实现hadoop+zookeeper高可用集群
本篇hadoop+zookeeper高可用是在上篇博客hadoop集群节点的搭建的基础上实现的
注意：Zookeeper 集群至少三台,总节点数为奇数个。
系统环境:
RHEL6.5 selinux and iptables is disabled
Hadoop 、jdk、zookeeper 程序使用 nfs 共享同步配置文件
本次实验安装包的版本：
| hadoop-2.7.3.tar.gz | zookeeper-3.4.9.tar.gz jdk-7u79-linux-x64.tar.gz |
|---|
server1作为master,server5作为备用master,server2、server3和server4作为集群服务器
实验环境 ：
| ip | 主机名 | 角色 |
|---|---|---|
| 172.25.245.1 | server1.example.com | NameNode 、 DFSZKFailoverController 、 ResourceManager |
| 172.25.245.2 | server2.example.com | DateNode 、 JournalNode 、 NodeManager 、 QuorumPeerMain |
| 172.25.245.3 | server3.example.com | DateNode 、 JournalNode 、 NodeManager 、 QuorumPeerMain |
| 172.25.245.4 | server4.example.com | DateNode 、 JournalNode 、 NodeManager 、 QuorumPeerMain |
| 172.25.245.5 | server5.example.com | NameNode 、 DFSZKFailoverController 、 ResourceManager |
搭建zookeeper集群：
1、配置新虚拟机server5
[root@server5 ~]# yum install nfs-utils -y ##安装nfs-utils
[root@server5 ~]# useradd -u 800 hadoop ## 新建hadoop用户
[root@server5 ~]# /etc/init.d/rpcbind start
[root@server5 ~]# /etc/init.d/nfs start
[root@server5 ~]# mount 172.25.254.1:/home/hadoop/ /home/hadoop ## 挂载server1端的目录
2、清理/tmp/*
[hadoop@server5 ~]$ rm -fr /tmp/*
3、编辑server1的/home/hadoop/zookeeper-3.4.9/conf下的配置文件，
因为1、3、4、5都使用了server1端的nfs文件系统，所以五台主机的/home/hadoop是同步的
[hadoop@server1 ~ ]$ cd /home/hadoop/zookeeper-3.4.9/conf
[hadoop@server1 conf]$ cp zoo_sample.cfg zoo.cfg
[hadoop@server1 conf]$ vim zoo.cfg ## 写入集群的三台设备
[hadoop@server1 conf]$ cat zoo.cfg | tail -n 3
server.1=172.25.254.2:2888:3888
server.2=172.25.254.3:2888:3888
server.3=172.25.254.4:2888:3888
4、三个集群节点：
[hadoop@server2 tmp]$ mkdir /tmp/zookeeper
[hadoop@server2 tmp]$ cd /tmp/zookeeper
[hadoop@server2 zookeeper]$ echo 1 > myid
[hadoop@server3 conf]$ mkdir /tmp/zookeeper
[hadoop@server3 conf]$ cd /tmp/zookeeper/
[hadoop@server3 zookeeper]$ echo 2 > myid
[hadoop@server4 tmp]$ mkdir /tmp/zookeeper
[hadoop@server4 tmp]$ cd /tmp/zookeeper
[hadoop@server4 zookeeper]$ echo 3 > myid
5、在各节点启动zookeeper服务,发现server3为leader
注意：需要同时把三个服务都开启，才能查看到状态！
在server2上：
[hadoop@server2 ~]$ cd zookeeper-3.4.9
[hadoop@server2 zookeeper-3.4.9]$ bin/zkServer.sh start
ZooKeeper JMX enabled by default
Using config: /home/hadoop/zookeeper-3.4.9/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
[hadoop@server2 zookeeper-3.4.9]$ bin/zkServer.sh status
ZooKeeper JMX enabled by default
Using config: /home/hadoop/zookeeper-3.4.9/bin/../conf/zoo.cfg
Mode: follower #####
在server3上
[hadoop@server3 ~]$ cd zookeeper-3.4.9
[hadoop@server3 zookeeper-3.4.9]$ bin/zkServer.sh start
ZooKeeper JMX enabled by default
Using config: /home/hadoop/zookeeper-3.4.9/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
[hadoop@server3 zookeeper-3.4.9]$ bin/zkServer.sh status
ZooKeeper JMX enabled by default
Using config: /home/hadoop/zookeeper-3.4.9/bin/../conf/zoo.cfg
Mode: leader #####
在server4上
[hadoop@server4 ~]$ cd zookeeper-3.4.9
[hadoop@server4 zookeeper-3.4.9]$ bin/zkServer.sh start
ZooKeeper JMX enabled by default
Using config: /home/hadoop/zookeeper-3.4.9/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
[hadoop@server4 zookeeper-3.4.9]$ bin/zkServer.sh status
ZooKeeper JMX enabled by default
Using config: /home/hadoop/zookeeper-3.4.9/bin/../conf/zoo.cfg
Mode: follower #####
二、配置hadoop
server1:
1、编辑core-site.xml文件
[hadoop@server1 ~]$ cd hadoop/etc/hadoop/
[hadoop@server1 hadoop]$ vim core-site.xml #指定hdfs的namenode为masters 指定zookeeper集群主机地址
