filter(lambda s:s[0].endswith("Spark")).toDF(["value"])
df_filter.show()
+-----------+
| value|
+-----------+
|Hello Spark|
+-----------+
# filter和broadcast混合使用
broads = sc.broadcast(["Hello","World"])
df_filter_broad = df_flat.filter(~col("value").isin(broads.value))
df_filter_broad.show()
+-----+
|value|
+-----+
|China|
|Spark|
+-----+
#distinct
df_distinct = df_flat.distinct()
df_distinct.show()
+-----+
|value|
+-----+
|World|
|China|
|Hello|
|Spark|
+-----+
#cache缓存
df.cache()
df.unpersist()
#sample抽样
dfsample = df.sample(False,0.6,0)
dfsample.show()
+-----------+
| value|
+-----------+
|Hello China|
|Hello Spark|
+-----------+
df2 = spark.createDataFrame([["Hello World"],["Hello Scala"],["Hello Spark"]]).toDF("value")
df2.show()
+-----------+
| value|
+-----------+
|Hello World|
|Hello Scala|
|Hello Spark|
+-----------+
#intersect交集
dfintersect = df.intersect(df2)
dfintersect.show()
+-----------+
| value|
+-----------+
|Hello Spark|
|Hello World|
+-----------+
#exceptAll补集
dfexcept = df.exceptAll(df2)
dfexcept.show()
+-----------+
| value|
+-----------+
|Hello China|
+-----------+
3，类Excel操作
可以对DataFrame进行增加列，删除列，重命名列，排序等操作，去除重复行，去除空行，就跟操作Excel表格一样。
df = spark.createDataFrame([
("LiLei",15,"male"),
("HanMeiMei",16,"female"),
("DaChui",17,"male"),
("RuHua",16,None)
]).toDF("name","age","gender")
df.show()
df.printSchema()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
| RuHua| 16| null|
+---------+---+------+
root
|-- name: string (nullable = true)
|-- age: long (nullable = true)
|-- gender: string (nullable = true)
#增加列
dfnew = df.withColumn("birthyear",-df["age"]+2020)
dfnew.show()
+---------+---+------+---------+
| name|age|gender|birthyear|
+---------+---+------+---------+
| LiLei| 15| male| 2005|
|HanMeiMei| 16|female| 2004|
| DaChui| 17| male| 2003|
| RuHua| 16| null| 2004|
+---------+---+------+---------+
#置换列的顺序
dfupdate = dfnew.select("name","age","birthyear","gender")
dfupdate.show()
#删除列
dfdrop = df.drop("gender")
dfdrop.show()
+---------+---+
| name|age|
+---------+---+
| LiLei| 15|
|HanMeiMei| 16|
| DaChui| 17|
| RuHua| 16|
+---------+---+
#重命名列
dfrename = df.withColumnRenamed("gender","sex")
dfrename.show()
+---------+---+------+
| name|age| sex|
+---------+---+------+
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
| RuHua| 16| null|
+---------+---+------+
#排序sort，可以指定升序降序
dfsorted = df.sort(df["age"].desc())
dfsorted.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| DaChui| 17| male|
| RuHua| 16| null|
|HanMeiMei| 16|female|
| LiLei| 15| male|
+---------+---+------+
#排序orderby,默认为升序,可以根据多个字段
dfordered = df.orderBy(df["age"].desc(),df["gender"].desc())
dfordered.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| DaChui| 17| male|
|HanMeiMei| 16|female|
| RuHua| 16| null|
| LiLei| 15| male|
+---------+---+------+
#去除nan值行
dfnotnan = df.na.drop()
dfnotnan.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
+---------+---+------+
#填充nan值
df_fill = df.na.fill("female")
df_fill.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
| RuHua| 16|female|
+---------+---+------+
#替换某些值
df_replace = df.na.replace({"":"female","RuHua":"SiYu"})
df_replace.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
| SiYu| 16| null|
+---------+---+------+
#去重，默认根据全部字段
df2 = df.unionAll(df)
df2.show()
dfunique = df2.dropDuplicates()
dfunique.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
| RuHua| 16| null|
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
| RuHua| 16| null|
+---------+---+------+
+---------+---+------+
| name|age|gender|
+---------+---+------+
| RuHua| 16| null|
| DaChui| 17| male|
|HanMeiMei| 16|female|
| LiLei| 15| male|
+---------+---+------+
#去重,根据部分字段
dfunique_part = df.dropDuplicates(["age"])
