We are working on redesigning our currently cluster. Our 40 ES data nodes are split between 2 data centers, 20 in each. Currently there are 3 master dedicated nodes, 2 in one DC, and 1 in the other. The goal with the two datacenters would be to have the primary shard in one DC and the replica in the other.
Theoretically, if 1 datacenter was to go offline, a new master election would fail because there wouldn't be enough nodes for a quorum. If we increased the number of masters this problem continues as 3,5,7 etc. all will have the same issue.
You can not create a symmetrically highly available cluster across just 2 data centres. You therefore need a dedicated master node in a third data center to act as an arbiter. Voting only nodes have recently been introduced and could be used as arbiter but if you are on an earlier version a dedicated master node will be your best bet.
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.