use a request-scoped cache to keep the metadata inheritence tree around for the whole request. This means we should only do one trip to Memcached/Mongo per course per request. This is expected to keep memory utilization down

common/djangoapps/request_cache/middleware.py

@@ -0,0 +1,20 @@
+import threading
+
+_request_cache_threadlocal = threading.local()
+_request_cache_threadlocal.data = {}
+
+class RequestCache(object):
+    @classmethod
+    def get_request_cache(cls):
+        return _request_cache_threadlocal
+            
+    def clear_request_cache(self):
+        _request_cache_threadlocal.data = {}
+
+    def process_request(self, request):
+        self.clear_request_cache()
+        return None
+
+    def process_response(self, request, response):
+        self.clear_request_cache()
+        return response