3.16 使用Zookeeper对HDFS HA配置自动故障转移及测试

一、说明

从上一节可看出，虽然搭建好了HA架构，但是只能手动进行active与standby的切换；

接下来看一下用zookeeper进行自动故障转移：

#
在启动HA之后，两个NameNode都是standby状态，可以利用zookeeper的选举功能，选出一个当Active

#
监控
    ZKFC
    FailoverController

二、配置

1、hdfs-site.xml

#”开启自动转移功能“，加入以下内容；

<property>
        <name>dfs.ha.automatic-failover.enabled</name>
        <value>true</value>
</property>

2、core-site.xml

#”设置故障转移的zookeeper集群“，加入以下内容；

<property>
        <name>ha.zookeeper.quorum</name>
        <value>master:2181,slave1:2181,slave2:2181</value>
</property>

3、关闭集群所有服务

#master
[root@master hadoop-2.5.0]# sbin/stop-dfs.sh

[root@master ~]# xcall jps
====== master jps ======
18719 Jps
====== slave1 jps ======
19150 Jps
====== slave2 jps ======
13595 Jps

#如果还有其他服务(zookeeper等)也要关闭；

4、同步配置文件

[root@master hadoop]# pwd
/opt/app/hadoop-2.5.0/etc/hadoop

[root@master hadoop]# scp -r hdfs-site.xml core-site.xml root@slave1:/opt/app/hadoop-2.5.0/etc/hadoop/

[root@master hadoop]# scp -r hdfs-site.xml core-site.xml root@slave2:/opt/app/hadoop-2.5.0/etc/hadoop/

5、启动zookeeper

#所有节点启动zookeeper
[root@master ~]# /opt/app/zookeeper-3.4.5/bin/zkServer.sh start

[root@slave1 ~]# /opt/app/zookeeper-3.4.5/bin/zkServer.sh start

[root@slave2 ~]# /opt/app/zookeeper-3.4.5/bin/zkServer.sh start

#查看
[root@master ~]# xcall jps
====== master jps ======
18824 Jps
18765 QuorumPeerMain
====== slave1 jps ======
19201 QuorumPeerMain
19263 Jps
====== slave2 jps ======
13646 QuorumPeerMain
13702 Jps

6、初始化HA在Zookeeper中状态

#master
[root@master hadoop-2.5.0]# bin/hdfs zkfc -formatZK

#
此时可以在slave1上用客户端连入zookeeper查看：
[root@slave1 zookeeper-3.4.5]# bin/zkCli.sh

[zk: localhost:2181(CONNECTED) 1] ls /
[zookeeper]

[zk: localhost:2181(CONNECTED) 2] ls /    #生成了hadoop-ha
[hadoop-ha, zookeeper]

7、启动HDFS服务

#master
[root@master hadoop-2.5.0]# sbin/start-dfs.sh 

#查看启动情况
[root@master ~]# xcall jps
====== master jps ======
19588 DFSZKFailoverController    #ZKFC监控进程
19087 NameNode
19193 DataNode
19393 JournalNode
18765 QuorumPeerMain
19662 Jps
====== slave1 jps ======
19743 DFSZKFailoverController    #ZKFC监控进程
19201 QuorumPeerMain
19800 Jps
19613 JournalNode
19521 DataNode
19443 NameNode
====== slave2 jps ======
13646 QuorumPeerMain
13850 DataNode
14014 Jps
13942 JournalNode

#查看nn1  nn2的状态
[root@master hadoop-2.5.0]# bin/hdfs haadmin -getServiceState nn1
19/04/18 10:34:50 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
active

[root@master hadoop-2.5.0]# bin/hdfs haadmin -getServiceState nn2
19/04/18 10:34:54 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
standby

#可见已经自动把nn1选举为active了，nn2为standby；在web中也可以看到；

8、测试故障自动转移

可以kill掉active状态的namenode，查看standby状态的namenode是否已经自动变为active了；