Spark+zookeeper搭建高可用集群

Master结点存在单点故障，所以要借助zookeeper，至少启动两台Master结点来实现高可用，配置方案比较简单

先停止所有Spark服务，然后安装zookeeper，并启动zookeeper

集群规划：

一、先安装Spark集群（Spark2.2.0安装教程）
二、安装zookeeper:(zookeeper安装教程)

三、高可用配置

在spark-env.sh上删掉SPARK_MASTER_IP配置项，并添加以下内容：

export SPARK_DAEMON_JAVA_OPTS="-Dspark.deploy.recoveryMode=ZOOKEEPER -Dspark.deploy.zookeeper.url=master.hadoop:2181,slave1.hadoop:2181,slave2.hadoop:2181 -Dspark.deploy.zookeeper.dir=/spark"
解释：

#-Dspark.deploy.recoverMode=ZOOKEEPER #代表发生故障使用zookeeper服务

#-Dspark.depoly.zookeeper.url=master.hadoop,slave1.hadoop,slave1.hadoop #主机名的名字

#-Dspark.deploy.zookeeper.dir=/spark #spark要在zookeeper上写数据时的保存目录

[root@master conf]# vi spark-env.sh 
export JAVA_HOME=/apps/jdk1.8.0_171
export SCALA_HOME=/apps/scala-2.11.7
#export HADOOP_HOME=/apps/hadoop-2.8.0/
#export HADOOP_CONF_DIR=/apps/hadoop-2.8.0/etc/hadoop
#export SPARK_MASTER_IP=master.hadoop
export SPARK_WORKER_MEMORY=512m
export SPARK_WORKER_CORES=2
export SPARK_WORKER_INSTANCES=1
export SPARK_DAEMON_JAVA_OPTS="-Dspark.deploy.recoveryMode=ZOOKEEPER -Dspark.deploy.zookeeper.url=master.hadoop:2181,slave1.hadoop:2181,slave2.hadoop:2181 -Dspark.deploy.zookeeper.dir=/spark"
 

然后修改slaves文件

[root@master conf]# vi slaves
# A Spark Worker will be started on each of the machines listed below.
master.hadoop
slave1.hadoop
slave2.hadoop
四、启动

先启动zookeeper

可以单台启动，也可以写一个启动脚本，集体启动。

在每台机器上执行该命令：

[root@master /]# zkServer.sh start
启动脚本：https://blog.csdn.net/nuc2015/article/details/81045941

启动后一个leader，其他的是flower

 

在第一台机器上启动spark

[root@master spark-2.2.0]# sbin/start-all.sh 
starting org.apache.spark.deploy.master.Master, logging to /apps/spark-2.2.0/logs/spark-root-org.apache.spark.deploy.master.Master-1-master.hadoop.out
master.hadoop: starting org.apache.spark.deploy.worker.Worker, logging to /apps/spark-2.2.0/logs/spark-root-org.apache.spark.deploy.worker.Worker-1-master.hadoop.out
slave1.hadoop: starting org.apache.spark.deploy.worker.Worker, logging to /apps/spark-2.2.0/logs/spark-root-org.apache.spark.deploy.worker.Worker-1-slave1.hadoop.out
slave2.hadoop: starting org.apache.spark.deploy.worker.Worker, logging to /apps/spark-2.2.0/logs/spark-root-org.apache.spark.deploy.worker.Worker-1-slave2.hadoop.out
[root@master spark-2.2.0]# jps
2321 Jps
2149 Worker
2028 QuorumPeerMain
2076 Master
[root@master spark-2.2.0]# 
在第二台机器上单独启动master

[root@slave1 spark-2.2.0]# sbin/start-master.sh 
查看web端口

第一个master为

Status: ALIVE

第二个master为

Status: STANDBY

搭建成功
————————————————