dfunique_part.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| DaChui| 17| male|
| LiLei| 15| male|
|HanMeiMei| 16|female|
+---------+---+------+
#简单聚合操作
dfagg = df.agg({"name":"count","age":"max"})
dfagg.show()
+-----------+--------+
|count(name)|max(age)|
+-----------+--------+
| 4| 17|
+-----------+--------+
#汇总信息
df_desc = df.describe()
df_desc.show()
+-------+------+-----------------+------+
|summary| name| age|gender|
+-------+------+-----------------+------+
| count| 4| 4| 3|
| mean| null| 16.0| null|
| stddev| null|0.816496580927726| null|
| min|DaChui| 15|female|
| max| RuHua| 17| male|
+-------+------+-----------------+------+
#频率超过0.5的年龄和性别
df_freq = df.stat.freqItems(("age","gender"),0.5)
df_freq.show()
+-------------+----------------+
|age_freqItems|gender_freqItems|
+-------------+----------------+
| [16]| [male]|
+-------------+----------------+4，类SQL表操作
类SQL表操作主要包括表查询(select，selectExpr，where)，表连接(join，union，unionAll)，表分组(groupby，agg，pivot)等操作。
df = spark.createDataFrame([
("LiLei",15,"male"),
("HanMeiMei",16,"female"),
("DaChui",17,"male"),
("RuHua",16,None)]).toDF("name","age","gender")
df.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
| RuHua| 16| null|
+---------+---+------+
#表查询select
dftest = df.select("name").limit(2)
dftest.show()
+---------+
| name|
+---------+
| LiLei|
|HanMeiMei|
+---------+
dftest = df.select("name",df["age"] + 1)
dftest.show()
+---------+---------+
| name|(age + 1)|
+---------+---------+
| LiLei| 16|
|HanMeiMei| 17|
| DaChui| 18|
| RuHua| 17|
+---------+---------+
#表查询select
dftest = df.select("name",-df["age"]+2020).toDF("name","birth_year")
dftest.show()
+---------+----------+
| name|birth_year|
+---------+----------+
| LiLei| 2005|
|HanMeiMei| 2004|
| DaChui| 2003|
| RuHua| 2004|
+---------+----------+
#表查询selectExpr,可以使用UDF函数，指定别名等
import datetime
spark.udf.register("getBirthYear",lambda age:datetime.datetime.now().year-age)
dftest = df.selectExpr("name", "getBirthYear(age) as birth_year" , "UPPER(gender) as gender" )
dftest.show()
+---------+----------+------+
| name|birth_year|gender|
+---------+----------+------+
| LiLei| 2005| MALE|
|HanMeiMei| 2004|FEMALE|
| DaChui| 2003| MALE|
| RuHua| 2004| null|
+---------+----------+------+
#表查询where, 指定SQL中的where字句表达式
dftest = df.where("gender='male' and age>15")
dftest.show()
+------+---+------+
| name|age|gender|
+------+---+------+
|DaChui| 17| male|
+------+---+------+
#表查询filter
dftest = df.filter(df["age"]>16)
dftest.show()
+------+---+------+
| name|age|gender|
+------+---+------+
|DaChui| 17| male|
+------+---+------+
#表查询filter
dftest = df.filter("gender ='male'")
dftest.show()
+------+---+------+
| name|age|gender|
+------+---+------+
| LiLei| 15| male|
|DaChui| 17| male|
+------+---+------+
#表连接join
dfscore = spark.createDataFrame([("LiLei","male",88),("HanMeiMei","female",90),("DaChui","male",50)]) \
.toDF("name","gender","score")
dfscore.show()
+---------+------+-----+
| name|gender|score|
+---------+------+-----+
| LiLei| male| 88|
|HanMeiMei|female| 90|
| DaChui| male| 50|
+---------+------+-----+
#表连接join,根据单个字段
dfjoin = df.join(dfscore.select("name","score"),"name")
dfjoin.show()
+---------+---+------+-----+
| name|age|gender|score|
+---------+---+------+-----+
| LiLei| 15| male| 88|
|HanMeiMei| 16|female| 90|
| DaChui| 17| male| 50|
+---------+---+------+-----+
#表连接join,根据多个字段
dfjoin = df.join(dfscore,["name","gender"])
dfjoin.show()
+---------+------+---+-----+
| name|gender|age|score|
+---------+------+---+-----+
|HanMeiMei|female| 16| 90|
| DaChui| male| 17| 50|
| LiLei| male| 15| 88|
+---------+------+---+-----+
#表连接join,根据多个字段
#可以指定连接方式为"inner","left","right","outer","semi","full","leftanti","anti"等多种方式
