The disclosure relates to a method and a state controller running in a Kubernetes system. The state controller being operative to assign labels to pods, the labels indicating services to which the pods are assigned and high-availability states of the pods; detect a failed pod having a label indicating a high-availability state of not ready; and reassign the label indicating the high-availability state of the failed pod to a healthy pod, thereby changing endpoints of services provided and service flows from the failed pod to the healthy pod.