spark 2.2.0 高可用搭建

一、概述

1.实验环境基于以前搭建的haoop HA；

2.spark HA所需要的zookeeper环境前文已经配置过，此处不再重复。

3.所需软件包为：scala-2.12.3.tgz、spark-2.2.0-bin-hadoop2.7.tar

4.主机规划

二、配置Scala

1.解压并拷贝

[root@bd1 ~]# tar -zxf scala-2.12.3.tgz 
[root@bd1 ~]# cp -r scala-2.12.3 /usr/local/

2.配置环境变量

[root@bd1 ~]# vim /etc/profile
export SCALA_HOME=/usr/local/scala
export PATH=:$SCALA_HOME/bin:$PATH
[root@bd1 ~]# source /etc/profile

3.验证

[root@bd1 ~]# scala -version
Scala code runner version 2.12.3 -- Copyright 2002-2017, LAMP/EPFL and Lightbend, Inc.

三、配置Spark

1.解压并拷贝

[root@bd1 ~]# tar -zxf spark-2.2.0-bin-hadoop2.7.tgz
[root@bd1 ~]# cp spark-2.2.0-bin-hadoop2.7 /usr/local/spark

2.配置环境变量

[root@bd1 ~]# vim /etc/profile
export SCALA_HOME=/usr/local/scala
export PATH=:$SCALA_HOME/bin:$PATH
[root@bd1 ~]# source /etc/profile

3.修改spark-env.sh    #文件不存在需要拷贝模板

[root@bd1 conf]# vim spark-env.sh
export JAVA_HOME=/usr/local/jdk
export HADOOP_HOME=/usr/local/hadoop
export HADOOP_CONF_DIR=/usr/local/hadoop/etc/hadoop
export SCALA_HOME=/usr/local/scala
export SPARK_DAEMON_JAVA_OPTS="-Dspark.deploy.recoveryMode=ZOOKEEPER -Dspark.deploy.zookeeper.url=bd4:2181,bd5:2181 -Dspark.deploy.zookeeper.dir=/spark"
export SPARK_WORKER_MEMORY=1g
export SPARK_WORKER_CORES=2
export SPARK_WORKER_INSTANCES=1

4.修改spark-defaults.conf    #文件不存在需要拷贝模板

[root@bd1 conf]# vim spark-defaults.conf
spark.master                     spark://master:7077
spark.eventLog.enabled           true
spark.eventLog.dir               hdfs://master:/user/spark/history
spark.serializer                 org.apache.spark.serializer.KryoSerializer

5.在HDFS文件系统中新建日志文件目录

hdfs dfs -mkdir -p /user/spark/history
hdfs dfs -chmod 777 /user/spark/history

6.修改slaves

[root@bd1 conf]# vim slaves
bd1
bd2
bd3
bd4
bd5

四、同步到其他主机

1.使用scp同步Scala到bd2-bd5

scp -r /usr/local/scala root@bd2:/usr/local/
scp -r /usr/local/scala root@bd3:/usr/local/
scp -r /usr/local/scala root@bd4:/usr/local/
scp -r /usr/local/scala root@bd5:/usr/local/

2.同步Spark到bd2-bd5

scp -r /usr/local/spark root@bd2:/usr/local/
scp -r /usr/local/spark root@bd3:/usr/local/
scp -r /usr/local/spark root@bd4:/usr/local/
scp -r /usr/local/spark root@bd5:/usr/local/

五、启动集群并测试HA

1.启动顺序为：zookeeper–>hadoop–>spark

2.启动spark

bd4：

[root@bd4 sbin]# cd /usr/local/spark/sbin/
[root@bd4 sbin]# ./start-all.sh 
starting org.apache.spark.deploy.master.Master, logging to /usr/local/spark/logs/spark-root-org.apache.spark.deploy.master.Master-1-bd4.out
bd4: starting org.apache.spark.deploy.worker.Worker, logging to /usr/local/spark/logs/spark-root-org.apache.spark.deploy.worker.Worker-1-bd4.out
bd2: starting org.apache.spark.deploy.worker.Worker, logging to /usr/local/spark/logs/spark-root-org.apache.spark.deploy.worker.Worker-1-bd2.out
bd3: starting org.apache.spark.deploy.worker.Worker, logging to /usr/local/spark/logs/spark-root-org.apache.spark.deploy.worker.Worker-1-bd3.out
bd5: starting org.apache.spark.deploy.worker.Worker, logging to /usr/local/spark/logs/spark-root-org.apache.spark.deploy.worker.Worker-1-bd5.out
bd1: starting org.apache.spark.deploy.worker.Worker, logging to /usr/local/spark/logs/spark-root-org.apache.spark.deploy.worker.Worker-1-bd1.out

[root@bd4 sbin]# jps
3153 DataNode
7235 Jps
3046 JournalNode
7017 Master
3290 NodeManager
7116 Worker
2958 QuorumPeerMain

bd5：

[root@bd5 sbin]# ./start-master.sh 
starting org.apache.spark.deploy.master.Master, logging to /usr/local/spark/logs/spark-root-org.apache.spark.deploy.master.Master-1-bd5.out

[root@bd5 sbin]# jps
3584 NodeManager
5602 RunJar
3251 QuorumPeerMain
8564 Master
3447 DataNode
8649 Jps
8474 Worker
3340 JournalNode

3.停掉bd4的Master进程

[root@bd4 sbin]# kill -9 7017
[root@bd4 sbin]# jps
3153 DataNode
7282 Jps
3046 JournalNode
3290 NodeManager
7116 Worker
2958 QuorumPeerMain

五、总结

一开始时想把Master放到bd1和bd2上，但是启动Spark后发现两个节点上都是Standby。然后修改配置文件转移到bd4和bd5上，才顺利运行。换言之Spark HA的Master必须位于Zookeeper集群上才能正常运行，即该节点上要有JournalNode这个进程。