Add some comments to areas that could be optimised.

synapse/handlers/sync.py

@@ -447,6 +447,9 @@ class SyncHandler(BaseHandler):
         )
         now_token = now_token.copy_and_replace("presence_key", presence_key)
 
+        # We now fetch all ephemeral events for this room in order to get
+        # this users current read receipt. This could almost certainly be
+        # optimised.
         _, all_ephemeral_by_room = yield self.ephemeral_by_room(
             sync_config, now_token
         )

synapse/push/__init__.py

@@ -26,7 +26,9 @@ import random
 
 logger = logging.getLogger(__name__)
 
-
+# Pushers could now be moved to pull out of the event_actions table instead
+# of listening on the event stream: this would avoid them having to run the
+# rules again.
 class Pusher(object):
     INITIAL_BACKOFF = 1000
     MAX_BACKOFF = 60 * 60 * 1000

synapse/push/bulk_push_rule_evaluator.py

@@ -59,6 +59,14 @@ def evaluator_for_room_id(room_id, hs, store):
 
 
 class BulkPushRuleEvaluator:
+    """
+    Runs push rules for all users in a room.
+    This is faster than running PushRuleEvaluator for each user because it
+    fetches all the rules for all the users in one (batched) db query
+    rarher than doing multiple queries per-user. It currently uses
+    the same logic to run the actual rules, but could be optimised further
+    (see https://matrix.org/jira/browse/SYN-562)
+    """
     def __init__(self, room_id, rules_by_user, display_names, users_in_room):
         self.room_id = room_id
         self.rules_by_user = rules_by_user