Spark集群架设详解

在这篇文章里，威廉将展示如何构建一个简单的Spark集群

我们采用VMware Workstation搭建数台虚拟机，具体信息如下：

物理机操作系统：Windows 7 Professional Service Pack 1 64位
虚拟机操作系统：Ubuntu 14.04 LTS 64位
虚拟机网络设置：
- 连接方式：NAT
- 子网IP：192.168.32.0
- 网关：192.168.32.2
- DNS：192.168.32.2

简单的Spark集群由至少一台Master和数台Slave构成；在本文中尝试构建3台虚拟机，1台用作Master，另外2台作Slave

接下来威廉就带大家详细的介绍部署的步骤

1. 虚拟机安装

在创建虚拟机前，先修改下Workstation的默认网络设置，点击编辑->虚拟网络编辑器

在这里需要注意的是，默认VMnet8网卡，也就是用于和物理机通信的网卡，是打开了DHCP服务的，我们需要按下图所示将其关闭，否则在后面设置静态IP的时候会产生冲突

下载虚拟机镜像，在VMware Workstation中创建虚拟机，威廉选取的是Ubuntu 14.04 LTS 64位的操作系统版本，大家可以自由选择，但注意如果是RedHat系列的Linux系统，在之后步骤用到的命令会有所不同

在本文中，我们一共需要创建3台虚拟机，1台用作master，2台用作slave，并修改相应的hostname为master和slave1，slave2

1.1 hostname配置

关于如何修改hostname，在man hostname中我们可以找到有这么一段：

The host name is usually set once at system startup in /etc/init.d/hostname.sh
 (normally by reading the contents of a file which contains the host name, 
 e.g.  /etc/hostname).

所以我们可以通过修改/etc/hostname文件来改变默认hostname，重启使之生效

1.2 创建统一用户

为所有虚拟机建立统一的用户，我们选取spark

sudo adduser spark

注：新创建的spark用户在使用sudo时可能出现not in the sudoers file的错误，解决方式参考1

1.3 JDK安装及JAVA_HOME设置

下载JDK并解压，编辑/home/spark/.profile文件，追加以下内容：

# set JAVA_HOME if not exist
if [ -n "$JAVA_HOME" ]; then
    JAVA_HOME="/home/spark/jdk1.7.0_75"
    export JAVA_HOME

    PATH="$JAVA_HOME:$PATH"
fi

2. 网络配置

2.1 固定IP配置

在spark集群中，master与slave之间将通过ssh相互通信，因此每台主机需要有固定的IP地址，并对应到相应的hostname

编辑/etc/network/interfaces，以master为例，追加以下内容：

# The primary network interface
auto eth0
iface eth0 inet static
address 192.168.32.130
netmask 255.255.255.0
gateway 192.168.32.2
dns-nameservers 192.168.32.2

输入以下命令使改动生效

sudo ifconfig eth0 down
sudo ifconfig eth0 up

slave主机修改的内容相似，只需把address改为192.168.32.131，192.168.32.132

2.2 hosts配置

编辑/etc/hosts，追加如下：

# spark hosts
192.168.32.130  master
192.168.32.131  slave1
192.168.32.132  slave2

输入以下命令测试修改是否生效

spark@master:~$ ping slave1
PING slave1 (192.168.32.131) 56(84) bytes of data.

3. ssh配置

检查ssh服务是否启动

ps -e | grep sshd

若没有启动，则需要先启动ssh服务

sudo service ssh start
# 若没有安装ssh服务器
sudo apt-get install openssh-server

生成ssh公钥密钥，密码为空，将会在/home/spark/.ssh文件夹下生成id_dsa(密钥), id_dsa.pub(公钥), authorized_keys

ssh-keygen -t dsa
# 密码留空

将master主机的id_dsa.pub内容追加到所有slave主机的authorized_keys文件内，通过以下命令测试master能否ssh无密码登陆所有slave

spark@master:~$ ssh slave1
Welcome to Ubuntu 14.04.1 LTS (GNU/Linux 3.13.0-32-generic x86_64)

 * Documentation:  https://help.ubuntu.com/

399 packages can be updated.
173 updates are security updates.

