diff --git a/cluster/saltbase/salt/kube-apiserver/init.sls b/cluster/saltbase/salt/kube-apiserver/init.sls
index c1eb9b24add..0238ecf85d2 100644
--- a/cluster/saltbase/salt/kube-apiserver/init.sls
+++ b/cluster/saltbase/salt/kube-apiserver/init.sls
@@ -22,6 +22,19 @@
     - mode: 644
 
 # Copy kube-apiserver manifest to manifests folder for kubelet.
+# Current containervm image by default has both docker and kubelet
+# running. But during cluster creation stage, docker and kubelet
+# could be overwritten completely, or restarted due to flag changes.
+# The ordering of salt states for service docker, kubelet and
+# master-addon below is very important to avoid the race between
+# salt restart docker or kubelet and kubelet start master components.
+# Without the ordering of salt states, when gce instance boot up,
+# configure-vm.sh will run and download the release. At the end of
+# boot, run-salt will installs kube-apiserver.manifest files to
+# kubelet config directory before the installation of proper version
+# kubelet. Please see
+# https://github.com/GoogleCloudPlatform/kubernetes/issues/10122#issuecomment-114566063
+# for detail explanation on this very issue.
 /etc/kubernetes/manifests/kube-apiserver.manifest:
   file.managed:
     - source: salt://kube-apiserver/kube-apiserver.manifest
@@ -31,6 +44,9 @@
     - mode: 644
     - makedirs: true
     - dir_mode: 755
+    - require:
+      - service: docker
+      - service: kubelet
 
 #stop legacy kube-apiserver service
 stop_kube-apiserver:
diff --git a/cluster/saltbase/salt/kube-controller-manager/init.sls b/cluster/saltbase/salt/kube-controller-manager/init.sls
index aa523c01928..cc61c3e084e 100644
--- a/cluster/saltbase/salt/kube-controller-manager/init.sls
+++ b/cluster/saltbase/salt/kube-controller-manager/init.sls
@@ -1,3 +1,9 @@
+# Copy kube-controller-manager manifest to manifests folder for kubelet.
+# The ordering of salt states for service docker, kubelet and
+# master-addon below is very important to avoid the race between
+# salt restart docker or kubelet and kubelet start master components.
+# Please see https://github.com/GoogleCloudPlatform/kubernetes/issues/10122#issuecomment-114566063
+# for detail explanation on this very issue.
 /etc/kubernetes/manifests/kube-controller-manager.manifest:
   file.managed:
     - source: salt://kube-controller-manager/kube-controller-manager.manifest
@@ -7,6 +13,9 @@
     - mode: 644
     - makedirs: true
     - dir_mode: 755
+    - require:
+      - service: docker
+      - service: kubelet
 
 /var/log/kube-controller-manager.log:
   file.managed:
diff --git a/cluster/saltbase/salt/kube-master-addons/init.sls b/cluster/saltbase/salt/kube-master-addons/init.sls
index 075a7a05113..91186052205 100644
--- a/cluster/saltbase/salt/kube-master-addons/init.sls
+++ b/cluster/saltbase/salt/kube-master-addons/init.sls
@@ -37,26 +37,10 @@ master-docker-image-tags:
   file.touch:
     - name: /srv/pillar/docker-images.sls
 
-# Current containervm image by default has both docker and kubelet
-# running. But during cluster creation stage, docker and kubelet
-# could be overwritten completely, or restarted due to flag changes.
-# The ordering of salt states for service docker, kubelet and
-# master-addon below is very important to avoid the race between
-# salt restart docker or kubelet and kubelet start master components.
-# Without the ordering of salt states, when gce instance boot up,
-# configure-vm.sh will run and download the release. At the end of
-# boot, run-salt will run kube-master-addons service which installs
-# master component manifest files to kubelet config directory before
-# the installation of proper version kubelet. Please see
-# https://github.com/GoogleCloudPlatform/kubernetes/issues/10122#issuecomment-114566063
-# for detail explanation on this very issue.
 kube-master-addons:
   service.running:
     - enable: True
     - restart: True
-    - require:
-      - service: docker
-      - service: kubelet
     - watch:
       - file: master-docker-image-tags
       - file: /etc/kubernetes/kube-master-addons.sh
diff --git a/cluster/saltbase/salt/kube-scheduler/init.sls b/cluster/saltbase/salt/kube-scheduler/init.sls
index d7965852362..6db90331d5e 100644
--- a/cluster/saltbase/salt/kube-scheduler/init.sls
+++ b/cluster/saltbase/salt/kube-scheduler/init.sls
@@ -1,4 +1,9 @@
 # Copy kube-scheduler manifest to manifests folder for kubelet.
+# The ordering of salt states for service docker, kubelet and
+# master-addon below is very important to avoid the race between
+# salt restart docker or kubelet and kubelet start master components.
+# Please see https://github.com/GoogleCloudPlatform/kubernetes/issues/10122#issuecomment-114566063
+# for detail explanation on this very issue.
 /etc/kubernetes/manifests/kube-scheduler.manifest:
   file.managed:
     - source: salt://kube-scheduler/kube-scheduler.manifest
@@ -8,6 +13,9 @@
     - mode: 644
     - makedirs: true
     - dir_mode: 755
+    - require:
+      - service: docker
+      - service: kubelet
 
 /var/log/kube-scheduler.log:
   file.managed: