Broker UAI resiliency and load balancing is achieved through the use of Multi-Replica Broker UAIs. The procedures and data involved in configuring a UAI Class to create Multi-Replica Broker UAIs can be found in UAI Classes. This page describes some of the reasons to use Multi-Replica Broker UAIs and some of the implications of doing so.
When a Broker UAI runs with multiple replicas, access to the broker remains channeled through a single external IP address, but the connections are load balanced and dispatched to multiple Kubernetes pods where the Broker UAI functionality is running. This has two beneficial effects:
NOTE When a Broker UAI pod terminates for any reason, all SSH sessions going through that pod are dropped. This is because the Broker UAI pods forward SSH sessions to the End-User UAIs, so they are always an active part of the connection.
The number of replicas a Multi-Replica UAI Broker should have is dictated primarily by the number of host nodes on which Broker UAIs can be deployed. From a load-balancing perspective, it makes sense to make the number of replicas equal to the number of available host nodes. From a resiliency perspective, that number could be considerably smaller (3 for example) on the assumption that multi-node failures or evictions are unlikely, and brokers that are evicted or restarted will start up relatively quickly elsewhere. It does not make sense from either perspective, however, to over-subscribe the available number of host nodes (except, perhaps during a temporary outage) since that will result in no additional resiliency and the potential for network traffic and resource consumption hot spots.