我们在使用Spark的时候主要是用来快速处理大批量的数据，那么实际开发和生产中会有哪些数据来源呢，我归类总结有：

• text
• csv
• json
• parquet
• jdbc
• hive
• kafka
• elasticsearch

接下来所有的测试是基于spark local模式，因为local模式便于测试不依赖spark集群环境。有一点要注意将代码运行在spark集群上时要将.master("local[*]")这行去掉，同时需要修改相应的路径名才能访问本地机器文件，以/tmp/people.txt文件为例：

local模式：/tmp/people.txt

集群模式：file:///tmp/people.txt 相当于local模式/tmp/people.txt

                    hdfs://master:8020/tmp/people.txt 分布式系统文件

在学习各种数据来源前先了解一种最基本的数据源，那就是数据集，也就是我们根据自身开发需求制造出来的数据，常常用在开发和测试一些简单功能上面。

开始编写代码制造数据集并形成dataframe显示出来

from pyspark.sql import SparkSession

if __name__ == '__main__':
    spark = SparkSession\
        .builder\
        .appName("loadDatas")\
        .master("local[*]")\
        .enableHiveSupport()\
        .getOrCreate()

    datas = [('Jack', 27), ('Rose', 24), ('Andy', 32)]
    df = spark.createDataFrame(datas, ['name', 'age'])
    df.show()
    # +----+---+
    # |name|age|
    # +----+---+
    # |Jack| 27|
    # |Rose| 24|
    # |Andy| 32|
    # +----+---+
    spark.stop()

text

数据源people.txt内容是

Jack 27
Rose 24
Andy 32

编写代码加载people.txt并通过sql显示出来

from pyspark.sql import SparkSession
from pyspark.sql import Row

if __name__ == '__main__':
    spark = SparkSession\
        .builder\
        .appName("loadTextData")\
        .master("local[*]")\
        .getOrCreate()
    lines = spark.sparkContext.textFile("/home/llh/data/people.txt")
    parts = lines.map(lambda line: line.split(" "))
    people = parts.map(lambda p: Row(name=p[0], age=p[1]))
    peopledf = spark.createDataFrame(people)
    peopledf.show()
    # +---+----+
    # |age|name|
    # +---+----+
    # | 27|Jack|
    # | 24|Rose|
    # | 32|Andy|
    # +---+----+
    peopledf.createOrReplaceTempView("people")
    namedf = spark.sql("select name from people where age < 30")
    namedf.show()
    # +----+
    # |name|
    # +----+
    # |Jack|
    # |Rose|
    # +----+
    spark.stop()

csv

数据源people.csv内容是

Jack,27
Rose,24
Andy,32

编写代码加载csv数据并显示出来

from pyspark.sql import SparkSession
from pyspark.sql.types import *
import pandas as pd

if __name__ == '__main__':
    spark = SparkSession\
        .builder\
        .appName("loadCsvData")\
        .master("local[*]")\
        .getOrCreate()
    # 方式一: 与Text生成的表头的另外一种形式
    schema = StructType([
        StructField("name", StringType(), True),
        StructField("age", IntegerType(), True)
    ])
    peopledf = spark.read.csv("/home/llh/data/people.csv", schema=schema)
    peopledf.show()
    # +----+---+
    # |name|age|
    # +----+---+
    # |Jack| 27|
    # |Rose| 24|
    # |Andy| 32|
    # +----+---+
    # 方式二: 该方式并未使用Spark
    data = pd.read_csv("/home/llh/data/people.csv", names=['name','age'])
    print(data.head())
    #    name  age
    # 0  Jack   27
    # 1  Rose   24
    # 2  Andy   32
    spark.stop()

json

数据源people.json内容是：

{"name":"Jack", "age":27}
{"name":"Rose", "age":24}
{"name":"Andy"}

编写代码加载json数据并通过接口显示

from pyspark.sql import SparkSession

if __name__ == '__main__':
    spark = SparkSession\
        .builder\
        .appName("loadJsonData")\
        .master("local[*]")\
        .getOrCreate()
    peopledf = spark.read.json("/home/llh/data/people.json")
    peopledf.show()
    # +----+----+
    # | age|name|
    # +----+----+
    # | 27 |Jack|
    # | 24 |Rose|
    # |null|Andy|
    # +----+----+
    peopledf.printSchema()
    # root
    # | -- age: long(nullable=true)
    # | -- name: string(nullable=true)
    peopledf.select('name').show()
    # +----+
    # |name|
    # +----+
    # |Jack|
    # |Rose|
    # |Andy|
    # +----+
    peopledf.select(peopledf['name'],peopledf['age']+1).show()
    # +----+---------+
    # |name|(age + 1)|
    # +----+---------+
    # |Jack|       28|
    # |Rose|       25|
    # |Andy|     null|
    # +----+---------+
    peopledf.filter(peopledf['age'] > 25).show()
    # +---+----+
    # |age|name|
    # +---+----+
    # | 27|Jack|
    # +---+----+
    peopledf.groupBy("age").count().show()
    # +----+-----+
    # | age|count|
    # +----+-----+
    # |null|    1|
    # |  27|    1|
    # |  24|    1|
    # +----+-----+
    spark.stop()

