Fix another logcontext leak in _persist_events

We need to run the errback in the sentinel context to avoid losing our own
context.

Also: add logging to runInteraction to help identify where "Starting db
connection from sentinel context" warnings are coming from

changelog.d/3606.misc

@@ -0,0 +1 @@
+Fix some random logcontext leaks.

synapse/storage/_base.py

@@ -311,6 +311,12 @@ class SQLBaseStore(object):
         after_callbacks = []
         exception_callbacks = []
 
+        if LoggingContext.current_context() == LoggingContext.sentinel:
+            logger.warn(
+                "Starting db txn '%s' from sentinel context",
+                desc,
+            )
+
         try:
             result = yield self.runWithConnection(
                 self._new_transaction,

synapse/storage/events.py

@@ -142,15 +142,14 @@ class _EventPeristenceQueue(object):
             try:
                 queue = self._get_drainining_queue(room_id)
                 for item in queue:
-                    # handle_queue_loop runs in the sentinel logcontext, so
-                    # there is no need to preserve_fn when running the
-                    # callbacks on the deferred.
                     try:
                         ret = yield per_item_callback(item)
+                    except Exception:
+                        with PreserveLoggingContext():
+                            item.deferred.errback()
+                    else:
                         with PreserveLoggingContext():
                             item.deferred.callback(ret)
-                    except Exception:
-                        item.deferred.errback()
             finally:
                 queue = self._event_persist_queues.pop(room_id, None)
                 if queue: