Hadoop2.7.1高可用环境搭建

Hadoop基础知识： http://hadoop.apache.org/docs/r1.0.4/cn/quickstart.html
注：本文中缺少的图可以下载附件DOC
Hadoop2.7.1集群搭建
1.系统配置
电脑1(Lenovo)，win7 64位系统，8G内存，此电脑虚拟机上运行name系统。
电脑1(Lenovo )，win7 64位系统，8G内存，此电脑虚拟机上运行standyname系统
电脑1(Lenovo)，win7 64位系统，8G内存，此电脑虚拟机上运行amrm系统
虚拟机：Vmware12.0
Hadoop2.7.1
Zookeeper3.4.6
2.集群规划
其具体规划如下：
JournalServer 应该单纯一台，slaves文件中为JournalNode（存储name的元数据）
journalServer and journalNode 中配置zookeeper，name和standy name
主机名 IP 安装软件 运行的进程

name	192.168.32.137	Jdk,hadoop
zookeeper	namenode、DFSZKFailoverController、datanode、jobhistorysever、NodeManager、JournalNode、QuorumPeerMain
sname	192.168.32.135	Jdk,hadoop
zookeeper	Namenode、DFSZKFailoverController,datanode、NodeManager、JournalNode、QuorumPeerMain
amrm	192.168.32.136	Jdk,hadoop
zookeeper	datanode、NodeManager、JournalNode、QuorumPeerMain,ResourceManager

说明：
在hadoop2.0中通常由两个name组成，一个处于active状态，另一个处于standby状态。Active name对外提供服务，而Standby name则不对外提供服务，仅同步activename的状态，以便能够在它失败时快速进行切换。hadoop2.0官方提供了两种HDFS HA的解决方案，一种是NFS，另一种是QJM。这里我们使用简单的QJM。在该方案中，主备name之间通过一组JournalNode同步元数据信息，一条数据只要成功写入多数JournalNode即认为写入成功。通常配置奇数个JournalNode
这里还配置了一个zookeeper集群，用于ZKFC（DFSZKFailoverController）故障转移，当
Active name挂掉了，会自动切换Standby name为standby状态。

1)在name,sname,amrm命令行vim /etc/hostname中分别设置name,sname,amrm的主机名，如下图所示：
 


2)在name,sname,amrm命令行vim /etc/hosts 中设置name,sname,amrm主机名和ip地址的对应关系，如下图所示：



3)验证各系统之间是否能够ping通。
4)安装SSH 并产生公私钥在name上：(可以copy ~/.ssh 到 sname和amrm，统一公私钥)

 ssh-keygen  -t  dsa  -P  ''  -f  ~/.ssh/id_dsa
 cat  ~/.ssh/id_dsa.pub  >>  ~/.ssh/authorized_keys

      拷贝公钥到sname，amrm做同样的动作（最好统一公私钥）

scp  -r  /root/.ssh   root@sname:/root/
scp  -r  /root/.ssh/id_dsa.pub  root@sname:/root/.ssh/id_dsa.pub
scp  -r  /root/.ssh/id_dsa.pub  root@amrm:/root/.ssh/id_dsa.pub

检查 ssh sname amrm 保证互相访问不需要密码 ,如果slaves文件中包括自己那么还要执行

 ssh name

---------------------------------------------------------------------------------------------------------------------------------------------
Scp 命令：

// scp from source to destination(local)
scp root@data:/root/.ssh/id_dsa.pub  ~/.ssh/data_dsa.pub
// scp from source(local) to destination
scp  -r  /root/.ssh/id_dsa.pub  root@amrm:/root/.ssh/id_dsa.pub

