【甘道夫】Hadoop2.2.0 NN HA具体配置+Client透明性试验【完整版】

引言：

前面转载过一篇团队兄弟【伊利丹】写的NN HA实验记录，我也基于他的环境实验了NN HA对于Client的透明性。

本篇文章记录的是亲自配置NN HA的具体全过程，以及全面測试HA对client訪问透明性的全过程。希望对大家有帮助。

---

实验环境：

Hadoop2.2.0的4节点集群。ZK节点3个（ZK节点数最好为奇数个），hosts文件和各节点角色分配例如以下：

hosts：

192.168.66.91 master

192.168.66.92 slave1

192.168.66.93 slave2

192.168.66.94 slave3

角色分配：

	Active NN	Standby NN	DN	JournalNode	Zookeeper	FailoverController
master	V			V	V	V
slave1		V	V	V	V	V
slave2			V	V	V
slave3			V

---

---

实验过程：

---

---

1.下载稳定版Zookeeper

http://mirrors.cnnic.cn/apache/zookeeper/stable/

并解压到hadoop集群某文件夹下，我放在了/home/yarn/下。

---

2.改动配置文件

配置文件在conf目录中，将zoo_sample.cfg改名为zoo.cfg。并对其做响应改动。下面是改动过后的zoo.cfg

# The number of milliseconds of each tick ZK之间，或者Client和ZK之间心跳的时间间隔

tickTime=2000

# The number of ticks that the initial 

# synchronization phase can take

initLimit=10

# The number of ticks that can pass between 

# sending a request and getting an acknowledgement

syncLimit=5

# the directory where the snapshot is stored. 

# do not use /tmp for storage, /tmp here is just 

# example sakes.  保存ZK数据的文件夹，请自行创建后在此处指定

dataDir=/home/yarn/Zookeeper/zoodata

# the port at which the clients will connect  客户端连接ZKserver的端口

clientPort=2181

# the maximum number of client connections.

# increase this if you need to handle more clients

#maxClientCnxns=60

#

# Be sure to read the maintenance section of the 

# administrator guide before turning on autopurge.

#

# http://zookeeper.apache.org/doc/current/zookeeperAdmin.html#sc_maintenance

#

# The number of snapshots to retain in dataDir

#autopurge.snapRetainCount=3

# Purge task interval in hours

# Set to "0" to disable auto purge feature

#autopurge.purgeInterval=1

#保存ZK日志的文件夹，请自行创建后在此处指定

dataLogDir=/home/yarn/Zookeeper/zoolog

#******下面部分均为分布式ZK集群使用******

#ZK集群初始化时。Follower节点须要和Leader节点通信，initLimit配置的是Leader最多等待多少个心跳

initLimit=5 

#Leader和Follower之间发送消息、请求和应答时。最多等待多少个心跳

syncLimit=2 

#server.A=B:C:D

#A是一个数字，表示这是第几号server

#B是当前server的ID或者主机名

#C是Followerserver与Leaderserver交换信息的port

#D是当Leader挂掉时，又一次选举Leader所使用的port

server.1=192.168.66.91:2888:3888 

server.2=192.168.66.92:2888:3888

server.3=192.168.66.93:2888:3888

#千万注意：接下来须要在各个几点的dataDir文件夹下建立myid文件，内容就是对应的A，也就是说，各个ZK节点的myid文件内容不同 !!!

---

3.改动各个节点的环境变量

在/etc/profile文件加入：

export ZOOKEEPER_HOME=/home/yarn/Zookeeper/zookeeper-3.4.6

并为PATH加上

$ZOOKEEPER_HOME/bin

注意：export ZOOKEEPER_HOME要在PATH的上方。

---

以下開始改动Hadoop的配置文件：

---

4.改动core-site.xml

<configuration>

<property>

  <name>fs.defaultFS</name>

  <value>hdfs://myhadoop</value>

  <description>注意：myhadoop为集群的逻辑名，需与hdfs-site.xml中的dfs.nameservices一致！</description>

</property>

<property>

  <name>hadoop.tmp.dir</name>

  <value>/home/yarn/Hadoop/hdfs2.0/tmp</value>

</property>

<property>

  <name>ha.zookeeper.quorum</name>

  <value>master:2181,slave1:2181,slave2:2181</value>

  <description>各个ZK节点的IP/host，及客户端连接ZK的port，该port需与zoo.cfg中的clientPort一致。</description>

</property>

</configuration>

---

5.改动hdfs-site.xml

<?

xml version="1.0" encoding="UTF-8"?>

<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>

<!--

  Licensed under the Apache License, Version 2.0 (the "License");

  you may not use this file except in compliance with the License.

  You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

  Unless required by applicable law or agreed to in writing, software

  distributed under the License is distributed on an "AS IS" BASIS,

  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.

  See the License for the specific language governing permissions and

  limitations under the License. See accompanying LICENSE file.

-->

<!-- Put site-specific property overrides in this file. -->

<configuration>

<property>

  <name>dfs.nameservices</name>

  <value>myhadoop</value>

  <description>

    Comma-separated list of nameservices.

    as same as fs.defaultFS in core-site.xml.

