diff --git a/releasenotes/notes/fix-grow-cluster-cache-5625d1a519a4ee01.yaml b/releasenotes/notes/fix-grow-cluster-cache-5625d1a519a4ee01.yaml
new file mode 100644
index 0000000..28828b6
--- /dev/null
+++ b/releasenotes/notes/fix-grow-cluster-cache-5625d1a519a4ee01.yaml
@@ -0,0 +1,5 @@
+---
+fixes:
+  - The current caching initializes the instance list for the grow
+    cluster manager each time a new process would instantiate the
+    manager.  Changed to manager to only initialize on creation.
diff --git a/trove_dashboard/content/database_clusters/cluster_manager.py b/trove_dashboard/content/database_clusters/cluster_manager.py
index 50f60d9..ffc10c6 100644
--- a/trove_dashboard/content/database_clusters/cluster_manager.py
+++ b/trove_dashboard/content/database_clusters/cluster_manager.py
@@ -42,17 +42,17 @@ def has_cluster(cluster_id):
 
 
 class ClusterInstanceManager(object):
-
-    instances = []
-
     def __init__(self, cluster_id):
         self.cluster_id = cluster_id
+        self.instances = []
 
     def get_instances(self):
+        if not hasattr(self, 'instances'):
+            self.instances = []
         return self.instances
 
     def get_instance(self, id):
-        for instance in self.instances:
+        for instance in self.get_instances():
             if instance.id == id:
                 return instance
         return None
@@ -63,7 +63,7 @@ class ClusterInstanceManager(object):
                                    volume, type, related_to, nics)
         self.instances.append(instance)
         update(self.cluster_id, self)
-        return self.instances
+        return self.get_instances()
 
     def delete_instance(self, id):
         instance = self.get_instance(id)
@@ -72,7 +72,8 @@ class ClusterInstanceManager(object):
             update(self.cluster_id, self)
 
     def clear_instances(self):
-        del self.instances[:]
+        self.instances = []
+        update(self.cluster_id, self)
 
 
 class ClusterInstance(object):