前提你得安装有Hadoop 我的版本hadoop2.3-cdh5.1.0

1、下载maven包

2、配置M2_HOME环境变量，配置maven 的bin目录到path路径

3、export MAVEN_OPTS="-Xmx2g -XX:MaxPermSize=512M -XX:ReservedCodeCacheSize=512m"

4、到官方下载spark-1.0.2.gz压缩包、解压

5、进入spark解压包目录

6、执行./make-distribution.sh --hadoop 2.3.0-cdh5.1.0 --with-yarn --tgz

7、漫长的等待

8、完成后会在当前目录下生成spark-1.0.2-bin-2.3.0-cdh5.1.0.tgz

9、复制到安装目录解压

10、配置conf下的配置文件

cp spark-env.sh.template spark-env.sh

vim spark-env.sh

配置参数：对应即可

export JAVA_HOME=/home/hadoop/jdk
export HADOOP_HOME=/home/hadoop/hadoop-2.3.0-cdh5.1.0
export HADOOP_CONF_DIR=/home/hadoop/hadoop-2.3.0-cdh5.1.0/etc/hadoop
export SPARK_YARN_APP_NAME=spark-on-yarn
export SPARK_EXECUTOR_INSTANCES=1
export SPARK_EXECUTOR_CORES=2
export SPARK_EXECUTOR_MEMORY=3500m
export SPARK_DRIVER_MEMORY=3500m
export SPARK_MASTER_IP=master
export SPARK_MASTER_PORT=7077
export SPARK_WORKER_CORES=2
export SPARK_WORKER_MEMORY=3500m
export SPARK_WORKER_INSTANCES=1

11、配置slaves

slave01
slave02
slave03
slave04
slave05

12、分发

拷贝spark安装目录到各个slave节点

13、启动

sbin/start-all.sh

14、运行实例

$SPARK_HOME/bin/spark-submit --class org.apache.spark.examples.SparkPi     --master yarn-client     --num-executors 3     --driver-memory 4g     --executor-memory 2g     --executor-cores 1     /home/hadoop/spark/lib/spark-examples-1.0.2-hadoop2.3.0-cdh5.1.0.jar     100

15、发送实例竟然没成功

在yarn监控界面点击日志出现一堆这些错误

INFO [main] org.apache.hadoop.ipc.Client: Retrying connect to server: 0.0.0.0/0.0.0.0:8030. Already tried 0 time(s).

INFO [main] org.apache.hadoop.ipc.Client: Retrying connect to server: 0.0.0.0/0.0.0.0:8030. Already tried 0 time(s).

INFO [main] org.apache.hadoop.ipc.Client: Retrying connect to server: 0.0.0.0/0.0.0.0:8030. Already tried 0 time(s).

INFO [main] org.apache.hadoop.ipc.Client: Retrying connect to server: 0.0.0.0/0.0.0.0:8030. Already tried 0 time(s).

16、解决问题

将spark目录下lib包的spark核心包拿到本地，发现里面有一个yarn-defaul.xml文件，打开发现

  <!-- Resource Manager Configs -->  <property>    <description>The hostname of the RM.</description>    <name>yarn.resourcemanager.hostname</name>    <value>0.0.0.0</value>  </property> 

可想而知，到本地找resorcemanager,如果运行节点不是在yarn节点的resourcemanager上运行，怎么可能找到呢

17、修改这个配置如下

  <!-- Resource Manager Configs -->  <property>    <description>The hostname of the RM.</description>    <name>yarn.resourcemanager.hostname</name>    <value>master</value>  </property> 

18、打包重新分发spark到各个节点

spark cdh5编译安装[spark-1.0.2 hadoop2.3.0 cdh5.1.0]