I'm asking if it's really necessary to disable replica shard allocation for each node as described in the documentation
As this is a cluster setting, why not disable replica shard allocation one time, upgrade all nodes and then re-enable replicas shard allocation at the end ?
You should re-enable it and allow the cluster to recover to green health after each node if possible. If you do not, your cluster health may end up red (i.e. some shards are completely unavailable) during the process, which rather defeats the purpose of doing a rolling upgrade.
I think it's pretty likely to be honest, unless you have stopped all write traffic for the duration of the upgrade. Again, this rather defeats the purpose of doing a rolling upgrade.
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.