when updating replicas for a sink, all pods will be rescheduled, which is not expected behaviour. Should fix this BUG and add related tests.