As an istio developer, we need to build a lot of enhance to keep service stable when pods down. Because we got a broken connect some time. I hope that K8s could cover that job like immediately deregistering when pods down. Does the kubeproxy watch the downing pod then redirect the traffic fast enough?
If the kubeproxy could stop unhealthy connect immediately, we believe it could work more smoothly. Then we could keep using service but not nacos. Otherwise we have to retry some times when it broken.
There’s no such thing as “immediately” - everything is async when an unplanned event happens.
We react as fast as we can, given API latency and readinessProbe periods and things like that.