@Mudassar Hussain The steps 1,2 and 3 are okay.You dont need to create directory2, when you enable a directory as snapshottable in this case /user/mudassar/snapdemo the snapshots will be created under this directory with .snapshot....... which makes it invisible when you run the hdfs dfs -ls command Let me demo on HDP 2.6 I will create your user on my local environment. As mudassar as the root user # adduser mudassar Switch to superuser and owner of HDFS # su - hdfs Create the snapshot demo notice the -p option as the root directory /user/mudassar doesn't exist yet Note: hdfs hadoop will be deprecated so use hdfs dfs command! $ hdfs dfs -mkdir -p /user/mudassar/snapdemo Validate directory $ hdfs dfs -ls /user/mudassar Found 1 items drwxr-xr-x - hdfs hdfs 0 2018-03-15 09:39 /user/mudassar/snapdemo Change ownership to mudassar $ hdfs dfs -chown mudassar /user/mudassar/snapdemo Validate change of ownership $ hdfs dfs -ls /user/mudassar Found 1 items drwxr-xr-x - mudassar hdfs 0 2018-03-15 09:39 /user/mudassar/snapdemo Make the directory snapshottable $ hdfs dfsadmin -allowSnapshot /user/mudassar/snapdemo Allowing snaphot on /user/mudassar/snapdemo succeeded Show all the snapshottable directories in your cluster a subcommand under hdfs $ hdfs lsSnapshottableDir drwxr-xr-x 0 mudassar hdfs 0 2018-03-15 09:39 0 65536 /user/mudassar/snapdemo Create 2 sample files in /tmp $ echo "Test one for snaphot No worries No worries I was worried you got stuck and didn't revert the HCC is full of solutions so" > /tmp/text1.txt $ echo "The default behavior is that only a superuser is allowed to access all the resources of the Kafka cluster, and no other user can access those resources" > /tmp/text2.txt Validate the files were created $cd /tmp $ls -lrt -rw-r--r-- 1 hdfs hadoop 121 Mar 15 10:04 text1.txt -rw-r--r-- 1 hdfs hadoop 152 Mar 15 10:04 text2.txt Copy the files from locall to HDFS $ hdfs dfs -put text1.txt /user/mudassar/snapdemo Create a snapshot of the file text1.txt $ hdfs dfs -createSnapshot /user/mudassar/snapdemo Created snapshot /user/mudassar/snapdemo/.snapshot/s20180315-101148.262 Note above the .snapshot directory which is a hidden system directory Show the snapshot of text1.txt $ hdfs dfs -ls /user/mudassar/snapdemo/.snapshot/s20180315-101148.262 Found 1 items -rw-r--r-- 3 hdfs hdfs 121 2018-03-15 10:10 /user/mudassar/snapdemo/.snapshot/s20180315-101148.262/text1.txt Copied the second file text2.txt from local /tmp to HDFS $ hdfs dfs -put text2.txt /user/mudassar/snapdemo Validation that the 2 files should be resent $ hdfs dfs -ls /user/mudassar/snapdemo Found 2 items -rw-r--r-- 3 hdfs hdfs 121 2018-03-15 10:10 /user/mudassar/snapdemo/text1.txt -rw-r--r-- 3 hdfs hdfs 152 2018-03-15 10:19 /user/mudassar/snapdemo/text2.txt Demo simulate loss of file text1.txt $ hdfs dfs -rm /user/mudassar/snapdemo/text1.txt Indeed file text1.txt was deleted ONLY text2.txt remains $ hdfs dfs -ls /user/mudassar/snapdemo Found 1 items -rw-r--r-- 3 hdfs hdfs 152 2018-03-15 10:19 /user/mudassar/snapdemo/text2.txt Restore the text1.txt $ hdfs dfs -cp -ptopax /user/mudassar/snapdemo/.snapshot/s20180315-101148.262/text1.txt /user/mudassar/snapdemo To use -ptopax this ensure the timestamp is restored you will need to set the dfs.namenode.accesstime.precision to default 1 hr which is 360000 seconds Check the original timestamp for text1.txt above !!! hdfs dfs -ls /user/mudassar/snapdemo Found 2 items -rw-r--r-- 3 hdfs hdfs 121 2018-03-15 10:10 /user/mudassar/snapdemo/text1.txt -rw-r--r-- 3 hdfs hdfs 152 2018-03-15 10:19 /user/mudassar/snapdemo/text2.txt In a nutshell, you don't need to create directory2 because when you run hdfs dfs -createSnapshot command it autocreates a directory under the original starting with .snapshot, that also saves you from extra steps of creating sort of a backup directory. I hope that explains it clearly this time ... View more

@Vivek Sabha Unfortunately, that is not possible with the current versions of HDP aka Hadoop 2.0 but Hadoop 3.0 current GA will support more than 2 NameNodes. See this doc which includes all the significant enhancements. The only viable option is to migrate a Namenode from one machine to another via Ambari, that should be the only solution for you and here is a HCC document by Kuldeep Kulkarni that you could use as a guide, unfortunately its 2 years old valid for Ambari - 2.2.2.X running HDP - 2.2.X-->/2.4.X Hope that helps ... View more