fix race condition on direct room

Author: shcherbak

build/Dockerfile

@@ -23,6 +23,10 @@ COPY workers.patch /tmp/workers.patch
 RUN patch $(ls /usr/local/lib/python$PYTHON_PKG_VER/site-packages/synapse/config/workers.py) \
     < /tmp/workers.patch \
     && rm /tmp/workers.patch
+COPY fix-empty-room-state.patch /tmp/fix-empty-room-state.patch
+RUN patch $(ls /usr/local/lib/python$PYTHON_PKG_VER/site-packages/synapse/storage/controllers/state.py) \
+    < /tmp/fix-empty-room-state.patch \
+    && rm /tmp/fix-empty-room-state.patch
 
 # stage 3 : apply e2e_room_keys.patch
 FROM pipwheels AS e2e_patch
@@ -41,8 +45,8 @@ COPY --from=workers_patch \
   /usr/local/lib/python${PYTHON_PKG_VER}/site-packages/synapse/config/workers.py \
   /usr/local/lib/python${PYTHON_PKG_VER}/site-packages/synapse/config/workers.py
 COPY --from=workers_patch \
-  /usr/local/lib/python$PYTHON_PKG_VER/site-packages/synapse/storage/databases/main/event_push_actions.py \
-  /usr/local/lib/python$PYTHON_PKG_VER/site-packages/synapse/storage/databases/main/event_push_actions.py
+  /usr/local/lib/python$PYTHON_PKG_VER/site-packages/synapse/storage/controllers/state.py \
+  /usr/local/lib/python$PYTHON_PKG_VER/site-packages/synapse/storage/controllers/state.py
 
 # stage 5 : base version with e2e_room_keys table optimization
 FROM main AS e2e

build/fix-empty-room-state.patch

@@ -0,0 +1,42 @@
+--- a/synapse/storage/controllers/state.py	2026-06-16 02:07:49.851487330 +0300
++++ b/synapse/storage/controllers/state.py	2026-06-16 02:07:13.674945849 +0300
+@@ -45,6 +45,7 @@
+ from synapse.util.caches import intern_string
+ from synapse.util.caches.descriptors import cached
+ from synapse.util.cancellation import cancellable
++from synapse.util.duration import Duration
+ from synapse.util.metrics import Measure
+
+ if TYPE_CHECKING:
+@@ -458,14 +459,25 @@
+             )
+
+         else:
+-            # no events in this room - so presumably no state
+-            state = {}
++            state = await self.get_current_state_ids(
++                room_id,
++                state_filter=state_filter,
++                await_full_state=await_full_state,
++            )
++
++            for _ in range(3):
++                await self.clock.sleep(Duration(seconds=1.5))
++                new_state = await self.get_current_state_ids(
++                    room_id,
++                    state_filter=state_filter,
++                    await_full_state=await_full_state,
++                )
++                if new_state == state:
++                    break
++                state = new_state
+
+-            # (erikj) This should be rarely hit, but we've had some reports that
+-            # we get more state down gappy syncs than we should, so let's add
+-            # some logging.
+             logger.info(
+-                "Failed to find any events in room %s at %s",
++                "Failed to find any events in room %s at %s; falling back to current state",
+                 room_id,
+                 stream_position.room_key,
+             )