讨论QQ：1586558083
正文
一、下载Spark安装包
1、从官网下载
http://spark.apache.org/downloads.html
2、从微软的镜像站下载
http://mirrors.hust.edu.cn/apache/
3、从清华的镜像站下载
https://mirrors.tuna.tsinghua.edu.cn/apache/
二、安装基础
1、Java8安装成功
2、zookeeper安装成功
3、hadoop2.7.5 HA安装成功
4、Scala安装成功（不安装进程也可以启动）
三、Spark安装过程
1、上传并解压缩
[hadoop@hadoop1 ~]$ ls
apps data exam inithive.conf movie spark-2.3.0-bin-hadoop2.7.tgz udf.jar
cookies data.txt executions json.txt projects student zookeeper.out
course emp hive.sql log sougou temp
[hadoop@hadoop1 ~]$ tar -zxvf spark-2.3.0-bin-hadoop2.7.tgz -C apps/
2、为安装包创建一个软连接
[hadoop@hadoop1 ~]$ cd apps/ [hadoop@hadoop1 apps]$ ls hadoop-2.7.5 hbase-1.2.6 spark-2.3.0-bin-hadoop2.7 zookeeper-3.4.10 zookeeper.out [hadoop@hadoop1 apps]$ ln -s spark-2.3.0-bin-hadoop2.7/ spark [hadoop@hadoop1 apps]$ ll 总用量 36 drwxr-xr-x. 10 hadoop hadoop 4096 3月 23 20:29 hadoop-2.7.5 drwxrwxr-x. 7 hadoop hadoop 4096 3月 29 13:15 hbase-1.2.6 lrwxrwxrwx. 1 hadoop hadoop 26 4月 20 13:48 spark -> spark-2.3.0-bin-hadoop2.7/ drwxr-xr-x. 13 hadoop hadoop 4096 2月 23 03:42 spark-2.3.0-bin-hadoop2.7 drwxr-xr-x. 10 hadoop hadoop 4096 3月 23 2017 zookeeper-3.4.10 -rw-rw-r--. 1 hadoop hadoop 17559 3月 29 13:37 zookeeper.out [hadoop@hadoop1 apps]$
3、进入spark/conf修改配置文件
（1）进入配置文件所在目录
[hadoop@hadoop1 ~]$ cd apps/spark/conf/ [hadoop@hadoop1 conf]$ ll 总用量 36 -rw-r--r--. 1 hadoop hadoop 996 2月 23 03:42 docker.properties.template -rw-r--r--. 1 hadoop hadoop 1105 2月 23 03:42 fairscheduler.xml.template -rw-r--r--. 1 hadoop hadoop 2025 2月 23 03:42 log4j.properties.template -rw-r--r--. 1 hadoop hadoop 7801 2月 23 03:42 metrics.properties.template -rw-r--r--. 1 hadoop hadoop 865 2月 23 03:42 slaves.template -rw-r--r--. 1 hadoop hadoop 1292 2月 23 03:42 spark-defaults.conf.template -rwxr-xr-x. 1 hadoop hadoop 4221 2月 23 03:42 spark-env.sh.template [hadoop@hadoop1 conf]$
（2）复制spark-env.sh.template并重命名为spark-env.sh，并在文件最后添加配置内容
[hadoop@hadoop1 conf]$ cp spark-env.sh.template spark-env.sh [hadoop@hadoop1 conf]$ vi spark-env.sh
export JAVA_HOME=/usr/local/jdk1.8.0_73 #export SCALA_HOME=/usr/share/scala export HADOOP_HOME=/home/hadoop/apps/hadoop-2.7.5 export HADOOP_CONF_DIR=/home/hadoop/apps/hadoop-2.7.5/etc/hadoop export SPARK_WORKER_MEMORY=500m export SPARK_WORKER_CORES=1 export SPARK_DAEMON_JAVA_OPTS="-Dspark.deploy.recoveryMode=ZOOKEEPER -Dspark.deploy.zookeeper.url=hadoop1:2181,hadoop2:2181,hadoop3:2181,hadoop4:2181 -Dspark.deploy.zookeeper.dir=/spark"
注：
#export SPARK_MASTER_IP=hadoop1 这个配置要注释掉。 集群搭建时配置的spark参数可能和现在的不一样，主要是考虑个人电脑配置问题，如果memory配置太大，机器运行很慢。 说明： -Dspark.deploy.recoveryMode=ZOOKEEPER #说明整个集群状态是通过zookeeper来维护的，整个集群状态的恢复也是通过zookeeper来维护的。就是说用zookeeper做了spark的HA配置，Master(Active)挂掉的话，Master(standby)要想变成Master（Active）的话，Master(Standby)就要像zookeeper读取整个集群状态信息，然后进行恢复所有Worker和Driver的状态信息，和所有的Application状态信息； -Dspark.deploy.zookeeper.url=hadoop1:2181,hadoop2:2181,hadoop3:2181,hadoop4:2181#将所有配置了zookeeper，并且在这台机器上有可能做master(Active)的机器都配置进来；（我用了4台，就配置了4台）