---------------------------------------------------------------------------------------------------------------------------------------------
注：scp 在ssh通的情况下用
错误：
   -1. Please contact your system administrator.
Add correct host key in /root/.ssh/known_hosts to get rid of this message.
Offending ECDSA key in /root/.ssh/known_hosts:2
  remove with:
  ssh-keygen -f "/root/.ssh/known_hosts" -R sname
执行：

ssh-keygen -f "/root/.ssh/known_hosts" -R sname

或删除

/root/.ssh/known_hosts

的第2行。

-2. Warning: the ECDSA host key for 'sname' differs from the key for the IP address '192.168.32.138'
Offending key for IP in /root/.ssh/known_hosts:2
Matching host key in /root/.ssh/known_hosts:5
Are you sure you want to continue connecting (yes/no)? yes
Welcome to Ubuntu 15.10 (GNU/Linux 4.2.0-16-generic x86_64)
* Documentation:  https://help.ubuntu.com/
82 packages can be updated.
42 updates are security updates.
Last login: Fri Dec 11 22:30:00 2015 from 192.168.32.138
解决：删除/root/.ssh/known_hosts的第2行。
-3. Your id_dsa is 755 cann’t used

chmod 700 ~/.ssh/id_dsa

(私钥文件权限)



5）关ip6
-1.

cat /proc/sys/net/ipv6/conf/all/disable_ipv6

       显示0说明ipv6开启，1说明关闭 

-2在 /etc/sysctl.conf 增加下面几行，并重启。

#disable IPv6
net.ipv6.conf.all.disable_ipv6 = 1
net.ipv6.conf.default.disable_ipv6 = 1
net.ipv6.conf.lo.disable_ipv6 = 1

     -3. sudo vim  /etc/default/grub
     -4. 将文件中的  

GRUB_CMDLINE_LINUX_DEFAULT="quiet spalsh" 

 
修改为
     

GRUB_CMDLINE_LINUX_DEFAULT="ipv6.disable=1 quiet splash"

     -5. wq保存后，运行sudo update-grub更新
     -6. 重启网络服务，禁用ipv6成功
       可以使用
                 

 ip a | grep inet6

       查看关闭情况，若没有结果则说明禁用IPv6成功

3.安装配置zookeeper集群
1)解压zookeeper压缩包到/hadoop

 tar –zxvf zookeeper-3.4.6.tar.gz   /hadoop
 mv /hadoop/zookeeper-3.4.6  /hadoop/zookeeper-3.4.6

2)在/hadoop/zookeeper-3.4.6/conf修改zookeeper配置zoo.cfg，具体配置如下图所示：




3)在/hadoop/zookeeper-3.4.6中设置创建tmp目录

Mkdir /hadoop/zookeeper-3.4.6/tmp

4)在/hadoop/zookeeper-3.4.6/tmp目录中创建空文件myid，并写入4

vim  /hadoop/zookeeper-3.4.6/tmp/myid。

5)将配置好的zookeeper拷贝到sname和amrm

scp -r /hadoop/zookeeper-3.4.6  root@sname:/hadoop/zookeeper-3.4.6
scp -r /hadoop/zookeeper-3.4.6  root@amrm:/hadoop/zookeeper-3.4.6

6)在sname和amrm中分别修改myid为2和3。
4.安装配置hadoop集群
1)解压hadoop压缩包到/hadoop

tar -zxvf hadoop-2.7.1.tar.gz  /hadoop

2)安装hadoop在~/.bashrc中配置hadoop的环境变量信息，如下图所示：

# the variable for hadoop
export JAVA_HOME=/usr/lib/java/jdk1.7.0_79
export JRE_HOME=${JAVA_HOME}/jre
export CLASS_PATH=.:${JAVA_HOME}/lib:${JRE_HOME}/lib
export HADOOP_HOME=/hadoop/hadoop-2.7.1
export ZOOKEEPER_HOME=/hadoop/zookeeper-3.4.6
export PATH=${JAVA_HOME}/bin:${HADOOP_HOME}/bin:${HADOOP_HOME}/sbin:${ZOOKEEPER_HOME}/bin:${PATH}
export HADOOP_MAPRED_HOME=${HADOOP_HOME}
export HADOOP_COMMON_HOME=${HADOOP_HOME}
export HADOOP_HDFS_HOME=${HADOOP_HOME}
export HADOOP_COMMON_LIB_NATIVE_DIR=${HADOOP}/lib/native
export YARN_HOME=${HADOOP_HOME}
export HADOOP_OPT="-Djava.library.path=${HADOOP_HOME}/lib/native"

