Refactor and convert `Linearizer` to async (#12357)

Refactor and convert `Linearizer` to async. This makes a `Linearizer`
cancellation bug easier to fix.

Also refactor to use an async context manager, which eliminates an
unlikely footgun where code that doesn't immediately use the context
manager could forget to release the lock.

Signed-off-by: Sean Quah <seanq@element.io>

changelog.d/12357.misc

@@ -0,0 +1 @@
+Refactor `Linearizer`, convert methods to async and use an async context manager.

synapse/federation/federation_server.py

@@ -188,7 +188,7 @@ class FederationServer(FederationBase):
     async def on_backfill_request(
         self, origin: str, room_id: str, versions: List[str], limit: int
     ) -> Tuple[int, Dict[str, Any]]:
-        with (await self._server_linearizer.queue((origin, room_id))):
+        async with self._server_linearizer.queue((origin, room_id)):
             origin_host, _ = parse_server_name(origin)
             await self.check_server_matches_acl(origin_host, room_id)
 
@@ -218,7 +218,7 @@ class FederationServer(FederationBase):
             Tuple indicating the response status code and dictionary response
             body including `event_id`.
         """
-        with (await self._server_linearizer.queue((origin, room_id))):
+        async with self._server_linearizer.queue((origin, room_id)):
             origin_host, _ = parse_server_name(origin)
             await self.check_server_matches_acl(origin_host, room_id)
 
@@ -529,7 +529,7 @@ class FederationServer(FederationBase):
         # in the cache so we could return it without waiting for the linearizer
         # - but that's non-trivial to get right, and anyway somewhat defeats
         # the point of the linearizer.
-        with (await self._server_linearizer.queue((origin, room_id))):
+        async with self._server_linearizer.queue((origin, room_id)):
             resp: JsonDict = dict(
                 await self._state_resp_cache.wrap(
                     (room_id, event_id),
@@ -883,7 +883,7 @@ class FederationServer(FederationBase):
     async def on_event_auth(
         self, origin: str, room_id: str, event_id: str
     ) -> Tuple[int, Dict[str, Any]]:
-        with (await self._server_linearizer.queue((origin, room_id))):
+        async with self._server_linearizer.queue((origin, room_id)):
             origin_host, _ = parse_server_name(origin)
             await self.check_server_matches_acl(origin_host, room_id)
 
@@ -945,7 +945,7 @@ class FederationServer(FederationBase):
         latest_events: List[str],
         limit: int,
     ) -> Dict[str, list]:
-        with (await self._server_linearizer.queue((origin, room_id))):
+        async with self._server_linearizer.queue((origin, room_id)):
             origin_host, _ = parse_server_name(origin)
             await self.check_server_matches_acl(origin_host, room_id)
 

synapse/handlers/appservice.py

@@ -330,10 +330,8 @@ class ApplicationServicesHandler:
                     continue
 
                 # Since we read/update the stream position for this AS/stream
-                with (
-                    await self._ephemeral_events_linearizer.queue(
-                        (service.id, stream_key)
-                    )
+                async with self._ephemeral_events_linearizer.queue(
+                    (service.id, stream_key)
                 ):
                     if stream_key == "receipt_key":
                         events = await self._handle_receipts(service, new_token)

synapse/handlers/device.py

@@ -833,7 +833,7 @@ class DeviceListUpdater:
     async def _handle_device_updates(self, user_id: str) -> None:
         "Actually handle pending updates."
 
-        with (await self._remote_edu_linearizer.queue(user_id)):
+        async with self._remote_edu_linearizer.queue(user_id):
             pending_updates = self._pending_updates.pop(user_id, [])
             if not pending_updates:
                 # This can happen since we batch updates

synapse/handlers/e2e_keys.py

@@ -118,7 +118,7 @@ class E2eKeysHandler:
             from_device_id: the device making the query. This is used to limit
                 the number of in-flight queries at a time.
         """
-        with await self._query_devices_linearizer.queue((from_user_id, from_device_id)):
+        async with self._query_devices_linearizer.queue((from_user_id, from_device_id)):
             device_keys_query: Dict[str, Iterable[str]] = query_body.get(
                 "device_keys", {}
             )
@@ -1386,7 +1386,7 @@ class SigningKeyEduUpdater:
         device_handler = self.e2e_keys_handler.device_handler
         device_list_updater = device_handler.device_list_updater
 
