loadbalancer: do not manage HAproxy service on BACKUP nodes

When a loadbalancer is in BACKUP mode, HAproxy can't bind the pools to
the VIP, since the VIP is not hosted. Puppet warns it when compiling in
the catalogue.
This patch ensure HAproxy is run where we Keepalived is in MASTER mode.

If a failover occurs and MASTER has changed, Keepalived & HAproxy are auto-migrated thanks to VRRP to another node. The worth case will only show warnings in Puppet catalog but HAproxy won't start anyway on the node.

Bug #237

spec/classes/cloud_loadbalancer_spec.rb

@@ -111,7 +111,7 @@ describe 'cloud::loadbalancer' do
       end
     end
 
-    context 'configure keepalived in backup' do
+    context 'when keepalived and HAproxy are in backup' do
       it 'configure vrrp_instance with BACKUP state' do
         should contain_keepalived__instance('1').with({
           'interface'     => params[:keepalived_public_interface],
@@ -123,6 +123,9 @@ describe 'cloud::loadbalancer' do
           'notify_backup' => '"/etc/init.d/haproxy stop"',
         })
       end # configure vrrp_instance with BACKUP state
+      it 'configure haproxy server without service managed' do
+        should contain_class('haproxy').with(:service_manage => false)
+      end # configure haproxy server
     end # configure keepalived in backup
 
     context 'configure keepalived in master' do
@@ -139,6 +142,9 @@ describe 'cloud::loadbalancer' do
           'notify_backup' => '"/etc/init.d/haproxy stop"',
         })
       end
+      it 'configure haproxy server with service managed' do
+        should contain_class('haproxy').with(:service_manage => true)
+      end # configure haproxy server
     end # configure keepalived in master
 
     context 'configure logrotate file' do