I wonder whether this is also true for the Client nodes. Do ES Client nodes also need half of the available memory for Lucene, since it seems that the Client nodes are mostly doing the merging from the Data nodes. What is recommended?
For client and tribe nodes (as well as master-only nodes), allocating half of OS memory for off-heap operations is not necessary. Merging results from the data nodes does not use off-heap memory.
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.