We want new nodes to join automatically using discovery-gce (via tags) without having to manually update unicast.hosts.
In tests, when both are set, unicast hosts take precedence and new nodes without unicast.hosts do not join automatically.
If we remove unicast.hosts from masters but leave it in data nodes, will restarted data nodes still find the masters without causing unassigned shards?
How exactly is master discovery precedence handled in 6.5.3 when both unicast.hosts and discovery-gce are configured? Does unicast always override plugin-based discovery?
For 7.x, where discovery.seed_hosts replaces unicast.hosts, is the behavior the same?
Looking for the recommended approach to migrate from unicast-based discovery to fully dynamic GCE tag-based discovery without risking cluster instability or shard loss
This version has not been supported for more than 5 years. You need to upgrade to a supported version as a matter of urgency. That’s the recommended approach.
In 7.x Elasticsearch will use addresses from all host providers, both discovery-gce and discovery.seed_hosts, which I think is what you want. I don’t remember how 6.x did it but it sounds like it doesn’t work how you want. An upgrade will fix that.
For 7.x, where discovery.seed_hosts replaces unicast.hosts, is the behavior the same?
In 6.x, I have 3 master nodes with added unicast.hosts with 3 mastre ip in elasticsearch.yml and i have discovery-gce plugin.
Now am trying to add 3 more masters with discovery-gce plugin to same cluster. but these 3 master nodes not automatically detecting old masternodes and joining cluster dynamically.
again, I have to add unicast.hosts line in elasticsearch.yml and need to restart elasticsearch service. then those nodes joining with cluster and now i have 6 master nodes.
but my requirement is if we have discovery-gce plugin enabled again don't want to include unicast.hosts line and restart the service. Once we create new nodes it should dynamically join with cluster by using discovery-gce plugin.
I have tested scenaio where we are using only discovery-gce plugin existing nodes and not added unicast.hosts line, so if we add new nodes with discovery-gce plugin enabled, new nodes automatically joining with cluster.
7.x, 8.x and 9.x all behave the same as each other. I don’t remember how 6.x behaved, it’s been too long, but I wouldn’t be surprised if it was different from 7.x and later.
Sure, the solution to this as things stand is to first upgrade to a supported version.
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.