5.配置hadoop
hadoop2.7.1的所有配置文件从存在/hadoop/hadoop-2.7.1/etc/hadoop之中。

cd /hadoop/hadoop-2.7.1/etc/hadoop

1)修改hadoop-env.sh 加入jdk家目录

export  JAVA_HOME=/usr/lib/java/jdk1.7.0_79

2)修改core-site.xml

<configuration>
		<!-- 指定hdfs的nameservice为ns -->
		<property>
			<name>fs.defaultFS</name>
			<value>hdfs://ns</value>
		</property>
		<!-- 指定hadoop临时目录  -->
		<property>
			<name>hadoop.tmp.dir</name>
			<value>/hadoop/tmp </value>
		</property>
		<!-- 指定zookeeper地址 -->
		<property>
			<name>ha.zookeeper.quorum</name>
			<value>name:2181,sname:2181,amrm:2181</value>
		</property>
	</configuration>

3)修改hdfs-site.xml  // 

<configuration>
		<!--指定hdfs的nameservice为ns，需要和core-site.xml中的保持一致 -->
		<property>
			<name>dfs.nameservices</name>
			<value>ns</value>
		</property>
		<!-- ns下面有两个name，分别是nm，snm -->
		<property>
			<name>dfs.ha.names.ns</name>
			<value>nm,snm</value>
		</property>
		<!-- nm的RPC通信地址 -->
		<property>
			<name>dfs.name.rpc-address.ns.nm</name>
			<value>name:9000</value>
		</property>
		<!-- nm的http通信地址 -->
		<property>
			<name>dfs.name.http-address.ns.nm</name>
			<value>name:50070</value>
		</property>
		<!-- snm的RPC通信地址 -->
		<property>
			<name>dfs.name.rpc-address.ns.snm</name>
			<value>sname:9000</value>
		</property>
		<!-- snm的http通信地址 -->
		<property>
			<name>dfs.name.http-address.ns.snm</name>
			<value>sname:50070</value>
		</property>
    <!-- hadoop.tmp.dir 在core-site.xml中设置这里不用设，否者则添加如下两个属性 -->
     <property>
			<name>dfs.name.name.dir</name>
			<value>/hadoop/dfs/name</value>
		</property>
     <property>
			<name>dfs.name.data.dir</name>
			<value>/hadoop/dfs/data</value>
		</property> 
		<!-- 指定name的元数据在JournalNode上的存放位置 加入amrm集群更健壮-->
		<property>
			<name>dfs.name.shared.edits.dir</name>	
         <value>qjournal://name:8485;sname:8485;amrm:8485/ns</value>
		</property>
		<!-- 指定JournalNode在本地磁盘存放数据的位置 -->
		<property>
			<name>dfs.journalnode.edits.dir</name>
			<value>/hadoop/journal</value>
		</property>
		<!-- 开启name失败自动切换 -->
		<property>
			<name>dfs.ha.automatic-failover.enabled</name>
			<value>true</value>
		</property>
		<!-- 配置失败自动切换实现方式 -->
		<property>
			<name>dfs.client.failover.proxy.provider.ns</name>
			<value>org.apache.hadoop.hdfs.server.name.ha.ConfiguredFailoverProxyProvider</value>
		</property>
		<!-- 配置隔离机制方法，多个机制用换行分割，即每个机制暂用一行-->
		<property>
			<name>dfs.ha.fencing.methods</name>
			<value>
				sshfence
				shell(/bin/true)
			</value>
		</property>
		<!-- 使用sshfence隔离机制时需要ssh免登陆 -->
		<property>
			<name>dfs.ha.fencing.ssh.private-key-files</name>
			<value>/root/.ssh/id_dsa</value>
		</property>
		<!-- 配置sshfence隔离机制超时时间 -->
		<property>
			<name>dfs.ha.fencing.ssh.connect-timeout</name>
			<value>30000</value>
		</property>
	</configuration>

4)修改mapred-site.xml

<configuration>
		<!-- 指定mr框架为yarn方式 -->
		<property>
			<name>mapreduce.framework.name</name>
			<value>yarn</value>
		</property>
     <!-- 启动historyserver  -->
            <property>
                 <name>mapreduce.jobhistory.address</name>
                 <value>name:10020</value>
            </property>

