Reduce log errors from heka

Heka depends on haproxy and keepaived being present to communicate
with ElasticSearch.  If we start ElasticSearch prior to haproxy and
keepalived, the number of errors are reduced in heka.

Change-Id: Id2c742ea572c6450a371421e21f34aa69355bb8b
Partial-Bug: #1560779

ansible/site.yml

@@ -8,6 +8,12 @@
         tags: ceph,
         when: enable_ceph | bool }
 
+- hosts: elasticsearch
+  roles:
+    - { role: elasticsearch,
+        tags: elasticsearch,
+        when: enable_central_logging | bool }
+
 - hosts:
     - cinder-api
     - glance-api
@@ -32,12 +38,6 @@
         tags: kibana,
         when: enable_central_logging | bool }
 
-- hosts: elasticsearch
-  roles:
-    - { role: elasticsearch,
-        tags: elasticsearch,
-        when: enable_central_logging | bool }
-
 - hosts: memcached
   roles:
     - { role: memcached,