-Dspark.deploy.zookeeper.dir=/spark 这里的dir和zookeeper配置文件zoo.cfg中的dataDir的区别？？？ -Dspark.deploy.zookeeper.dir是保存spark的元数据，保存了spark的作业运行状态； zookeeper会保存spark集群的所有的状态信息，包括所有的Workers信息，所有的Applactions信息，所有的Driver信息,如果集群
（3）复制slaves.template成slaves
[hadoop@hadoop1 conf]$ cp slaves.template slaves
[hadoop@hadoop1 conf]$ vi slaves
添加如下内容
hadoop1
hadoop2
hadoop3
hadoop4
（4）将安装包分发给其他节点
[hadoop@hadoop1 ~]$ cd apps/ [hadoop@hadoop1 apps]$ scp -r spark-2.3.0-bin-hadoop2.7/ hadoop2:$PWD [hadoop@hadoop1 apps]$ scp -r spark-2.3.0-bin-hadoop2.7/ hadoop3:$PWD [hadoop@hadoop1 apps]$ scp -r spark-2.3.0-bin-hadoop2.7/ hadoop4:$PWD
创建软连接
[hadoop@hadoop2 ~]$ cd apps/ [hadoop@hadoop2 apps]$ ls hadoop-2.7.5 hbase-1.2.6 spark-2.3.0-bin-hadoop2.7 zookeeper-3.4.10 [hadoop@hadoop2 apps]$ ln -s spark-2.3.0-bin-hadoop2.7/ spark [hadoop@hadoop2 apps]$ ll 总用量 16 drwxr-xr-x 10 hadoop hadoop 4096 3月 23 20:29 hadoop-2.7.5 drwxrwxr-x 7 hadoop hadoop 4096 3月 29 13:15 hbase-1.2.6 lrwxrwxrwx 1 hadoop hadoop 26 4月 20 19:26 spark -> spark-2.3.0-bin-hadoop2.7/ drwxr-xr-x 13 hadoop hadoop 4096 4月 20 19:24 spark-2.3.0-bin-hadoop2.7 drwxr-xr-x 10 hadoop hadoop 4096 3月 21 19:31 zookeeper-3.4.10 [hadoop@hadoop2 apps]$
4、配置环境变量
所有节点均要配置
[hadoop@hadoop1 spark]$ vi ~/.bashrc
#Spark
export SPARK_HOME=/home/hadoop/apps/spark
export PATH=$PATH:$SPARK_HOME/bin
保存并使其立即生效
[hadoop@hadoop1 spark]$ source ~/.bashrc
四、启动
1、先启动zookeeper集群
所有节点均要执行
[hadoop@hadoop1 ~]$ zkServer.sh start
ZooKeeper JMX enabled by default
Using config: /home/hadoop/apps/zookeeper-3.4.10/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
[hadoop@hadoop1 ~]$ zkServer.sh status
ZooKeeper JMX enabled by default
Using config: /home/hadoop/apps/zookeeper-3.4.10/bin/../conf/zoo.cfg
Mode: follower
[hadoop@hadoop1 ~]$
2、在启动HDFS集群
任意一个节点执行即可
[hadoop@hadoop1 ~]$ start-dfs.sh
3、在启动Spark集群
在一个节点上执行
[hadoop@hadoop1 ~]$ cd apps/spark/sbin/ [hadoop@hadoop1 sbin]$ start-all.sh
4、查看进程
5、问题
查看进程发现spark集群只有hadoop1成功启动了Master进程，其他3个节点均没有启动成功，需要手动启动，进入到/home/hadoop/apps/spark/sbin目录下执行以下命令，3个节点都要执行
[hadoop@hadoop2 ~]$ cd ~/apps/spark/sbin/
[hadoop@hadoop2 sbin]$ start-master.sh
6、执行之后再次查看进程
Master进程和Worker进程都以启动成功
五、验证
1、查看Web界面Master状态
hadoop1是ALIVE状态，hadoop2、hadoop3和hadoop4均是STANDBY状态
hadoop1节点
hadoop2节点
hadoop3
hadoop4
2、验证HA的高可用
手动干掉hadoop1上面的Master进程，观察是否会自动进行切换
干掉hadoop1上的Master进程之后，再次查看web界面
hadoo1节点，由于Master进程被干掉，所以界面无法访问
hadoop2节点，Master被干掉之后，hadoop2节点上的Master成功篡位成功，成为ALIVE状态
hadoop3节点
hadoop4节点
六、执行Spark程序on standalone
1、执行第一个Spark程序
[hadoop@hadoop3 ~]$ /home/hadoop/apps/spark/bin/spark-submit
> --class org.apache.spark.examples.SparkPi
> --master spark://hadoop1:7077
> --executor-memory 500m
> --total-executor-cores 1
> /home/hadoop/apps/spark/examples/jars/spark-examples_2.11-2.3.0.jar
> 100
其中的spark://hadoop1:7077是下图中的地址
运行结果
2、启动spark shell
[hadoop@hadoop1 ~]$ /home/hadoop/apps/spark/bin/spark-shell > --master spark://hadoop1:7077 > --executor-memory 500m > --total-executor-cores 1
参数说明：
--master spark://hadoop1:7077 指定Master的地址
