Description of problem:
'oc rollout pause' will erase 'Running' pods in special situation(depends on the number of replicas&maxSurge&maxUnavailable). I found the rule is: if running pods' number bigger than unavailable pods' number, the Running pods will not be erased after 'oc rollout pause', such as: replicas=5 maxSurge=1 maxUnavailable=1, after edit with nonexist image and pause, the pods sitatuion is：4 Running and 2 ErrImagePull. if running pods' number smaller than unavailable pods' number, the running pods will be erased after rollout pause, such as: replicas=5 maxSurge=1 maxUnavailable=4,  after edit with nonexist image, there are: 1 running, 5 ErrImagePull pods, then run 'oc rollout pause', and there are 5 ErrImagePull pods only, the Running pods be erased.


Version-Release number of selected component (if applicable):

openshift v3.6.28
kubernetes v1.5.2+43a9be4
etcd 3.1.0

How reproducible:
Always

Steps to Reproduce:
1. create deployment(#oc create -f hello-openshift.yaml) with file:

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: hello-openshift
spec:
  replicas: 2
  template:
    metadata:
      labels:
        app: hello-openshift
    spec:
      containers:
      - name: hello-openshift
        image: openshift/hello-openshift
        ports:
        - containerPort: 80
  strategy:
    rollingUpdate:
      maxSurge: 1
      maxUnavailable: 1
    type: RollingUpdate

2. there are 2 pods in Running status after step 1

# oc get pods
NAME                               READY     STATUS    RESTARTS   AGE
hello-openshift-2049296760-c9d37   1/1       Running   0          6s
hello-openshift-2049296760-ftpgz   1/1       Running   0          6s

3. change the image to 'nonexist' in deployment 

#oc edit deployment 

spec: 
  containers:
    - image: openshift/hello-openshift     ===change to==> nonexist

[root@host-8-x-x ~]# oc edit deployment
deployment "hello-openshift" edited
# oc get pods
NAME                               READY     STATUS         RESTARTS   AGE
hello-openshift-2049296760-c9d37   1/1       Running        0          1m
hello-openshift-2758264543-mr9q8   0/1       ErrImagePull   0          32s
hello-openshift-2758264543-s3t4b   0/1       ErrImagePull   0          32s


4. # oc rollout pause deployment/hello-openshift
deployment "hello-openshift" paused
# oc get pods
NAME                               READY     STATUS             RESTARTS   AGE
hello-openshift-2758264543-mr9q8   0/1       ImagePullBackOff   0          53s
hello-openshift-2758264543-s3t4b   0/1       ImagePullBackOff   0          53s

obviously, the Running pods 'hello-openshift-2049296760-c9d37' is missed.

5.  resume the deployment, still could not find the Running pod:

# oc rollout resume deployment/hello-openshift
deployment "hello-openshift" resumed
[root@host-8-x-x ~]# oc get pods
NAME                               READY     STATUS             RESTARTS   AGE
hello-openshift-2758264543-133sf   0/1       ImagePullBackOff   0          14m
hello-openshift-2758264543-39t5j   0/1       ImagePullBackOff   0          14m
hello-openshift-2758264543-8hk0j   0/1       ImagePullBackOff   0          14m
hello-openshift-2758264543-f3h7q   0/1       ImagePullBackOff   0          14m
hello-openshift-2758264543-rtffw   0/1       ImagePullBackOff   0          14m
[root@host-8-x-x~]# 

Actual results:

'oc rollout pause' will erase  Running pods in special situation(depends on the number of replicas&maxSurge&maxUnavailable) , 


Expected results:

'oc rollout pause' will not effect erase the number of Running and unavailable pods as expects