Spark Standalone模式应用程序开发


　　在本博客的《Spark快速入门指南(Quick Start Spark)》文章中简单地介绍了如何通过Spark shell来快速地运用API。本文将介绍如何快速地利用Spark提供的API开发Standalone模式的应用程序。Spark支持三种程序语言的开发：Scala (利用SBT进行编译), Java (利用Maven进行编译)以及Python。下面我将分别用Scala、Java和Python开发同样功能的程序：

一、Scala版本：

程序如下：

01package scala
02/**
03 * User: 过往记忆
04 * Date: 14-6-10
05 * Time: 下午11:37
06 * bolg: http://www.iteblog.com
07 * 本文地址：http://www.iteblog.com/archives/1041
08 * 过往记忆博客，专注于hadoop、hive、spark、shark、flume的技术博客，大量的干货
09 * 过往记忆博客微信公共帐号：iteblog_hadoop
10 */
11import org.apache.spark.SparkContext
12import org.apache.spark.SparkConf
13object Test {
14    def main(args: Array[String]) {
15      val logFile = "file:///spark-bin-0.9.1/README.md"
16      val conf = new SparkConf().setAppName("Spark Application in Scala")
17      val sc = new SparkContext(conf)
18      val logData = sc.textFile(logFile, 2).cache()
19      val numAs = logData.filter(line => line.contains("a")).count()
20      val numBs = logData.filter(line => line.contains("b")).count()
21      println("Lines with a: %s, Lines with b: %s".format(numAs, numBs))
22    }
23  }
24}

为了编译这个文件，需要创建一个xxx.sbt文件，这个文件类似于pom.xml文件，这里我们创建一个scala.sbt文件，内容如下：

1name := "Spark application in Scala"
2version := "1.0"
3scalaVersion := "2.10.4"
4libraryDependencies += "org.apache.spark" %% "spark-core" % "1.0.0"
5resolvers += "Akka Repository" at "http://repo.akka.io/releases/"

编译：

1# sbt/sbt package
2[info] Done packaging.
3[success] Total time: 270 s, completed Jun 11, 2014 1:05:54 AM
二、Java版本
01/**
02 * User: 过往记忆
03 * Date: 14-6-10
04 * Time: 下午11:37
05 * bolg: http://www.iteblog.com
06 * 本文地址：http://www.iteblog.com/archives/1041
07 * 过往记忆博客，专注于hadoop、hive、spark、shark、flume的技术博客，大量的干货
08 * 过往记忆博客微信公共帐号：iteblog_hadoop
09 */
10/* SimpleApp.java */
11import org.apache.spark.api.java.*;
12import org.apache.spark.SparkConf;
13import org.apache.spark.api.java.function.Function;
14 
15public class SimpleApp {
16    public static void main(String[] args) {
17        String logFile = "file:///spark-bin-0.9.1/README.md";
18        SparkConf conf =new SparkConf().setAppName("Spark Application in Java");
19        JavaSparkContext sc = new JavaSparkContext(conf);
20        JavaRDD<String> logData = sc.textFile(logFile).cache();
21 
22        long numAs = logData.filter(new Function<String, Boolean>() {
23            public Boolean call(String s) { return s.contains("a"); }
24        }).count();
25 
26        long numBs = logData.filter(new Function<String, Boolean>() {
27            public Boolean call(String s) { return s.contains("b"); }
28        }).count();
29 
30        System.out.println("Lines with a: " + numAs +",lines with b: " + numBs);
31    }
32}

本程序分别统计README.md文件中包含a和b的行数。本项目的pom.xml文件内容如下：

01<?xml version="1.0" encoding="UTF-8"?>
02<project xmlns="http://maven.apache.org/POM/4.0.0"
03         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
04         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0
05 
06http://maven.apache.org/xsd/maven-4.0.0.xsd">
07 
08    <modelVersion>4.0.0</modelVersion>
09 
10    <groupId>spark</groupId>
11    <artifactId>spark</artifactId>
12    <version>1.0</version>
13 
14    <dependencies>
15        <dependency>
16            <groupId>org.apache.spark</groupId>
17            <artifactId>spark-core_2.10</artifactId>
18            <version>1.0.0</version>
19        </dependency>
20    </dependencies>
21</project>

利用Maven来编译这个工程：

1# mvn install
2[INFO] ------------------------------------------------------------------------
3[INFO] BUILD SUCCESS
4[INFO] ------------------------------------------------------------------------
5[INFO] Total time: 5.815s
6[INFO] Finished at: Wed Jun 11 00:01:57 CST 2014
7[INFO] Final Memory: 13M/32M
8[INFO] ------------------------------------------------------------------------
三、Python版本
01#
02# User: 过往记忆
03# Date: 14-6-10
04# Time: 下午11:37
05# bolg: http://www.iteblog.com
06# 本文地址：http://www.iteblog.com/archives/1041
07# 过往记忆博客，专注于hadoop、hive、spark、shark、flume的技术博客，大量的干货
08# 过往记忆博客微信公共帐号：iteblog_hadoop
09#
10from pyspark import SparkContext
11 
12logFile = "file:///spark-bin-0.9.1/README.md"
13sc = SparkContext("local", "Spark Application in Python")
14logData = sc.textFile(logFile).cache()
15 
16numAs = logData.filter(lambda s: 'a' in s).count()
17numBs = logData.filter(lambda s: 'b' in s).count()
18 
19print "Lines with a: %i, lines with b: %i" % (numAs, numBs)
四、测试运行

本程序的程序环境是Spark 1.0.0，单机模式，测试如下：
1、测试Scala版本的程序

1# bin/spark-submit --class "scala.Test"  \
2                   --master local[4]    \
3              target/scala-2.10/simple-project_2.10-1.0.jar
4 
514/06/11 01:07:53 INFO spark.SparkContext: Job finished:
6count at Test.scala:18, took 0.019705 s
7Lines with a: 62, Lines with b: 35

2、测试Java版本的程序

1# bin/spark-submit --class "SimpleApp"  \
2                   --master local[4]    \
3              target/spark-1.0-SNAPSHOT.jar
4 
514/06/11 00:49:14 INFO spark.SparkContext: Job finished:
6count at SimpleApp.java:22, took 0.019374 s
7Lines with a: 62, lines with b: 35

3、测试Python版本的程序

1# bin/spark-submit --master local[4]    \
2                simple.py
3 
4Lines with a: 62, lines with b: 35

本文地址：《Spark Standalone模式应用程序开发》：http://www.iteblog.com/archives/1041，过往记忆，大量关于Hadoop、Spark等个人原创技术博客本博客文章除特别声明，全部都是原创！

尊重原创，转载请注明： 转载自过往记忆（http://www.iteblog.com/）
本文链接地址: 《Spark Standalone模式应用程序开发》（http://www.iteblog.com/archives/1041）
E-mail:wyphao.2007@163.com    

相关内容