Last login: Sun May 10 07:09:45 2015 from master
spark@slave1:~$ 

4. spark安装

登陆 https://spark.apache.org/downloads.html 下载对应的spark版本， 我们选择下载预编译版本spark-1.3.1-bin-hadoop2.6.tgz，解压到任意不含空格的目录，需保证每台主机都采用相同的目录

配置master主机的$SPARK_HOME/conf/slaves文件，写入所有slave node的hostname

slave1
slave2

配置所有主机的$SPARK_HOME/conf/spark-env.sh，写入$JAVA_HOME2和$SPARK_MASTER_IP3

export JAVA_HOME=/home/spark/jdk1.7.0_75
export SPARK_MASTER_IP=192.168.32.130

5. 启动spark

$SPARK_HOME/sbin/start-all.sh

查看master主机$SPARK_HOME/logs/spark-spark-org.apache.spark.deploy.master.Master-1-master.out

Spark Command: /home/spark/jdk1.7.0_75/bin/java -cp /home/spark/spark-1.3.1-bin-hadoop2.6/sbin/../conf:/home/spark/spark-1.3.1-bin-hadoop2.6/lib/spark-assembly-1.3.1-hadoop2.6.0.jar:/home/spark/spark-1.3.1-bin-hadoop2.6/lib/datanucleus-core-3.2.10.jar:/home/spark/spark-1.3.1-bin-hadoop2.6/lib/datanucleus-rdbms-3.2.9.jar:/home/spark/spark-1.3.1-bin-hadoop2.6/lib/datanucleus-api-jdo-3.2.6.jar -XX:MaxPermSize=128m -Dspark.akka.logLifecycleEvents=true -Xms512m -Xmx512m org.apache.spark.deploy.master.Master --ip 192.168.32.130 --port 7077 --webui-port 8080
========================================

Using Spark's default log4j profile: org/apache/spark/log4j-defaults.properties
15/05/11 05:45:58 INFO Master: Registered signal handlers for [TERM, HUP, INT]
15/05/11 05:45:58 WARN Utils: Your hostname, master resolves to a loopback address: 127.0.1.1; using 192.168.32.130 instead (on interface eth0)
15/05/11 05:45:58 WARN Utils: Set SPARK_LOCAL_IP if you need to bind to another address
15/05/11 05:46:04 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
15/05/11 05:46:07 INFO SecurityManager: Changing view acls to: spark
15/05/11 05:46:07 INFO SecurityManager: Changing modify acls to: spark
15/05/11 05:46:07 INFO SecurityManager: SecurityManager: authentication disabled; ui acls disabled; users with view permissions: Set(spark); users with modify permissions: Set(spark)
15/05/11 05:46:15 INFO Slf4jLogger: Slf4jLogger started
15/05/11 05:46:18 INFO Remoting: Starting remoting
15/05/11 05:46:22 INFO Remoting: Remoting started; listening on addresses :[akka.tcp://sparkMaster@192.168.32.130:7077]
15/05/11 05:46:22 INFO Remoting: Remoting now listens on addresses: [akka.tcp://sparkMaster@192.168.32.130:7077]
15/05/11 05:46:23 INFO Utils: Successfully started service 'sparkMaster' on port 7077.
15/05/11 05:46:34 INFO Server: jetty-8.y.z-SNAPSHOT
15/05/11 05:46:36 INFO AbstractConnector: Started SelectChannelConnector@master:6066
15/05/11 05:46:36 INFO Utils: Successfully started service on port 6066.
15/05/11 05:46:36 INFO StandaloneRestServer: Started REST server for submitting applications on port 6066
15/05/11 05:46:36 INFO Master: Starting Spark master at spark://192.168.32.130:7077
15/05/11 05:46:36 INFO Master: Running Spark version 1.3.1
15/05/11 05:46:37 INFO Server: jetty-8.y.z-SNAPSHOT
15/05/11 05:46:37 INFO AbstractConnector: Started SelectChannelConnector@0.0.0.0:8080
15/05/11 05:46:37 INFO Utils: Successfully started service 'MasterUI' on port 8080.
15/05/11 05:46:37 INFO MasterWebUI: Started MasterWebUI at http://master:8080
15/05/11 05:46:53 INFO Master: Registering worker slave2:60027 with 1 cores, 512.0 MB RAM
15/05/11 05:46:53 INFO Master: Registering worker slave1:59125 with 1 cores, 512.0 MB RAM

