we find an api service sometimes get timeout return after we update it yesterday.
everything about this api service looks like ok on marathon page. then we login the server which the service instance is running and check its ipvs table.
126.96.36.199 is the current container address which is running. but 188.8.131.52 is the container address which has been killed when marathon update the api service.
so when virtual server access 184.108.40.206 its always timeout.
we try to delete by ipvsadm but after a while the dead server comeback again. maybe navstar has persistent storage with ipvs table and always sync from each other in cluster.