一、集群的规划
Zookeeper集群：
192.168.176.131 （bigdata112）
192.168.176.132 （bigdata113）
192.168.176.135 （bigdata114）
Hadoop集群：
192.168.176.131 （bigdata112） NameNode1 ResourceManager1 Journalnode
192.168.176.132 （bigdata113） NameNode2 ResourceManager2 Journalnode
192.168.176.135 （bigdata114） DataNode1 NodeManager1
192.168.176.136 （bigdata115） DataNode2 NodeManager2
二、准备工作
1、安装JDK
2、配置环境变量
3、配置免密码登录
4、配置主机名
三、配置Zookeeper（在192.168.176.131安装）
在主节点（bigdata112）上配置ZooKeeper
（*）配置/training/zookeeper-3.4.6/conf/zoo.cfg文件
dataDir=/training/zookeeper-3.4.6/tmp
server.1=bigdata112:2888:3888
server.2=bigdata113:2888:3888
server.3=bigdata114:2888:3888（*）在/training/zookeeper-3.4.6/tmp目录下创建一个myid的空文件echo 1 > /training/zookeeper-3.4.6/tmp/myid
（*）将配置好的zookeeper拷贝到其他节点，同时修改各自的myid文件
scp -r /training/zookeeper-3.4.6/ bigdata113:/training
scp -r /training/zookeeper-3.4.6/ bigdata114:/training（*）分别修改113和114上/training/zookeeper-3.4.6/tmp/myid为2和3
四、安装Hadoop集群（在bigdata112上安装）
1、修改hadoo-env.shexport JAVA_HOME=/training/jdk1.8.0_144
2、修改core-site.xml
fs.defaultFS
hdfs://ns1
hadoop.tmp.dir
/training/hadoop-2.7.3/tmp
ha.zookeeper.quorum
bigdata112:2181,bigdata113:2181,bigdata114:2181
3、修改hdfs-site.xml（配置这个nameservice中有几个namenode）
dfs.replication
2
dfs.webhdfs.enabled
true
dfs.nameservices
ns1
dfs.ha.namenodes.ns1
nn1,nn2
dfs.namenode.rpc-address.ns1.nn1
bigdata112:9000
dfs.namenode.http-address.ns1.nn1
bigdata112:50070
dfs.namenode.rpc-address.ns1.nn2
bigdata113:9000
dfs.namenode.http-address.ns1.nn2
bigdata113:50070
dfs.namenode.shared.edits.dir
qjournal://bigdata112:8485;bigdata113:8485;/ns1
dfs.journalnode.edits.dir
/training/hadoop-2.7.3/journal
dfs.ha.automatic-failover.enabled
true
dfs.client.failover.proxy.provider.ns1
org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider
dfs.ha.fencing.methods
sshfence
shell(/bin/true)
dfs.ha.fencing.ssh.private-key-files
/root/.ssh/id_rsa
dfs.ha.fencing.ssh.connect-timeout
30000
4、修改mapred-site.xml
mapreduce.framework.name
yarn
5、修改yarn-site.xml
yarn.resourcemanager.ha.enabled
true
yarn.resourcemanager.cluster-id
yrc
yarn.resourcemanager.ha.rm-ids
rm1,rm2
yarn.resourcemanager.hostname.rm1
bigdata112
yarn.resourcemanager.hostname.rm2
bigdata113
yarn.resourcemanager.zk-address
bigdata112:2181,bigdata113:2181,bigdata114:2181
yarn.nodemanager.aux-services
mapreduce_shuffle
6、修改slaves
bigdata14
bigdata157、将配置好的hadoop拷贝到其他节点
scp -r /training/hadoop-2.7.3/ root@bigdata113:/training/
scp -r /training/hadoop-2.7.3/ root@bigdata114:/training/
scp -r /training/hadoop-2.7.3/ root@bigdata115:/training/五、启动Zookeeper集群
进到zk的安装目录的bin目录下：
启动
./zkServer.sh start
查看状态
./zkServer.sh status六、在bigdata112和bigdata113上启动journalnode
hadoop-daemon.sh start journalnode
七、格式化HDFS（在bigdata112上执行）
- 格式化HDFS
hdfs namenode -format
2.将112上这台的/training/hadoop-2.7.3/tmp/dfs拷贝到bigdata13的/training/hadoop-2.7.3/tmp/dfs下
scp -r /training/hadoop-2.7.3/tmp/dfs/* root@bigdata113:/training/hadoop-2.7.3/tmp/dfs/3.格式化zookeeper
hdfs zkfc -formatZK日志：
17/07/13 00:34:33 INFO ha.ActiveStandbyElector: Successfully created /hadoop-ha/ns1 in ZK.八、在bigdata12上启动Hadoop集群
start-all.sh
日志：
Starting namenodes on [bigdata12 bigdata13]
bigdata12: starting namenode, logging to /root/training/hadoop-2.4.1/logs/hadoop-root-namenode-hadoop113.out
bigdata13: starting namenode, logging to /root/training/hadoop-2.4.1/logs/hadoop-root-namenode-hadoop112.out
bigdata14: starting datanode, logging to /root/training/hadoop-2.4.1/logs/hadoop-root-datanode-hadoop115.out
bigdata15: starting datanode, logging to /root/training/hadoop-2.4.1/logs/hadoop-root-datanode-hadoop114.out
bigdata13: starting zkfc, logging to /root/training/hadoop-2.7.3/logs/hadoop-root-zkfc-bigdata13.out
bigdata12: starting zkfc, logging to /root/training/hadoop-2.7.3/logs/hadoop-root-zkfc-bigdata12.out
bigdata113上的ResourceManager需要单独启动 命令：yarn-daemon.sh start resourcemanager九、问题延伸
1、隔离机制和隔离级别
（*）关系型数据库：如果不考虑事务隔离级别，造成脏读、不可重复读、幻读
（*）HDFS的HA：如果不考虑隔离机制（隔离级别），造成脑裂的问题
2、什么是脑裂的问题？
脑裂的问题，针对的是数据节点（DataNode）
由于某种原因，造成了整个HDFS中存在多个active的NameDode，这时候DataNode就不知道谁是真正的NameNode。