2.编辑hdfs-site.xml文件：
dfs.replication
3
指定 hdfs 的 nameservices 为 masters,和 core-site.xml 文件中的设置保持一致
dfs.nameservices
masters
masters 下面有两个 namenode 节点,分别是 h1 和 h2 (名称可自定义)
dfs.ha.namenodes.masters
h1,h2
指定 h1 节点的 rpc 通信地址
dfs.namenode.rpc-address.masters.h1
172.25.254.1:9000
指定 h1 节点的 http 通信地址
dfs.namenode.http-address.masters.h1
172.25.254.1:50070
指定 h2 节点的 rpc 通信地址
dfs.namenode.rpc-address.masters.h2
172.25.254.5:9000
指定 h2 节点的 http 通信地址
dfs.namenode.http-address.masters.h2
172.25.254.5:50070
指定 NameNode 元数据在 JournalNode 上的存放位置
dfs.namenode.shared.edits.dir
qjournal://172.25.254.2:8485;172.25.254.3:8485;172.25.254.4:8485/masters
指定 JournalNode 在本地磁盘存放数据的位置
dfs.journalnode.edits.dir
/tmp/journaldata
开启 NameNode 失败自动切换
dfs.ha.automatic-failover.enabled
true
配置失败自动切换实现方式
dfs.client.failover.proxy.provider.masters
org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider
配置隔离机制方法,每个机制占用一行
dfs.ha.fencing.methods
sshfence
shell(/bin/true)
使用 sshfence 隔离机制时需要 ssh 免密码
dfs.ha.fencing.ssh.private-key-files
/home/hadoop/.ssh/id_rsa
配置 sshfence 隔离机制超时时间
dfs.ha.fencing.ssh.connect-timeout
30000
3.编辑文件slaves:
[hadoop@server1 hadoop]$ vim slaves
4、启动 hdfs 集群(按顺序启动)
在三个 DN 上依次启动 zookeeper 集群
[hadoop@server2 zookeeper-3.4.9]$ bin/zkServer.sh start
在三个 DN 上依次启动 journalnode(第一次启动 hdfs 必须先启动 journalnode)
[hadoop@server2 hadoop]$ sbin/hadoop-daemon.sh start journalnode
starting journalnode, logging to /home/hadoop/hadoop-2.7.3/logs/hadoop-hadoop-journalnode-server2.out
[hadoop@server3 hadoop]$ sbin/hadoop-daemon.sh start journalnode
starting journalnode, logging to /home/hadoop/hadoop-2.7.3/logs/hadoop-hadoop-journalnode-server3.out
[hadoop@server4 hadoop]$ sbin/hadoop-daemon.sh start journalnode
starting journalnode, logging to /home/hadoop/hadoop-2.7.3/logs/hadoop-hadoop-journalnode-server4.out
在server1格式化 HDFS 集群
Namenode 数据默认存放在/tmp,需要把数据拷到 h2
[hadoop@server1 hadoop]$ bin/hdfs namenode -format
[hadoop@server1 hadoop]$ scp -r /tmp/hadoop-hadoop 172.25.254.5:/tmp
格式化 zookeeper (只需在 h1 上执行即可)
[hadoop@server1 hadoop]$ bin/hdfs zkfc -formatZK
启动 hdfs 集群(只需在 h1 上执行即可)
[hadoop@server1 hadoop]$ sbin/start-dfs.sh
在server2进入命令行：
[hadoop@server2 zookeeper-3.4.9]$ bin/zkCli.sh
[zk: localhost:2181(CONNECTED) 0] ls /
[hadoop-ha, zookeeper]
[zk: localhost:2181(CONNECTED) 1] ls
[zk: localhost:2181(CONNECTED) 2] ls /hadoop-ha/masters
[ActiveBreadCrumb, ActiveStandbyElectorLock]
[zk: localhost:2181(CONNECTED) 3] ls /hadoop-ha/masters/Active
Node does not exist: /hadoop-ha/masters/Active
[zk: localhost:2181(CONNECTED) 4] ls /hadoop-ha/masters/ActiveBreadCrumb
[]
[zk: localhost:2181(CONNECTED) 5] get /hadoop-ha/masters/ActiveBreadCrumb
mastersh2server5 �F(�> ## active为server5
cZxid = 0x200000009
ctime = Tue Nov 20 20:05:25 CST 2018
mZxid = 0x200000009
mtime = Tue Nov 20 20:05:25 CST 2018
pZxid = 0x200000009
cversion = 0
dataVersion = 0
aclVersion = 0
ephemeralOwner = 0x0
dataLength = 28
numChildren = 0
测试：
发现server5为active server1为stanby
server5端杀掉namenode进程：
杀掉 h2主机的 namenode 进程后依然可以访问,此时 h1转为 active 状态接管 namenode
刷新发现active跳到server1上：实现高可用
[hadoop@server5 tmp]$ jps
1561 DFSZKFailoverController
1464 NameNode
2088 Jps
[hadoop@server5 tmp]$ kill -9 1464
[hadoop@server5 tmp]$ jps
2098 Jps
1561 DFSZKFailoverController
重新开启server5的namenode:
[hadoop@server5 hadoop]$ sbin/hadoop-daemon.sh start namenode
starting namenode, logging to /home/hadoop/hadoop-2.7.3/logs/hadoop-hadoop-namenode-server1.out
刷新发现server5变成active
