From 356615d91c82275cf5102602faa524800b8592cc Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Fri, 24 Mar 2023 17:13:37 +0000
Subject: [PATCH 1/2] Prune old typing notifications

Rather than keeping them around forever in memory, slowing things down.

Fixes #11750.
---
 synapse/handlers/typing.py | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/synapse/handlers/typing.py b/synapse/handlers/typing.py
index 3f656ea4f508..39ae44ea95fc 100644
--- a/synapse/handlers/typing.py
+++ b/synapse/handlers/typing.py
@@ -52,6 +52,11 @@ class RoomMember:
 FEDERATION_PING_INTERVAL = 40 * 1000
 
 
+# How long to remember a typing notification happened in a room before
+# forgetting about it.
+FORGET_TIMEOUT = 10 * 60 * 1000
+
+
 class FollowerTypingHandler:
     """A typing handler on a different process than the writer that is updated
     via replication.
@@ -83,7 +88,10 @@ def __init__(self, hs: "HomeServer"):
         self.wheel_timer: WheelTimer[RoomMember] = WheelTimer(bucket_size=5000)
         self._latest_room_serial = 0
 
+        self._rooms_updated: Set[str] = set()
+
         self.clock.looping_call(self._handle_timeouts, 5000)
+        self.clock.looping_call(self._prune_old_typing, FORGET_TIMEOUT)
 
     def _reset(self) -> None:
         """Reset the typing handler's data caches."""
@@ -92,6 +100,8 @@ def _reset(self) -> None:
         # map room IDs to sets of users currently typing
         self._room_typing = {}
 
+        self._rooms_updated = set()
+
         self._member_last_federation_poke = {}
         self.wheel_timer = WheelTimer(bucket_size=5000)
 
@@ -178,6 +188,7 @@ def process_replication_rows(
             prev_typing = self._room_typing.get(row.room_id, set())
             now_typing = set(row.user_ids)
             self._room_typing[row.room_id] = now_typing
+            self._rooms_updated.add(row.room_id)
 
             if self.federation:
                 run_as_background_process(
@@ -209,6 +220,19 @@ async def _send_changes_in_typing_to_remotes(
     def get_current_token(self) -> int:
         return self._latest_room_serial
 
+    def _prune_old_typing(self) -> None:
+        """Prune rooms that haven't seen typing updates since last time.
+
+        This is safe to do as clients should time out old typing notifications.
+        """
+        stale_rooms = self._room_serials.keys() - self._rooms_updated
+
+        for room_id in stale_rooms:
+            self._room_serials.pop(room_id, None)
+            self._room_typing.pop(room_id, None)
+
+        self._rooms_updated = set()
+
 
 class TypingWriterHandler(FollowerTypingHandler):
     def __init__(self, hs: "HomeServer"):
@@ -388,6 +412,7 @@ def _push_update_local(self, member: RoomMember, typing: bool) -> None:
         self._typing_stream_change_cache.entity_has_changed(
             member.room_id, self._latest_room_serial
         )
+        self._rooms_updated.add(member.room_id)
 
         self.notifier.on_new_event(
             StreamKeyType.TYPING, self._latest_room_serial, rooms=[member.room_id]

From b4fc42f2f071e5e09efcedee7af5b252598fc3a1 Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Mon, 27 Mar 2023 10:37:35 +0100
Subject: [PATCH 2/2] Newsfile

---
 changelog.d/15332.bugfix | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 changelog.d/15332.bugfix

diff --git a/changelog.d/15332.bugfix b/changelog.d/15332.bugfix
new file mode 100644
index 000000000000..ca6fb1d2fd79
--- /dev/null
+++ b/changelog.d/15332.bugfix
@@ -0,0 +1 @@
+Fix bug in worker mode where on a rolling restart of workers the "typing" worker would consume 100% CPU until it got restarted.