基于Hadoop2.2.0安装Spark 1.0

基于Hadoop2.2.0安装spark1.0

 

 

转至元数据结尾

 

• Created by spark.long 昨天7:33 上午

转至元数据起始

 

基于Hadoop2.2.0安装Spark 1.0

基础

服务器的版本：ubuntu 12.04

JDK

下载

tar -zxvf ... -C /usr/local
ln -s /usr/local/jdk1.7.0_60 /usr/local/jvm

设置环境变量

export JAVA_HOME=/usr/local/jdk
export JRE_HOME=${JAVA_HOME}/jre 
export CLASSPATH=.:${JAVA_HOME}/lib:${JRE_HOME}/lib 
export PATH=$JAVA_HOME/bin:$PATH

Scala

下载

wget http://www.scala-lang.org/files/archive/scala-2.10.3.tgz
tar -zxf scala-2.10.3.tgz -C /usr/local/
ln -s /usr/local/scala-2.10.3 /usr/local/scala

设置环境变量

export SCALA_HOME=/usr/local/scala
export PATH=$SCALA_HOME/bin:$PATH

maven

下载

#wget http://mirror.bit.edu.cn/apache/maven/maven-3/3.1.1/binaries/apache-maven-3.1.1-bin.tar.gz

1. tar xvzf apache-maven-3.1.1-bin.tar.gz -C /usr/local # ln -s /usr/local/apache-maven-3.1.1 /usr/local/maven

   设置环境变量

   export MAVEN_HOME=/usr/local/maven
   export PATH=$MAVEN_HOME/bin:$PATH

   PROTOBUF 安装（主节点）

   安装依赖包

   apt-get install g++ autoconf automake libtool cmake zlib1g-dev pkg-config libssl-dev make

   下载安装

   wget https://protobuf.googlecode.com/files/protobuf-2.5.0.tar.gz
2. tar xvzf protobuf-2.5.0.tar.gz# ./configure --prefix=/usr/local/protobuf# make && make install

   设置环境变量

   export PATH=/usr/local/protobuf/bin:$PATH

   Hadoop2.2.0安装

   基础

   http://blog.changecong.com/2013/10/ubuntu-%E7%BC%96%E8%AF%91%E5%AE%89%E8%A3%85-hadoop-2-2-0/ http://blog.csdn.net/licongcong_0224/article/details/12972889

   创建用户并设置无密码登录

   nanenode

   
   useradd -s /bin/bash -m hadoop
   mkdir -p /home/hadoop/.ssh
   chown hadoop.hadoop /home/hadoop/.ssh 
   su - hadoop
   ssh-keygen -t rsa 
   #如果主节点也当datanode的话，
   执行：cat id_rsa.pub > authorized_keys 
   

   datanode

   useradd -s /bin/bash -m hadoop
   mkdir -p /home/hadoop/.ssh
   chown hadoop.hadoop /home/hadoop/.ssh 
   
   echo "主节点id_rsa.pub的内容" > id_rsa.pub
   cat /home/hadoop/.ssh/id_rsa.pub >> /home/hadoop/.ssh/authorized_keys 
   
   

   每个节点关闭防火墙

   ufw disable

   Hadoop安装问题汇总

   [ERROR] class file for org.mortbay.component.AbstractLifeCycle not found

        解决方法：编辑hadoop-common-project/hadoop-auth/pom.xml文件，添加依赖：
        <dependency>
         <groupId>org.mortbay.jetty</groupId>       <artifactId>jetty-util</artifactId>      <scope>test</scope>    </dependency>

   注意hosts，192.168.137.100 namenode对应

   Spark1.0安装

   下载源码 spark1.0

   进入下载网页：http://spark.apache.org/downloads.html
   wget http://d3kbcqa49mib13.cloudfront.net/spark-1.0.0.tgz

   配置

   解压、配置路径

   sudo tar zxf spark-1.0.0.tgz
   sudo cp spark-1.0.0 /usr/local/ -rf
   cd /usr/local
   sudo ln -s spark-1.0.0/ spark 
   

   设置变量

   sudo /etc/profile:
   export SPARK_HOME=/usr/local/spark 
   export SCALA_HOME=/usr/local/scala
   export PATH=$SCALA_HOME/bin:$PATH
   source /etc/profile 
   