-        with (await self._remote_edu_linearizer.queue(user_id)):
+        async with self._remote_edu_linearizer.queue(user_id):
             pending_updates = self._pending_updates.pop(user_id, [])
             if not pending_updates:
                 # This can happen since we batch updates

synapse/handlers/e2e_room_keys.py

@@ -83,7 +83,7 @@ class E2eRoomKeysHandler:
 
         # we deliberately take the lock to get keys so that changing the version
         # works atomically
-        with (await self._upload_linearizer.queue(user_id)):
+        async with self._upload_linearizer.queue(user_id):
             # make sure the backup version exists
             try:
                 await self.store.get_e2e_room_keys_version_info(user_id, version)
@@ -126,7 +126,7 @@ class E2eRoomKeysHandler:
         """
 
         # lock for consistency with uploading
-        with (await self._upload_linearizer.queue(user_id)):
+        async with self._upload_linearizer.queue(user_id):
             # make sure the backup version exists
             try:
                 version_info = await self.store.get_e2e_room_keys_version_info(
@@ -187,7 +187,7 @@ class E2eRoomKeysHandler:
         # TODO: Validate the JSON to make sure it has the right keys.
 
         # XXX: perhaps we should use a finer grained lock here?
-        with (await self._upload_linearizer.queue(user_id)):
+        async with self._upload_linearizer.queue(user_id):
 
             # Check that the version we're trying to upload is the current version
             try:
@@ -332,7 +332,7 @@ class E2eRoomKeysHandler:
         # TODO: Validate the JSON to make sure it has the right keys.
 
         # lock everyone out until we've switched version
-        with (await self._upload_linearizer.queue(user_id)):
+        async with self._upload_linearizer.queue(user_id):
             new_version = await self.store.create_e2e_room_keys_version(
                 user_id, version_info
             )
@@ -359,7 +359,7 @@ class E2eRoomKeysHandler:
         }
         """
 
-        with (await self._upload_linearizer.queue(user_id)):
+        async with self._upload_linearizer.queue(user_id):
             try:
                 res = await self.store.get_e2e_room_keys_version_info(user_id, version)
             except StoreError as e:
@@ -383,7 +383,7 @@ class E2eRoomKeysHandler:
             NotFoundError: if this backup version doesn't exist
         """
 
-        with (await self._upload_linearizer.queue(user_id)):
+        async with self._upload_linearizer.queue(user_id):
             try:
                 await self.store.delete_e2e_room_keys_version(user_id, version)
             except StoreError as e:
@@ -413,7 +413,7 @@ class E2eRoomKeysHandler:
             raise SynapseError(
                 400, "Version in body does not match", Codes.INVALID_PARAM
             )
-        with (await self._upload_linearizer.queue(user_id)):
+        async with self._upload_linearizer.queue(user_id):
             try:
                 old_info = await self.store.get_e2e_room_keys_version_info(
                     user_id, version

synapse/handlers/federation.py

@@ -151,7 +151,7 @@ class FederationHandler:
                 return. This is used as part of the heuristic to decide if we
                 should back paginate.
         """
-        with (await self._room_backfill.queue(room_id)):
+        async with self._room_backfill.queue(room_id):
             return await self._maybe_backfill_inner(room_id, current_depth, limit)
 
     async def _maybe_backfill_inner(

synapse/handlers/federation_event.py

@@ -224,7 +224,7 @@ class FederationEventHandler:
                     len(missing_prevs),
                     shortstr(missing_prevs),
                 )
-                with (await self._room_pdu_linearizer.queue(pdu.room_id)):
+                async with self._room_pdu_linearizer.queue(pdu.room_id):
                     logger.info(
                         "Acquired room lock to fetch %d missing prev_events",
                         len(missing_prevs),

synapse/handlers/message.py

@@ -851,7 +851,7 @@ class EventCreationHandler:
         # a situation where event persistence can't keep up, causing
         # extremities to pile up, which in turn leads to state resolution
         # taking longer.
-        with (await self.limiter.queue(event_dict["room_id"])):
+        async with self.limiter.queue(event_dict["room_id"]):
             if txn_id and requester.access_token_id:
                 existing_event_id = await self.store.get_event_id_from_transaction_id(
                     event_dict["room_id"],

synapse/handlers/presence.py

@@ -1030,7 +1030,7 @@ class PresenceHandler(BasePresenceHandler):
             is_syncing: Whether or not the user is now syncing
             sync_time_msec: Time in ms when the user was last syncing
         """
-        with (await self.external_sync_linearizer.queue(process_id)):
+        async with self.external_sync_linearizer.queue(process_id):
             prev_state = await self.current_state_for_user(user_id)
 
             process_presence = self.external_process_to_current_syncs.setdefault(
@@ -1071,7 +1071,7 @@ class PresenceHandler(BasePresenceHandler):
 