--executor-memory 500m:指定每个worker可用内存为500m
--total-executor-cores 1: 指定整个集群使用的cup核数为1个
注意：
如果启动spark shell时没有指定master地址，但是也可以正常启动spark shell和执行spark shell中的程序，其实是启动了spark的local模式，该模式仅在本机启动一个进程，没有与集群建立联系。
Spark Shell中已经默认将SparkContext类初始化为对象sc。用户代码如果需要用到，则直接应用sc即可
Spark Shell中已经默认将SparkSQl类初始化为对象spark。用户代码如果需要用到，则直接应用spark即可
3、 在spark shell中编写WordCount程序
（1）编写一个hello.txt文件并上传到HDFS上的spark目录下
[hadoop@hadoop1 ~]$ vi hello.txt
[hadoop@hadoop1 ~]$ hadoop fs -mkdir -p /spark
[hadoop@hadoop1 ~]$ hadoop fs -put hello.txt /spark
hello.txt的内容如下
you,jump
i,jump
you,jump
i,jump
jump
（2）在spark shell中用scala语言编写spark程序
scala> sc.textFile("/spark/hello.txt").flatMap(_.split(",")).map((_,1)).reduceByKey(_+_).saveAsTextFile("/spark/out")
说明：
sc是SparkContext对象，该对象是提交spark程序的入口
textFile("/spark/hello.txt")是hdfs中读取数据
flatMap(_.split(" "))先map再压平
map((_,1))将单词和1构成元组
reduceByKey(_+_)按照key进行reduce，并将value累加
saveAsTextFile("/spark/out")将结果写入到hdfs中
（3）使用hdfs命令查看结果
[hadoop@hadoop2 ~]$ hadoop fs -cat /spark/out/p*
(jump,5)
(you,2)
(i,2)
[hadoop@hadoop2 ~]$
七、 执行Spark程序on YARN
1、前提
成功启动zookeeper集群、HDFS集群、YARN集群
2、启动Spark on YARN
[hadoop@hadoop1 bin]$ spark-shell --master yarn --deploy-mode client
报错如下：
报错原因：内存资源给的过小，yarn直接kill掉进程，则报rpc连接失败、ClosedChannelException等错误。
解决方法：
先停止YARN服务，然后修改yarn-site.xml，增加如下内容
<property>
<name>yarn.nodemanager.vmem-check-enabled</name>
<value>false</value>
<description>Whether virtual memory limits will be enforced for containers</description>
</property>
<property>
<name>yarn.nodemanager.vmem-pmem-ratio</name>
<value>4</value>
<description>Ratio between virtual memory to physical memory when setting memory limits for containers</description>
</property>
将新的yarn-site.xml文件分发到其他Hadoop节点对应的目录下，最后在重新启动YARN。
重新执行以下命令启动spark on yarn
[hadoop@hadoop1 hadoop]$ spark-shell --master yarn --deploy-mode client
启动成功
3、打开YARN的web界面
打开YARN WEB页面：http://hadoop4:8088 可以看到Spark shell应用程序正在运行
单击ID号链接，可以看到该应用程序的详细信息
单击“ApplicationMaster”链接
4、运行程序
scala> val array = Array(1,2,3,4,5)
array: Array[Int] = Array(1, 2, 3, 4, 5)
scala> val rdd = sc.makeRDD(array)
rdd: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[0] at makeRDD at <console>:26
scala> rdd.count
res0: Long = 5
scala>
再次查看YARN的web界面
查看executors
5、执行Spark自带的示例程序PI
[hadoop@hadoop1 ~]$ spark-submit --class org.apache.spark.examples.SparkPi
> --master yarn
> --deploy-mode cluster
> --driver-memory 500m
> --executor-memory 500m
> --executor-cores 1
> /home/hadoop/apps/spark/examples/jars/spark-examples_2.11-2.3.0.jar
> 10
执行过程
[hadoop@hadoop1 ~]$ spark-submit --class org.apache.spark.examples.SparkPi
> --master yarn
> --deploy-mode cluster
> --driver-memory 500m
> --executor-memory 500m
> --executor-cores 1
> /home/hadoop/apps/spark/examples/jars/spark-examples_2.11-2.3.0.jar
> 10
2018-04-21 17:57:32 WARN NativeCodeLoader:62 - Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
2018-04-21 17:57:34 INFO ConfiguredRMFailoverProxyProvider:100 - Failing over to rm2