   组件依赖

   apt-get install unzip

   build Spark

   cd $SPARK_HOME
   export MAVEN_OPTS="-Xmx2g -XX:MaxPermSize=512M -XX:ReservedCodeCacheSize=512m"
   #mvn -Dyarn.version=2.2.0 -Dhadoop.version=2.2.0 -Pnew-yarn -DskipTests package
   mvn -Pyarn -Phadoop-2.2 -Dhadoop.version=2.2.0 -DskipTests clean package 
   #sudo ./sbt/sbt assembly
   SPARK_HADOOP_VERSION=2.2.0 SPARK_YARN=true ./sbt/sbt assembly 
   

   将编译后的文件分发到各数据节点中

   1../bin/ 
   2../sbin/
   3. ./assemble/...
   4../conf/ 
   scp -r spark/sbin [email protected]:/home/hadoop/spark
   scp -r spark/bin [email protected]:/home/hadoop/spark
   scp -r spark/conf [email protected]:/home/hadoop/spark
   scp -r spark/assembly/target/scala-2.10/spark-assembly-1.0.0-hadoop2.2.0.jar [email protected]:/home/hadoop/spark/assembly/target/scala-2.10

   测行测试

   build成功后，会在spark目录下生成两个文件，分别为：
   1. examples/target/scala-2.10/spark-examples-1.0.0-hadoop2.2.0.jar
   2. assembly/target/scala-2.10/spark-assembly-1.0.0-hadoop2.2.0.jar

   配置运行脚本

   1. sudo cp conf/spark-env.sh.template conf/spark-env.sh
   并添加如下内容：
   export JAVA_HOME=/usr/local/jvm/java
   export SCALA_HOME=/usr/local/scala
   export HADOOP_HOME=/home/hadoop/hadoop 
   2. sudo cp conf/log4j.properties.template conf/log4j.properties 
   3.cd $SPARK_HOME 
   vi conf/slaves
   添加数据节点：datanode1

   配置Spark路径

   vi /etc/profile
   export SPARK_HOME=/usr/local/spark
   export PATH=$PATH:$SPARK_HOME/bin 
   source /etc/profile

   运行测试

   本地模式

   普通集群模式

   结合HDFS模式

   ./bin/spark-shell 
   var file = sc.textFile("hdfs://namenode:9000/user/spark/hdfs.cmd");
   val count = file.flatMap(line => line.split(" ")).map(word => (word, 1)).reduceByKey(+)
   count.collect()

   基于Yarn模式

   在Spark目录mkdir test && cd test
   vi run_spark_shell.sh 
   SPARK_JAR=../assembly/target/scala-2.10/spark-assembly-1.0.0-hadoop2.2.0.jar \
   spark-class org.apache.spark.deploy.yarn.Client \
   --jar ../examples/target/scala-2.10/spark-examples-1.0.0-hadoop2.2.0.jar \
   --class org.apache.spark.examples.SparkPi \
   --num-workers 3 \
   --master-memory 1g \
   --worker-memory 1g \
   --worker-cores 1 
   
   

   运行状态监控

   在启动./bin/spark-she ll时，会显示如下日志：
   14/06/23 14:36:43 INFO SparkUI: Started SparkUI at http://namenode:4040
   则可访问：http://192.168.137.100:4040/

   Shark安装

   添加DataNode

   无密码登录

   修改hostname及hadoop/slaves,spark/slaves文件

   同步hadoop文件

   同步spark文件

   同步环境变量/etc/profile

   安装问题汇总

   spark目录必须为hadoop权限下的目录

   chown hadoop.hadoop /usr/local/spark

   SCALA_HOME not set

   build Spark的时候，用的是Spark帐户，Scala_Home是有设置的，但还会报这个错误，但切换到root帐户下，就没有出现这个问题。 
   

   JAVA_HOME is not set

   在conf/ spark-env.sh上添加export JAVA_HOME=/usr/local/jvm/java即可

   wrap: java.lang.reflect.InvocationTargetException: PermGen space

   出现这个错误，内存不够，应该是jvm参数引起的，给jvm分配的空间太小了。我用官方推荐的export MAVEN_OPTS="-Xmx2g -XX:MaxPermSize=512M -XX:ReservedCodeCacheSize=512m"
   ，还会报此错误，我改成MaxPermSize=1g，解决此问题。

   WARN NativeCodeLoader: Unable to load native-hadoop library for your platform

   
   

   WARN TaskSchedulerImpl: Initial job has not accepted any resources

   worker内存不足引起的，可以修改每个datanode的worker数量，和每个worker的内存 
   

   附录

   1. Spark On YARN 环境搭建：http://sofar.blog.51cto.com/353572/1352713