HDFS failover启动错误Parent znode does not exist

现象

HDFS配置了高可用，NameNode和DataNode状态看起来均正常，但两个failover进程异常。
查看failover日志，报错信息如下，

tail -100f /var/log/hadoop-hdfs/hadoop-cmf-hdfs-FAILOVERCONTROLLER-hxpernode01.esgyn.cn.log.out

2019-10-24 19:51:13,396 FATAL org.apache.hadoop.ha.ZKFailoverController: Unable to start failover controller. Parent znode does not exist.
Run with -formatZK flag to initialize ZooKeeper.
2019-10-24 19:51:13,397 INFO org.apache.zookeeper.ZooKeeper: Session: 0x36dfd9a6aa60005 closed
2019-10-24 19:51:13,397 INFO org.apache.zookeeper.ClientCnxn: EventThread shut down
2019-10-24 19:51:13,397 INFO org.apache.hadoop.hdfs.tools.DFSZKFailoverController: SHUTDOWN_MSG: 
/************************************************************
SHUTDOWN_MSG: Shutting down DFSZKFailoverController at hxpernode01.esgyn.cn/10.19.41.49
************************************************************/

解决

从Zookeeper查看没有hadoop-ha节点。因此需要执行“初始化自动故障转换Znode”，如下图