2018-04-21 17:57:34 INFO Client:54 - Requesting a new application from cluster with 4 NodeManagers
2018-04-21 17:57:34 INFO Client:54 - Verifying our application has not requested more than the maximum memory capability of the cluster (8192 MB per container)
2018-04-21 17:57:34 INFO Client:54 - Will allocate AM container, with 884 MB memory including 384 MB overhead
2018-04-21 17:57:34 INFO Client:54 - Setting up container launch context for our AM
2018-04-21 17:57:34 INFO Client:54 - Setting up the launch environment for our AM container
2018-04-21 17:57:34 INFO Client:54 - Preparing resources for our AM container
2018-04-21 17:57:36 WARN Client:66 - Neither spark.yarn.jars nor spark.yarn.archive is set, falling back to uploading libraries under SPARK_HOME.
2018-04-21 17:57:39 INFO Client:54 - Uploading resource file:/tmp/spark-93bd68c9-85de-482e-bbd7-cd2cee60e720/__spark_libs__8262081479435245591.zip -> hdfs://myha01/user/hadoop/.sparkStaging/application_1524303370510_0005/__spark_libs__8262081479435245591.zip
2018-04-21 17:57:44 INFO Client:54 - Uploading resource file:/home/hadoop/apps/spark/examples/jars/spark-examples_2.11-2.3.0.jar -> hdfs://myha01/user/hadoop/.sparkStaging/application_1524303370510_0005/spark-examples_2.11-2.3.0.jar
2018-04-21 17:57:44 INFO Client:54 - Uploading resource file:/tmp/spark-93bd68c9-85de-482e-bbd7-cd2cee60e720/__spark_conf__2498510663663992254.zip -> hdfs://myha01/user/hadoop/.sparkStaging/application_1524303370510_0005/__spark_conf__.zip
2018-04-21 17:57:44 INFO SecurityManager:54 - Changing view acls to: hadoop
2018-04-21 17:57:44 INFO SecurityManager:54 - Changing modify acls to: hadoop
2018-04-21 17:57:44 INFO SecurityManager:54 - Changing view acls groups to:
2018-04-21 17:57:44 INFO SecurityManager:54 - Changing modify acls groups to:
2018-04-21 17:57:44 INFO SecurityManager:54 - SecurityManager: authentication disabled; ui acls disabled; users with view permissions: Set(hadoop); groups with view permissions: Set(); users with modify permissions: Set(hadoop); groups with modify permissions: Set()
2018-04-21 17:57:44 INFO Client:54 - Submitting application application_1524303370510_0005 to ResourceManager
2018-04-21 17:57:44 INFO YarnClientImpl:273 - Submitted application application_1524303370510_0005
2018-04-21 17:57:45 INFO Client:54 - Application report for application_1524303370510_0005 (state: ACCEPTED)
2018-04-21 17:57:45 INFO Client:54 -
client token: N/A
diagnostics: N/A
ApplicationMaster host: N/A
ApplicationMaster RPC port: -1
queue: default
start time: 1524304664749
final status: UNDEFINED
tracking URL: http://hadoop4:8088/proxy/application_1524303370510_0005/
user: hadoop
2018-04-21 17:57:46 INFO Client:54 - Application report for application_1524303370510_0005 (state: ACCEPTED)
2018-04-21 17:57:47 INFO Client:54 - Application report for application_1524303370510_0005 (state: ACCEPTED)
2018-04-21 17:57:48 INFO Client:54 - Application report for application_1524303370510_0005 (state: ACCEPTED)
2018-