Quick Start: https://spark.apache.org/docs/latest/quick-start.html
在Spark 2.0之前,Spark的编程接口为RDD (Resilient Distributed Dataset)。而在2.0之后,RDDs被Dataset替代。Dataset很像RDD,但是有更多优化。RDD仍然支持,不过强烈建议切换到Dataset,以获得更好的性能。
RDD文档: https://spark.apache.org/docs/latest/rdd-programming-guide.html
Dataset文档: https://spark.apache.org/docs/latest/sql-programming-guide.html
一、最简单的Spark Shell交互分析
scala> val textFile = spark.read.textFile("README.md") # 构建一个Dataset textFile: org.apache.spark.sql.Dataset[String] = [value: string] scala> textFile.count() # Dataset的简单计算 res0: Long = 104 scala> val linesWithSpark = textFile.filter(line => line.contain("Spark")) # 由现有Dataset生成新Dataset res1: org.apache.spark.sql.Dataset[String] = [value: string] # 等价于: # res1 = new Dataset() # for line in textFile: # if line.contain("Spark"): # res1.append(line) # linesWithSpark = res1 scala> linesWithSpark.count() res2: Long = 19 # 可以将多个操作串行起来 scala> textFile.filter(line => line.contain("Spark")).count() res3: Long = 19
进一步的Dataset分析:
scala> textFile.map(line => line.split(" ").size).reduce((a,b) => if (a > b) a else b) res12: Int = 16 # 其实map和reduce就是两个普通的算子,不要被MapReduce中一个map配一个reduce、先map后reduce的思想所束缚 # map算子就是对Dataset的元素X计算fun(X),并且将所有f(X)作为新的Dataset返回 # reduce算子其实就是通过两两计算fun(X,Y)=Z,将Dataset中的所有元素归约为1个值 # 也可以引入库进行计算 scala> import java.lang.Math import java.lang.Math scala> textFile.map(line => line.split(" ").size).reduce((a, b) => Math.max(a, b)) res14: Int = 16 # 还可以使用其他算子 scala> val wordCounts = textFile.flatMap(line => line.split(" ")).groupByKey(identity).count() # flatMap算子也是对Dataset的每个元素X执行fun(X)=Y,只不过map的res是 # res.append(Y),如[[Y11, Y12], [Y21, Y22]],结果按元素区分 # 而flatMap是 # res += Y,如[Y11, Y12, Y21, Y22],各元素结果合在一起 # groupByKey算子将Dataset的元素X作为参数传入进行计算f(X),并以f(X)作为key进行分组,返回值为KeyValueGroupedDataset类型 # 形式类似于(key: k; value: X1, X2, ...),不过KeyValueGroupedDataset不是一个Dataset,value列表也不是一个array # 注意:这里的textFile和textFile.flatMap都是Dataset,不是RDD,groupByKey()中可以传func;如果以sc.textFile()方法读文件,得到的是RDD,groupByKey()中间不能传func # identity就是函数 x => x,即返回自身的函数 # KeyValueGroupedDataset的count()方法返回(key, len(value))列表,结果是Dataset类型 scala> wordCounts.collect() res37: Array[(String, Long)] = Array((online,1), (graphs,1), ... # collect操作:将分布式存储在集群上的RDD/Dataset中的所有数据都获取到driver端
数据的cache:
scala> linesWithSpark.cache() # in-memory cache,让数据在分布式内存中缓存 res38: linesWithSpark.type = [value: string] scala> linesWithSpark.count() res41: Long = 19
二、最简单的独立Spark任务(spark-submit提交)
需提前安装sbt,sbt是scala的编译工具(Scala Build Tool),类似java的maven。
brew install sbt
1)编写SimpleApp.scala
import org.apache.spark.sql.SparkSession object SimpleApp { def main(args: Array[String]) { val logFile = "/Users/dxm/work-space/spark-2.4.5-bin-hadoop2.7/README.md" val spark = SparkSession.builder.appName("Simple Application").getOrCreate() val logData = spark.read.textFile(logFile).cache() val numAs = logData.filter(line => line.contains("a")).count() # 包含字母a的行数 val numBs = logData.filter(line => line.contains("b")).count() # 包含字母b的行数 println(s"Lines with a: $numAs, Lines with b: $numBs") spark.stop() } }
2)编写sbt依赖文件build.sbt
name := "Simple Application" version := "1.0" scalaVersion := "2.12.10" libraryDependencies += "org.apache.spark" %% "spark-sql" % "2.4.5"
其中,"org.apache.spark" %% "spark-sql" % "2.4.5"这类库名可以在网上查到,例如https://mvnrepository.com/artifact/org.apache.spark/spark-sql_2.10/1.0.0
3)使用sbt打包
目录格式如下,如果SimpleApp.scala和build.sbt放在一个目录下会编不出来
$ find . . ./build.sbt ./src ./src/main ./src/main/scala ./src/main/scala/SimpleApp.scala
sbt目录格式要求见官方文档 https://www.scala-sbt.org/1.x/docs/Directories.html
src/ main/ resources/scala/ scala-2.12/ java/ test/ resources scala/ scala-2.12/ java/
使用sbt打包
# 打包 $ sbt package ... [success] Total time: 97 s (01:37), completed 2020-6-10 10:28:24 # jar包位于 target/scala-2.12/simple-application_2.12-1.0.jar
4)提交并执行Spark任务
$ bin/spark-submit --class "SimpleApp" --master spark://xxx:7077 ../scala-tests/SimpleApp/target/scala-2.12/simple-application_2.12-1.0.jar # 报错:Caused by: java.lang.ClassNotFoundException: scala.runtime.LambdaDeserialize # 参考:https://stackoverflow.com/questions/47172122/classnotfoundexception-scala-runtime-lambdadeserialize-when-spark-submit # 这是spark版本和scala版本不匹配导致的
查询spark所使用的scala的版本
$ bin/spark-shell --master spark://xxx:7077 scala> util.Properties.versionString res0: String = version 2.11.12
修改build.sbt:
scalaVersion := "2.11.12"
从下载页也可验证,下载的spark 2.4.5使用的是scala 2.11
重新sbt package,产出位置变更为target/scala-2.11/simple-application_2.11-1.0.jar
再次spark-submit,成功
$ bin/spark-submit --class "SimpleApp" --master spark://xxx:7077 ../scala-tests/SimpleApp/target/scala-2.11/simple-application_2.11-1.0.jar Lines with a: 61, Lines with b: 30