Description of problem:
In HA env the node label node-role.kubernetes.io/master=true is lost after instance stop and start.
Non-HA env, e.g. a master and a node, doesn't reproduce it after master stop and start.

Version-Release number of selected component (if applicable):
openshift v3.9.41

How reproducible:
Always

Steps to Reproduce:
1. First, check nodes
[root@ip-172-18-7-74 ~]# oc get no
NAME                            STATUS    ROLES     AGE       VERSION
ip-172-18-1-221.ec2.internal    Ready     master    52m       v1.9.1+a0ce1bc657
ip-172-18-13-114.ec2.internal   Ready     compute   52m       v1.9.1+a0ce1bc657
ip-172-18-2-68.ec2.internal     Ready     compute   52m       v1.9.1+a0ce1bc657
ip-172-18-4-42.ec2.internal     Ready     master    52m       v1.9.1+a0ce1bc657
ip-172-18-7-74.ec2.internal     Ready     master    52m       v1.9.1+a0ce1bc657
2. Stop the instance ip-172-18-7-74.ec2.internal in AWS console.
3. Minutes later, start the instance in AWS console.
4. Check nodes again.
[root@ip-172-18-7-74 ~]# oc get no
NAME                            STATUS    ROLES     AGE       VERSION
ip-172-18-1-221.ec2.internal    Ready     master    1h        v1.9.1+a0ce1bc657
ip-172-18-13-114.ec2.internal   Ready     compute   1h        v1.9.1+a0ce1bc657
ip-172-18-2-68.ec2.internal     Ready     compute   1h        v1.9.1+a0ce1bc657
ip-172-18-4-42.ec2.internal     Ready     master    1h        v1.9.1+a0ce1bc657
ip-172-18-7-74.ec2.internal     Ready     <none>    18m       v1.9.1+a0ce1bc657

[root@ip-172-18-7-74 ~]# oc label --list no ip-172-18-7-74.ec2.internal | grep role
role=node

Actual results:
4. Node label node-role.kubernetes.io/master=true is lost

Expected results:
4. The label still exists

Additional info: