第一章，安装spark集群，测试和运行wordcount 案例

大家好，又见面了，我是你们的朋友全栈君。

1.1下载spark

地址：https://mirrors.tuna.tsinghua.edu.cn/apache/spark/spark-2.3.0/

1.2上传解压；

xiaoye@ubuntu:~/Downloads$ ls
apache-activemq-5.15.3-bin.tar.gz  hive-0.13.1-cdh5.2.0.tar.gz
flume-ng-1.6.0-cdh5.10.1.tar.gz    spark-2.3.0-bin-hadoop2.7.tgz
hadoop-2.5.0-cdh5.2.0.tar.gz       zookeeper-3.4.5-cdh5.10.2.tar.gz

xiaoye@ubuntu:~/Downloads$ tar -zxvf spark-2.3.0-bin-hadoop2.7.tgz -C ../

1.3为安装包建立一个软连接：

xiaoye@ubuntu:~$ ln -s spark-2.3.0-bin-hadoop2.7/ spark

1.4进入spark/conf修改配置文件

xiaoye@ubuntu:~$ cd spark
xiaoye@ubuntu:~/spark$ cd conf
xiaoye@ubuntu:~/spark/conf$ ls
docker.properties.template  metrics.properties.template   spark-env.sh.template
fairscheduler.xml.template  slaves.template
log4j.properties.template   spark-defaults.conf.template

xiaoye@ubuntu:~/spark/conf$ 

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

export JAVA_HOME=/usr/lib/jvm/java-1.8.0-openjdk-amd64
export HADOOP_HOME=/home/xiaoye/hadoop
export HADOOP_CONF_DIR=/home/xiaoye/hadoop/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=ubuntu:2181,ubuntu2:2181,ubuntu3:2181 -Dspark.deploy.zookeeper.dir=/spark"

解释：

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=

buntu:2181,ubuntu2:2181,ubuntu3:2181 

#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3台，就配置了3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 

复制slaves.template成slaves:

xiaoye@ubuntu:~/spark/conf$ cp slaves.template  slaves

修改slaves:

xiaoye@ubuntu:~/spark/conf$ vi slaves

ubuntu
ubuntu2
ubuntu3

1.5 一台机子配置完后分发给其他节点

xiaoye@ubuntu:~$ scp -r spark-2.3.0-bin-hadoop2.7/ xiaoye@192.168.26.141:/home/xiaoye

xiaoye@ubuntu:~$ scp -r spark-2.3.0-bin-hadoop2.7/ xiaoye@192.168.26.142:/home/xiaoye

1.6为每台机子创建软连接

xiaoye@ubuntu2:~$ ln -s spark-2.3.0-bin-hadoop2.7/ spark

xiaoye@ubuntu3:~$ ln -s spark-2.3.0-bin-hadoop2.7/ spark

1.7为每台机子配置环境变量

xiaoye@ubuntu:~$ vim .bashrc

#Spark
export SPARK_HOME=/home/xiaoye/spark
export PATH=$PATH:$SPARK_HOME/bin

配置好后立即生效：

xiaoye@ubuntu:~$ source .bashrc

2试着启动spark集群

2.1启动zookeeper集群：

每台机子执行如下操作：

xiaoye@ubuntu:~$ ./zookeeper/sbin/zkServer.sh  start
JMX enabled by default
Using config: /home/xiaoye/zookeeper/sbin/../conf/zoo.cfg
Starting zookeeper … STARTED
xiaoye@ubuntu:~$ ./zookeeper/sbin/zkServer.sh  status
JMX enabled by default
Using config: /home/xiaoye/zookeeper/sbin/../conf/zoo.cfg

Mode: leader

2.2启动hdfs:

在一个节点执行以下命令即可：

xiaoye@ubuntu:~$ ./hadoop/sbin/start-dfs.sh 
18/06/09 21:36:31 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Starting namenodes on [ubuntu ubuntu2]
ubuntu: Warning: Permanently added 'ubuntu,192.168.26.140' (ECDSA) to the list of known hosts.
ubuntu2: Warning: Permanently added 'ubuntu2,192.168.26.141' (ECDSA) to the list of known hosts.
ubuntu2: starting namenode, logging to /home/xiaoye/hadoop/logs/hadoop-xiaoye-namenode-ubuntu2.out
ubuntu: starting namenode, logging to /home/xiaoye/hadoop/logs/hadoop-xiaoye-namenode-ubuntu.out
ubuntu2: Warning: Permanently added 'ubuntu2,192.168.26.141' (ECDSA) to the list of known hosts.
ubuntu: Warning: Permanently added 'ubuntu,192.168.26.140' (ECDSA) to the list of known hosts.
ubuntu3: Warning: Permanently added 'ubuntu3,192.168.26.142' (ECDSA) to the list of known hosts.
ubuntu: starting datanode, logging to /home/xiaoye/hadoop/logs/hadoop-xiaoye-datanode-ubuntu.out
ubuntu2: starting datanode, logging to /home/xiaoye/hadoop/logs/hadoop-xiaoye-datanode-ubuntu2.out
ubuntu3: starting datanode, logging to /home/xiaoye/hadoop/logs/hadoop-xiaoye-datanode-ubuntu3.out
Starting journal nodes [ubuntu ubuntu2 ubuntu3]
ubuntu3: Warning: Permanently added 'ubuntu3,192.168.26.142' (ECDSA) to the list of known hosts.
ubuntu: Warning: Permanently added 'ubuntu,192.168.26.140' (ECDSA) to the list of known hosts.
ubuntu2: Warning: Permanently added 'ubuntu2,192.168.26.141' (ECDSA) to the list of known hosts.
ubuntu: starting journalnode, logging to /home/xiaoye/hadoop/logs/hadoop-xiaoye-journalnode-ubuntu.out
ubuntu3: starting journalnode, logging to /home/xiaoye/hadoop/logs/hadoop-xiaoye-journalnode-ubuntu3.out
ubuntu2: starting journalnode, logging to /home/xiaoye/hadoop/logs/hadoop-xiaoye-journalnode-ubuntu2.out
18/06/09 21:36:58 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Starting ZK Failover Controllers on NN hosts [ubuntu ubuntu2]
ubuntu2: Warning: Permanently added 'ubuntu2,192.168.26.141' (ECDSA) to the list of known hosts.
ubuntu: Warning: Permanently added 'ubuntu,192.168.26.140' (ECDSA) to the list of known hosts.
ubuntu2: starting zkfc, logging to /home/xiaoye/hadoop/logs/hadoop-xiaoye-zkfc-ubuntu2.out
ubuntu: starting zkfc, logging to /home/xiaoye/hadoop/logs/hadoop-xiaoye-zkfc-ubuntu.out
xiaoye@ubuntu:~$ jps
\2746 NameNode
3252 DFSZKFailoverController
2573 QuorumPeerMain
2861 DataNode
3324 Jps
3060 JournalNode

2.3启动spark

在三个节点分别执行以下命令：

xiaoye@ubuntu:~$ ./spark/sbin/start-all.sh 
org.apache.spark.deploy.master.Master running as process 3359.  Stop it first.
localhost: Warning: Permanently added 'localhost' (ECDSA) to the list of known hosts.
localhost: starting org.apache.spark.deploy.worker.Worker, logging to /home/xiaoye/spark/logs/spark-xiaoye-org.apache.spark.deploy.worker.Worker-1-ubuntu.out

好了之后jps分别查看启动的进程：

leader节点：

xiaoye@ubuntu:~$ jps
2746 NameNode
3359 Master
3252 DFSZKFailoverController
3550 Worker
2573 QuorumPeerMain
2861 DataNode
3583 Jps
3060 JournalNode

其他两个从节点：

xiaoye@ubuntu2:~$ jps
2488 JournalNode
2748 Worker
2784 Jps
2654 Master
2229 QuorumPeerMain
2377 DataNode

xiaoye@ubuntu3:~$ jps
2510 Master
2582 Worker
2298 DataNode
2402 JournalNode
2665 Jps
2198 QuorumPeerMain

3，验证是否成功启动

在web上查看：

4，验证HA高可用

动干掉ubuntu上面的Master进程，观察是否会自动进行切换

xiaoye@ubuntu:~$ jps
3656 Jps
2746 NameNode
3359 Master
3252 DFSZKFailoverController
2573 QuorumPeerMain
2861 DataNode
3060 JournalNode
xiaoye@ubuntu:~$ kill -9 3359
xiaoye@ubuntu:~$ jps
2746 NameNode
3252 DFSZKFailoverController
2573 QuorumPeerMain
2861 DataNode
3666 Jps
3060 JournalNode

查看web：

再看看其他两个节点有没有成功上位成live的：

ubuntu3自动成为live节点说明成功。

5，执行spark程序 on standalone

5.1执行第一个spark程序，执行的是spark自带的案例：

在master下执行：

xiaoye@ubuntu3:~$ /home/xiaoye/spark/bin/spark-submit --class org.apache.spark.examples.SparkPi --master spark://ubuntu3:7077 --executor-memory 500m --total-executor-cores 1 /home/xiaoye/spark/examples/jars/spark-examples_2.11-2.3.0.jar  100