            <property>
                  <name>mapreduce.jobhistory.webapp.address</name>
                  <value>name:19888</value>
            </property>
            <!--dir为分布式文件系统中的文件目录，启动时先启动dfs，在启动historyserver -->
            <property>
                   <name>mapreduce.jobhistory.intermediate-done-dir</name>
                   <value>/history/indone</value>
            </property>
            <!--dir为分布式文件系统中的文件目录，启动时先启动dfs，在启动historyserver -->
            <property>
                  <name>mapreduce.jobhistory.done-dir</name>
                  <value>/history/done</value>
           </property>
	</configuration>	

5)修改yarn-site.xml

<configuration>
		<!-- 指定resourcemanager地址 -->
		<property>
			<name>yarn.resourcemanager.hostname</name>
			<value>amrm</value>
		</property>
     <!--ResourceManager 对客户端暴露的地址。
      客户端通过该地址向RM提交应用程序，杀死应用程序等-->
		<property>
			<name>yarn.resourcemanager.address</name>
			<value>${yarn.resourcemanager.hostname}:8032</value>
		</property>
   <!--ResourceManager 对ApplicationMaster暴露的访问地址。
      ApplicationMaster通过该地址向RM申请资源、释放资源等。-->
		<property>
			<name>yarn.resourcemanager.scheduler.address</name>
			<value>${yarn.resourcemanager.hostname}:8030</value>
		</property>
  <!-- ResourceManager 对NodeManager暴露的地址。
          NodeManager通过该地址向RM汇报心跳，领取任务等。-->
		<property>
			<name>yarn.resourcemanager.resource-tracker.address</name>
			<value>${yarn.resourcemanager.hostname}:8031</value>
		</property>
   <!--ResourceManager 对管理员暴露的访问地址。
          管理员通过该地址向RM发送管理命令等。默认值：${yarn.resourcemanager.hostname}:8033-->
		<property>
			<name>yarn.resourcemanager.admin.address</name>
			<value>${yarn.resourcemanager.hostname}:8033</value>
		</property>
    <!--ResourceManager对外web ui地址-->
		<property>
			<name>yarn.resourcemanager.webapp.address</name>
			<value>${yarn.resourcemanager.hostname}:8088</value>
		</property>

		<!-- 指定nodemanager启动时加载server的方式为shuffle server -->
		<property>
			<name>yarn.nodemanager.aux-services</name>
			<value>mapreduce_shuffle</value>
		</property>
	</configuration>

6)修改slaves
slaves是指定子节点的位置，因为要在name上启动HDFS、在amrm启动yarn，所以name上的slaves文件指定的是datanode的位置，amrm上的slaves文件指定的是nodemanager的位置

cd /hadoop/hadoop-2.7.1/tmp/hadoop/etc/hadoop/
vim slaves
name
sname
amrm

注：name中slaves为amrm和journalnode的地址，amrm中slaves为nodeamananger的地址。
6.将配置好的hadoop拷贝到sname和amrm

scp  -r /hadoop/hadoop-2.7.1/tmp root@amrm:/hadoop/hadoop-2.7.1/tmp

scp  -r /hadoop/hadoop-2.7.1/tmp  root@sname:/hadoop/hadoop-2.7.1/tmp/
scp  -r /hadoop/hadoop-2.7.1/tmp  root@amrm:/hadoop/hadoop-2.7.1/tmp/

*********************注意：以下操作必须严格按照顺序*****************************
7.启动zookeeper集群，(在name,sname,amrm的/hadoop/hadoop-2.7.1/tmp/zk/bin/里开启)

cd /hadoop/hadoop-2.7.1/tmp/zk/bin  // 按顺序启动name，sname,amrm
./zkServer.sh start(启动zookeeper节点)
./zkServer.sh status(查看zookeeper状态)

