I'm using Kubernetes Continuous Deploy Plugin to deploy and upgrade a Deployment on my Kubernetes Cluster. I'm using pipeline and this is the Jenkinsfile:

pipeline {
    environment {
        JOB_NAME = "${JOB_NAME}".replace("-deploy", "")
        REGISTRY = "my-docker-registry"
    }
    agent any
    stages {
        stage('Fetching kubernetes config files') {
            steps {
                git 'git_url_of_k8s_configurations'
            }
        }
        stage('Deploy on kubernetes') {
            steps {
                kubernetesDeploy(
                    kubeconfigId: 'k8s-default-namespace-config-id',
                    configs: 'deployment.yml',
                    enableConfigSubstitution: true
                )
            }
        }
    }
}

Deployment.yml instead is:

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: ${JOB_NAME}
spec:
  replicas: 1
  template:
    metadata:
      labels:
        build_number: ${BUILD_NUMBER}
        app: ${JOB_NAME}
        role: rolling-update
    spec:
      containers:
      - name: ${JOB_NAME}-container        
        image: ${REGISTRY}/${JOB_NAME}:latest
        ports:
        - containerPort: 8080
        envFrom:
            - configMapRef:
                name: postgres
      imagePullSecrets:
      - name: regcred
  strategy:
    type: RollingUpdate

In order to let Kubernetes understand that Deployment is changed ( so to upgrade it and pods ) I used the Jenkins build number as annotation:

...
metadata:
  labels:
    build_number: ${BUILD_NUMBER}
...

The problem or my misunderstanding:

If Deployment does not exists on Kubernetes, all works good, creating one Deployment and one ReplicaSet.

If Deployment still exists and an upgrade is applied, Kubernetes creates a new ReplicaSet:

Before first deploy

First deploy

Second deploy

Third deploy

As you can see, each new Jenkins deploy will update corretly the deployment but creates a new ReplicaSet without removing the old one.

What could be the issue?