一、服役新数据节点

1、需求
       随着公司业务的增长，数据量越来越大，原有的数据节点的容量已经不能满足存储数据的需求，需要在原有集群基础上动态添加新的数据节点。

2、环境准备
(1) 在 hadoop153 主机上再克隆一台 hadoop154 主机

(2) 修改 hadoop153 的ip地址和主机名称

(3) 修改 root 与 test 用户的 xsync 分发脚本，可以分发到 hadoop154 上

(4) 在 hadoop151 上修改主机名与ip映射，并用 root 用户分发

(5) 配置 NameNode 与 ResourceManager 这两台机器可以 ssh 登录 hadoop154

(6) 删除原来HDFS文件系统留存的文件(/opt/module/hadoop-2.7.2/data 和 log)

3、添加白名单

(1) 在 NameNode 的 /opt/module/hadoop-2.7.2/etc/hadoop 目录下创建 dfs.hosts 文件
添加到白名单的主机节点，都允许访问NameNode，不在白名单的主机节点，都会被退出

[test@hadoop151 hadoop]$ touch dfs.hosts
[test@hadoop151 hadoop]$ vim dfs.hosts 
[test@hadoop151 hadoop]$ cat dfs.hosts 
hadoop151
hadoop152
hadoop153
hadoop154

(2) 在 NameNode 的 hdfs-site.xml 配置文件中增加 dfs.hosts 属性

<property>
	<name>dfs.hosts</name>
	<value>/opt/module/hadoop-2.7.2/etc/hadoop/dfs.hosts</value>
</property>

(3) 配置文件分发

[test@hadoop151 hadoop]$ xsync hdfs-site.xml 
fname=hdfs-site.xml
pdir=/opt/module/hadoop-2.7.2/etc/hadoop
------------------- hadoop151 --------------
sending incremental file list

sent 36 bytes  received 12 bytes  96.00 bytes/sec
total size is 1468  speedup is 30.58
------------------- hadoop152 --------------
sending incremental file list
hdfs-site.xml

sent 87 bytes  received 49 bytes  272.00 bytes/sec
total size is 1468  speedup is 10.79
------------------- hadoop153 --------------
sending incremental file list
hdfs-site.xml

sent 87 bytes  received 49 bytes  90.67 bytes/sec
total size is 1468  speedup is 10.79
------------------- hadoop154 --------------
sending incremental file list
hdfs-site.xml

sent 851 bytes  received 43 bytes  1788.00 bytes/sec
total size is 1468  speedup is 1.64

(4) 刷新 NameNode
在 NameNode 节点上 hadoop151 执行如下命令：

[test@hadoop151 hadoop]$ hdfs dfsadmin -refreshNodes
Refresh nodes successful

(5) 刷新 ResourceManager
在 ResourceManager 节点 hadoop152 执行如下命令：

[test@hadoop152 ~]$ yarn rmadmin -refreshNodes
20/02/04 13:48:19 INFO client.RMProxy: Connecting to ResourceManager at hadoop152/192.168.1.152:8033

(6) 在 hadoop154 上单点启动 datanode 与 nodemanager

[test@hadoop154 hadoop-2.7.2]$ hadoop-daemon.sh start datanode
starting datanode, logging to /opt/module/hadoop-2.7.2/logs/hadoop-test-datanode-hadoop154.out
[test@hadoop154 hadoop-2.7.2]$ yarn-daemon.sh start nodemanager
starting nodemanager, logging to /opt/module/hadoop-2.7.2/logs/yarn-test-nodemanager-hadoop154.out

(7) web端查看

(8) 如果数据不均衡，可以用命令实现集群的再平衡

[test@hadoop151 hadoop]$ start-balancer.sh
starting balancer, logging to /opt/module/hadoop-2.7.2/logs/hadoop-test-balancer-hadoop151.out
Time Stamp               Iteration#  Bytes Already Moved  Bytes Left To Move  Bytes Being Moved

二、退役数据节点

采用黑名单退役的方式，在黑名单上面的所有主机都会被强制退出。

1、在 NameNode 上创建 dfs.hosts.exclude 文件
添加要退役的主机节点名称

[test@hadoop151 hadoop]$ pwd
/opt/module/hadoop-2.7.2/etc/hadoop
[test@hadoop151 hadoop]$ touch dfs.hosts,exclude
[test@hadoop151 hadoop]$ cat dfs.hosts.exclude 
hadoop154

2、修改 hdfs-site.xml 并分发

<property>
	  <name>dfs.hosts.exclude</name>
      <value>/opt/module/hadoop-2.7.2/etc/hadoop/dfs.hosts.exclude</value>
</property>

[test@hadoop151 hadoop]$ xsync hdfs-site.xml 
fname=hdfs-site.xml
pdir=/opt/module/hadoop-2.7.2/etc/hadoop
------------------- hadoop151 --------------
sending incremental file list

sent 36 bytes  received 12 bytes  32.00 bytes/sec
total size is 1626  speedup is 33.88
------------------- hadoop152 --------------
sending incremental file list
hdfs-site.xml

sent 313 bytes  received 49 bytes  724.00 bytes/sec
total size is 1626  speedup is 4.49
------------------- hadoop153 --------------
sending incremental file list
hdfs-site.xml

sent 313 bytes  received 49 bytes  724.00 bytes/sec
total size is 1626  speedup is 4.49
------------------- hadoop154 --------------
sending incremental file list
hdfs-site.xml

sent 313 bytes  received 49 bytes  724.00 bytes/sec
total size is 1626  speedup is 4.49

3、刷新 NameNode、刷新 ResourceManager

[test@hadoop151 hadoop]$ hdfs dfsadmin -refreshNodes
Refresh nodes successful
[test@hadoop152 ~]$ yarn rmadmin -refreshNodes
20/02/04 14:52:23 INFO client.RMProxy: Connecting to ResourceManager at hadoop152/192.168.1.152:8033

4、检查Web浏览器，退役节点的状态为decommission in progress(退役中)，说明数据节点正在复制块到其他节点

5、等待退役节点状态为 decommissioned(所有块已经复制完成)，停止该节点及节点资源管理器。这里要注意：如果副本数是3，服役的节点小于等于3，是不能退役成功的，需要修改副本数后才能退役

6、如果数据不均衡，可以用命令实现集群的再平衡

[test@hadoop151 hadoop]$ start-balancer.sh
starting balancer, logging to /opt/module/hadoop-2.7.2/logs/hadoop-test-balancer-hadoop151.out
Time Stamp               Iteration#  Bytes Already Moved  Bytes Left To Move  Bytes Being Moved

注：不允许白名单和黑名单中同时出现同一个主机名称。