对于PyCharm,需要作如下设置:
1、安装pyspark,它会自动安装py4j
2、在edit configuration中,add content root,选择spark下载包的python/pyspark/lib下的pyspark.zip和py4j.zip两个包;
代码实例:
from pyspark.sql import Row from pyspark.sql import SparkSession logFile = "file:///Users/peishuaishuai/tmp/sparktest.txt" # Should be some file on your system spark = SparkSession.builder.appName("SimpleApp").getOrCreate() input = spark.read.text(logFile).rdd.map( lambda x: str(x[0]).split("\t") ).filter( lambda x: len(x) == 2 ).map( lambda x: Row(name=x[0], grade=int(x[1])) ) schemaData = spark.createDataFrame(input) schemaData.createOrReplaceTempView("tb") print(schemaData.count()) schemaData.printSchema() datas = spark.sql("select name,sum(grade) from tb group by name").rdd.map( lambda x: "\t".join([x[0], str(x[1])]) ) datas.repartition(3).saveAsTextFile("file:///Users/peishuaishuai/tmp/sparktest_output") spark.stop()
输入数据为:
name1 11 name2 12 name3 13 name4 14 name5 15 name1 16 name2 17 name3 18 name4 19 name5 20 name11 21 name12 22 name1 23 name2 24 name3 25 name4 26 name5 27 name18 28 name19 29 name20 30 name21 31 name1 32 name2 33 name3 34 name4 35 name5 36 name27 37 name28 38 name29 39 name1 40 name2 41 name3 42 name4 43
输出 print结果为:
33 root |-- grade: long (nullable = true) |-- name: string (nullable = true)
文件中内容为:
name3 132 name19 29 name2 127 name12 22 name11 21 name20 30 name28 38 name27 37 name5 98 name29 39 name21 31 name4 137 name1 122 name18 28
pyspark开发起来,有点问题就是当级联过多的时候,类型可能丢失,导致代码没有提示,这点很不爽。
其实对比了python、scala、java,我觉得编写大型的spark代码,用Java是最靠谱的,因为它强类型,代码提示很爽很直观。