Spark在Hadoop集群上的配置(spark-1.1.0-bin-hadoop2.4)

运行Spark服务，需要在每个节点上部署Spark。

可以先从主节点上将配置修改好，然后把Spark直接scp到其他目录。

关键配置

修改conf/spark-env.sh文件：

export JAVA_HOME=/usr/java/latest
export HADOOP_CONF_DIR=/opt/hadoop-2.4.1/etc/hadoop/
export SPARK_MASTER_IP=master

以上是必要的几个配置，详细的配置说明，请参见官网的Document。

修改conf/slaves，slave节点配置，将worker节点的主机名直接添加进去即可。

启动集群

sbin/start-all.sh

jps查看本机java进程，主节点应该有Master进程，worker节点应该有个Worker进程。

WebUI地址：http://master:8080

测试Spark，bin/run-example SparkPi,正常的话，可以看到以下测试结果：

...
14/11/11 22:11:25 INFO scheduler.TaskSetManager: Finished task 0.0 in stage 0.0 (TID 0) in 4052 ms on localhost (1/2)
14/11/11 22:11:25 INFO scheduler.DAGScheduler: Stage 0 (reduce at SparkPi.scala:35) finished in 4.130 s
14/11/11 22:11:25 INFO scheduler.TaskSetManager: Finished task 1.0 in stage 0.0 (TID 1) in 69 ms on localhost (2/2)
14/11/11 22:11:25 INFO scheduler.TaskSchedulerImpl: Removed TaskSet 0.0, whose tasks have all completed, from pool
14/11/11 22:11:25 INFO spark.SparkContext: Job finished: reduce at SparkPi.scala:35, took 4.613856515 s
Pi is roughly 3.1431
14/11/11 22:11:26 INFO handler.ContextHandler: stopped o.e.j.s.ServletContextHandler{/metrics/json,null}
14/11/11 22:11:26 INFO handler.ContextHandler: stopped o.e.j.s.ServletContextHandler{/stages/stage/kill,null}
14/11/11 22:11:26 INFO handler.ContextHandler: stopped o.e.j.s.ServletContextHandler{/,null}
14/11/11 22:11:26 INFO handler.ContextHandler: stopped o.e.j.s.ServletContextHandler{/static,null}
14/11/11 22:11:26 INFO handler.ContextHandler: stopped o.e.j.s.ServletContextHandler{/executors/json,null}
...