         Used when the process has stopped/disappeared.
         """
-        with (await self.external_sync_linearizer.queue(process_id)):
+        async with self.external_sync_linearizer.queue(process_id):
             process_presence = self.external_process_to_current_syncs.pop(
                 process_id, set()
             )

synapse/handlers/read_marker.py

@@ -40,7 +40,7 @@ class ReadMarkerHandler:
         the read marker has changed.
         """
 
-        with await self.read_marker_linearizer.queue((room_id, user_id)):
+        async with self.read_marker_linearizer.queue((room_id, user_id)):
             existing_read_marker = await self.store.get_account_data_for_room_and_type(
                 user_id, room_id, "m.fully_read"
             )

synapse/handlers/room.py

@@ -883,7 +883,7 @@ class RoomCreationHandler:
         #
         # we also don't need to check the requester's shadow-ban here, as we
         # have already done so above (and potentially emptied invite_list).
-        with (await self.room_member_handler.member_linearizer.queue((room_id,))):
+        async with self.room_member_handler.member_linearizer.queue((room_id,)):
             content = {}
             is_direct = config.get("is_direct", None)
             if is_direct:

synapse/handlers/room_member.py

@@ -515,8 +515,8 @@ class RoomMemberHandler(metaclass=abc.ABCMeta):
 
         # We first linearise by the application service (to try to limit concurrent joins
         # by application services), and then by room ID.
-        with (await self.member_as_limiter.queue(as_id)):
-            with (await self.member_linearizer.queue(key)):
+        async with self.member_as_limiter.queue(as_id):
+            async with self.member_linearizer.queue(key):
                 result = await self.update_membership_locked(
                     requester,
                     target,

synapse/handlers/sso.py

@@ -430,7 +430,7 @@ class SsoHandler:
         # grab a lock while we try to find a mapping for this user. This seems...
         # optimistic, especially for implementations that end up redirecting to
         # interstitial pages.
-        with await self._mapping_lock.queue(auth_provider_id):
+        async with self._mapping_lock.queue(auth_provider_id):
             # first of all, check if we already have a mapping for this user
             user_id = await self.get_sso_user_by_remote_user_id(
                 auth_provider_id,

synapse/push/bulk_push_rule_evaluator.py

@@ -397,7 +397,7 @@ class RulesForRoom:
             self.room_push_rule_cache_metrics.inc_hits()
             return self.data.rules_by_user
 
-        with (await self.linearizer.queue(self.room_id)):
+        async with self.linearizer.queue(self.room_id):
             if state_group and self.data.state_group == state_group:
                 logger.debug("Using cached rules for %r", self.room_id)
                 self.room_push_rule_cache_metrics.inc_hits()

synapse/replication/tcp/client.py

@@ -451,7 +451,7 @@ class FederationSenderHandler:
             # service for robustness? Or could we replace it with an assertion that
             # we're not being re-entered?
 
-            with (await self._fed_position_linearizer.queue(None)):
+            async with self._fed_position_linearizer.queue(None):
                 # We persist and ack the same position, so we take a copy of it
                 # here as otherwise it can get modified from underneath us.
                 current_position = self.federation_position

synapse/rest/media/v1/media_repository.py

@@ -258,7 +258,7 @@ class MediaRepository:
         # We linearize here to ensure that we don't try and download remote
         # media multiple times concurrently
         key = (server_name, media_id)
-        with (await self.remote_media_linearizer.queue(key)):
+        async with self.remote_media_linearizer.queue(key):
             responder, media_info = await self._get_remote_media_impl(
                 server_name, media_id
             )
@@ -294,7 +294,7 @@ class MediaRepository:
         # We linearize here to ensure that we don't try and download remote
         # media multiple times concurrently
         key = (server_name, media_id)
-        with (await self.remote_media_linearizer.queue(key)):
+        async with self.remote_media_linearizer.queue(key):
             responder, media_info = await self._get_remote_media_impl(
                 server_name, media_id
             )
@@ -850,7 +850,7 @@ class MediaRepository:
 
