大數據開發之hadoop HA配置

hadoop HA

一、原理

      HA高可用

      存儲

           當editlog發生變化時，則直接寫入JournalNode，以用來分享給其他的NameNode

二、安裝部署

       step1、配置zookeeper

       step2、配置hadoop

       hadoop-env.sh

       core-site.xml

           <configuration>

                 <property>

                 <name>fs.defaultFS</name>

                 <value>hdfs://mycluster</value>

</property>

 

                 <property>

                 <name>hadoop.tmp.dir</name>

                 <value>/opt/modules/cdh/hadoop-2.5.0-cdh5.3.6/data</value>

                 </property>

</configuration>

 

           zookeeper地址

                 <property>

                 <name>ha.zookeeper.quorum</name>                              <value>hlh001:2181,hlh002:2181,hlh003:2181</value>

                 </property>

 

       hdfs-site.xml

                 <configuration>

                      <!-- 指定數據冗餘份數 -->

                       <property>

                      <name>dfs.replication</name>

                      <value>3</value>

                       </property>

 

                      <!-- 完全分佈式集羣名稱 -->

                       <property>

                      <name>dfs.nameservices</name>

                      <value>mycluster</value>

                       </property>

 

                      <!-- 集羣中NameNode節點都有哪些 -->

                       <property>

                      <name>dfs.ha.namenodes.mycluster</name>

                      <value>nn1,nn2</value>

                       </property>

 

                      <!-- nn1的RPC通信地址 -->

                       <property>

                      <name>dfs.namenode.rpc-address.mycluster.nn1</name>

                      <value>z04:8020</value>

                       </property>

 

                      <!-- nn2的RPC通信地址 -->

                       <property>

                      <name>dfs.namenode.rpc-address.mycluster.nn2</name>

                      <value>z05:8020</value>

                       </property>

 

                      <!-- nn1的http通信地址 -->

                       <property>

                      <name>dfs.namenode.http-address.mycluster.nn1</name>

                      <value>z04:50070</value>

                       </property>

 

                      <!-- nn2的http通信地址 -->

                       <property>

                      <name>dfs.namenode.http-address.mycluster.nn2</name>

                      <value>z05:50070</value>

                       </property>

 

                 <!-- 指定NameNode元數據在JournalNode上的存放位置 -->

                       <property>

                      <name>dfs.namenode.shared.edits.dir</name>

                            <value>qjournal://z04:8485;z05:8485;z06:8485/mycluster</value>

                       </property>

 

                      <!-- 配置隔離機制，即同一時刻只能有一臺服務器對外響應 -->

                       <property>

                      <name>dfs.ha.fencing.methods</name>

                      <value>sshfence</value>

                       </property>

 

                      <!-- 使用隔離機制時需要ssh無祕鑰登錄-->

                       <property>

                      <name>dfs.ha.fencing.ssh.private-key-files</name>

                      <value>/home/z/.ssh/id_rsa</value>

                       </property>

 

                      <!-- 聲明journalnode服務器存儲目錄-->

                       <property>

                      <name>dfs.journalnode.edits.dir</name>

                      <value>/opt/modules/cdh/hadoop-2.5.0-cdh5.3.6/data/jn</value>

                       </property>

 

                      <!-- 關閉權限檢查-->

                       <property>

                      <name>dfs.permissions.enable</name>

                      <value>false</value>

                       </property>

 

                      <!-- 訪問代理類：client，mycluster，active配置失敗自動切換實現方式-->

                       <property>

                     <name>dfs.client.failover.proxy.provider.mycluster</name>

                      <value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value>

                 </property>

                

</configuration>

故障自動轉移

      <property>

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

      </property>

 

yarn-site.xml

      <configuration>

 

<!-- Site specific YARN configuration properties -->

           <property>

               <name>yarn.nodemanager.aux-services</name>

                <value>mapreduce_shuffle</value>

              </property>

 

            <property>

               <name>yarn.log-aggregation-enable</name>

                   <value>true</value>

           </property>

     

            <property>

               <name>yarn.log.server.url</name>

                <value>http://z01:19888/jobhistory/logs/</value>

           </property>

 

              <property>

                   <name>yarn.log-aggregation.retain-seconds</name>

               <value>86400</value>

           </property>

 

            <!--啓用resourcemanager ha-->

           <property>

                  <name>yarn.resourcemanager.ha.enabled</name>

                   <value>true</value>

              </property>

 

           <!--聲明兩臺resourcemanager的地址-->

           <property>

                   <name>yarn.resourcemanager.cluster-id</name>

                <value>cluster-yarn1</value>

           </property>

 

               <property>

               <name>yarn.resourcemanager.ha.rm-ids</name>

                <value>rm1,rm2</value>

           </property>

 

           <property>

                  <name>yarn.resourcemanager.hostname.rm1</name>

               <value>hlh002</value>

            </property>

 

           <property>

                  <name>yarn.resourcemanager.hostname.rm2</name>

                  <value>hlh003</value>

            </property>

 

           <!--指定zookeeper集羣的地址-->

           <property>

               <name>yarn.resourcemanager.zk-address</name>

                <value>hlh001:2181,hlh002:2181,hlh003:2181</value>

           </property>

 

           <!--啓用自動恢復-->

               <property>

                  <name>yarn.resourcemanager.recovery.enabled</name>

                <value>true</value>

           </property>

 

           <!--指定resourcemanager的狀態信息存儲在zookeeper集羣-->

            <property>

 <name>yarn.resourcemanager.store.class</name>     <value>org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore</value>

              </property>

</configuration>