Cloudera CDH5 RM HA功能验证

简介: 最新的Cloudera CDH5.0.0 beta版本已经支撑RM的HA， 笔者为此简单验证了RM HA的功能. 后续将持续解析其HA的道理，以及其与社区RM HA的差别. 

 

 

集群属下与RM failover功能性验证 

 

1. 硬件筹办 
   

   四台机械， bj1， bj3， bj4， bj5 筹办好响应的景象（包含ssh互通， java景象）. 
   

   角色申明， bj1为rm1， bj3为rm2， bj4和bj4为slave. 
   

   Zookeeper属下在bj1上. 
   

2. Hadoop版本筹办http://archive.cloudera.com/cdh5/cdh/5/ 响应的CDH5版本hadoop-2.2.0-cdh5.0.0-beta-1.tar.gz（包含属下包和原代码），然后属下到每台slave中. 
   
3. Zookeeper安装在bj1， 最新Zookeeper， 解压后设备 conf/zoo.cfg文件， 然后启动. 
   

   [yuling.sh＠v125050024 ~]￥ cd zookeeper-3.4.3/ 
   

   [yuling.sh＠v125050024 zookeeper-3.4.3]￥ cp conf/zoo_sample.cfg conf/zoo.cfg 
   

   [yuling.sh＠v125050024 zookeeper-3.4.3]￥ bin/zkServer.sh start 
   

4. 设备文件筹办，参考（https://www.cloudera.com/content/cloudera-content/cloudera-docs/CDH5/latest/CDH5-High-Availability-Guide/cdh5hag_cfg_RM_HA.html）. 
   
   1. etc/hadoop/slaves 
      

      bj4 
      

      bj5 
      

   2. etc/hadoop/hdfs-site.xml 
      

   <property> 
   

   <name>fs.default.name</name> 
   

   <value>hdfs://bj1:9000</value> 
   

   </property> 
   

   1. etc/hadoop/mapred-site.xml 
      

      <property> 
      

      <name>mapreduce.framework.name</name> 
      

      <value>yarn</value> 
      

      </property> 
      

   2. etc/hadoop/yarn-site.xml设备如下 
      

   除了yarn.resourcemanager.ha.id须要稍作批改外， 其它设备都可以一样. 
   

   <!-- Resource Manager Configs --> 
   

   <property> 
   

   <name>yarn.resourcemanager.connect.retry-interval.ms</name> 
   

   <value>2000</value> 
   

   </property> 
   

   <property> 
   

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

   <value>true</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.ha.automatic-failover.enabled</name> 
   

   <value>true</value> 
   

   </property> 
   

   <property> 
   

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

   <value>rm1，rm2</value> 
   

   </property> 
   

   <property> 
   

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

   <value>rm2</value> <!—注释， rm1上设备为rm1， rm2上设备rm2--> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.store.class</name> 
   

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

   </property> 
   

   <property> 
   

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

   <value>bj1:2181</value> 
   

   </property> 
   

   
   ?

   <property> 
   

   <name>ha.zookeeper.quorum</name> 
   

   <value>bj1:2181</value> 
   

   </property> 
   

   
   ?

   <property> 
   

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

   <value>true</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.app.mapreduce.am.scheduler.connection.wait.interval-ms</name> 
   

   <value>5000</value> 
   

   </property> 
   

   <!-- RM1 configs --> 
   

   <property> 
   

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

   <value>bj1:23140</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.scheduler.address.rm1</name> 
   

   <value>bj1:23130</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.webapp.address.rm1</name> 
   

   <value>bj1:23188</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.resource-tracker.address.rm1</name> 
   

   <value>bj1:23125</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.admin.address.rm1</name> 
   

   <value>bj1:23141</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.ha.admin.address.rm1</name> 
   

   <value>bj1:23142</value> 
   

   </property> 
   

   <!-- RM2 configs --> 
   

   <property> 
   

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

   <value>bj3:23140</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.scheduler.address.rm2</name> 
   

   <value>bj3:23130</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.webapp.address.rm2</name> 
   

   <value>bj3:23188</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.resource-tracker.address.rm2</name> 
   

   <value>bj3:23125</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.admin.address.rm2</name> 
   

   <value>bj3:23141</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.resourcemanager.ha.admin.address.rm2</name> 
   

   <value>bj3:23142</value> 
   

   </property> 
   

   <!-- Node Manager Configs --> 
   

   <property> 
   

   <description>Address where the localizer IPC is.</description> 
   

   <name>yarn.nodemanager.localizer.address</name> 
   

   <value>0.0.0.0:23344</value> 
   

   </property> 
   

   <property> 
   

   <description>NM Webapp address.</description> 
   

   <name>yarn.nodemanager.webapp.address</name> 
   

   <value>0.0.0.0:23999</value> 
   

   </property> 
   

   <property> 
   

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

   <value>mapreduce_shuffle</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.nodemanager.aux-services.mapreduce.shuffle.class</name> 
   

   <value>org.apache.hadoop.mapred.ShuffleHandler</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.nodemanager.local-dirs</name> 
   

   <value>/tmp/pseudo-dist/yarn/local</value> 
   

   </property> 
   

   <property> 
   

   <name>yarn.nodemanager.log-dirs</name> 
   

   <value>/tmp/pseudo-dist/yarn/log</value> 
   

   </property> 
   

   <property> 
   

   <name>mapreduce.shuffle.port</name> 
   

   <value>23080</value> 
   

   </property> 
   

5. 起首启动HDFS 
   

   bin/hadoop namenode –format 
   

   sbin/start-dfs.sh 
   

   网页上查看Namenode:　 http://bj1:50070/dfshealth.jsp 
   

 

1. 启动Yarn 
   

   rm1上启动resourcemanager 
   

   sbin/yarn-daemon.sh start resourcemanager 
   

 

rm2上启动resourcemanager 

sbin/yarn-daemon.sh start resourcemanager 

?

slave启动NodeManager 

????sbin/yarn-daemons.sh start nodemanager 

查看rm1和mr2的网页. http://bj1:23188/cluster 和 http://bj3:23188/cluster 此中active RM的网页可以查看， stanby的RM无法查看网页. 

注: 若是yarn.resourcemanager.ha.automatic-failover.enabled设置为false， 则须要手动设置此中一个RM为active，负责两个RM都为standby. 

1. 提交一个sleep功课测试 
   

   bin/hadoop jar share/hadoop/mapreduce/hadoop-mapreduce-client-jobclient-2.2.0-cdh5.0.0-beta-1.jar sleep -m 1000 
   

   然后可以到网页上查看功课运行景象 
   

2. 在功课运行过程中kill掉active的RM过程， 这时辰打开standby RM的网页，可以看到刚才提交的功课持续运行. 
   

   [yuling.sh＠v125050024 hadoop-2.2.0-cdh5.0.0-beta-1]￥ jps 
   

   31333 ResourceManager 
   

   31671 Jps 
   

   29502 NameNode 
   

   25375 QuorumPeerMain 
   

   [yuling.sh＠v125050024 hadoop-2.2.0-cdh5.0.0-beta-1]￥ kill 31333