parquet

这种格式数据一般存放在hdfs上，用一般编辑器打开会显示一堆乱码

编写代码加载parquet数据并显示出来

from pyspark.sql import SparkSession

if __name__ == '__main__':
    spark = SparkSession\
        .builder\
        .appName("loadParquetData")\
        .master("local[*]")\
        .getOrCreate()
    peopledf = spark.read.parquet("/home/llh/data/people.parquet")
    peopledf.createOrReplaceTempView("people")
    namedf = spark.sql("select name from people where age < 30")
    namedf.show()
    # +----+
    # |name|
    # +----+
    # |Jack|
    # |Rose|
    # +----+
    spark.stop()
    

jdbc

jdbc可以包含mysql、oracle、tidb等，我们这里以mysql为例，数据库是test，表为people

编写代码加载mysql数据库并显示

from pyspark.sql import SparkSession

if __name__ == '__main__':
    spark = SparkSession\
        .builder\
        .appName("loadJdbcData")\
        .master("local[*]")\
        .getOrCreate()
    peopledf = spark.read\
        .format("jdbc")\
        .option("url", "jdbc:mysql://localhost:3306/test")\
        .option("driver", "com.mysql.jdbc.Driver")\
        .option("dbtable", "(select * from people) tmp")\
        .option("user", "root")\
        .option("password", "1")\
        .load()
    peopledf.show()
    # +----+---+
    # |name|age|
    # +----+---+
    # |Jack| 27|
    # |Rose| 24|
    # |Andy| 32|
    # +----+---+
    spark.stop()

运行时可以会报找不到mysql驱动：java.lang.ClassNotFoundException: com.mysql.jdbc.Driver，解决办法是mysql驱动下载一个驱动放到pyspark安装目录jars下，默认在/usr/local/lib/python3.7/site_package/pyspark/jars/

hive

hive数据存放文件分隔符是一种特殊符号"^A"，而且一般的spark配置了hive数据库信息，所以可以直接读取hive数据库

编写代码加载people.hive到people表中并显示出来

from pyspark.sql import SparkSession

if __name__ == '__main__':
    spark = SparkSession\
        .builder\
        .appName("loadHiveData")\
        .master("local[*]")\
        .enableHiveSupport()\
        .getOrCreate()
    spark.sql("create table if not exists people (name string, age int) using hive")
    spark.sql("load data local inpath '/home/llh/data/people.hive' into table people")

    spark.sql("select * from people").show()
    # +----+---+
    # |name|age|
    # +----+---+
    # |Jack| 27|
    # |Rose| 24|
    # |Andy| 32|
    # +----+---+
    spark.stop()

kafka

kafka与spark结合常用于实时项目，也就是spark streaming后续会单独写

elasticsearch

es与mysql等数据库类似

编写代码加载并显示出来

from pyspark.sql import SparkSession

if __name__ == '__main__':
    spark = SparkSession\
        .builder\
        .appName("loadEsData")\
        .master("local[*]")\
        .enableHiveSupport()\
        .getOrCreate()
    peopledf = spark.read\
        .format("org.elasticsearch.spark.sql")\
        .option("es.nodes", "localhost")\
        .option("es.port", 9200)\
        .option("es.resource", "people/data")\
        .load()
    peopledf.registerTempTable("people")
    spark.sql("select * from people").show()
    # +----+---+
    # |name|age|
    # +----+---+
    # |Jack| 27|
    # |Rose| 24|
    # |Andy| 32|
    # +----+---+
    spark.stop()

以上是比较常用的数据来源，当然还有一些比如hbase、phoenix等等...掌握上面的几种再举一反三问题不大。

Spark学习目录：

• Spark学习实例1(Python)：单词统计 Word Count
• Spark学习实例2(Python)：加载数据源Load Data Source
• Spark学习实例3(Python)：保存数据Save Data
• Spark学习实例4(Python)：RDD转换 Transformations
• Spark学习实例5(Python)：RDD执行 Actions
• Spark学习实例6(Python)：共享变量Shared Variables
• Spark学习实例7(Python)：RDD、DataFrame、DataSet相互转换
• Spark学习实例8(Python)：输入源实时处理 Input Sources Streaming
• Spark学习实例9(Python)：窗口操作 Window Operations