8.启动journalnode，(在name,sname,amrm的/hadoop/hadoop-2.7.1/tmp/hadoop/sbin里启动)//在name中启动即可 非hadoop-daemon.sh

hadoop-daemons.sh start journalnode

jps(依次在每个节点中查看各节点是否多了Journalnode进程)
9.格式化HDFS，在name上执行格式化命令

hdfs  namenode  -format  ns

格式化后会在根据core-site.xml中的hadoop.tmp.dir配置生成个文件，这里我配置的是/hadoop/hadoop-2.7.1/tmp，然后将/hadoop/hadoop-2.7.1/tmp拷贝到sname和amrm的/hadoop/hadoop-2.7.1/tmp下。

scp  -r /hadoop/hadoop-2.7.1/dfs  root@sname:/hadoop/hadoop-2.7.1
scp  -r /hadoop/hadoop-2.7.1/dfs  root@amrm:/hadoop/hadoop-2.7.1

注：格式化生成的目录不要轻易删除，否者启动回报不一致异常
10.格式化ZK，在name上执行格式化命令

hdfs  zkfc  -formatZK

11.启动HDFS，在name的/hadoop/hadoop-2.7.1/tmp/hadoop/sbin中执行start-dfs.sh命令

cd  /hadoop/hadoop-2.7.1/sbin/
start-dfs.sh

启动之后，分别进入name,sname,amrm中jps，查看是否多了name 和 DFSZKFailoverController两个进程(name,sname)
12.启动 historyserver  在name中的/hadoop/hadoop-2.7.1/tmp/hadoop/sbin中执行，

hdfs dfs -mkdir /history
hdfs dfs -mkdir /history/indone
hdfs dfs -mkdir /history/done
mr-jobhistory-daemon.sh  start historyserver

13.启动YARN
在 amrm 中的/hadoop/hadoop-2.7.1/tmp/hadoop/sbin中执行start-yarn.sh命令

cd  /hadoop/hadoop-2.7.1/sbin/
start-yarn.sh

是在amrm上执行start-yarn.sh，把name和resourcemanager分开是因为性能问题，因为他们都要占用大量资源，所以把他们分开了，他们分开了就要分别在不同的机器上启动
14.到此，hadoop2.7.0的配置完毕，可以通过浏览器访问来查看部署是否成功
   1) http://192.168.32.137:50070 namenode
     


   2) http://192.168.32.136:8088  resourcemanager




    3) http://192.168.32.137:19888  jobhistroysever




15.执行job
  

1)hdfs  dfs  -mkdir /test
   2)hdfs  dfs  -mkdir /test/input
   3)hdfs  dfs  -put  etc/hadoop/*.xml  /test/input
   4)hadoop jar  share/hadoop/mapreduce/hadoop-mapreduce-examples-2.7.1.jar grep /test/input /test/output 'dfs[a-z.]+'
     [img]http://dl2.iteye.com/upload/attachment/0122/5366/817ca7cf-a5b0-307c-bdd8-40ede16677f4.png[/img]        
   5) hdfs  dfs  -get /test/output   output  //当前目录
   6) cat   output/* 查看结果

备注：另外一种查看结果的方式

  hdfs dfs -cat /test/output/*

查看job状态：









Jobhistorysever：







16.关闭hadoop
在amrm中
           
          

 stop-yarn.sh

在name中
        

  mr-jobhistory-daemon.sh stop historyserver
            stop-dfs.sh

17.

hadoop dfsadmin -safemode leave

注：以上过程有什么问题，可以查看相关日志文件
相关异常
1. org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.ipc.StandbyException): Operation category READ is not supported in state standby
因为掉电，导致hadoop 的HA 出现 org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.ipc.StandbyException): Operation category READ is not supported in state standby 此问题，原来从web 页面查看，是两个节点都变成了standy，所以要切换
              

 hdfs haadmin -transitionToActive --forcemanual nm

2. org.apache.hadoop.ipc.Client: Retrying connect to server: amrm/192.168.32.136:8032. Already tried 8 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS)
telnet: Unable to connect to remote host: Connection refused  Ubuntu 15.10
查看能否ping通，查看端口是否开放，如果能ping通，同时端口开放，用如下命令查看系统端口监听
netstat -ntulp
确保local Address的地址为0.0.0.0 或192.168.32.137。
解决办法 修改/etc/hosts 地址映射





