spark安装与部署 spark概述
- spark平台结构
文章图片
spark统一栈 - spark官网
- 下载配置jdk, scala, sbt, maven;
- 下载配置spark
- 修改~/.bash_profile
export JAVA_HOME=$HOME/jdk1.7.0_79`export PATH=$JAVA_HOME/bin:$PATH`export CLASSPATH=.:$JAVA_HOME/lib/dt.jar:$JAVA_HOME/lib/tools.jarexport SCALA_HOME=$HOME/scala/scala-2.10.5export SPARK_HOME=$HOME/spark-1.4.0-bin-hadoop2.6export HADOOP_HOME=$HOME/hadoop-2.6.0export HADOOP_CONF_DIR=$HOME/hadoop-2.6.0/etc/hadoopexport MAVEN_HOME=$HOME/apache-maven-3.2.5export SBT_HOME=$HOME/sbtexport PATH=$PATH:$SCALA_HOME/bin:$SPARK_HOME/bin:$HADOOP_HOME/bin:$HADOOP_HOME/sbin:$MAVEN_HOME/bin:$SBT_HOME/bin source ~/.bash_profile
- 准备三台机器，系统为ubuntu 14.04LTS，配置好ip，共享文件夹（其中可能发生权限不够的问题，需要将当前用户加入到vsboxsf组下面去），mac下可以免密码登录三台机器，分别为: gg01, ggg02, ggg03。
- 准备软件，需要jdk，sbt，scala，maven，spark
- 配置路径：
export JAVA_HOME=$HOME/jdk1.7.0_79
export CLASSPATH=.:$JAVA_HOME/lib/dt.jar:$JAVA_HOME/lib/tools.jar
export SCALA_HOME=$HOME/scala-2.10.5
export SPARK_HOME=$HOME/spark-1.4.0-bin-hadoop2.6
export MAVEN_HOME=$HOME/apache-maven-3.2.5
export SBT_HOME=$HOME/sbt
export PATH=$PATH:$SCALA_HOME/bin:$SPARK_HOME/bin:$MAVEN_HOME/bin:$SBT_HOME/bin
```
然后
source ~/.bashrc，查看软件是否准备好了（java -version之类的）- 修改spark的环境，添加spark－env.sh，添加路径
export SCALA_HOME=/home/sunyonggang/scala-2.10.5
export SPARK_WORKER_MEMORY=1G
export JAVA_HOME=/home/sunyonggang/jdk1.7.0_79
```
然后给添加的slaves文件添加主机
gg01，然后可以直接启动了。- 将
gg01的配置复制到ggg02, ggg03中去，启动集群
sunyonggang@gg01:~/spark-1.4.0-bin-hadoop2.6$ jps
5498 Jps
5374 Worker
5174 Master
sunyonggang@ggg02:~/spark-1.4.0-bin-hadoop2.6/conf$ jps
4055 Jps
3979 Worker
查看只有一个worker，那是由于我集群中的三台机器的eth0的ip地址都是一样的，将虚拟机改为桥接网络，自定义ip。
- Ubuntu使用固定ip：将网络连接换为桥接方式，发现gg01的ip地址变为了192.168.199.255（这个是broadcast地址重复），需要修改，另外两台没有问题，可以直接使用，但为了统一，建议修改ip地址，具体查看ubuntu如何固定ip（已保存evernote，或者自己搜索下）。
- 启动过程中还是只有一个worker，查看错误：
```
16/04/06 19:14:19 INFO Worker: Retrying connection to master (attempt # 1)
【spark安装与部署】16/04/06 19:14:19 INFO Worker: Connecting to master akka.tcp://sparkMaster@gg01:7077/user/Master...
