和分布式文件系统和NoSQL数据库相比而言，spark集群的安装配置还算是比较简单的：

很多教程提到要安装java和scala，但我发现spark最新版本是包含scala的，JRE采用linux内嵌的版本也是可以的！

1. 在主节点（bluejoe0）上安装spark1.1.0：
   wget http://mirror.bit.edu.cn/apache/spark/spark-1.1.0/spark-1.1.0-bin-hadoop2.3.tgz
   tar -zxvf spark-1.1.0-bin-hadoop2.3.tgz
   ln -s spark-1.1.0-bin-hadoop2.3 spark
2. 启动spark-shell：
   cd /usr/local/spark/bin
   ./spark-shell
   可以看到spark已经自带了scala 2.10：
   
3. 输入测试程序：
   scala> val data = Array(1, 2, 3, 4, 5)
   data: Array[Int] = Array(1, 2, 3, 4, 5)

   scala> val distData = sc.parallelize(data)
   distData: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[0] at parallelize at <console>:14

   scala> distData.reduce(_+_)

4. 可以观察4040端口：

   

5. 也可以测试PI的计算：
   ./bin/run-example SparkPi
   14/11/23 16:08:25 INFO SparkContext: Job finished: reduce at SparkPi.scala:35, took 1.008332384 s
   Pi is roughly 3.1403
6. 也可以采用spark-submit来提交任务：
   ./bin/spark-submit --class org.apache.spark.examples.SparkPi --master local[6] /usr/local/spark/lib/spark-examples-1.1.0-hadoop2.3.0.jar 1000
   14/11/23 16:07:30 INFO SparkContext: Job finished: reduce at SparkPi.scala:35, took 46.220537186 s
   Pi is roughly 3.14172056
7. 现在安装几个从节点，scp spark.tgz文件到其它节点，如：bluejoe4,bluejoe5,bluejoe9
8. 注意设置好ssh无密码登录；
9. 修改conf/slaves
   # A Spark Worker will be started on each of the machines listed below.
   bluejoe4
   bluejoe5
   bluejoe9
10. 在bluejoe0上启动spark集群：
    ./sbin/start-all.sh
    此时可以在浏览器上观察到3个从节点的情况：
    
11. 再测试在集群上计算PI的程序：
    ./bin/spark-submit --class org.apache.spark.examples.SparkPi --master spark://bluejoe0:7077 /usr/local/spark/lib/spark-examples-1.1.0-hadoop2.3.0.jar 1000
    14/11/23 16:05:00 INFO SparkContext: Job finished: reduce at SparkPi.scala:35, took 26.322514766 s
    Pi is roughly 3.14159516

    此时观察浏览器的显示：