运行Hadoop的集群如何动态增加与删除节点

问题导读：
1.增加节点后，Hadoop不会自动的按照新的备份系数进行调整，该如何处理？

2.增加和删除节点，都需要修改哪些配置文件？

1. 安装配置节点
2. 在配置过程中需要在所有的Hadoop服务器上更新以下三项配置
$ sudo vim /etc/hadoop/conf/slaves

hadoop-node-1
hadoop-node-2
hadoop-node-3
复制代码

$ sudo vim /etc/hosts

10.6.1.150 hadoop-master
10.6.1.151 hadoop-node-1
10.6.1.152 hadoop-node-2
10.6.1.153 hadoop-node-3
复制代码

$ sudo vim /etc/hadoop/conf/hdfs-site.xml

<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>

<configuration>
<property>
<name>dfs.data.dir</name>
<value>/data/hdfs</value>
</property>
<property>
<name>dfs.replication</name>
<value>3</value>
</property>
<property>
<name>dfs.datanode.max.xcievers</name>
<value>4096</value>
</property>
</configuration>
复制代码

3. 启动datanode与tasktracker
dongguo@hadoop-node-3:~$ sudo /etc/init.d/hadoop-0.20-datanode start
dongguo@hadoop-node-3:~$ sudo /etc/init.d/hadoop-0.20-tasktracker start

4. 检查新增节点是否已经Live
通过WEB管理界面查看

http://10.6.1.150:50070/dfsnodelist.jsp?whatNodes=LIVE

5.应用新的备份系数dfs.replication

5.1 检查目前的备份系数
dongguo@hadoop-master:~$ sudo -u hdfs hadoop fs -lsr /dongguo
-rw-r--r-- 2 hdfs supergroup 33 2012-10-07 22:02 /dongguo/hello.txt

结果行中的第2列是备份系数（注：文件夹信息存储在namenode节点上，没有备份，故文件夹的备份系数是横杠-）
目前文件的备份系数仍是之前设置的参数2，Hadoop不会自动的按照新的备份系数进行调整。

dongguo@hadoop-master:~$ sudo -u hdfs hadoop fsck /

12/10/10 21:18:32 INFO security.UserGroupInformation: JAAS Configuration already set up for Hadoop, not re-installing.
FSCK started by hdfs (auth:SIMPLE) from /10.6.1.150 for path / at Wed Oct 10 21:18:33 CST 2012
.................Status: HEALTHY
Total size:        7786 B
Total dirs:        46
Total files:        17
Total blocks (validated):        17 (avg. block size 458 B)
Minimally replicated blocks:        17 (100.0 %)
Over-replicated blocks:        0 (0.0 %)
Under-replicated blocks:        0 (0.0 %)
Mis-replicated blocks:         0 (0.0 %)
Default replication factor:        2
Average block replication:        2.0
Corrupt blocks:         0
Missing replicas:         0 (0.0 %)
Number of data-nodes:         4
Number of racks:         1
FSCK ended at Wed Oct 10 21:18:33 CST 2012 in 48 milliseconds
The filesystem under path '/' is HEALTHY
复制代码

通过 hadoop fsck / 也可以方便的看到Average block replication的值仍然为旧值2，该值我们可以手动的进行动态修改。
而Default replication factor则需要重启整个Hadoop集群才能修改，但实际影响系统的还是Average block replication的值，因此并非一定要修改默认值。

5.2 修改hdfs文件备份系数，把/ 目录下所有文件备份系数设置为3
dongguo@hadoop-master:~$ sudo -u hdfs hadoop dfs -setrep -w 3 -R /

12/10/10 21:22:35 INFO security.UserGroupInformation: JAAS Configuration already set up for Hadoop, not re-installing.
Replication 3 set: hdfs://hadoop-master/dongguo/hello.txt
Replication 3 set: hdfs://hadoop-master/hbase/-ROOT-/70236052/.oldlogs/hlog.1349695889266
Replication 3 set: hdfs://hadoop-master/hbase/-ROOT-/70236052/.regioninfo
Replication 3 set: hdfs://hadoop-master/hbase/-ROOT-/70236052/info/7670471048629837399
Replication 3 set: hdfs://hadoop-master/hbase/.META./1028785192/.oldlogs/hlog.1349695889753
Replication 3 set: hdfs://hadoop-master/hbase/.META./1028785192/.regioninfo
Replication 3 set: hdfs://hadoop-master/hbase/.META./1028785192/info/7438047560768966146
Waiting for hdfs://hadoop-master/dongguo/hello.txt .... done
Waiting for hdfs://hadoop-master/hbase/-ROOT-/70236052/.oldlogs/hlog.1349695889266 ... done
Waiting for hdfs://hadoop-master/hbase/-ROOT-/70236052/.regioninfo ... done
Waiting for hdfs://hadoop-master/hbase/-ROOT-/70236052/info/7670471048629837399 ... done
Waiting for hdfs://hadoop-master/hbase/.META./1028785192/.oldlogs/hlog.1349695889753 ... done
Waiting for hdfs://hadoop-master/hbase/.META./1028785192/.regioninfo ... done
Waiting for hdfs://hadoop-master/hbase/.META./1028785192/info/7438047560768966146 ... done
...
复制代码

