Master component and kubelet container upgrade fixes

* Fixes an issue where apiserver and friends (controller manager, scheduler) were prevented from restarting after manifests/secrets are changed. This occurred when a replaced kubelet doesn't reconcile new master manifests, which caused old master component versions to linger during deployment. In my case this was causing upgrades from k8s 1.6/1.7 -> k8s 1.8 to fail
* Improves transitions from kubelet container to host kubelet by preventing issues where kubelet container reappeared during the deployment

roles/kubernetes/master/handlers/main.yml

@@ -13,6 +13,24 @@
     - Master | wait for kube-scheduler
     - Master | wait for kube-controller-manager
 
+- name: Master | Restart apiserver
+  command: /bin/true
+  notify:
+    - Master | Remove apiserver container
+    - Master | wait for the apiserver to be running
+
+- name: Master | Restart kube-scheduler
+  command: /bin/true
+  notify:
+    - Master | Remove scheduler container
+    - Master | wait for kube-scheduler
+
+- name: Master | Restart kube-controller-manager
+  command: /bin/true
+  notify:
+    - Master | Remove controller manager container
+    - Master | wait for kube-controller-manager
+
 - name: Master | reload systemd
   command: systemctl daemon-reload
 
@@ -21,6 +39,15 @@
     name: kubelet
     state: restarted
 
+- name: Master | Remove apiserver container
+  shell: "docker ps -af name=k8s_kube-apiserver* -q | xargs --no-run-if-empty docker rm -f"
+
+- name: Master | Remove scheduler container
+  shell: "docker ps -af name=k8s_kube-scheduler* -q | xargs --no-run-if-empty docker rm -f"
+
+- name: Master | Remove controller manager container
+  shell: "docker ps -af name=k8s_kube-controller-manager* -q | xargs --no-run-if-empty docker rm -f"
+
 - name: Master | wait for kube-scheduler
   uri:
     url: http://localhost:10251/healthz