             # TODO: Should we delete from the backup store
 
-            with (await self.remote_media_linearizer.queue(key)):
+            async with self.remote_media_linearizer.queue(key):
                 full_path = self.filepaths.remote_media_filepath(origin, file_id)
                 try:
                     os.remove(full_path)

synapse/state/__init__.py

@@ -573,7 +573,7 @@ class StateResolutionHandler:
         """
         group_names = frozenset(state_groups_ids.keys())
 
-        with (await self.resolve_linearizer.queue(group_names)):
+        async with self.resolve_linearizer.queue(group_names):
             cache = self._state_cache.get(group_names, None)
             if cache:
                 return cache

synapse/storage/databases/main/roommember.py

@@ -888,7 +888,7 @@ class RoomMemberWorkerStore(EventsWorkerStore):
             return frozenset(cache.hosts_to_joined_users)
 
         # Since we'll mutate the cache we need to lock.
-        with (await self._joined_host_linearizer.queue(room_id)):
+        async with self._joined_host_linearizer.queue(room_id):
             if state_entry.state_group == cache.state_group:
                 # Same state group, so nothing to do. We've already checked for
                 # this above, but the cache may have changed while waiting on

synapse/util/async_helpers.py

@@ -18,7 +18,7 @@ import collections
 import inspect
 import itertools
 import logging
-from contextlib import asynccontextmanager, contextmanager
+from contextlib import asynccontextmanager
 from typing import (
     Any,
     AsyncIterator,
@@ -29,7 +29,6 @@ from typing import (
     Generic,
     Hashable,
     Iterable,
-    Iterator,
     List,
     Optional,
     Set,
@@ -342,7 +341,7 @@ class Linearizer:
 
     Example:
 
-        with await limiter.queue("test_key"):
+        async with limiter.queue("test_key"):
             # do some work.
 
     """
@@ -383,95 +382,53 @@ class Linearizer:
         # non-empty.
         return bool(entry.deferreds)
 
-    def queue(self, key: Hashable) -> defer.Deferred:
-        # we avoid doing defer.inlineCallbacks here, so that cancellation works correctly.
-        # (https://twistedmatrix.com/trac/ticket/4632 meant that cancellations were not
-        # propagated inside inlineCallbacks until Twisted 18.7)
+    def queue(self, key: Hashable) -> AsyncContextManager[None]:
+        @asynccontextmanager
+        async def _ctx_manager() -> AsyncIterator[None]:
+            entry = await self._acquire_lock(key)
+            try:
+                yield
+            finally:
+                self._release_lock(key, entry)
+
+        return _ctx_manager()
+
+    async def _acquire_lock(self, key: Hashable) -> _LinearizerEntry:
+        """Acquires a linearizer lock, waiting if necessary.
+
+        Returns once we have secured the lock.
+        """
         entry = self.key_to_defer.setdefault(
             key, _LinearizerEntry(0, collections.OrderedDict())
         )
 
-        # If the number of things executing is greater than the maximum
-        # then add a deferred to the list of blocked items
-        # When one of the things currently executing finishes it will callback
-        # this item so that it can continue executing.
-        if entry.count >= self.max_count:
-            res = self._await_lock(key)
-        else:
+        if entry.count < self.max_count:
+            # The number of things executing is less than the maximum.
             logger.debug(
                 "Acquired uncontended linearizer lock %r for key %r", self.name, key
             )
             entry.count += 1
-            res = defer.succeed(None)
-
-        # once we successfully get the lock, we need to return a context manager which
-        # will release the lock.
-
-        @contextmanager
-        def _ctx_manager(_: None) -> Iterator[None]:
-            try:
-                yield
-            finally:
-                logger.debug("Releasing linearizer lock %r for key %r", self.name, key)
-
-                # We've finished executing so check if there are any things
-                # blocked waiting to execute and start one of them
-                entry.count -= 1
-
-                if entry.deferreds:
-                    (next_def, _) = entry.deferreds.popitem(last=False)
-
-                    # we need to run the next thing in the sentinel context.
-                    with PreserveLoggingContext():
-                        next_def.callback(None)
-                elif entry.count == 0:
-                    # We were the last thing for this key: remove it from the
-                    # map.
-                    del self.key_to_defer[key]
-
-        res.addCallback(_ctx_manager)
-        return res
-
-    def _await_lock(self, key: Hashable) -> defer.Deferred:
-        """Helper for queue: adds a deferred to the queue
-
-        Assumes that we've already checked that we've reached the limit of the number
-        of lock-holders we allow. Creates a new deferred which is added to the list, and
-        adds some management around cancellations.
-
-        Returns the deferred, which will callback once we have secured the lock.
-
-        """
-        entry = self.key_to_defer[key]
+            return entry
 
