Fix a bug introduced in Synapse v1.0.0 whereby device list updates would not be sent to remote homeservers if there were too many to send at once. (#11729)

Co-authored-by: Brendan Abolivier <babolivier@matrix.org>

1 files changed, 7 insertions, 1 deletions

diff --git a/synapse/storage/databases/main/devices.py b/synapse/storage/databases/main/devices.py
index 273adb61fd..324bd5f879 100644
--- a/synapse/storage/databases/main/devices.py
+++ b/synapse/storage/databases/main/devices.py
@@ -270,6 +270,10 @@ class DeviceWorkerStore(SQLBaseStore):
         # The most recent request's opentracing_context is used as the
         # context which created the Edu.
 
+        # This is the stream ID that we will return for the consumer to resume
+        # following this stream later.
+        last_processed_stream_id = from_stream_id
+
         query_map = {}
         cross_signing_keys_by_user = {}
         for user_id, device_id, update_stream_id, update_context in updates:
@@ -295,6 +299,8 @@ class DeviceWorkerStore(SQLBaseStore):
                 if update_stream_id > previous_update_stream_id:
                     query_map[key] = (update_stream_id, update_context)
 
+            last_processed_stream_id = update_stream_id
+
         results = await self._get_device_update_edus_by_remote(
             destination, from_stream_id, query_map
         )
@@ -307,7 +313,7 @@ class DeviceWorkerStore(SQLBaseStore):
             # FIXME: remove this when enough servers have upgraded
             results.append(("org.matrix.signing_key_update", result))
 
-        return now_stream_id, results
+        return last_processed_stream_id, results
 
     def _get_device_updates_by_remote_txn(
         self,