You can see some previous discussion on a similar topic here. Though it makes me wonder if you are not using persistent volumes for the masters as the cluster should be able to recover even if all masters are shut down. But if the masters lost their data then this would be expected.