diff --git a/cluster/saltbase/salt/docker/docker-healthcheck b/cluster/saltbase/salt/docker/docker-healthcheck
new file mode 100755
index 00000000000..9167567bb04
--- /dev/null
+++ b/cluster/saltbase/salt/docker/docker-healthcheck
@@ -0,0 +1,44 @@
+#!/bin/bash
+
+# Copyright 2015 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This script is intended to be run periodically, to check the health
+# of docker.  If it detects a failure, it will restart docker using systemctl.
+
+if timeout 10 docker version > /dev/null; then
+  exit 0
+fi
+
+echo "docker failed"
+echo "Giving docker 30 seconds grace before restarting"
+sleep 30
+
+if timeout 10 docker version > /dev/null; then
+  echo "docker recovered"
+  exit 0
+fi
+
+echo "docker still down; triggering docker restart"
+systemctl restart docker
+
+echo "Waiting 60 seconds to give docker time to start"
+sleep 60
+
+if timeout 10 docker version > /dev/null; then
+  echo "docker recovered"
+  exit 0
+fi
+
+echo "docker still failing"
diff --git a/cluster/saltbase/salt/docker/docker-healthcheck.service b/cluster/saltbase/salt/docker/docker-healthcheck.service
new file mode 100644
index 00000000000..b4ecfaf8f72
--- /dev/null
+++ b/cluster/saltbase/salt/docker/docker-healthcheck.service
@@ -0,0 +1,9 @@
+[Unit]
+Description=Run docker-healthcheck once
+
+[Service]
+Type=oneshot
+ExecStart=/opt/kubernetes/helpers/docker-healthcheck
+
+[Install]
+WantedBy=multi-user.target
diff --git a/cluster/saltbase/salt/docker/docker-healthcheck.timer b/cluster/saltbase/salt/docker/docker-healthcheck.timer
new file mode 100644
index 00000000000..3d252e4464e
--- /dev/null
+++ b/cluster/saltbase/salt/docker/docker-healthcheck.timer
@@ -0,0 +1,9 @@
+[Unit]
+Description=Trigger docker-healthcheck periodically
+
+[Timer]
+OnUnitInactiveSec=10s
+Unit=docker-healthcheck.service
+
+[Install]
+WantedBy=multi-user.target
diff --git a/cluster/saltbase/salt/docker/docker-prestart b/cluster/saltbase/salt/docker/docker-prestart
new file mode 100755
index 00000000000..ea23d6d7237
--- /dev/null
+++ b/cluster/saltbase/salt/docker/docker-prestart
@@ -0,0 +1,22 @@
+#!/bin/bash
+
+# Copyright 2015 The Kubernetes Authors All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This script is intended to be run before we start Docker.
+
+# cleanup docker network checkpoint to avoid running into known issue
+# of docker (https://github.com/docker/docker/issues/18283)
+rm -rf /var/lib/docker/network
+
diff --git a/cluster/saltbase/salt/docker/docker.service b/cluster/saltbase/salt/docker/docker.service
index aa535375ec8..a5e02c9f6e5 100644
--- a/cluster/saltbase/salt/docker/docker.service
+++ b/cluster/saltbase/salt/docker/docker.service
@@ -1,10 +1,11 @@
 [Unit]
 Description=Docker Application Container Engine
-Documentation=http://docs.docker.com
+Documentation=https://docs.docker.com
 After=network.target docker.socket
 Requires=docker.socket
 
 [Service]
+Type=notify
 EnvironmentFile={{ environment_file }}
 ExecStart=/usr/bin/docker daemon -H fd:// "$DOCKER_OPTS"
 MountFlags=slave
@@ -14,7 +15,7 @@ LimitCORE=infinity
 Restart=always
 RestartSec=2s
 StartLimitInterval=0
+ExecStartPre=/opt/kubernetes/helpers/docker-prestart
 
 [Install]
 WantedBy=multi-user.target
-
diff --git a/cluster/saltbase/salt/docker/init.sls b/cluster/saltbase/salt/docker/init.sls
index 7352bc6a5b7..6e1f5a85739 100644
--- a/cluster/saltbase/salt/docker/init.sls
+++ b/cluster/saltbase/salt/docker/init.sls
@@ -51,6 +51,13 @@ docker:
 
 {% if pillar.get('is_systemd') %}
 
+/opt/kubernetes/helpers/docker-prestart:
+  file.managed:
+    - source: salt://docker/docker-prestart
+    - user: root
+    - group: root
+    - mode: 755
+
 {{ pillar.get('systemd_system_path') }}/docker.service:
   file.managed:
     - source: salt://docker/docker.service
@@ -60,6 +67,8 @@ docker:
     - mode: 644
     - defaults:
         environment_file: {{ environment_file }}
+    - require:
+      - file: /opt/kubernetes/helpers/docker-prestart
 
 # The docker service.running block below doesn't work reliably
 # Instead we run our script which e.g. does a systemd daemon-reload
@@ -297,9 +306,16 @@ docker-upgrade:
       - file: /var/cache/docker-install/{{ override_deb }}
 {% endif %} # end override_docker_ver != ''
 
