Why prefer Least Connections over Round Robin, and when do sticky sessions hurt you?

Round Robin assumes uniform request cost; with long-lived or heterogeneous requests it overloads slow backends while idle ones sit empty, so Least Connections (or latency-aware EWMA) balances real load better.

Sticky sessions concentrate a heavy user on one node, defeat even distribution, and prevent draining a node cleanly during deploys. They also create correctness cliffs: if the sticky node dies, those users lose state, which is why externalizing state and dropping stickiness is the more robust path.