In the current release of Elasticsearch, the dicovery.zen.minimum_master_nodes settings have been removed. I have a cluster of 5 nodes cross data center (yes I KNOW it's not a good idea but I don't really have the choice), 2 on one Data Center and the 3 others on the other one. Setting the dicovery.zen.minimum_master_nodes settings to 3 assured me to still be UP in case of a split brain.
How I'm I supposed to this now with the cluster.initial_master_nodes only as it is only for the very first time the cluster is started?
Elasticsearch (versions 7 and later) manages the size of the voting set of master nodes itself, so you don't have to worry about this any more. I made a short video about this which shows how to observe what's going on under the hood:
Hey @DavidTurner thanks for your response.
Yes, I understand what you mean. But let's assume that my cluster is UP. I have node A, B, C on Data Center 1 and nodes D, E on Data Center 2. They already proceeded to a vote and the master nodes are A, B, C and D.
What will happen if I loose connexion between Data Center 1 and 2?
The cluster might become unavailable, or may survive in one or other data centre (but not both). When the connection is restored the cluster will recover.
I'm not sure from your reply whether this was the answer you're looking for or not, but if not I'd be happy to hear more about what we're missing here. Note that this hasn't changed in version 7 - the answer I gave applies to earlier versions too.
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.