其中–master spark://ubuntu3:7077是下图的位置的结果：

运行结果：

至于含义我们以后再讨论

5.2 启动spark shell

xiaoye@ubuntu3:~$ /home/xiaoye/spark/bin/spark-shell \
> --master spark://ubuntu3:7077 \
> --executor-memory 500m
> --total-executor-cores 1 

参数说明：

–master spark://ubuntu3:7077 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

 xiaoye@ubuntu3:~$ 
xiaoye@ubuntu3:~$ /home/xiaoye/spark/bin/spark-shell \
> --master spark://ubuntu3:7077 \
> --executor-memory 500m
--^H^H2018-06-09 23:21:53 WARN  Utils:66 - Your hostname, ubuntu3 resolves to a loopback address: 127.0.1.1; using 192.168.26.142 instead (on interface ens33)
2018-06-09 23:21:53 WARN  Utils:66 - Set SPARK_LOCAL_IP if you need to bind to another address
2018-06-09 23:21:55 WARN  NativeCodeLoader:62 - Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Setting default log level to "WARN".
To adjust logging level use sc.setLogLevel(newLevel). For SparkR, use setLogLevel(newLevel).
Spark context Web UI available at http://ubuntu3:4040
Spark context available as 'sc' (master = spark://ubuntu3:7077, app id = app-20180609232223-0001).
Spark session available as 'spark'.
Welcome to
      ____              __
     / __/__  ___ _____/ /__
    _\ \/ _ \/ _ `/ __/  '_/
   /___/ .__/\_,_/_/ /_/\_\   version 2.3.0
      /_/
         
Using Scala version 2.11.8 (OpenJDK 64-Bit Server VM, Java 1.8.0_162)
Type in expressions to have them evaluated.
Type :help for more information.

scala> --

scala> 

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

5.3在spark shell中编写word count 程序

创建一个文件，随便写点内容：

xiaoye@ubuntu3:~$ cat classes/aa.txt 
hello world!
aa aa
d
d 
dg
g

然后上传到hdfs中：

xiaoye@ubuntu3:~/hadoop$ ./hadoop/ fs -mkdir -p /spark
-bash: ./hadoop/: Is a directory
xiaoye@ubuntu3:~/hadoop$ hadoop fs -mkdir -p /spark
18/06/09 23:32:33 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
xiaoye@ubuntu3:~/hadoop$ hadoop fs -ls /spark
18/06/09 23:33:12 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
xiaoye@ubuntu3:~/hadoop$ hadoop fs -put classes/aa.txt /spark
18/06/09 23:34:04 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
put: `classes/aa.txt': No such file or directory
xiaoye@ubuntu3:~/hadoop$ ls
bin             etc                  hadoop   lib      sbin   tmp
bin-mapreduce1  examples             include  libexec  share
cloudera        examples-mapreduce1  journal  logs     src
xiaoye@ubuntu3:~/hadoop$ cd ..
xiaoye@ubuntu3:~$ hadoop fs -put classes/aa.txt /spark
18/06/09 23:34:33 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
xiaoye@ubuntu3:~$ hadoop fs -ls /spark
18/06/09 23:34:48 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Found 1 items
-rw-r--r--   2 xiaoye supergroup         29 2018-06-09 23:34 /spark/aa.txt

5.4在spark shell中用scala编写spark程序，按空格分割数据

scala> sc.textFile("/spark/aa.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/aa.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

执行完后，查看hdfs的执行结果：

xiaoye@ubuntu3:~$ hadoop fs -cat /spark/out/p*
18/06/09 23:52:42 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
(d,2)
(aa,2)
(hello,1)
(world!,1)
(dg,1)
(g,1)

执行结果跟我们相要的结果是一样的。

6，执行spark程序 on Yarn

首先要成功启动zookeeper集群、HDFS集群、YARN集群

启动spark on Yarn

xiaoye@ubuntu3:~$

 spark-shell --master yarn --deploy-mode client

启动时间比较长，成功启动：

启动的进程如下：

xiaoye@ubuntu3:~$ jps
2298 DataNode
5150 ExecutorLauncher
3950 ResourceManager
2402 JournalNode
5375 Jps
5275 CoarseGrainedExecutorBackend
5274 CoarseGrainedExecutorBackend
4092 NodeManager
2198 QuorumPeerMain
4424 SparkSubmit