Tested on gcp, there are 4 failureDomains a, b, c, f in CPMS, remove one a, a new master will be created in f. If readd f to CPMS, instance will be moved back from f to a

Always

Before update cpms.
      failureDomains:
        gcp:
        - zone: us-central1-a
        - zone: us-central1-b
        - zone: us-central1-c
        - zone: us-central1-f
$ oc get machine                  
NAME                              PHASE     TYPE            REGION        ZONE            AGE
zhsungcp22-4glmq-master-2         Running   n2-standard-4   us-central1   us-central1-c   3h4m
zhsungcp22-4glmq-master-hzsf2-0   Running   n2-standard-4   us-central1   us-central1-b   90m
zhsungcp22-4glmq-master-plch8-1   Running   n2-standard-4   us-central1   us-central1-a   11m
zhsungcp22-4glmq-worker-a-cxf5w   Running   n2-standard-4   us-central1   us-central1-a   3h
zhsungcp22-4glmq-worker-b-d5vzm   Running   n2-standard-4   us-central1   us-central1-b   3h
zhsungcp22-4glmq-worker-c-4d897   Running   n2-standard-4   us-central1   us-central1-c   3h

1. Delete failureDomain "zone: us-central1-a" in cpms, new machine Running in zone f.
      failureDomains:
        gcp:
        - zone: us-central1-b
        - zone: us-central1-c
        - zone: us-central1-f 
$ oc get machine              
NAME                              PHASE     TYPE            REGION        ZONE            AGE
zhsungcp22-4glmq-master-2         Running   n2-standard-4   us-central1   us-central1-c   3h19m
zhsungcp22-4glmq-master-b7pdl-1   Running   n2-standard-4   us-central1   us-central1-f   13m
zhsungcp22-4glmq-master-hzsf2-0   Running   n2-standard-4   us-central1   us-central1-b   106m
zhsungcp22-4glmq-worker-a-cxf5w   Running   n2-standard-4   us-central1   us-central1-a   3h16m
zhsungcp22-4glmq-worker-b-d5vzm   Running   n2-standard-4   us-central1   us-central1-b   3h16m
zhsungcp22-4glmq-worker-c-4d897   Running   n2-standard-4   us-central1   us-central1-c   3h16m
2. Add failureDomain "zone: us-central1-a" again, new machine running in zone a, the machine in zone f will be deleted.
      failureDomains:
        gcp:
        - zone: us-central1-a
        - zone: us-central1-f
        - zone: us-central1-c
        - zone: us-central1-b
$ oc get machine                          
NAME                              PHASE     TYPE            REGION        ZONE            AGE
zhsungcp22-4glmq-master-2         Running   n2-standard-4   us-central1   us-central1-c   3h35m
zhsungcp22-4glmq-master-5kltp-1   Running   n2-standard-4   us-central1   us-central1-a   12m
zhsungcp22-4glmq-master-hzsf2-0   Running   n2-standard-4   us-central1   us-central1-b   121m
zhsungcp22-4glmq-worker-a-cxf5w   Running   n2-standard-4   us-central1   us-central1-a   3h32m
zhsungcp22-4glmq-worker-b-d5vzm   Running   n2-standard-4   us-central1   us-central1-b   3h32m
zhsungcp22-4glmq-worker-c-4d897   Running   n2-standard-4   us-central1   us-central1-c   3h32m  

Instance is moved back from f to a

Instance shouldn't be moved back from f to a

https://issues.redhat.com//browse/OCPBUGS-7366