-# Default docker systemd unit file doesn't use an EnvironmentFile; replace it with one that does.
 {% if pillar.get('is_systemd') %}
 
+/opt/kubernetes/helpers/docker-prestart:
+  file.managed:
+    - source: salt://docker/docker-prestart
+    - user: root
+    - group: root
+    - mode: 755
+
+# Default docker systemd unit file doesn't use an EnvironmentFile; replace it with one that does.
 {{ pillar.get('systemd_system_path') }}/docker.service:
   file.managed:
     - source: salt://docker/docker.service
@@ -309,6 +325,8 @@ docker-upgrade:
     - mode: 644
     - defaults:
         environment_file: {{ environment_file }}
+    - require:
+      - file: /opt/kubernetes/helpers/docker-prestart
 
 # The docker service.running block below doesn't work reliably
 # Instead we run our script which e.g. does a systemd daemon-reload
@@ -316,7 +334,7 @@ docker-upgrade:
 # TODO: Fix this
 fix-service-docker:
   cmd.wait:
-    - name: /opt/kubernetes/helpers/services bounce docker
+    - name: /opt/kubernetes/helpers/services enable docker
     - watch:
       - file: {{ pillar.get('systemd_system_path') }}/docker.service
       - file: {{ environment_file }}
@@ -325,30 +343,77 @@ fix-service-docker:
       - cmd: docker-upgrade
 {% endif %}
 
+/opt/kubernetes/helpers/docker-healthcheck:
+  file.managed:
+    - source: salt://docker/docker-healthcheck
+    - user: root
+    - group: root
+    - mode: 755
+
+{{ pillar.get('systemd_system_path') }}/docker-healthcheck.service:
+  file.managed:
+    - source: salt://docker/docker-healthcheck.service
+    - template: jinja
+    - user: root
+    - group: root
+    - mode: 644
+
+{{ pillar.get('systemd_system_path') }}/docker-healthcheck.timer:
+  file.managed:
+    - source: salt://docker/docker-healthcheck.timer
+    - template: jinja
+    - user: root
+    - group: root
+    - mode: 644
+
+# Tell systemd to load the timer
+fix-systemd-docker-healthcheck-timer:
+  cmd.wait:
+    - name: /opt/kubernetes/helpers/services bounce docker-healthcheck.timer
+    - watch:
+      - file: {{ pillar.get('systemd_system_path') }}/docker-healthcheck.timer
+
+# Trigger a first run of docker-healthcheck; needed because the timer fires 10s after the previous run.
+fix-systemd-docker-healthcheck-service:
+  cmd.wait:
+    - name: /opt/kubernetes/helpers/services bounce docker-healthcheck.service
+    - watch:
+      - file: {{ pillar.get('systemd_system_path') }}/docker-healthcheck.service
+    - require:
+      - cmd: fix-service-docker
+
 {% endif %}
 
 docker:
-  service.running:
 # Starting Docker is racy on aws for some reason.  To be honest, since Monit
 # is managing Docker restart we should probably just delete this whole thing
 # but the kubernetes components use salt 'require' to set up a dag, and that
 # complicated and scary to unwind.
+# On AWS, we use a trick now... we don't start the docker service through Salt.
+# Kubelet or our health checker will start it.  But we use service.enabled,
+# so we still have a `service: docker` node for our DAG.
 {% if grains.cloud is defined and grains.cloud == 'aws' %}
-    - enable: False
+  service.enabled:
 {% else %}
+  service.running:
     - enable: True
 {% endif %}
+# If we put a watch on this, salt will try to start the service.
+# We put the watch on the fixer instead
+{% if not pillar.get('is_systemd') %}
     - watch:
       - file: {{ environment_file }}
 {% if override_docker_ver != '' %}
       - cmd: docker-upgrade
 {% endif %}
+{% endif %}
+    - require:
+      - file: {{ environment_file }}
+{% if override_docker_ver != '' %}
+      - cmd: docker-upgrade
+{% endif %}
 {% if pillar.get('is_systemd') %}
-      - file: {{ pillar.get('systemd_system_path') }}/docker.service
-{% endif %}
-{% if override_docker_ver != '' %}
-    - require:
-      - cmd: docker-upgrade
+      - cmd: fix-service-docker
 {% endif %}
 {% endif %} # end grains.os_family != 'RedHat'
 
diff --git a/cluster/saltbase/salt/salt-helpers/services b/cluster/saltbase/salt/salt-helpers/services
index f55b9b39c77..bc8db58f326 100644
--- a/cluster/saltbase/salt/salt-helpers/services
+++ b/cluster/saltbase/salt/salt-helpers/services
@@ -63,6 +63,9 @@ elif [[ "${ACTION}" == "down" ]]; then
   reload_state
   disable_service
   stop_service
+elif [[ "${ACTION}" == "enable" ]]; then
+  reload_state
+  enable_service
 else
   echo "Unknown action: ${ACTION}"
   exit 1