Several `inactiveAdded` nodes were present in the cluster. i.e., the user had added nodes to the cluster but hadn't rebalanced. So, these nodes would not service any requests until a rebalance is performed.
After a few hours, the user removed these `inactiveAdded` nodes and the latency issue resolved itself.
When looking through the SDK logs, we observe a lot of requests going to a few `inactiveAdded` nodes and failing