Spark笔记整理：spark单机安装部署分布式集群与HA安装部署+spark源码编译

Posted 2020-11-01

[TOC]

---

spark单机安装部署

1.安装scala
解压：tar -zxvf soft/scala-2.10.5.tgz -C app/
重命名：mv scala-2.10.5/ scala
配置到环境变量：
export SCALA_HOME=/home/uplooking/app/scala
export PATH=$PATH:$SCALA_HOME/bin
# 虽然spark本身自带scala，但还是建议安装

2.安装单机版spark
解压：tar -zxvf soft/spark-1.6.2-bin-hadoop2.6.tgz -C app/
重命名：mv spark-1.6.2-bin-hadoop2.6/ spark
配置到环境变量：
export SPARK_HOME=/home/uplooking/app/spark
export PATH=$PATH:$SPARK_HOME/bin:$SPARK_HOME/sbin
测试：
运行一个简单的spark程序
spark-shell
sc.textFile("./hello").flatMap(_.split(" ")).map((_, 1)).reduceByKey(_+_).collect.foreach(println)

完全分布式安装

修改spark-env.sh
    1、cd /home/uplooking/app/spark/conf
    2、cp spark-env.sh.template spark-env.sh
    3、vi spark-env.sh
    export JAVA_HOME=/opt/jdk
    export SCALA_HOME=/home/uplooking/app/scala
    export SPARK_MASTER_IP=uplooking01
    export SPARK_MASTER_PORT=7077
    export SPARK_WORKER_CORES=1
    export SPARK_WORKER_INSTANCES=1
    export SPARK_WORKER_MEMORY=1g
    export HADOOP_CONF_DIR=/home/uplooking/app/hadoop/etc/hadoop
修改slaves配置文件
    添加两行记录
    uplooking02
    uplooking03
部署到uplooking02和uplooking03这两台机器上(这两台机器需要提前安装scala)
    scp -r /home/uplooking/app/scala [email protected]:/home/uplooking/app
    scp -r /home/uplooking/app/scala [email protected]:/home/uplooking/app
    ----
    scp -r /home/uplooking/app/spark [email protected]:/home/uplooking/app
    scp -r /home/uplooking/app/spark [email protected]:/home/uplooking/app
    ---在uplooking02和uplooking03上加载好环境变量,需要source生效
    scp /home/uplooking/.bash_profile [email protected]:/home/uplooking
    scp /home/uplooking/.bash_profile [email protected]:/home/uplooking
启动
    修改事宜
        为了避免和hadoop中的start/stop-all.sh脚本发生冲突，将spark/sbin/start/stop-all.sh重命名
        mv start-all.sh start-spark-all.sh
        mv stop-all.sh stop-spark-all.sh
    启动
        sbin/start-spark-all.sh
        会在我们配置的主节点uplooking01上启动一个进程Master
        会在我们配置的从节点uplooking02上启动一个进程Worker
        会在我们配置的从节点uplooking03上启动一个进程Worker
    简单的验证
        启动spark-shell
        bin/spark-shell
        scala> sc.textFile("hdfs://ns1/data/hello").flatMap(_.split(" ")).map((_, 1)).reduceByKey(_+_).collect.foreach(println)
        我们发现spark非常快速的执行了这个程序，计算出我们想要的结果

    一个端口:8080/4040
        8080-->spark集群的访问端口，类似于hadoop中的50070和8088的综合
        4040-->sparkUI的访问地址
        7077-->hadoop中的9000端口

基于zookeeper的HA配置

最好在集群停止的时候来做
第一件事
    注释掉spark-env.sh中两行内容
        #export SPARK_MASTER_IP=uplooking01
        #export SPARK_MASTER_PORT=7077
第二件事
    在spark-env.sh中加一行内容
        export SPARK_DAEMON_JAVA_OPTS="-Dspark.deploy.recoveryMode=ZOOKEEPER -Dspark.deploy.zookeeper.url=uplooking01:2181,uplooking02:2181,uplooking03:2181 -Dspark.deploy.zookeeper.dir=/spark"
    解释
        spark.deploy.recoveryMode设置成 ZOOKEEPER
        spark.deploy.zookeeper.urlZooKeeper URL
        spark.deploy.zookeeper.dir ZooKeeper 保存恢复状态的目录，缺省为 /spark
重启集群
    在任何一台spark节点上启动start-spark-all.sh
    手动在集群中其他从节点上再启动master进程：sbin/start-master.sh -->在uplooking02
通过浏览器方法 uplooking01:8080 /uplooking02:8080-->Status: STANDBY Status: ALIVE
    验证HA，只需要手动停掉master上spark进程Master，等一会slave01上的进程Master状态会从STANDBY编程ALIVE

# 注意，如果在uplooking02上启动，此时uplooking02也会是master，而uplooking01则都不是，
# 因为配置文件中并没有指定master，只指定了slave
# spark-start-all.sh也包括了start-master.sh的操作，所以才会在该台机器上也启动master.