  </description>

</property>

<property>

  <name>dfs.ha.namenodes.myhadoop</name>

  <value>nn1,nn2</value>

  <description>

    The prefix for a given nameservice, contains a comma-separated

    list of namenodes for a given nameservice (eg EXAMPLENAMESERVICE).

  </description>

</property>

<property>

  <name>dfs.namenode.rpc-address.myhadoop.nn1</name>

  <value>master:8020</value>

  <description>

    RPC address for nomenode1 of hadoop-test

  </description>

</property>

<property>

  <name>dfs.namenode.rpc-address.myhadoop.nn2</name>

  <value>slave1:8020</value>

  <description>

    RPC address for nomenode2 of hadoop-test

  </description>

</property>

<property>

  <name>dfs.namenode.http-address.myhadoop.nn1</name>

  <value>master:50070</value>

  <description>

    The address and the base port where the dfs namenode1 web ui will listen on.

  </description>

</property>

<property>

  <name>dfs.namenode.http-address.myhadoop.nn2</name>

  <value>slave1:50070</value>

  <description>

    The address and the base port where the dfs namenode2 web ui will listen on.

  </description>

</property>

<property>  

  <name>dfs.namenode.servicerpc-address.myhadoop.n1</name>  

  <value>master:53310</value>  

</property>  

<property>  

  <name>dfs.namenode.servicerpc-address.myhadoop.n2</name>  

  <value>slave1:53310</value>  

</property>

<property>

  <name>dfs.namenode.name.dir</name>

  <value>file:///home/yarn/Hadoop/hdfs2.0/name</value>

  <description>Determines where on the local filesystem the DFS name node

      should store the name table(fsimage).  If this is a comma-delimited list

      of directories then the name table is replicated in all of the

      directories, for redundancy. </description>

</property>

<property>

  <name>dfs.namenode.shared.edits.dir</name>

  <value>qjournal://slave1:8485;slave2:8485;slave3:8485/hadoop-journal</value>

  <description>A directory on shared storage between the multiple namenodes

  in an HA cluster. This directory will be written by the active and read

  by the standby in order to keep the namespaces synchronized. This directory

  does not need to be listed in dfs.namenode.edits.dir above. It should be

  left empty in a non-HA cluster.

  </description>

</property>

<property>

  <name>dfs.datanode.data.dir</name>

  <value>file:///home/yarn/Hadoop/hdfs2.0/data</value>

  <description>Determines where on the local filesystem an DFS data node

  should store its blocks.  If this is a comma-delimited

  list of directories, then data will be stored in all named

  directories, typically on different devices.

  Directories that do not exist are ignored.

  </description>

</property>

<property>

  <name>dfs.ha.automatic-failover.enabled</name>

  <value>true</value>

  <description>

    Whether automatic failover is enabled. See the HDFS High

    Availability documentation for details on automatic HA

    configuration.

  </description>

</property>

<property>

  <name>dfs.journalnode.edits.dir</name>

  <value>/home/yarn/Hadoop/hdfs2.0/journal/</value>

</property>

<property>  

  <name>dfs.client.failover.proxy.provider.myhadoop</name>                        

  <value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value>

  <description>Configure the name of the Java class which will be used by the DFS Client to determine which NameNode is the current Active, and therefore which NameNode is currently serving client requests. 

这个类是Client的訪问代理。是HA特性对于Client透明的关键！

</description>  

</property>  

      

<property>      

  <name>dfs.ha.fencing.methods</name>      

  <value>sshfence</value>  

  <description>how to communicate in the switch process</description>

</property>  

    

<property>      

  <name>dfs.ha.fencing.ssh.private-key-files</name>      

  <value>/home/yarn/.ssh/id_rsa</value>

  <description>the location stored ssh key</description>

</property>  

  

<property>  

  <name>dfs.ha.fencing.ssh.connect-timeout</name>  

  <value>1000</value>  

</property>  

  

