spark集群1.6.2升级为2.0.2

参考文档

http://dblab.xmu.edu.cn/blog/1187-2/ Spark 2.0分布式集群环境搭建

http://blog.csdn.net/andy572633/article/details/7211546 linux下杀死进程（kill）的N种方法

更改原有spark目录名

可提前stop-all

将原有的spark改为spark1.6.2

sudo mv /usr/local/spark  /usr/local/saprk1.6.2

Master安装spark2.0.2

下载安装包 spark-2.0.2-bin-without-hadoop.tgz

sudo tar -zxf ~/下载/spark-2.0.2-bin-without-hadoop.tgz -C /usr/local/
cd /usr/local
sudo mv ./spark-2.0.2-bin-without-hadoop/ ./spark
sudo chown -R hadoop ./spark

vim ~/.bashrc

在.bashrc添加如下配置：

export SPARK_HOME=/usr/local/spark
export PATH=$PATH:$SPARK_HOME/bin:$SPARK_HOME/sbin

Master spark配置

slaves

将 slaves.template 拷贝到 slaves

cd /usr/local/spark/
cp ./conf/slaves.template ./conf/slaves

添加slaves

spark-env.sh

将 spark-env.sh.template 拷贝到 spark-env.sh

cp ./conf/spark-env.sh.template ./conf/spark-env.sh

添加内容如下：

export SPARK_DIST_CLASSPATH=$(/usr/local/hadoop/bin/hadoop classpath)
export HADOOP_CONF_DIR=/usr/local/hadoop/etc/hadoop
export SPARK_MASTER_IP=xxx (具体ip)

分发到slaves

将spark文件夹发送到各个node上：

cd /usr/local/
tar -zcf ~/spark.master.tar.gz ./spark
cd ~
scp ./spark.master.tar.gz n01:/home/hadoop
scp ./spark.master.tar.gz n02:/home/hadoop
scp ./spark.master.tar.gz n03:/home/hadoop
scp ./spark.master.tar.gz n04:/home/hadoop
scp ./spark.master.tar.gz n05:/home/hadoop
scp ./spark.master.tar.gz n06:/home/hadoop
scp ./spark.master.tar.gz n07:/home/hadoop
scp ./spark.master.tar.gz n08:/home/hadoop
scp ./spark.master.tar.gz n09:/home/hadoop
scp ./spark.master.tar.gz n10:/home/hadoop

在n01…n10上执行：

sudo rm -rf /usr/local/spark/
sudo tar -zxf ~/spark.master.tar.gz -C /usr/local
sudo chown -R hadoop /usr/local/spark

启动spark集群

cd /usr/local/spark/
sbin/start-master.sh

打开http://m01:8080,如下



左上角可看到已升级为spark2.0.2

可能出现的问题

如果jps出现多个Master进程(可通过web ui 8080 8081 等端口反馈)

1.先停止当前spark-all

2.jps找到多余的Master PID

3.然后kill终结此Master进程

ps -ef | grep master
kill -s 9 master_pid

其中-s 9 制定了传递给进程的信号是９，即强制、尽快终止进程。

世界清静.