In HDFS rolling upgrade page at a high level I am seeing the steps mentioned like below
But in cloudera documentation of Hadoop to prepare the cluster for upgrade it is just saying to enter NN in safemode & save namespace
sudo -u hdfs hdfs dfsadmin -safemode enter sudo -u hdfs hdfs dfsadmin -saveNamespace
The first step in Cloudera documentation (-safemode enter) is for NN to enter "read-only" mode, where it stops accepting new changes to the filesystem or blocks. This is done to make sure you get a consistent copy of the in-memory fsimage file. The next step (-savenamespace) is basically writing that consistent copy of in-memory fsimage to the fsimage on disk, so that just in case anything wrong during the upgrade, your data is not lost.
The first command on the HDFS rolling upgrade page (-rollingUpgrade prepare) does the same thing in the background.
When you shutdown or restart HDFS, NN comes out of safemode automatically.