Alex 的 Hadoop 菜鸟教程: 第17课 更快速的MapReduce - Spark

时间:2022-09-24 21:32:39

原文地址: http://blog.csdn.net/nsrainbow/article/details/43735737  最新课程请关注原作者博客,获得更好的显示体验


声明

  • 本文基于Centos6.x + CDH 5.x

Spark是什么

Spark是Apache的*项目。项目背景是 Hadoop 的 MapReduce 太挫太慢了,于是有人就做了Spark,目前Spark声称在内存中比Hadoop快100倍,在磁盘上比Hadoop快10倍。

安装Spark

spark有5个组件
  • spark-core: spark核心包
  • spark-worker: spark-worker用的脚本
  • spark-master: spark-master用的脚本
  • spark-python: Spark的Python客户端
  • spark-history-server: 任务历史服务
开始安装Spark

安装组件包

我挑选host1作为master 和 worker,所以在host1上安装以下包
sudo yum install spark-core spark-master spark-worker spark-python
host2 作为 history-server 和 worker
sudo yum install spark-core spark-worker spark-history-server spark-python

配置Spark

Spark支持两种模式
  • 独立模式:  在独立模式, Spark使用一个 Master 服务来运行任务。
  • YARN模式: 在YARN模式, YARN ResourceManager 代替了Spark Master。Job还是由NodeManager运行。YARN 模式搭建会比较复杂,但是它支持安全机制,并且跟YARN集群的配合更好。
本教程中使用独立模式

编辑每一台安装了Spark机器上的 /etc/spark/conf/spark-env.sh 修改master所在机器的机器名,在这个教程中就是host1
###
### === IMPORTANT ===
### Change the following to specify a real cluster‘s Master host
###
export STANDALONE_SPARK_MASTER_HOST=‘host1‘
注意: 包裹host1的符号也要换成单引号

创建Spark History Server需要的hdfs文件夹 /user/spark/applicationHistory/
$ sudo -u hdfs hadoop fs -mkdir /user/spark 
$ sudo -u hdfs hadoop fs -mkdir /user/spark/applicationHistory
$ sudo -u hdfs hadoop fs -chown -R spark:spark /user/spark
$ sudo -u hdfs hadoop fs -chmod 1777 /user/spark/applicationHistory
在Spark客户端,在本例中就是host2,创建一份新的配置文件
cp /etc/spark/conf/spark-defaults.conf.template /etc/spark/conf/spark-defaults.conf
把下面这两行增加到/etc/spark/conf/spark-defaults.conf 里面去
spark.eventLog.dir=/user/spark/applicationHistory
spark.eventLog.enabled=true
在所有的机器上复制hdfs-site.xml到 /etc/spark/conf 下
cp /etc/hadoop/conf/hdfs-site.xml /etc/spark/conf/

启动Spark

在host1上启动master服务

sudo service spark-master start

在其他节点上启动woker服务,本教程中就是 host1 和 host2

sudo service spark-worker start

在其中一个节点上启动history服务,本教程中用host2启动history
sudo service spark-history-server start

启动顺序

  1. master
  2. worker
  3. history-server
打开浏览器访问 http://host1:18080 可以看到Spark的管理界面
Alex 的 Hadoop 菜鸟教程: 第17课 更快速的MapReduce - Spark

使用Spark

使用 spark-shell 命令进入spark shell
[root@host1 impala]# spark-shell
2015-02-10 09:02:07,059 INFO [main] spark.SecurityManager (Logging.scala:logInfo(59)) - Changing view acls to: root
2015-02-10 09:02:07,069 INFO [main] spark.SecurityManager (Logging.scala:logInfo(59)) - Changing modify acls to: root
2015-02-10 09:02:07,070 INFO [main] spark.SecurityManager (Logging.scala:logInfo(59)) - SecurityManager: authentication disabled; ui acls disabled; users with view permissions: Set(root); users with modify permissions: Set(root)
2015-02-10 09:02:07,072 INFO [main] spark.HttpServer (Logging.scala:logInfo(59)) - Starting HTTP Server
2015-02-10 09:02:07,217 INFO [main] server.Server (Server.java:doStart(272)) - jetty-8.y.z-SNAPSHOT
2015-02-10 09:02:07,350 INFO [main] server.AbstractConnector (AbstractConnector.java:doStart(338)) - Started SocketConnector@0.0.0.0:59058
2015-02-10 09:02:07,352 INFO [main] util.Utils (Logging.scala:logInfo(59)) - Successfully started service ‘HTTP class server‘ on port 59058.
Welcome to
____ __
/ __/__ ___ _____/ /__
_\ \/ _ \/ _ `/ __/ ‘_/
/___/ .__/\_,_/_/ /_/\_\ version 1.2.0
/_/

Using Scala version 2.10.4 (Java HotSpot(TM) 64-Bit Server VM, Java 1.7.0_25)
...
2015-02-10 09:02:21,572 INFO [main] storage.BlockManagerMaster (Logging.scala:logInfo(59)) - Registered BlockManager
2015-02-10 09:02:22,472 INFO [main] scheduler.EventLoggingListener (Logging.scala:logInfo(59)) - Logging events to file:/user/spark/applicationHistory/local-1423530140986
2015-02-10 09:02:22,672 INFO [main] repl.SparkILoop (Logging.scala:logInfo(59)) - Created spark context..
Spark context available as sc.

scala>
我们来开始玩一下Spark。还是做之前用YARN做的wordcount任务,看看Spark如何完成这项任务。

STEP1

创建测试文本

$ echo "Hello World Bye World" > file0
$ echo "Hello Hadoop Goodbye Hadoop" > file1
$ hdfs dfs -mkdir -p /user/spark/wordcount/input
$ hdfs dfs -put file* /user/spark/wordcount/input

STEP2

进入 spark-shell 运行 wordcount任务脚本
val file = sc.textFile("hdfs://mycluster/user/spark/wordcount/input")
val counts = file.flatMap(line => line.split(" ")).map(word => (word, 1)).reduceByKey(_ + _)
counts.saveAsTextFile("hdfs://mycluster/user/spark/wordcount/output")
这回不用写java代码了,简单了好多。这里用的是Scala语言。
Spark支持 Java, Scale, Python 三种语言,但是对Scala的支持最全,建议开始用java来写,后期还是熟悉下Scala比较好。

STEP3

我们去看下结果,我用Pig看下结果
grunt> ls
hdfs://mycluster/user/spark/wordcount/input <dir>
hdfs://mycluster/user/spark/wordcount/output <dir>
grunt> cd output
grunt> ls
hdfs://mycluster/user/spark/wordcount/output/_SUCCESS<r 2> 0
hdfs://mycluster/user/spark/wordcount/output/part-00000<r 2> 8
hdfs://mycluster/user/spark/wordcount/output/part-00001<r 2> 10
hdfs://mycluster/user/spark/wordcount/output/part-00002<r 2> 33
grunt> cat part-00000
(Bye,1)
grunt> cat part-00001
(World,2)
grunt> cat part-00002
(Goodbye,1)
(Hello,2)
(Hadoop,2)

更深入的学习请看手册Spark Programming Guide , 另外这个手册写的真不错。

参考资料