<property>  

  <name>dfs.namenode.handler.count</name>  

  <value>8</value>  

</property> 

</configuration>

---

6.将改动好的core-site.xml和hdfs-site.xml复制到各个hadoop节点。

---

7.启动

（1）启动ZK

在全部的ZK节点运行命令：

zkServer.sh start

查看各个ZK的从属关系：

yarn@master:~$ zkServer.sh status

JMX enabled by default

Using config: /home/yarn/Zookeeper/zookeeper-3.4.6/bin/../conf/zoo.cfg

Mode: follower

yarn@slave1:~$ zkServer.sh status

JMX enabled by default

Using config: /home/yarn/Zookeeper/zookeeper-3.4.6/bin/../conf/zoo.cfg

Mode: follower

yarn@slave2:~$ zkServer.sh status

JMX enabled by default

Using config: /home/yarn/Zookeeper/zookeeper-3.4.6/bin/../conf/zoo.cfg

Mode: leader

注意：

哪个ZK节点会成为leader是随机的，第一次实验时slave2成为了leader，第二次实验时slave1成为了leader！

此时，在各个节点都能够查看到ZK进程：

yarn@master:~$ jps

3084 QuorumPeerMain

3212 Jps

（2）格式化ZK（仅第一次须要做）

随意ZK节点上运行：

hdfs zkfc -formatZK

（3）启动ZKFC

ZookeeperFailoverController是用来监控NN状态。协助实现主备NN切换的，所以只在主备NN节点上启动即可：

hadoop-daemon.sh start zkfc

启动后我们能够看到ZKFC进程：

yarn@master:~$ jps

3084 QuorumPeerMain

3292 Jps

3247 DFSZKFailoverController

（4）启动用于主备NN之间同步元数据信息的共享存储系统JournalNode

參见角色分配表，在各个JN节点上启动：

hadoop-daemon.sh start journalnode

启动后在各个JN节点都能够看到JournalNode进程：

yarn@master:~$ jps

3084 QuorumPeerMain

3358 Jps

3325 JournalNode

3247 DFSZKFailoverController

（5）格式化并启动主NN

格式化：

hdfs namenode -format

注意：仅仅有第一次启动系统时需格式化。请勿反复格式化！

在主NN节点运行命令启动NN:

hadoop-daemon.sh start namenode

启动后能够看到NN进程：

yarn@master:~$ jps

3084 QuorumPeerMain

3480 Jps

3325 JournalNode

3411 NameNode

3247 DFSZKFailoverController

（6）在备NN上同步主NN的元数据信息

hdfs namenode -bootstrapStandby

下面是正常运行时的最后部分日志：

Re-format filesystem in Storage Directory /home/yarn/Hadoop/hdfs2.0/name ? (Y or N) Y

14/06/15 10:09:08 INFO common.Storage: Storage directory /home/yarn/Hadoop/hdfs2.0/name has been successfully formatted.

14/06/15 10:09:09 INFO namenode.TransferFsImage: Opening connection to http://master:50070/getimage?getimage=1&txid=935&storageInfo=-47:564636372:0:CID-d899b10e-10c9-4851-b60d-3e158e322a62

14/06/15 10:09:09 INFO namenode.TransferFsImage: Transfer took 0.11s at 63.64 KB/s

14/06/15 10:09:09 INFO namenode.TransferFsImage: Downloaded file fsimage.ckpt_0000000000000000935 size 7545 bytes.

14/06/15 10:09:09 INFO util.ExitUtil: Exiting with status 0

14/06/15 10:09:09 INFO namenode.NameNode: SHUTDOWN_MSG: 

/************************************************************

SHUTDOWN_MSG: Shutting down NameNode at slave1/192.168.66.92

************************************************************/

（7）启动备NN

在备NN上运行命令：

hadoop-daemon.sh start namenode

（8）设置主NN（这一步能够省略，这是在设置手动切换NN时的步骤。ZK已经自己主动选择一个节点作为主NN了）

到眼下为止，事实上HDFS还不知道谁是主NN，能够通过监控页面查看，两个节点的NN都是Standby状态。

以下我们须要在主NN节点上运行命令激活主NN：

hdfs haadmin -transitionToActive nn1

（9）在主NN上启动Datanode

在[nn1]上，启动全部datanode

hadoop-daemons.sh start datanode

---

8.效果验证1--主备自己主动切换

眼下的主NN是192.168.0.91

备NN是192.168.0.92

我在主NN上kill掉NameNode进程：

yarn@master:~$ jps

5161 NameNode

5085 JournalNode

5438 Jps

4987 DFSZKFailoverController

4904 QuorumPeerMain

yarn@master:~$ kill 5161

yarn@master:~$ jps

5451 Jps

5085 JournalNode

4987 DFSZKFailoverController

4904 QuorumPeerMain

此时，主NN监控页面无法訪问：

备NN自己主动切换为主NN：

---

9.效果验证2--HA对shell的透明性

訪问逻辑名myhadoop，运行命令查看文件夹结构，不受影响：

yarn@slave3:~$ hadoop dfs -ls hdfs://myhadoop/

DEPRECATED: Use of this script to execute hdfs command is deprecated.

Instead use the hdfs command for it.

Found 3 items

drwxr-xr-x   - yarn supergroup          0 2014-03-20 00:10 hdfs://myhadoop/home

drwxrwx---   - yarn supergroup          0 2014-03-17 20:11 hdfs://myhadoop/tmp

drwxr-xr-x   - yarn supergroup          0 2014-03-17 20:15 hdfs://myhadoop/workspace

---

10.效果验证3--HA对Client程序的透明性

使用自己写的HdfsDAO.java測试，程序中将指向HDFS的path设置为：

private static final String HDFS = "hdfs://myhadoop/";

先ping myhadoop确保没有配置hosts。然后执行程序，一切正常：

yarn@master:~$ ping myhadoop

ping: unknown host myhadoop

yarn@master:~$ hadoop jar Desktop/hatest.jar HdfsDAO 

ls: /

==========================================================

name: hdfs://myhadoop/home, folder: true, size: 0

name: hdfs://myhadoop/tmp, folder: true, size: 0

name: hdfs://myhadoop/workspace, folder: true, size: 0

==========================================================