dfjoin = df.join(dfscore,["name","gender"],"right")
dfjoin.show()
+---------+------+---+-----+
| name|gender|age|score|
+---------+------+---+-----+
|HanMeiMei|female| 16| 90|
| DaChui| male| 17| 50|
| LiLei| male| 15| 88|
+---------+------+---+-----+
dfjoin = df.join(dfscore,["name","gender"],"outer")
dfjoin.show()
+---------+------+---+-----+
| name|gender|age|score|
+---------+------+---+-----+
|HanMeiMei|female| 16| 90|
| DaChui| male| 17| 50|
| LiLei| male| 15| 88|
| RuHua| null| 16| null|
+---------+------+---+-----+
#表连接，灵活指定连接关系
dfmark = dfscore.withColumnRenamed("gender","sex")
dfmark.show()
+---------+------+-----+
| name| sex|score|
+---------+------+-----+
| LiLei| male| 88|
|HanMeiMei|female| 90|
| DaChui| male| 50|
+---------+------+-----+
dfjoin = df.join(dfmark,(df["name"] == dfmark["name"]) & (df["gender"]==dfmark["sex"]),
"inner")
dfjoin.show()
+---------+---+------+---------+------+-----+
| name|age|gender| name| sex|score|
+---------+---+------+---------+------+-----+
|HanMeiMei| 16|female|HanMeiMei|female| 90|
| DaChui| 17| male| DaChui| male| 50|
| LiLei| 15| male| LiLei| male| 88|
+---------+---+------+---------+------+-----+
#表合并union
dfstudent = spark.createDataFrame([("Jim",18,"male"),("Lily",16,"female")]).toDF("name","age","gender")
dfstudent.show()
+----+---+------+
|name|age|gender|
+----+---+------+
| Jim| 18| male|
|Lily| 16|female|
+----+---+------+
dfunion = df.union(dfstudent)
dfunion.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 15| male|
|HanMeiMei| 16|female|
| DaChui| 17| male|
| RuHua| 16| null|
| Jim| 18| male|
| Lily| 16|female|
+---------+---+------+
#表分组 groupBy
from pyspark.sql import functions as F
dfgroup = df.groupBy("gender").max("age")
dfgroup.show()
+------+--------+
|gender|max(age)|
+------+--------+
| null| 16|
|female| 16|
| male| 17|
+------+--------+
#表分组后聚合，groupBy,agg
dfagg = df.groupBy("gender").agg(F.mean("age").alias("mean_age"),
F.collect_list("name").alias("names"))
dfagg.show()
+------+--------+---------------+
|gender|mean_age| names|
+------+--------+---------------+
| null| 16.0| [RuHua]|
|female| 16.0| [HanMeiMei]|
| male| 16.0|[LiLei, DaChui]|
+------+--------+---------------+
#表分组聚合，groupBy,agg
dfagg = df.groupBy("gender").agg(F.expr("avg(age)"),F.expr("collect_list(name)"))
dfagg.show()
+------+--------+------------------+
|gender|avg(age)|collect_list(name)|
+------+--------+------------------+
| null| 16.0| [RuHua]|
|female| 16.0| [HanMeiMei]|
| male| 16.0| [LiLei, DaChui]|
+------+--------+------------------+
#表分组聚合，groupBy,agg
df.groupBy("gender","age").agg(F.collect_list(col("name"))).show()
+------+---+------------------+
|gender|age|collect_list(name)|
+------+---+------------------+
| male| 15| [LiLei]|
| male| 17| [DaChui]|
|female| 16| [HanMeiMei]|
| null| 16| [RuHua]|
+------+---+------------------+
#表分组后透视，groupBy,pivot
dfstudent = spark.createDataFrame([("LiLei",18,"male",1),("HanMeiMei",16,"female",1),
("Jim",17,"male",2),("DaChui",20,"male",2)]).toDF("name","age","gender","class")
dfstudent.show()
dfstudent.groupBy("class").pivot("gender").max("age").show()
+---------+---+------+-----+
| name|age|gender|class|
+---------+---+------+-----+
| LiLei| 18| male| 1|
|HanMeiMei| 16|female| 1|
| Jim| 17| male| 2|
| DaChui| 20| male| 2|
+---------+---+------+-----+
+-----+------+----+
|class|female|male|
+-----+------+----+
| 1| 16| 18|
| 2| null| 20|
+-----+------+----+
#窗口函数
df = spark.createDataFrame([("LiLei",78,"class1"),("HanMeiMei",87,"class1"),
("DaChui",65,"class2"),("RuHua",55,"class2")]) \
.toDF("name","score","class")
df.show()
dforder = df.selectExpr("name","score","class",
"row_number() over (partition by class order by score desc) as order")
