Why? Istio-ingressgateway is often deployed with HPA. The expectation is to load rebalance when replicas are scaling up. However, 1) Some L4 based lbs in front of istio-ingressgateway. Those lb can see the new istio-ingressgateway replicas but cannot migrate load. 2) Grpc streaming essentially doesn't migrate unless connection is destroying. A reasonable http max_connection_duration could highly improve the users' experience in above conditions.