I can not understand the behavior of a Deployment. It always uses the wrong version of ReplicaSets in my case. First i did

kubectl create -f [filename]

apiVersion: apps/v1beta2
kind: Deployment
metadata:
  labels:
    app: kube-state-metrics
  name: kube-state-metrics
  namespace: monitoring
spec:
  replicas: 1
  selector:
    matchLabels:
      app: kube-state-metrics
  strategy:
    type: Recreate
  template:
    metadata:
      labels:
        app: kube-state-metrics

But а pod could not start because the master node has a taint. I changed the deployment file and added a toleration:

spec:
  tolerations:
  - key: node-role.kubernetes.io/master
    operator: Equal
    effect: NoSchedule

kubectl replace -f [filename]

I got two revisions, but NewReplicaSet was set to old revision and OldReplicaSet to modified version respectively. Hmmm...

I deleted the deployment and called "create" again. The situation did not improve.

OldReplicaSets:  <none>
NewReplicaSet:   kube-state-metrics-59b7dccd55 (1/1 replicas created)
Events:
Type    Reason             Age   From                   Message
----    ------             ----  ----                   -------
Normal  ScalingReplicaSet  14m   deployment-controller  Scaled up replica set kube-state-metrics-69c88bb67b to 1
Normal  ScalingReplicaSet  14m   deployment-controller  Scaled down replica set kube-state-metrics-69c88bb67b to 0
Normal  ScalingReplicaSet  13m   deployment-controller  Scaled up replica set kube-state-metrics-59b7dccd55 to 1

OldReplicaSets is none, but actually is used. And again NewReplicaSets is wrong. In addition, the deployment has two revisions:

REVISION  CHANGE-CAUSE
1         kubectl create --filename=manifests
2         kubectl create --filename=manifests

How to fix it?