16/04/06 19:14:19 WARN Remoting: Tried to associate with unreachable remote address [akka.tcp://sparkMaster@gg01:7077]. Address is now gated for 5000 ms, all messages to this address will be delivered to dead letters. Reason: Connection refused: gg01/192.168.199.150:7077
```
首先排除ssh的问题，然后修改hosts文件中的本机对应的ip，对于gg01来说：
>
>192.168.199.150 gg01
重新启动
文章图片
master alive 11.配置好zk，启动各个节点的zk：
sunyonggang@gg01:~/zookeeper-3.4.6$ jps
7808 QuorumPeerMain
7833 Jps
12.修改spark的配置文件，将指定master改为zookeeper选举模式
export SCALA_HOME=/home/sunyonggang/scala-2.10.5
.#export SPARK_MASTER_IP=gg01
export SPARK_WORKER_MEMORY=1G
export JAVA_HOME=/home/sunyonggang/jdk1.7.0_79
export SPARK_DAEMON_JAVA_OPTS="-Dspark.deploy.recoveryMode=ZOOKEEPER -Dspark.deploy.zookeeper.url=gg01:2181,ggg02:2181,ggg03:2181 -Dspark.deploy.zookeeper.dir=/spark"
配置好zk之后一定要注意启动zk，如果不启动，直接启动spark就报错。
13.在gg01中启动集群，在ggg03中启动standby的master服务
文章图片
master alive
文章图片
standy by 然后kill掉gg01中的master进程
文章图片
stand by to master 到此，Spark HA搭建完毕。
14.使用spark－submit提交一个任务，这边我选用的是examples中的一个PI的计算：
sunyonggang@ggg03:~/spark-1.4.0-bin-hadoop2.6/examples$ cat simple.sbt
name := "Example"version := "1.0"scalaVersion := "2.10.5"libraryDependencies += "org.apache.spark" %% "spark-core" % "1.4.0"sunyonggang@ggg03:~/spark-1.4.0-bin-hadoop2.6/examples$ sbt package
[info] Set current project to Example (in build file:/home/sunyonggang/spark-1.4.0-bin-hadoop2.6/examples/)
[info] Updating {file:/home/sunyonggang/spark-1.4.0-bin-hadoop2.6/examples/}examples...
[info] Resolving org.fusesource.jansi#jansi;
1.4 ...
[info] Done updating.
[info] Compiling 1 Scala source to /home/sunyonggang/spark-1.4.0-bin-hadoop2.6/examples/target/scala-2.10/classes...
[info] Packaging /home/sunyonggang/spark-1.4.0-bin-hadoop2.6/examples/target/scala-2.10/example_2.10-1.0.jar ...
[info] Done packaging.
[success] Total time: 53 s, completed Apr 7, 2016 12:19:06 PM
15.运行jar包：
./bin/spark-submit --class org.apache.spark.examples.SparkPi --master spark://gg01:7077 /home/sunyonggang/spark-1.4.0-bin-hadoop2.6/examples/target/scala-2.10/example_2.10-1.0.jar运行结果：
16/04/07 12:27:25 INFO BlockManagerMasterEndpoint: Registering block manager 192.168.199.150:48388 with 267.3 MB RAM, BlockManagerId(2, 192.168.199.150, 48388)
16/04/07 12:27:26 INFO BlockManagerInfo: Added broadcast_0_piece0 in memory on 192.168.199.146:35760 (size: 1202.0 B, free: 267.3 MB)
16/04/07 12:27:26 INFO BlockManagerInfo: Added broadcast_0_piece0 in memory on 192.168.199.150:48388 (size: 1202.0 B, free: 267.3 MB)
16/04/07 12:27:26 INFO TaskSetManager: Finished task 0.0 in stage 0.0 (TID 0) in 1176 ms on 192.168.199.146 (1/2)
16/04/07 12:27:26 INFO DAGScheduler: ResultStage 0 (reduce at SparkPi.scala:35) finished in 1.846 s
16/04/07 12:27:26 INFO TaskSetManager: Finished task 1.0 in stage 0.0 (TID 1) in 1094 ms on 192.168.199.150 (2/2)
16/04/07 12:27:26 INFO DAGScheduler: Job 0 finished: reduce at SparkPi.scala:35, took 2.715414 s
16/04/07 12:27:26 INFO TaskSchedulerImpl: Removed TaskSet 0.0, whose tasks have all completed, from pool
Pi is roughly 3.14122
16/04/07 12:27:26 INFO SparkUI: Stopped Spark web UI at http://192.168.199.145:4040