dforder.show()
+---------+-----+------+
| name|score| class|
+---------+-----+------+
| LiLei| 78|class1|
|HanMeiMei| 87|class1|
| DaChui| 65|class2|
| RuHua| 55|class2|
+---------+-----+------+
+---------+-----+------+-----+
| name|score| class|order|
+---------+-----+------+-----+
| DaChui| 65|class2| 1|
| RuHua| 55|class2| 2|
|HanMeiMei| 87|class1| 1|
| LiLei| 78|class1| 2|
+---------+-----+------+-----+六，DataFrame的SQL交互
将DataFrame注册为临时表视图或者全局表视图后，可以使用sql语句对DataFrame进行交互。
不仅如此，还可以通过SparkSQL对Hive表直接进行增删改查等操作。
1，注册视图后进行SQL交互
#注册为临时表视图, 其生命周期和SparkSession相关联
df = spark.createDataFrame([("LiLei",18,"male"),("HanMeiMei",17,"female"),("Jim",16,"male")],
("name","age","gender"))
df.show()
df.createOrReplaceTempView("student")
dfmale = spark.sql("select * from student where gender='male'")
dfmale.show()
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 18| male|
|HanMeiMei| 17|female|
| Jim| 16| male|
+---------+---+------+
+-----+---+------+
| name|age|gender|
+-----+---+------+
|LiLei| 18| male|
| Jim| 16| male|
+-----+---+------+
#注册为全局临时表视图,其生命周期和整个Spark应用程序关联
df.createOrReplaceGlobalTempView("student")
query = """
select t.gender
, collect_list(t.name) as names
from global_temp.student t
group by t.gender
""".strip("\n")
spark.sql(query).show()
#可以在新的Session中访问
spark.newSession().sql("select * from global_temp.student").show()
+------+------------+
|gender| names|
+------+------------+
|female| [HanMeiMei]|
| male|[LiLei, Jim]|
+------+------------+
+---------+---+------+
| name|age|gender|
+---------+---+------+
| LiLei| 18| male|
|HanMeiMei| 17|female|
| Jim| 16| male|
+---------+---+------+
2，对Hive表进行增删改查操作
#删除hive表
query = "DROP TABLE IF EXISTS students"
spark.sql(query)
#建立hive分区表
#(注：不可以使用中文字段作为分区字段)
query = """CREATE TABLE IF NOT EXISTS `students`
(`name` STRING COMMENT '姓名',
`age` INT COMMENT '年龄'
)
PARTITIONED BY ( `class` STRING COMMENT '班级', `gender` STRING COMMENT '性别')
""".replace("\n"," ")
spark.sql(query)
##动态写入数据到hive分区表
spark.conf.set("hive.exec.dynamic.partition.mode", "nonstrict") #注意此处有一个设置操作
dfstudents = spark.createDataFrame([("LiLei",18,"class1","male"),
("HanMeimei",17,"class2","female"),
("DaChui",19,"class2","male"),
("Lily",17,"class1","female")]).toDF("name","age","class","gender")
dfstudents.show()
#动态写入分区
dfstudents.write.mode("overwrite").format("hive")\
.partitionBy("class","gender").saveAsTable("students")
#写入到静态分区
dfstudents = spark.createDataFrame([("Jim",18,"class3","male"),
("Tom",19,"class3","male")]).toDF("name","age","class","gender")
dfstudents.createOrReplaceTempView("dfclass3")
#INSERT INTO 尾部追加, INSERT OVERWRITE TABLE 覆盖分区
query = """
INSERT OVERWRITE TABLE `students`
PARTITION(class='class3',gender='male')
SELECT name,age from dfclass3
""".replace("\n"," ")
spark.sql(query)
#写入到混合分区
dfstudents = spark.createDataFrame([("David",18,"class4","male"),
("Amy",17,"class4","female"),
("Jerry",19,"class4","male"),
("Ann",17,"class4","female")]).toDF("name","age","class","gender")
dfstudents.createOrReplaceTempView("dfclass4")
query = """
INSERT OVERWRITE TABLE `students`
PARTITION(class='class4',gender)
SELECT name,age,gender from dfclass4
""".replace("\n"," ")
spark.sql(query)
#读取全部数据
dfdata = spark.sql("select * from students")
dfdata.show()
+---------+---+------+------+
| name|age| class|gender|
+---------+---+------+------+
| Ann| 17|class4|female|
| Amy| 17|class4|female|
|HanMeimei| 17|class2|female|
| DaChui| 19|class2| male|
| LiLei| 18|class1| male|
| Lily| 17|class1|female|
| Jerry| 19|class4| male|
| David| 18|class4| male|
| Jim| 18|class3| male|
| Tom| 19|class3| male|
+---------+---+------+------+
- EOF -
看完本文有收获？请转发分享给更多人
关注「大数据与机器学习文摘」，成为Top 1%
点赞和在看就是最大的支持❤️