Spark源码编译

安装好maven后，并且配置好本地的spark仓库（不然编译时依赖从网上下载会很慢），
然后就可以在spark源码目录执行下面的命令：
mvn -Pyarn -Dhadoop.version=2.6.4 -Dyarn.version=2.6.4 -DskipTests clean package

编译成功后输出如下：

......
[INFO] ------------------------------------------------------------------------
[INFO] Reactor Summary:
[INFO] 
[INFO] Spark Project Parent POM ........................... SUCCESS [  3.617 s]
[INFO] Spark Project Test Tags ............................ SUCCESS [ 17.419 s]
[INFO] Spark Project Launcher ............................. SUCCESS [ 12.102 s]
[INFO] Spark Project Networking ........................... SUCCESS [ 11.878 s]
[INFO] Spark Project Shuffle Streaming Service ............ SUCCESS [  7.324 s]
[INFO] Spark Project Unsafe ............................... SUCCESS [ 16.326 s]
[INFO] Spark Project Core ................................. SUCCESS [04:31 min]
[INFO] Spark Project Bagel ................................ SUCCESS [ 11.671 s]
[INFO] Spark Project GraphX ............................... SUCCESS [ 55.420 s]
[INFO] Spark Project Streaming ............................ SUCCESS [02:03 min]
[INFO] Spark Project Catalyst ............................. SUCCESS [02:40 min]
[INFO] Spark Project SQL .................................. SUCCESS [03:38 min]
[INFO] Spark Project ML Library ........................... SUCCESS [03:56 min]
[INFO] Spark Project Tools ................................ SUCCESS [ 15.726 s]
[INFO] Spark Project Hive ................................. SUCCESS [02:30 min]
[INFO] Spark Project Docker Integration Tests ............. SUCCESS [ 11.961 s]
[INFO] Spark Project REPL ................................. SUCCESS [ 42.913 s]
[INFO] Spark Project YARN Shuffle Service ................. SUCCESS [  8.391 s]
[INFO] Spark Project YARN ................................. SUCCESS [ 42.013 s]
[INFO] Spark Project Assembly ............................. SUCCESS [02:06 min]
[INFO] Spark Project External Twitter ..................... SUCCESS [ 19.155 s]
[INFO] Spark Project External Flume Sink .................. SUCCESS [ 22.164 s]
[INFO] Spark Project External Flume ....................... SUCCESS [ 26.228 s]
[INFO] Spark Project External Flume Assembly .............. SUCCESS [  3.838 s]
[INFO] Spark Project External MQTT ........................ SUCCESS [ 33.132 s]
[INFO] Spark Project External MQTT Assembly ............... SUCCESS [  7.937 s]
[INFO] Spark Project External ZeroMQ ...................... SUCCESS [ 17.900 s]
[INFO] Spark Project External Kafka ....................... SUCCESS [ 37.597 s]
[INFO] Spark Project Examples ............................. SUCCESS [02:39 min]
[INFO] Spark Project External Kafka Assembly .............. SUCCESS [ 10.556 s]
[INFO] ------------------------------------------------------------------------
[INFO] BUILD SUCCESS
[INFO] ------------------------------------------------------------------------
[INFO] Total time: 31:22 min
[INFO] Finished at: 2018-04-24T18:33:58+08:00
[INFO] Final Memory: 89M/1440M
[INFO] ------------------------------------------------------------------------

然后就可以在下面的目录中看到编译成功的文件：

[[email protected] scala-2.10]$ pwd
/home/uplooking/compile/spark-1.6.2/assembly/target/scala-2.10
[[email protected] scala-2.10]$ ls -lh
总用量 135M
-rw-rw-r-- 1 uplooking uplooking 135M 4月  24 18:28 spark-assembly-1.6.2-hadoop2.6.4.jar

在已经安装的spark的lib目录下也可以看到该文件：

[[email protected] lib]$ ls -lh
总用量 291M
-rw-r--r-- 1 uplooking uplooking 332K 6月  22 2016 datanucleus-api-jdo-3.2.6.jar
-rw-r--r-- 1 uplooking uplooking 1.9M 6月  22 2016 datanucleus-core-3.2.10.jar
-rw-r--r-- 1 uplooking uplooking 1.8M 6月  22 2016 datanucleus-rdbms-3.2.9.jar
-rw-r--r-- 1 uplooking uplooking 6.6M 6月  22 2016 spark-1.6.2-yarn-shuffle.jar
-rw-r--r-- 1 uplooking uplooking 173M 6月  22 2016 spark-assembly-1.6.2-hadoop2.6.0.jar
-rw-r--r-- 1 uplooking uplooking 108M 6月  22 2016 spark-examples-1.6.2-hadoop2.6.0.jar