diff --git a/rabbitmq/meta/heka.yml b/rabbitmq/meta/heka.yml
index b3855be..73515f8 100644
--- a/rabbitmq/meta/heka.yml
+++ b/rabbitmq/meta/heka.yml
@@ -75,6 +75,16 @@
         window: 120
         periods: 0
         function: avg
+    rabbitmq_check:
+      description: 'RabbitMQ cannot be checked'
+      severity: down
+      rules:
+      - metric: rabbitmq_check
+        relational_operator: '=='
+        threshold: 0
+        window: 60
+        periods: 0
+        function: last
   alarm:
     rabbitmq_server_disk:
       alerting: enabled
@@ -96,6 +106,10 @@
       - rabbitmq_queue_warning
       dimension:
         service: rabbitmq-cluster
+    rabbitmq_check:
+      alerting: enabled
+      triggers:
+      - rabbitmq_check
 aggregator:
   alarm_cluster:
     rabbitmq_cluster:
@@ -109,3 +123,24 @@
       - rabbitmq_server_queue
       dimension:
         service: rabbitmq
+    rabbitmq_service:
+      # A check failure on a single node doesn't mean that the whole cluster
+      # is down, this is why a 'hostname' group_by and 'majority_of_members'
+      # policy are used here
+      policy: majority_of_members
+      group_by: hostname
+      match:
+        member: rabbitmq_check
+      members:
+      - rabbitmq_check
+      dimension:
+        service: rabbitmq
+    rabbitmq:
+      policy: highest_severity
+      match:
+        service: rabbitmq
+      members:
+      - rabbitmq_cluster
+      - rabbitmq_service
+      dimension:
+        cluster_name: rabbitmq