Kessel services (SPs and internally) using gRPC are not distributing requests across multiple API pods, resulting in uneven load distribution and potential performance bottlenecks.

Because gRPC uses persistent connections, all requests from a single client get routed to the same backend pod instead of being distributed across available replicas. Traditional load balancing works at the connection level, not the request level, which means it cannot balance individual gRPC calls within a connection.

We need to determine the best route forward: either implement client-side load balancing where gRPC clients manage their own request distribution, or deploy an L7-aware service mesh/proxy that can inspect and route individual gRPC requests (3scale?)