Batching in the user directory import (#4900)

changelog.d/4900.feature

@@ -0,0 +1 @@
+The user directory has been rewritten to make it faster, with less chance of falling behind on a large server.

synapse/storage/user_directory.py

@@ -32,6 +32,11 @@ TEMP_TABLE = "_temp_populate_user_directory"
 
 
 class UserDirectoryStore(BackgroundUpdateStore):
+
+    # How many records do we calculate before sending it to
+    # add_users_who_share_private_rooms?
+    SHARE_PRIVATE_WORKING_SET = 500
+
     def __init__(self, db_conn, hs):
         super(UserDirectoryStore, self).__init__(db_conn, hs)
 
@@ -218,6 +223,14 @@ class UserDirectoryStore(BackgroundUpdateStore):
                             user_set = (user_id, other_user_id)
                             to_insert.add(user_set)
 
+                            # If it gets too big, stop and write to the database
+                            # to prevent storing too much in RAM.
+                            if len(to_insert) >= self.SHARE_PRIVATE_WORKING_SET:
+                                yield self.add_users_who_share_private_room(
+                                    room_id, to_insert
+                                )
+                                to_insert.clear()
+
                     if to_insert:
                         yield self.add_users_who_share_private_room(room_id, to_insert)
                         to_insert.clear()