Client is experiencing high number of connections and the primary reason for this is that they have 140 services deployed across 120 machines and with a single KV endpoint the number of connections would be 16800. They do use more connections for certain services and hence the connection count is around 20000 and would increase if they deploy to more app servers, thereby causing scaling bottleneck from a connection perspective. The workload on the cluster is very small ( <10000 ops per sec). While there are opportunities to improve the client code and tune the connections, we also have an opportunity to optimize our SDK. If we can provide an option to not have a persistent connection, then for small workload use cases that span across large numbers of app servers the connection count would remain low.