1. Turn off the cluster
stop-dfs.sh
stop-yarn.sh
2. The cluster each VERSION datanode deleted node
(VERSION datanodeid which recorded information of the path / hadoop / tmp / dfs / data / current)
cd /usr/local/hadoop/tmp/dfs/data/current/
ls
rm -f VERSION
3. Do hdfs namenode -format (format statement)
hdfs namenode -format
4. Restart the cluster
start-dfs.sh
start-yarn.sh