pod restart makes it sound like a network thing to me... like there's an attempt to talk to a pod, the gRPC call hangs for some reason, and it's not rectified until it's removed from the pool