可以看到Hadoop对所有文件的备份系数进行了刷新

5.3 再次检查备份系数的情况
dongguo@hadoop-master:~$ sudo -u hdfs hadoop fsck /

12/10/10 21:23:26 INFO security.UserGroupInformation: JAAS Configuration already set up for Hadoop, not re-installing.
FSCK started by hdfs (auth:SIMPLE) from /10.6.1.150 for path / at Wed Oct 10 21:23:27 CST 2012
.................Status: HEALTHY
Total size:        7786 B
Total dirs:        46
Total files:        17
Total blocks (validated):        17 (avg. block size 458 B)
Minimally replicated blocks:        17 (100.0 %)
Over-replicated blocks:        0 (0.0 %)
Under-replicated blocks:        0 (0.0 %)
Mis-replicated blocks:         0 (0.0 %)
Default replication factor:        2
Average block replication:        3.0
Corrupt blocks:         0
Missing replicas:         0 (0.0 %)
Number of data-nodes:         4
Number of racks:         1
FSCK ended at Wed Oct 10 21:23:27 CST 2012 in 11 milliseconds
The filesystem under path '/' is HEALTHY
复制代码

可以看到已经变成了新的备份系数"3"

5.4 测试一下创建新的文件时是否能集成新的备份系数
dongguo@hadoop-master:~$ sudo -u hdfs hadoop fs -copyFromLocal mysql-connector-java-5.1.22.tar.gz /dongguo
dongguo@hadoop-master:~$ sudo -u hdfs hadoop fs -lsr /dongguo

-rw-r--r-- 3 hdfs supergroup 33 2012-10-07 22:02 /dongguo/hello.txt
-rw-r--r-- 3 hdfs supergroup 4028047 2012-10-10 21:28 /dongguo/mysql-connector-java-5.1.22.tar.gz
复制代码

可以看到新上传的文件的备份系数是"3"

6 对HDFS中的文件进行负载均衡
dongguo@hadoop-master:~$ sudo -u hdfs hadoop balancer

Time Stamp Iteration# Bytes Already Moved Bytes Left To Move Bytes Being Moved
12/10/10 21:30:25 INFO net.NetworkTopology: Adding a new node: /default-rack/10.6.1.153:50010
12/10/10 21:30:25 INFO net.NetworkTopology: Adding a new node: /default-rack/10.6.1.150:50010
12/10/10 21:30:25 INFO net.NetworkTopology: Adding a new node: /default-rack/10.6.1.152:50010
12/10/10 21:30:25 INFO net.NetworkTopology: Adding a new node: /default-rack/10.6.1.151:50010
12/10/10 21:30:25 INFO balancer.Balancer: 0 over utilized nodes:
12/10/10 21:30:25 INFO balancer.Balancer: 0 under utilized nodes: 
The cluster is balanced. Exiting...
Balancing took 1.006 seconds
复制代码

至此，Hadoop集群的动态增加就已经完成了。
下面，我开始对Hadoop集群的节点进行动态的删除。

Hadoop集群节点的动态删除
1. 使用新增的节点
尽可能的在HDFS中产生一些测试数据，并通过Hive执行一些Job以便使新的节点也执行MapReduce并行计算。
这样做的原因是尽可能的模拟线上的环境，因为线上环境在进行删除节点之前肯定是有很多数据和Job执行过的。

2. 修改core-site.xml
dongguo@hadoop-master:~$ sudo vim /etc/hadoop/conf/core-site.xml

<property>
<name>dfs.hosts.exclude</name>
<value>/etc/hadoop/conf/exclude</value>
<description>Names a file that contains a list of hosts that are
not permitted to connect to the namenode. The full pathname of the
file must be specified. If the value is empty, no hosts are
excluded.</description>
</property>
复制代码

3. 修改hdfs-site.xml
dongguo@hadoop-master:~$ sudo vim /etc/hadoop/conf/hdfs-site.xml

<property>
<name>dfs.hosts.exclude</name>
<value>/etc/hadoop/conf/exclude</value>
<description>Names a file that contains a list of hosts that are
not permitted to connect to the namenode. The full pathname of the
file must be specified. If the value is empty, no hosts are
excluded.</description>
</property>
复制代码

4. 创建/etc/hadoop/conf/exclude
dongguo@hadoop-master:~$ sudo vim /etc/hadoop/conf/exclude

hadoop-node-3
复制代码

在文件中增加需要删除的节点，一行一个，我这里仅需要写入新增的hadoop-node-3做测试。

5. 降低备份系数
在我的测试环境中，目前节点为4台，备份系数为3，如果去掉一台的话备份系数就与节点数相同了，而Hadoop是不允许的。
通常备份系数不需要太高，可以是服务器总量的1/3左右即可，Hadoop默认的数值是3。

