In Tableau Server version 2020.2 and later, load balancing for extract-based queries has improved and may result in faster load times for extract-based dashboards. Large deployments that are extract heavy and have a high volume of dashboard loads will probably see the most improvement, particularly if they have Hyper running on standalone nodes. Smaller deployments where Hyper is running on nodes that are shared with other server processes may also see performance improvements.
Hyper logs a server health metric about the amount of resources Hyper is consuming and also takes into account load from other Tableau processes that may be running on the same server node. Based on this information, extract queries will be sent to a node that has available resources to process the query. In addition to evaluating system resources, the load balancer improves the chance that the node you are routed to has the extract already cached. This reduces the number of duplicate extracts across nodes and improves memory and I/O usage.
To use this feature, the Cache Server process must be active. For more information, see Tableau Server Cache Server. If Cache Server process is not enabled, load balancing will automatically revert to the previous functionality.
This feature is turned on by default. To disable it, use the following tsm commands:
tsm configuration set -k hyper_standalone_consistent.hashing.enabled -v false
tsm configuration set -k hyper_standalone.health.enabled -v false
Apply the changes using the following tsm command:
tsm pending-changes apply
For more information, see tsm configuration set Options.