查看slave主机$SPARK_HOME/logs/spark-spark-org.apache.spark.deploy.worker.Worker-1-slave1.out

Spark Command: /home/spark/jdk1.7.0_75/bin/java -cp /home/spark/spark-1.3.1-bin-hadoop2.6/sbin/../conf:/home/spark/spark-1.3.1-bin-hadoop2.6/lib/spark-assembly-1.3.1-hadoop2.6.0.jar:/home/spark/spark-1.3.1-bin-hadoop2.6/lib/datanucleus-core-3.2.10.jar:/home/spark/spark-1.3.1-bin-hadoop2.6/lib/datanucleus-api-jdo-3.2.6.jar:/home/spark/spark-1.3.1-bin-hadoop2.6/lib/datanucleus-rdbms-3.2.9.jar -XX:MaxPermSize=128m -Dspark.akka.logLifecycleEvents=true -Xms512m -Xmx512m org.apache.spark.deploy.worker.Worker spark://192.168.32.130:7077
========================================

Using Spark's default log4j profile: org/apache/spark/log4j-defaults.properties
15/05/11 08:24:27 INFO Worker: Registered signal handlers for [TERM, HUP, INT]
15/05/11 08:24:27 WARN Utils: Your hostname, slave1 resolves to a loopback address: 127.0.1.1; using 192.168.32.131 instead (on interface eth0)
15/05/11 08:24:27 WARN Utils: Set SPARK_LOCAL_IP if you need to bind to another address
15/05/11 08:24:30 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
15/05/11 08:24:31 INFO SecurityManager: Changing view acls to: spark
15/05/11 08:24:31 INFO SecurityManager: Changing modify acls to: spark
15/05/11 08:24:31 INFO SecurityManager: SecurityManager: authentication disabled; ui acls disabled; users with view permissions: Set(spark); users with modify permissions: Set(spark)
15/05/11 08:24:33 INFO Slf4jLogger: Slf4jLogger started
15/05/11 08:24:33 INFO Remoting: Starting remoting
15/05/11 08:24:34 INFO Remoting: Remoting started; listening on addresses :[akka.tcp://sparkWorker@slave1:51940]
15/05/11 08:24:34 INFO Remoting: Remoting now listens on addresses: [akka.tcp://sparkWorker@slave1:51940]
15/05/11 08:24:34 INFO Utils: Successfully started service 'sparkWorker' on port 51940.
15/05/11 08:24:35 INFO Worker: Starting Spark worker slave1:51940 with 1 cores, 512.0 MB RAM
15/05/11 08:24:35 INFO Worker: Running Spark version 1.3.1
15/05/11 08:24:35 INFO Worker: Spark home: /home/spark/spark-1.3.1-bin-hadoop2.6
15/05/11 08:24:35 INFO Server: jetty-8.y.z-SNAPSHOT
15/05/11 08:24:35 INFO AbstractConnector: Started SelectChannelConnector@0.0.0.0:8081
15/05/11 08:24:35 INFO Utils: Successfully started service 'WorkerUI' on port 8081.
15/05/11 08:24:35 INFO WorkerWebUI: Started WorkerWebUI at http://slave1:8081
15/05/11 08:24:36 INFO Worker: Connecting to master akka.tcp://sparkMaster@192.168.32.130:7077/user/Master...
15/05/11 08:24:37 INFO Worker: Successfully registered with master spark://192.168.32.130:7077

spark提供了UI用于管理集群，登陆192.168.32.130:8080，我们看到两台slave主机已成功连接

slave主机也有类似的UI用来管理任务，登陆192.168.32.131:8011

至此，威廉给大家展示了如何创建简单的spark集群，当在配置过程中遇到问题时，可以向spark的官方邮件组[email protected]提问题，或者访问这个较为便捷的网页http://apache-spark-user-list.1001560.n3.nabble.com/，再者就是万能的谷歌，度娘啦

---

1. 默认只有随系统创建的用户有sudo权限，编辑/etc/sudoers文件，定位到root ALL=(ALL:ALL) ALL行，在下面添加一行spark ALL=(ALL:ALL) ALL ↩
2. 若不配置JAVA_HOME，在启动spark时slave node会出现JAVA_HOME not found 的问题 ↩
3. 若不配置SPARK_MASTER_IP，在启动spark时slave node会出现 Connection refused: master/192.168.32.130:7077 的问题 ↩