How-to: Enable yarn ResourceManager HA

Update yarn-site.xml. Add following in yarn-site.xml(master.chff.dc and slave01.chff.dc are resourcemanager nodes):

        <property>

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

                <value>true</value>

        </property>

        <property>

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

                <value>true</value>

        </property>

        <!-- 使嵌入式自动故障转移。HA环境启动，与 ZKRMStateStore 配合 处理fencing -->

        <property>

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

                <value>true</value>

        </property>

        <!-- 集群名称，确保HA选举时对应的集群 -->

        <property>

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

        <value>yarn-cluster</value>

        </property>

        <property>

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

                <value>rm1,rm2</value>

        </property>

        <property>

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

                <value>master.chff.dc</value>

        </property>

        <property>

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

                <value>slave01.chff.dc</value>

        </property>

        <property>

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

                <value>org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairScheduler</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>

        <!-- ZKRMStateStore 配置 -->

        <property>

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

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

        </property>

        <property>

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

                <value>master.chff.dc:2183,slave01.chff.dc:2183,slave.chff.dc:2183</value>

        </property>

        <property>

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

                <value>master.chff.dc:2183,slave01.chff.dc:2183,slave.chff.dc:2183</value>

        </property>

        <!-- Client访问RM的RPC地址 (applications manager interface) -->

        <property>

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

                <value>master.chff.dc:23140</value>

        </property>

        <property>

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

                <value>slave01.chff.dc:23140</value>

        </property>

        <!-- AM访问RM的RPC地址(scheduler interface) -->

        <property>

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

                <value>master.chff.dc:23130</value>

        </property>

        <property>

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

                <value>slave01.chff.dc:23130</value>

        </property>

        <!-- RM admin interface -->

        <property>

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

                <value>master.chff.dc:23141</value>

        </property>

        <property>

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

                <value>slave01.chff.dc:23141</value>

        </property>

        <!--NM访问RM的RPC端口 -->

        <property>

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

                <value>master.chff.dc:23125</value>

        </property>

        <property>

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

                <value>slave01.chff.dc:23125</value>

        </property>

        <!-- RM web application 地址 -->

        <property>

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

                <value>master.chff.dc:8088</value>

        </property>

        <property>

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

                <value>slave01.chff.dc:8088</value>

        </property>

        <property>

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

                <value>master.chff.dc:8089</value>

        </property>

        <property>

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

                <value>slave01.chff.dc:8089</value>

        </property>
start yarn on two nodes:

start-yarn.sh
Check status:

[hadoop@master deploy_scripts]$ yarn rmadmin -getServiceState rm1

active

[hadoop@master deploy_scripts]$ yarn rmadmin -getServiceState rm2

standby