Run things as background processes

This fixes #3518, and ensures that we get useful logs and metrics for lots of
things that happen in the background.

(There are certainly more things that happen in the background; these are just
the common ones I've found running a single-process synapse locally).

synapse/storage/events.py

@@ -33,12 +33,13 @@ from synapse.api.errors import SynapseError
 # these are only included to make the type annotations work
 from synapse.events import EventBase  # noqa: F401
 from synapse.events.snapshot import EventContext  # noqa: F401
+from synapse.metrics.background_process_metrics import run_as_background_process
 from synapse.storage.events_worker import EventsWorkerStore
 from synapse.types import RoomStreamToken, get_domain_from_id
 from synapse.util.async import ObservableDeferred
 from synapse.util.caches.descriptors import cached, cachedInlineCallbacks
 from synapse.util.frozenutils import frozendict_json_encoder
-from synapse.util.logcontext import PreserveLoggingContext, make_deferred_yieldable
+from synapse.util.logcontext import make_deferred_yieldable
 from synapse.util.logutils import log_function
 from synapse.util.metrics import Measure
 
@@ -155,11 +156,8 @@ class _EventPeristenceQueue(object):
                     self._event_persist_queues[room_id] = queue
                 self._currently_persisting_rooms.discard(room_id)
 
-        # set handle_queue_loop off on the background. We don't want to
-        # attribute work done in it to the current request, so we drop the
-        # logcontext altogether.
-        with PreserveLoggingContext():
-            handle_queue_loop()
+        # set handle_queue_loop off in the background
+        run_as_background_process("persist_events", handle_queue_loop)
 
     def _get_drainining_queue(self, room_id):
         queue = self._event_persist_queues.setdefault(room_id, deque())