3. org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.ipc.StandbyException): Operation category WRITE is not supported in state standby
name 处于standby状态



4. org.apache.hadoop.mapreduce.lib.input.InvalidInputException: Input path does not exist: hdfs://sname:9000/user/root/grep-temp-1382738569
这个是由于map的产生的文件放在分布式文件系统/user/${username}中新建

 hdfs  dfs  -mkdir /user
 hdfs  dfs  -mkdir /user/${username}

5. java.io.IOException: Unknown Job job_1450012188054_0001 at org.apache.hadoop.mapreduce.v2.hs.HistoryClientService$HSClientProtocolHandler.verifyAndGetJob(HistoryClientService.java:218)
at org.apache.hadoop.mapreduce.v2.hs.HistoryClientService$HSClientProtocolHandler.getCounters(HistoryClientService.java:232) at org.apache.hadoop.mapreduce.v2.api.impl.pb.service.MRClientProtocolPBServiceImpl.getCounters(MRClientProtocolPBServiceImpl.java:159) at org.apache.hadoop.yarn.proto.MRClientProtocol$MRClientProtocolService$2.callBlockingMethod(MRClientProtocol.java:281)
at org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:616)
at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:969)
at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:2049)
at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:2045)
at java.security.AccessController.doPrivileged(Native Method)
at javax.security.auth.Subject.doAs(Subject.java:415)
at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1657)
at org.apache.hadoop.ipc.Server$Handler.run(Server.java:2043)



解决办法

：hdfs dfs -chmod -R  777  /history

6



解决方式：在/etc/hosts中，添加jamel地址映射。
注：
1.Job 成功的显示输出结果

15/12/13 22:17:44 INFO mapreduce.Job: Job job_1450012188054_0002 completed successfully
15/12/13 22:17:45 INFO mapreduce.Job: Counters: 50
	File System Counters
		FILE: Number of bytes read=493
		FILE: Number of bytes written=1176179
		FILE: Number of read operations=0
		FILE: Number of large read operations=0
		FILE: Number of write operations=0
		HDFS: Number of bytes read=33949
		HDFS: Number of bytes written=663
		HDFS: Number of read operations=30
		HDFS: Number of large read operations=0
		HDFS: Number of write operations=2
	Job Counters 
		Killed map tasks=3
		Launched map tasks=12
		Launched reduce tasks=1
		Data-local map tasks=12
		Total time spent by all maps in occupied slots (ms)=1450715
		Total time spent by all reduces in occupied slots (ms)=112387
		Total time spent by all map tasks (ms)=1450715
		Total time spent by all reduce tasks (ms)=112387
		Total vcore-seconds taken by all map tasks=1450715
		Total vcore-seconds taken by all reduce tasks=112387
		Total megabyte-seconds taken by all map tasks=1485532160
		Total megabyte-seconds taken by all reduce tasks=115084288
	Map-Reduce Framework
		Map input records=926
		Map output records=17
		Map output bytes=508
		Map output materialized bytes=541
		Input split bytes=969
		Combine input records=17
		Combine output records=15
		Reduce input groups=15
		Reduce shuffle bytes=541
		Reduce input records=15
		Reduce output records=15
		Spilled Records=30
		Shuffled Maps =9
		Failed Shuffles=0
		Merged Map outputs=9
		GC time elapsed (ms)=67395
		CPU time spent (ms)=15090
		Physical memory (bytes) snapshot=1492398080
		Virtual memory (bytes) snapshot=6682742784
		Total committed heap usage (bytes)=1178963968
	Shuffle Errors
		BAD_ID=0
		CONNECTION=0
		IO_ERROR=0
		WRONG_LENGTH=0
		WRONG_MAP=0
		WRONG_REDUCE=0
	File Input Format Counters 
		Bytes Read=32980
	File Output Format Counters 
		Bytes Written=663

2.本文所搭建的是高可用对于namenode而言，而RM HA可以访问如下地址：
ResourceMananger HA 访问-
http://hadoop.apache.org/docs/stable/hadoop-yarn/hadoop-yarn-site/ResourceManagerHA.html、