Skip to main content

The Tableau docs around using a load balancer make it sound like adding one is just for improving reliability of the overall cluster. I want to make sure this is the case for a deployment I am doing. Between the docs and slack channels I am receiving contradictory answers on this question though.

 

Docs:

Example: Install and Configure a Three-Node HA Cluster - Tableau

In the example page it states (bold added):

"At this point, all three nodes have gateways, which are used to route requests to available server processes. All gateways are active, but to further reduce the potential for downtime in the cluster, you can configure a load balancer."

This appears to mean there is not a requirement for a load balancer.

Slack server channel:

 

Some answers I received about a required load balancer in the slack channels contradict this, stating that if not using a load balancer all traffic will only hit a single node or that authentication may not pass through the initial node to any other node picking up the traffic.

I went to my own research using a load script and the workgroup database.

 

My script captured all the workbooks/views I have on server then randomly navigated to each of them 100x roughly, Using the workgroup database I could see load times and also that each worker received http requests. See the image from my monitoring workbook. From my own research it appears that traffic is flowing fine to each of my nodes as expected and in line with tableau documentation.

Load Balancer requirement​Can someone definitively tell me if a load balancer is required in a multi-node cluster? Thanks!

2 answers
  1. Aug 7, 2023, 7:23 PM

    @Eric Summers​ 

    Hi, defitely if you are looking for an HA environment, you would need a load balancer. However, it is not necessary to have a load balancer in a multinode environment:

     

    "You can enhance the reliability of Tableau Server by running gateways on multiple nodes, and configuring a load balancer to distribute requests across the gateways. Unlike the repository process, which can be active or passive, all gateway processes are active. If one gateway in a cluster becomes unavailable, the load balancer stops sending requests to it. The load balancer algorithm you choose determines how the gateways will route client requests."

     

    So, basically, if you have multiple gateways, you are creating a desired duplicity of processes so you can make your environment more reliable.

     

    So, is it necessary? No, is it recommended? Yes.

     

    A small example is when you have a primary node with no backgrounders, and a second node specialized for extract refreshes and a third node specialized for flow taks. In this type of multi node installs, it does not make sense to have a gateway in the second node or third node, and the install would work with gateways only in the first node.

     

    If this post resolves the question, would you be so kind to "Select as Best"?. This will help other users find the same answer/resolution and help community keep track of answered questions. Thank you.

     

    Regards,

     

    Diego Martinez

    Tableau Visionary and Forums Ambassador

0/9000