$ kubectl --context pm -n service-ratelimit get pods -o wide
NAME                                        READY   STATUS    RESTARTS   AGE     IP              NODE                                               NOMINATED NODE   READINESS GATES
service-ratelimit-canary-666d66cc75-cmbnw   6/6     Running   0          12m     IPa   ip-....ap-northeast-2.compute.internal    <none>           <none>
service-ratelimit-canary-666d66cc75-vbbc4   6/6     Running   0          2m6s    IPb     ip-....ap-northeast-2.compute.internal    <none>           <none>

$ istioctl --context ph pc endpoint url-shortener-admin-staging-7c8f9d8d5-dmc5z.url-shortener-admin | grep 'IPb'
No result

		if oie, exists := emap[nie.Address]; exists {
				// If endpoint exists already, we should push if it's health status changes.
				if oie.HealthStatus != nie.HealthStatus {
					needPush = true
				}
				newIstioEndpoints = append(newIstioEndpoints, nie)
			} else {
				// If the endpoint does not exist in shards that means it is a
				// new endpoint. Always send new endpoints even if they are not healthy.
				// This is OK since we disable panic threshold when SendUnhealthyEndpoints is enabled.
				// Without SendUnhealthyEndpoints we do not need this; headless services will trigger the push in the Kubernetes controller.
				if features.SendUnhealthyEndpoints.Load() {
					needPush = true
				}
				newIstioEndpoints = append(newIstioEndpoints, nie)
			}

1.20, but happens from 1.19.4