Retry failed master startup once

Master startup can fail when ec2 transparently reallocates the block
storage, causing etcd writes to temporarily fail. Retry failures blindly
just
once to allow time for this transient condition to to resolve and for
systemd
to restart the master (which will eventually succeed).

https://github.com/coreos/etcd/issues/3864
https://github.com/openshift/origin/issues/6065
https://github.com/openshift/origin/issues/6447

roles/openshift_master/tasks/main.yml

@@ -168,10 +168,21 @@
 - include: set_loopback_context.yml
   when: openshift.common.version_gte_3_2_or_1_2
 
+# TODO: Master startup can fail when ec2 transparently reallocates the block
+# storage, causing etcd writes to temporarily fail. Retry failures blindly just
+# once to allow time for this transient condition to to resolve and for systemd
+# to restart the master (which will eventually succeed).
+#
+# https://github.com/coreos/etcd/issues/3864
+# https://github.com/openshift/origin/issues/6065
+# https://github.com/openshift/origin/issues/6447
 - name: Start and enable master
   service: name={{ openshift.common.service_type }}-master enabled=yes state=started
   when: not openshift_master_ha | bool
   register: start_result
+  until: not start_result | failed
+  retries: 1
+  delay: 60
   notify: Verify API Server
 
 - name: Check for non-HA master service presence