I am trying to setup Kibana over two node cluster of Elasticsearch as a part of ELK Stack and came across to the information that i simply can't do that (cause Kibana don't have such feature till yet ) ; Got a term Client Node as a solution for this but i have one architectural question related to that . Kindly suggest what if my client node only goes down , i'll loose the entire interface from my ES cluster ; Any high availability solution to that ?
Yeah , Saw it .. but seems like required feature is not there , request is still open since last two years , even contributors are suggesting to use non-data node (may be client node ) but no one is talking about availability there , what if that single node goes down with any circumstances ?
There is nothing wrong with using a client node together with Kibana, but if you want to avoid it being a single point of failure, a load balancer might be an alternative. You can also set up two Kibana instances and load-balance in front of these.
Ok , Using LB on top of ES nodes sounds good but didn't get the concept of using two Kibana instaces ;
one more thing , If i want to use LB say HA-Proxy so it should be on isolated machine i mean apart from the KIbana instances right ?
Logstash, Beats and most clients libraries are able to connect to more than one Elasticsearch node, so do not require any load balancer ahead of Elasticsearch. Exactly how you set up your load balancer of choice in a HA fashion I am probably not the right person to answer.
As you asked about a HA solution, the Kibana process in itself is also a single point of failure if you only have one, which is why I mentioned the option of having 2 with a load balancer in front.
Ohh.. actually I have two node ES cluster and i want to use Kibana over them for visualization , since Kibana can link with single ES server only so i came across the solution as client node ..however you have raised a valid point to bring Kibana in HA mode too but my prior concern is to fetch data from ES cluster using KIbana any suggestion for that ?
A 2-node Elasticsearch cluster can also never be fully HA as Elasticsearch requires a majority of modes to be present in order to elect a master. This is why I always recommend a minimum cluster size of 3 nodes. You don't however necessarily need three big data nodes, as two data nodes that are master eligible can be complemented by a third, smaller dedicated master node that can act as a tie-breaker.
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.