下面，我们将备份系数从3降低到2

5.1 在所有的Hadoop服务器上更新以下配置
$ sudo vim /etc/hadoop/conf/hdfs-site.xml

<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>

<configuration>
<property>
<name>dfs.data.dir</name>
<value>/data/hdfs</value>
</property>
<property>
<name>dfs.replication</name>
<value>2</value>
</property>
<property>
<name>dfs.datanode.max.xcievers</name>
<value>4096</value>
</property>
</configuration>
复制代码

5.2 修改hdfs文件备份系数，把/ 目录下所有文件备份系数设置为2
dongguo@hadoop-master:~$ sudo -u hdfs hadoop dfs -setrep -w 2 -R /

遇到的疑问：
在进行文件备份系数的降低时，能够很快的进行Replication set，但是在Waiting for的过程中却很长时间没有完成。
最终只能手动Ctrl+C中断，个人猜测在这个过程中HDFS正视图对数据文件进行操作，在删除一个副本容量的数据。
因此，我们应该对dfs.replication的数值做出很好的规划，尽量避免需要降低该数值的情况出现。

6. 动态刷新配置
dongguo@hadoop-master:~$ sudo -u hdfs hadoop dfsadmin -refreshNodes

7. 检查节点的处理状态
通过WEB管理界面查看
Decommissioning(退役中)

http://10.6.1.150:50070/dfsnodelist.jsp?whatNodes=DECOMMISSIONING

Dead(已经下线)

http://10.6.1.150:50070/dfsnodelist.jsp?whatNodes=DEAD

可以看到，节点已经经历了退役的过程并成功的下线了。

需要注意的是：
在删除节点时一定要停止所有Hadoop的Job，否则程序还会向要删除的节点同步数据，这样也会导致Decommission的过程一直无法完成。

8. 检查进程状态
这时我们查看进程状态，可以发现datanode进程已经被自动中止了
dongguo@hadoop-node-3:~$ sudo /etc/init.d/hadoop-0.20-datanode status
hadoop-0.20-datanode is not running.

而Tasktracker进程还在，需要我们手动中止
dongguo@hadoop-node-3:~$ sudo /etc/init.d/hadoop-0.20-tasktracker status
hadoop-0.20-tasktracker is running
dongguo@hadoop-node-3:~$ sudo /etc/init.d/hadoop-0.20-tasktracker stop
Stopping Hadoop tasktracker daemon: stopping tasktracker
hadoop-0.20-tasktracker.

此时，即使我们手动启动datanode，也是不能成功的，日志中会显示UnregisteredDatanodeException的错误。
dongguo@hadoop-node-3:~$ sudo /etc/init.d/hadoop-0.20-datanode start

Starting Hadoop datanode daemon: starting datanode, logging to /usr/lib/hadoop-0.20/logs/hadoop-hadoop-datanode-hadoop-node-3.out
ERROR. Could not start Hadoop datanode daemon
复制代码

dongguo@hadoop-node-3:~$ tailf /var/log/hadoop/hadoop-hadoop-datanode-hadoop-node-3.log


2012-10-11 19:33:22,084 ERROR org.apache.hadoop.hdfs.server.datanode.DataNode: org.apache.hadoop.ipc.RemoteException: org.apache.hadoop.hdfs.protocol.UnregisteredDatanodeException: Data node hadoop-node-3:50010 is attempting to report storage ID DS-500645823-10.6.1.153-50010-1349941031723. Node 10.6.1.153:50010 is expected to serve this storage.
at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.getDatanode(FSNamesystem.java:4547)
at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.verifyNodeRegistration(FSNamesystem.java:4512)
at org.apache.hadoop.hdfs.server.namenode.FSNamesystem.registerDatanode(FSNamesystem.java:2355)
at org.apache.hadoop.hdfs.server.namenode.NameNode.register(NameNode.java:932)
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:39)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
at java.lang.reflect.Method.invoke(Method.java:597)
at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:557)
at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1434)
at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1430)
at java.security.AccessController.doPrivileged(Native Method)
at javax.security.auth.Subject.doAs(Subject.java:396)
at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1157)
at org.apache.hadoop.ipc.Server$Handler.run(Server.java:1428)
at org.apache.hadoop.ipc.Client.call(Client.java:1107)
at org.apache.hadoop.ipc.RPC$Invoker.invoke(RPC.java:226)
at $Proxy4.register(Unknown Source)
at org.apache.hadoop.hdfs.server.datanode.DataNode.register(DataNode.java:717)
at org.apache.hadoop.hdfs.server.datanode.DataNode.runDatanodeDaemon(DataNode.java:1519)
at org.apache.hadoop.hdfs.server.datanode.DataNode.createDataNode(DataNode.java:1586)
at org.apache.hadoop.hdfs.server.datanode.DataNode.secureMain(DataNode.java:1711)
at org.apache.hadoop.hdfs.server.datanode.DataNode.main(DataNode.java:1728)
2012-10-11 19:33:22,097 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: SHUTDOWN_MSG: 
/************************************************************
SHUTDOWN_MSG: Shutting down DataNode at hadoop-node-3/10.6.1.153
************************************************************/
复制代码