Elastic Load Balancer

Elastic Stack Elasticsearch
1

Hi,
I am new to elastic, trying to set up my cluster.
I have 3 master nodes, 3 data nodes and 1 client node, each one of them runs on a different machine.

My question is, is there a default load balancer that gets the transport and sends it to the appropriate nodes?

If I want to send data to the cluster, to which address should I send it, as each data node has a different IP address?
Do I have to use an external LB for that?

Sorry for the basic question... any help would be appreciated!

2

Hi @El-k, Elasticsearch nodes will try to balance load naturally between themselves and you should not need to route requests to a specific node manually. Any node has the ability to effectively function as an internal load balancer across the cluster.

We do generally recommend that you don't route requests to a dedicated master node, as those nodes need to fulfil additional duties.

For more information, see:

3

Thanks, @Sander_Philipse . Just to make sure I got it right:

Assuming that Elasticsearch nodes perform an internal load balancer,
can I choose one random data-node IP address
and always access the cluster endpoint using 'http://:9200/'?

Will the cluster then direct the request to the right node?

4

The node that receives the request will coordinate the execution. This will often involve multiple nodes in the cluster.

All official clients, including Beats and Logstash, support specifying a number of nodes in the connection string and will send requests to all nodes listed.

What you are proposing is bad for primarily 2 reasons:

  • Coordinating a request is not free and adds load to the node. Having only a sinle node coordinating all requests can cause load imbalance and performance problems. Sometimes a number of dedicated coordinating only nodes are used as the entry point to the cluster, but that is generally not needed for small clusters.
  • If the clients are only able to connect to one node that node is a single point of failure. If you need to restart it, e.g. due to an uprade, or it crashes the clients will experience an outage.
5

Thanks, @Christian_Dahlqvist . This is the answer I was looking for!
At the end, decided to create an internal LB to manage traffic to data nodes with health checks.

6

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.