+        # Otherwise, the number of things executing is at the maximum and we have to
+        # add a deferred to the list of blocked items.
+        # When one of the things currently executing finishes it will callback
+        # this item so that it can continue executing.
         logger.debug("Waiting to acquire linearizer lock %r for key %r", self.name, key)
 
         new_defer: "defer.Deferred[None]" = make_deferred_yieldable(defer.Deferred())
         entry.deferreds[new_defer] = 1
 
-        def cb(_r: None) -> "defer.Deferred[None]":
-            logger.debug("Acquired linearizer lock %r for key %r", self.name, key)
-            entry.count += 1
-
-            # if the code holding the lock completes synchronously, then it
-            # will recursively run the next claimant on the list. That can
-            # relatively rapidly lead to stack exhaustion. This is essentially
-            # the same problem as http://twistedmatrix.com/trac/ticket/9304.
-            #
-            # In order to break the cycle, we add a cheeky sleep(0) here to
-            # ensure that we fall back to the reactor between each iteration.
-            #
-            # (This needs to happen while we hold the lock, and the context manager's exit
-            # code must be synchronous, so this is the only sensible place.)
-            return self._clock.sleep(0)
-
-        def eb(e: Failure) -> Failure:
+        try:
+            await new_defer
+        except Exception as e:
             logger.info("defer %r got err %r", new_defer, e)
             if isinstance(e, CancelledError):
                 logger.debug(
-                    "Cancelling wait for linearizer lock %r for key %r", self.name, key
+                    "Cancelling wait for linearizer lock %r for key %r",
+                    self.name,
+                    key,
                 )
-
             else:
                 logger.warning(
                     "Unexpected exception waiting for linearizer lock %r for key %r",
@@ -481,10 +438,43 @@ class Linearizer:
 
             # we just have to take ourselves back out of the queue.
             del entry.deferreds[new_defer]
-            return e
+            raise
 
-        new_defer.addCallbacks(cb, eb)
-        return new_defer
+        logger.debug("Acquired linearizer lock %r for key %r", self.name, key)
+        entry.count += 1
+
+        # if the code holding the lock completes synchronously, then it
+        # will recursively run the next claimant on the list. That can
+        # relatively rapidly lead to stack exhaustion. This is essentially
+        # the same problem as http://twistedmatrix.com/trac/ticket/9304.
+        #
+        # In order to break the cycle, we add a cheeky sleep(0) here to
+        # ensure that we fall back to the reactor between each iteration.
+        #
+        # This needs to happen while we hold the lock. We could put it on the
+        # exit path, but that would slow down the uncontended case.
+        await self._clock.sleep(0)
+
+        return entry
+
+    def _release_lock(self, key: Hashable, entry: _LinearizerEntry) -> None:
+        """Releases a held linearizer lock."""
+        logger.debug("Releasing linearizer lock %r for key %r", self.name, key)
+
+        # We've finished executing so check if there are any things
+        # blocked waiting to execute and start one of them
+        entry.count -= 1
+
+        if entry.deferreds:
+            (next_def, _) = entry.deferreds.popitem(last=False)
+
+            # we need to run the next thing in the sentinel context.
+            with PreserveLoggingContext():
+                next_def.callback(None)
+        elif entry.count == 0:
+            # We were the last thing for this key: remove it from the
+            # map.
+            del self.key_to_defer[key]
 
 
 class ReadWriteLock:

tests/util/test_linearizer.py

@@ -46,7 +46,7 @@ class LinearizerTestCase(unittest.TestCase):
         unblock_d: "Deferred[None]" = Deferred()
 
         async def task() -> None:
-            with await linearizer.queue(key):
+            async with linearizer.queue(key):
                 acquired_d.callback(None)
                 await unblock_d
 
@@ -125,7 +125,7 @@ class LinearizerTestCase(unittest.TestCase):
 
         async def func(i: int) -> None:
             with LoggingContext("func(%s)" % i) as lc:
-                with (await linearizer.queue(key)):
+                async with linearizer.queue(key):
                     self.assertEqual(current_context(), lc)
 
                 self.assertEqual(current_context(), lc)