环境:CentOS 6.4, Hadoop 1.1.2, JDK 1.7, Spark 0.7.2, Scala 2.9.3
yum search openjdk-devel sudo yum install java-1.7.0-openjdk-devel.x86_64 /usr/sbin/alternatives --config java /usr/sbin/alternatives --config javac sudo vim /etc/profile # add the following lines at the end export JAVA_HOME=/usr/lib/jvm/java-1.7.0-openjdk-1.7.0.19.x86_64 export JRE_HOME=$JAVA_HOME/jre export PATH=$PATH:$JAVA_HOME/bin export CLASSPATH=.:$JAVA_HOME/lib/dt.jar:$JAVA_HOME/lib/tools.jar # save and exit vim # make the bash profile take effect immediately $ source /etc/profile # test $ java -version
Spark 0.7.2 依赖 Scala 2.9.3, 我们必须要安装Scala 2.9.3.
下载 scala-2.9.3.tgz 并 保存到home目录.
1 $ tar -zxf scala-2.9.3.tgz 2 $ sudo mv scala-2.9.3 /usr/lib 3 $ sudo vim /etc/profile 4 # add the following lines at the end 5 export SCALA_HOME=/usr/lib/scala-2.9.3 6 export PATH=$PATH:$SCALA_HOME/bin 7 # save and exit vim 8 #make the bash profile take effect immediately 9 source /etc/profile 10 # test 11 $ scala -version
下载预编译好的Spark, spark-0.7.2-prebuilt-hadoop1.tgz.
如果你想从零开始编译,则下载源码包,但是我不建议你这么做,因为有一个Maven仓库,twitter4j.org, 被墙了,导致编译时需要FQ,非常麻烦。如果你有DIY精神,并能顺利FQ,则可以试试这种方式。
$ tar -zxf spark-0.7.2-prebuilt-hadoop1.tgz
$ vim ~/.bash_profile
# add the following lines at the end
export SPARK_EXAMPLES_JAR=$HOME/spark-0.7.2/examples/target/scala-2.9.3/spark-examples_2.9.3-0.7.2.jar
# save and exit vim
#make the bash profile take effect immediately
$ source /etc/profile
这一步其实最关键,很不幸的是,官方文档和网上的博客,都没有提及这一点。我是偶然看到了这两篇帖子,Running SparkPi, Null pointer exception when running ./run spark.examples.SparkPi local,才补上了这一步,之前死活都无法运行SparkPi。
$ vim ~/.bash_profile
# add the following lines at the end
export SPARK_HOME=$HOME/spark-0.7.2
export PATH=$PATH:$SPARK_HOME/bin
# save and exit vim
#make the bash profile take effect immediately
$ source /etc/profile
$ cd ~/spark-0.7.2
$ ./run spark.examples.SparkPi local
用VMware Workstation 创建三台CentOS 虚拟机,hostname分别设置为 master, slave01, slave02,设置SSH无密码登陆,安装hadoop,然后启动hadoop集群。参考我的这篇博客,在CentOS上安装Hadoop.
在三台机器上都要安装 Scala 2.9.3 , 按照第2节的步骤。JDK在安装Hadoop时已经安装了。
解压
$ tar -zxf spark-0.7.2-prebuilt-hadoop1.tgz
设置SPARK_EXAMPLES_JAR 环境变量
$ vim ~/.bash_profile
# add the following lines at the end
export SPARK_EXAMPLES_JAR=$HOME/spark-0.7.2/examples/target/scala-2.9.3/spark-examples_2.9.3-0.7.2.jar
# save and exit vim
#make the bash profile take effect immediately
$ source /etc/profile
在 in conf/spark-env.sh
中设置SCALA_HOME
$ cd ~/spark-0.7.2/conf
$ mv spark-env.sh.template spark-env.sh
$ vim spark-env.sh
# add the following line
export SCALA_HOME=/usr/lib/scala-2.9.3
# save and exit
在conf/slaves
, 添加Spark worker的hostname, 一行一个。
$ vim slaves
slave01
slave02
# save and exit
(可选)设置 SPARK_HOME环境变量,并将SPARK_HOME/bin加入PATH
$ vim ~/.bash_profile
# add the following lines at the end
export SPARK_HOME=$HOME/spark-0.7.2
export PATH=$PATH:$SPARK_HOME/bin
# save and exit vim
#make the bash profile take effect immediately
$ source /etc/profile
既然master上的这个文件件已经配置好了,把它拷贝到所有的worker。注意,三台机器spark所在目录必须一致,因为master会登陆到worker上执行命令,master认为worker的spark路径与自己一样。
$ cd
$ scp -r spark-0.7.2 dev@slave01:~
$ scp -r spark-0.7.2 dev@slave02:~
按照第5.3节设置SPARK_EXAMPLES_JAR
环境变量,配置文件不用配置了,因为是直接从master复制过来的,已经配置好了。
在master上执行
$ cd ~/spark-0.7.2
$ bin/start-all.sh
检测进程是否启动
$ jps
11055 Jps
2313 SecondaryNameNode
2409 JobTracker
2152 NameNode
4822 Master
浏览master的web UI(默认http://localhost:8080). 这是你应该可以看到所有的word节点,以及他们的CPU个数和内存等信息。 ##5.6 运行SparkPi例子
$ cd ~/spark-0.7.2
$ ./run spark.examples.SparkPi spark://master:7077
(可选)运行自带的例子,SparkLR 和 SparkKMeans.
#Logistic Regression
#./run spark.examples.SparkLR spark://master:7077
#kmeans
$ ./run spark.examples.SparkKMeans spark://master:7077 ./kmeans_data.txt 2 1
$ cd ~/spark-0.7.2
$ hadoop fs -put README.md .
$ MASTER=spark://master:7077 ./spark-shell
scala> val file = sc.textFile("hdfs://master:9000/user/dev/README.md")
scala> val count = file.flatMap(line => line.split(" ")).map(word => (word, 1)).reduceByKey(_+_)
scala> count.collect()
$ cd ~/spark-0.7.2
$ bin/stop-all.sh